xref: /netbsd-src/sys/dev/ccd.c (revision 1b9578b8c2c1f848eeb16dabbfd7d1f0d9fdefbd)
1 /*	$NetBSD: ccd.c,v 1.141 2011/07/04 16:06:45 joerg 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 1.141 2011/07/04 16:06:45 joerg 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 		if ((error = VOP_GETATTR(vpp[ix], &va, l->l_cred)) != 0) {
314 #ifdef DEBUG
315 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
316 				printf("%s: %s: getattr failed %s = %d\n",
317 				    cs->sc_xname, ci->ci_path,
318 				    "error", error);
319 #endif
320 			goto out;
321 		}
322 		ci->ci_dev = va.va_rdev;
323 
324 		/*
325 		 * Get partition information for the component.
326 		 */
327 		error = VOP_IOCTL(vpp[ix], DIOCGPART, &dpart,
328 		    FREAD, l->l_cred);
329 		if (error) {
330 #ifdef DEBUG
331 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
332 				 printf("%s: %s: ioctl failed, error = %d\n",
333 				     cs->sc_xname, ci->ci_path, error);
334 #endif
335 			goto out;
336 		}
337 
338 /*
339  * This diagnostic test is disabled (for now?) since not all port supports
340  * on-disk BSD disklabel.
341  */
342 #if 0 /* def DIAGNOSTIC */
343 		/* Check fstype field of component. */
344 		if (dpart.part->p_fstype != FS_CCD)
345 			printf("%s: WARNING: %s: fstype %d != FS_CCD\n",
346 			    cs->sc_xname, ci->ci_path, dpart.part->p_fstype);
347 #endif
348 
349 		/*
350 		 * Calculate the size, truncating to an interleave
351 		 * boundary if necessary.
352 		 */
353 		maxsecsize =
354 		    ((dpart.disklab->d_secsize > maxsecsize) ?
355 		    dpart.disklab->d_secsize : maxsecsize);
356 		size = dpart.part->p_size;
357 		if (cs->sc_ileave > 1)
358 			size -= size % cs->sc_ileave;
359 
360 		if (size == 0) {
361 #ifdef DEBUG
362 			if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
363 				printf("%s: %s: size == 0\n",
364 				    cs->sc_xname, ci->ci_path);
365 #endif
366 			error = ENODEV;
367 			goto out;
368 		}
369 
370 		if (minsize == 0 || size < minsize)
371 			minsize = size;
372 		ci->ci_size = size;
373 		cs->sc_size += size;
374 	}
375 
376 	/*
377 	 * Don't allow the interleave to be smaller than
378 	 * the biggest component sector.
379 	 */
380 	if ((cs->sc_ileave > 0) &&
381 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
382 #ifdef DEBUG
383 		if (ccddebug & (CCDB_FOLLOW|CCDB_INIT))
384 			printf("%s: interleave must be at least %d\n",
385 			    cs->sc_xname, (maxsecsize / DEV_BSIZE));
386 #endif
387 		error = EINVAL;
388 		goto out;
389 	}
390 
391 	/*
392 	 * If uniform interleave is desired set all sizes to that of
393 	 * the smallest component.
394 	 */
395 	if (cs->sc_flags & CCDF_UNIFORM) {
396 		for (ci = cs->sc_cinfo;
397 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
398 			ci->ci_size = minsize;
399 
400 		cs->sc_size = cs->sc_nccdisks * minsize;
401 	}
402 
403 	/*
404 	 * Construct the interleave table.
405 	 */
406 	ccdinterleave(cs);
407 
408 	/*
409 	 * Create pseudo-geometry based on 1MB cylinders.  It's
410 	 * pretty close.
411 	 */
412 	ccg->ccg_secsize = DEV_BSIZE;
413 	ccg->ccg_ntracks = 1;
414 	ccg->ccg_nsectors = 1024 * (1024 / ccg->ccg_secsize);
415 	ccg->ccg_ncylinders = cs->sc_size / ccg->ccg_nsectors;
416 
417 	/*
418 	 * Create thread to handle deferred I/O.
419 	 */
420 	cs->sc_zap = false;
421 	error = kthread_create(PRI_BIO, KTHREAD_MPSAFE, NULL, ccdthread,
422 	    cs, &cs->sc_thread, "%s", cs->sc_xname);
423 	if (error) {
424 		printf("ccdinit: can't create thread: %d\n", error);
425 		goto out;
426 	}
427 
428 	/*
429 	 * Only now that everything is set up can we enable the device.
430 	 */
431 	mutex_enter(cs->sc_iolock);
432 	cs->sc_flags |= CCDF_INITED;
433 	mutex_exit(cs->sc_iolock);
434 	kmem_free(tmppath, MAXPATHLEN);
435 	return (0);
436 
437  out:
438 	for (ix = 0; ix < path_alloced; ix++) {
439 		kmem_free(cs->sc_cinfo[ix].ci_path,
440 		    cs->sc_cinfo[ix].ci_pathlen);
441 	}
442 	kmem_free(cs->sc_cinfo, cs->sc_nccdisks * sizeof(struct ccdcinfo));
443 	kmem_free(tmppath, MAXPATHLEN);
444 	return (error);
445 }
446 
447 static void
448 ccdinterleave(struct ccd_softc *cs)
449 {
450 	struct ccdcinfo *ci, *smallci;
451 	struct ccdiinfo *ii;
452 	daddr_t bn, lbn;
453 	int ix;
454 	u_long size;
455 
456 #ifdef DEBUG
457 	if (ccddebug & CCDB_INIT)
458 		printf("ccdinterleave(%p): ileave %d\n", cs, cs->sc_ileave);
459 #endif
460 	/*
461 	 * Allocate an interleave table.
462 	 * Chances are this is too big, but we don't care.
463 	 */
464 	size = (cs->sc_nccdisks + 1) * sizeof(struct ccdiinfo);
465 	cs->sc_itable = kmem_zalloc(size, KM_SLEEP);
466 
467 	/*
468 	 * Trivial case: no interleave (actually interleave of disk size).
469 	 * Each table entry represents a single component in its entirety.
470 	 */
471 	if (cs->sc_ileave == 0) {
472 		bn = 0;
473 		ii = cs->sc_itable;
474 
475 		for (ix = 0; ix < cs->sc_nccdisks; ix++) {
476 			/* Allocate space for ii_index. */
477 			ii->ii_indexsz = sizeof(int);
478 			ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
479 			ii->ii_ndisk = 1;
480 			ii->ii_startblk = bn;
481 			ii->ii_startoff = 0;
482 			ii->ii_index[0] = ix;
483 			bn += cs->sc_cinfo[ix].ci_size;
484 			ii++;
485 		}
486 		ii->ii_ndisk = 0;
487 #ifdef DEBUG
488 		if (ccddebug & CCDB_INIT)
489 			printiinfo(cs->sc_itable);
490 #endif
491 		return;
492 	}
493 
494 	/*
495 	 * The following isn't fast or pretty; it doesn't have to be.
496 	 */
497 	size = 0;
498 	bn = lbn = 0;
499 	for (ii = cs->sc_itable; ; ii++) {
500 		/* Allocate space for ii_index. */
501 		ii->ii_indexsz = sizeof(int) * cs->sc_nccdisks;
502 		ii->ii_index = kmem_alloc(ii->ii_indexsz, KM_SLEEP);
503 
504 		/*
505 		 * Locate the smallest of the remaining components
506 		 */
507 		smallci = NULL;
508 		for (ci = cs->sc_cinfo;
509 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
510 			if (ci->ci_size > size &&
511 			    (smallci == NULL ||
512 			     ci->ci_size < smallci->ci_size))
513 				smallci = ci;
514 
515 		/*
516 		 * Nobody left, all done
517 		 */
518 		if (smallci == NULL) {
519 			ii->ii_ndisk = 0;
520 			break;
521 		}
522 
523 		/*
524 		 * Record starting logical block and component offset
525 		 */
526 		ii->ii_startblk = bn / cs->sc_ileave;
527 		ii->ii_startoff = lbn;
528 
529 		/*
530 		 * Determine how many disks take part in this interleave
531 		 * and record their indices.
532 		 */
533 		ix = 0;
534 		for (ci = cs->sc_cinfo;
535 		     ci < &cs->sc_cinfo[cs->sc_nccdisks]; ci++)
536 			if (ci->ci_size >= smallci->ci_size)
537 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
538 		ii->ii_ndisk = ix;
539 		bn += ix * (smallci->ci_size - size);
540 		lbn = smallci->ci_size / cs->sc_ileave;
541 		size = smallci->ci_size;
542 	}
543 #ifdef DEBUG
544 	if (ccddebug & CCDB_INIT)
545 		printiinfo(cs->sc_itable);
546 #endif
547 }
548 
549 /* ARGSUSED */
550 static int
551 ccdopen(dev_t dev, int flags, int fmt, struct lwp *l)
552 {
553 	int unit = ccdunit(dev);
554 	struct ccd_softc *cs;
555 	struct disklabel *lp;
556 	int error = 0, part, pmask;
557 
558 #ifdef DEBUG
559 	if (ccddebug & CCDB_FOLLOW)
560 		printf("ccdopen(0x%"PRIx64", 0x%x)\n", dev, flags);
561 #endif
562 	if (unit >= numccd)
563 		return (ENXIO);
564 	cs = &ccd_softc[unit];
565 
566 	mutex_enter(&cs->sc_dvlock);
567 
568 	lp = cs->sc_dkdev.dk_label;
569 
570 	part = DISKPART(dev);
571 	pmask = (1 << part);
572 
573 	/*
574 	 * If we're initialized, check to see if there are any other
575 	 * open partitions.  If not, then it's safe to update
576 	 * the in-core disklabel.  Only read the disklabel if it is
577 	 * not already valid.
578 	 */
579 	if ((cs->sc_flags & (CCDF_INITED|CCDF_VLABEL)) == CCDF_INITED &&
580 	    cs->sc_dkdev.dk_openmask == 0)
581 		ccdgetdisklabel(dev);
582 
583 	/* Check that the partition exists. */
584 	if (part != RAW_PART) {
585 		if (((cs->sc_flags & CCDF_INITED) == 0) ||
586 		    ((part >= lp->d_npartitions) ||
587 		     (lp->d_partitions[part].p_fstype == FS_UNUSED))) {
588 			error = ENXIO;
589 			goto done;
590 		}
591 	}
592 
593 	/* Prevent our unit from being unconfigured while open. */
594 	switch (fmt) {
595 	case S_IFCHR:
596 		cs->sc_dkdev.dk_copenmask |= pmask;
597 		break;
598 
599 	case S_IFBLK:
600 		cs->sc_dkdev.dk_bopenmask |= pmask;
601 		break;
602 	}
603 	cs->sc_dkdev.dk_openmask =
604 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
605 
606  done:
607 	mutex_exit(&cs->sc_dvlock);
608 	return (error);
609 }
610 
611 /* ARGSUSED */
612 static int
613 ccdclose(dev_t dev, int flags, int fmt, struct lwp *l)
614 {
615 	int unit = ccdunit(dev);
616 	struct ccd_softc *cs;
617 	int part;
618 
619 #ifdef DEBUG
620 	if (ccddebug & CCDB_FOLLOW)
621 		printf("ccdclose(0x%"PRIx64", 0x%x)\n", dev, flags);
622 #endif
623 
624 	if (unit >= numccd)
625 		return (ENXIO);
626 	cs = &ccd_softc[unit];
627 
628 	mutex_enter(&cs->sc_dvlock);
629 
630 	part = DISKPART(dev);
631 
632 	/* ...that much closer to allowing unconfiguration... */
633 	switch (fmt) {
634 	case S_IFCHR:
635 		cs->sc_dkdev.dk_copenmask &= ~(1 << part);
636 		break;
637 
638 	case S_IFBLK:
639 		cs->sc_dkdev.dk_bopenmask &= ~(1 << part);
640 		break;
641 	}
642 	cs->sc_dkdev.dk_openmask =
643 	    cs->sc_dkdev.dk_copenmask | cs->sc_dkdev.dk_bopenmask;
644 
645 	if (cs->sc_dkdev.dk_openmask == 0) {
646 		if ((cs->sc_flags & CCDF_KLABEL) == 0)
647 			cs->sc_flags &= ~CCDF_VLABEL;
648 	}
649 
650 	mutex_exit(&cs->sc_dvlock);
651 	return (0);
652 }
653 
654 static bool
655 ccdbackoff(struct ccd_softc *cs)
656 {
657 
658 	/* XXX Arbitrary, should be a uvm call. */
659 	return uvmexp.free < (uvmexp.freemin >> 1) &&
660 	    disk_isbusy(&cs->sc_dkdev);
661 }
662 
663 static void
664 ccdthread(void *cookie)
665 {
666 	struct ccd_softc *cs;
667 
668 	cs = cookie;
669 
670 #ifdef DEBUG
671  	if (ccddebug & CCDB_FOLLOW)
672  		printf("ccdthread: hello\n");
673 #endif
674 
675 	mutex_enter(cs->sc_iolock);
676 	while (__predict_true(!cs->sc_zap)) {
677 		if (bufq_peek(cs->sc_bufq) == NULL) {
678 			/* Nothing to do. */
679 			cv_wait(&cs->sc_push, cs->sc_iolock);
680 			continue;
681 		}
682 		if (ccdbackoff(cs)) {
683 			/* Wait for memory to become available. */
684 			(void)cv_timedwait(&cs->sc_push, cs->sc_iolock, 1);
685 			continue;
686 		}
687 #ifdef DEBUG
688  		if (ccddebug & CCDB_FOLLOW)
689  			printf("ccdthread: dispatching I/O\n");
690 #endif
691 		ccdstart(cs);
692 		mutex_enter(cs->sc_iolock);
693 	}
694 	cs->sc_thread = NULL;
695 	mutex_exit(cs->sc_iolock);
696 #ifdef DEBUG
697  	if (ccddebug & CCDB_FOLLOW)
698  		printf("ccdthread: goodbye\n");
699 #endif
700 	kthread_exit(0);
701 }
702 
703 static void
704 ccdstrategy(struct buf *bp)
705 {
706 	int unit = ccdunit(bp->b_dev);
707 	struct ccd_softc *cs = &ccd_softc[unit];
708 
709 	/* Must be open or reading label. */
710 	KASSERT(cs->sc_dkdev.dk_openmask != 0 ||
711 	    (cs->sc_flags & CCDF_RLABEL) != 0);
712 
713 	mutex_enter(cs->sc_iolock);
714 	/* Synchronize with device init/uninit. */
715 	if (__predict_false((cs->sc_flags & CCDF_INITED) == 0)) {
716 		mutex_exit(cs->sc_iolock);
717 #ifdef DEBUG
718  		if (ccddebug & CCDB_FOLLOW)
719  			printf("ccdstrategy: unit %d: not inited\n", unit);
720 #endif
721  		bp->b_error = ENXIO;
722  		bp->b_resid = bp->b_bcount;
723  		biodone(bp);
724 		return;
725 	}
726 
727 	/* Defer to thread if system is low on memory. */
728 	bufq_put(cs->sc_bufq, bp);
729 	if (__predict_false(ccdbackoff(cs))) {
730 		mutex_exit(cs->sc_iolock);
731 #ifdef DEBUG
732  		if (ccddebug & CCDB_FOLLOW)
733  			printf("ccdstrategy: holding off on I/O\n");
734 #endif
735 		return;
736 	}
737 	ccdstart(cs);
738 }
739 
740 static void
741 ccdstart(struct ccd_softc *cs)
742 {
743 	daddr_t blkno;
744 	int wlabel;
745 	struct disklabel *lp;
746 	long bcount, rcount;
747 	struct ccdbuf *cbp;
748 	char *addr;
749 	daddr_t bn;
750 	vnode_t *vp;
751 	buf_t *bp;
752 
753 	KASSERT(mutex_owned(cs->sc_iolock));
754 
755 	disk_busy(&cs->sc_dkdev);
756 	bp = bufq_get(cs->sc_bufq);
757 	KASSERT(bp != NULL);
758 
759 #ifdef DEBUG
760 	if (ccddebug & CCDB_FOLLOW)
761 		printf("ccdstart(%s, %p)\n", cs->sc_xname, bp);
762 #endif
763 
764 	/* If it's a nil transfer, wake up the top half now. */
765 	if (bp->b_bcount == 0)
766 		goto done;
767 
768 	lp = cs->sc_dkdev.dk_label;
769 
770 	/*
771 	 * Do bounds checking and adjust transfer.  If there's an
772 	 * error, the bounds check will flag that for us.  Convert
773 	 * the partition relative block number to an absolute.
774 	 */
775 	blkno = bp->b_blkno;
776 	wlabel = cs->sc_flags & (CCDF_WLABEL|CCDF_LABELLING);
777 	if (DISKPART(bp->b_dev) != RAW_PART) {
778 		if (bounds_check_with_label(&cs->sc_dkdev, bp, wlabel) <= 0)
779 			goto done;
780 		blkno += lp->d_partitions[DISKPART(bp->b_dev)].p_offset;
781 	}
782 	mutex_exit(cs->sc_iolock);
783 	bp->b_rawblkno = blkno;
784 
785 	/* Allocate the component buffers and start I/O! */
786 	bp->b_resid = bp->b_bcount;
787 	bn = bp->b_rawblkno;
788 	addr = bp->b_data;
789 	for (bcount = bp->b_bcount; bcount > 0; bcount -= rcount) {
790 		cbp = ccdbuffer(cs, bp, bn, addr, bcount);
791 		rcount = cbp->cb_buf.b_bcount;
792 		bn += btodb(rcount);
793 		addr += rcount;
794 		vp = cbp->cb_buf.b_vp;
795 		if ((cbp->cb_buf.b_flags & B_READ) == 0) {
796 			mutex_enter(vp->v_interlock);
797 			vp->v_numoutput++;
798 			mutex_exit(vp->v_interlock);
799 		}
800 		(void)VOP_STRATEGY(vp, &cbp->cb_buf);
801 	}
802 	return;
803 
804  done:
805 	disk_unbusy(&cs->sc_dkdev, 0, 0);
806 	cv_broadcast(&cs->sc_stop);
807 	cv_broadcast(&cs->sc_push);
808 	mutex_exit(cs->sc_iolock);
809 	bp->b_resid = bp->b_bcount;
810 	biodone(bp);
811 }
812 
813 /*
814  * Build a component buffer header.
815  */
816 static struct ccdbuf *
817 ccdbuffer(struct ccd_softc *cs, struct buf *bp, daddr_t bn, void *addr,
818     long bcount)
819 {
820 	struct ccdcinfo *ci;
821 	struct ccdbuf *cbp;
822 	daddr_t cbn, cboff;
823 	u_int64_t cbc;
824 	int ccdisk;
825 
826 #ifdef DEBUG
827 	if (ccddebug & CCDB_IO)
828 		printf("ccdbuffer(%p, %p, %" PRId64 ", %p, %ld)\n",
829 		       cs, bp, bn, addr, bcount);
830 #endif
831 	/*
832 	 * Determine which component bn falls in.
833 	 */
834 	cbn = bn;
835 	cboff = 0;
836 
837 	/*
838 	 * Serially concatenated
839 	 */
840 	if (cs->sc_ileave == 0) {
841 		daddr_t sblk;
842 
843 		sblk = 0;
844 		for (ccdisk = 0, ci = &cs->sc_cinfo[ccdisk];
845 		    cbn >= sblk + ci->ci_size;
846 		    ccdisk++, ci = &cs->sc_cinfo[ccdisk])
847 			sblk += ci->ci_size;
848 		cbn -= sblk;
849 	}
850 	/*
851 	 * Interleaved
852 	 */
853 	else {
854 		struct ccdiinfo *ii;
855 		int off;
856 
857 		cboff = cbn % cs->sc_ileave;
858 		cbn /= cs->sc_ileave;
859 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++)
860 			if (ii->ii_startblk > cbn)
861 				break;
862 		ii--;
863 		off = cbn - ii->ii_startblk;
864 		if (ii->ii_ndisk == 1) {
865 			ccdisk = ii->ii_index[0];
866 			cbn = ii->ii_startoff + off;
867 		} else {
868 			ccdisk = ii->ii_index[off % ii->ii_ndisk];
869 			cbn = ii->ii_startoff + off / ii->ii_ndisk;
870 		}
871 		cbn *= cs->sc_ileave;
872 		ci = &cs->sc_cinfo[ccdisk];
873 	}
874 
875 	/*
876 	 * Fill in the component buf structure.
877 	 */
878 	cbp = CCD_GETBUF();
879 	KASSERT(cbp != NULL);
880 	buf_init(&cbp->cb_buf);
881 	cbp->cb_buf.b_flags = bp->b_flags;
882 	cbp->cb_buf.b_oflags = bp->b_oflags;
883 	cbp->cb_buf.b_cflags = bp->b_cflags;
884 	cbp->cb_buf.b_iodone = ccdiodone;
885 	cbp->cb_buf.b_proc = bp->b_proc;
886 	cbp->cb_buf.b_dev = ci->ci_dev;
887 	cbp->cb_buf.b_blkno = cbn + cboff;
888 	cbp->cb_buf.b_data = addr;
889 	cbp->cb_buf.b_vp = ci->ci_vp;
890 	cbp->cb_buf.b_objlock = ci->ci_vp->v_interlock;
891 	if (cs->sc_ileave == 0)
892 		cbc = dbtob((u_int64_t)(ci->ci_size - cbn));
893 	else
894 		cbc = dbtob((u_int64_t)(cs->sc_ileave - cboff));
895 	cbp->cb_buf.b_bcount = cbc < bcount ? cbc : bcount;
896 
897 	/*
898 	 * context for ccdiodone
899 	 */
900 	cbp->cb_obp = bp;
901 	cbp->cb_sc = cs;
902 	cbp->cb_comp = ccdisk;
903 
904 	BIO_COPYPRIO(&cbp->cb_buf, bp);
905 
906 #ifdef DEBUG
907 	if (ccddebug & CCDB_IO)
908 		printf(" dev 0x%"PRIx64"(u%lu): cbp %p bn %" PRId64 " addr %p"
909 		       " bcnt %d\n",
910 		    ci->ci_dev, (unsigned long) (ci-cs->sc_cinfo), cbp,
911 		    cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
912 		    cbp->cb_buf.b_bcount);
913 #endif
914 
915 	return (cbp);
916 }
917 
918 /*
919  * Called at interrupt time.
920  * Mark the component as done and if all components are done,
921  * take a ccd interrupt.
922  */
923 static void
924 ccdiodone(struct buf *vbp)
925 {
926 	struct ccdbuf *cbp = (struct ccdbuf *) vbp;
927 	struct buf *bp = cbp->cb_obp;
928 	struct ccd_softc *cs = cbp->cb_sc;
929 	int count;
930 
931 #ifdef DEBUG
932 	if (ccddebug & CCDB_FOLLOW)
933 		printf("ccdiodone(%p)\n", cbp);
934 	if (ccddebug & CCDB_IO) {
935 		printf("ccdiodone: bp %p bcount %d resid %d\n",
936 		       bp, bp->b_bcount, bp->b_resid);
937 		printf(" dev 0x%"PRIx64"(u%d), cbp %p bn %" PRId64 " addr %p"
938 		       " bcnt %d\n",
939 		       cbp->cb_buf.b_dev, cbp->cb_comp, cbp,
940 		       cbp->cb_buf.b_blkno, cbp->cb_buf.b_data,
941 		       cbp->cb_buf.b_bcount);
942 	}
943 #endif
944 
945 	if (cbp->cb_buf.b_error != 0) {
946 		bp->b_error = cbp->cb_buf.b_error;
947 		printf("%s: error %d on component %d\n",
948 		       cs->sc_xname, bp->b_error, cbp->cb_comp);
949 	}
950 	count = cbp->cb_buf.b_bcount;
951 	buf_destroy(&cbp->cb_buf);
952 	CCD_PUTBUF(cbp);
953 
954 	/*
955 	 * If all done, "interrupt".
956 	 */
957 	mutex_enter(cs->sc_iolock);
958 	bp->b_resid -= count;
959 	if (bp->b_resid < 0)
960 		panic("ccdiodone: count");
961 	if (bp->b_resid == 0) {
962 		/*
963 		 * Request is done for better or worse, wakeup the top half.
964 		 */
965 		if (bp->b_error != 0)
966 			bp->b_resid = bp->b_bcount;
967 		disk_unbusy(&cs->sc_dkdev, (bp->b_bcount - bp->b_resid),
968 		    (bp->b_flags & B_READ));
969 		if (!disk_isbusy(&cs->sc_dkdev)) {
970 			if (bufq_peek(cs->sc_bufq) != NULL) {
971 				cv_broadcast(&cs->sc_push);
972 			}
973 			cv_broadcast(&cs->sc_stop);
974 		}
975 		mutex_exit(cs->sc_iolock);
976 		biodone(bp);
977 	} else
978 		mutex_exit(cs->sc_iolock);
979 }
980 
981 /* ARGSUSED */
982 static int
983 ccdread(dev_t dev, struct uio *uio, int flags)
984 {
985 	int unit = ccdunit(dev);
986 	struct ccd_softc *cs;
987 
988 #ifdef DEBUG
989 	if (ccddebug & CCDB_FOLLOW)
990 		printf("ccdread(0x%"PRIx64", %p)\n", dev, uio);
991 #endif
992 	if (unit >= numccd)
993 		return (ENXIO);
994 	cs = &ccd_softc[unit];
995 
996 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
997 	if ((cs->sc_flags & CCDF_INITED) == 0)
998 		return (ENXIO);
999 
1000 	return (physio(ccdstrategy, NULL, dev, B_READ, minphys, uio));
1001 }
1002 
1003 /* ARGSUSED */
1004 static int
1005 ccdwrite(dev_t dev, struct uio *uio, int flags)
1006 {
1007 	int unit = ccdunit(dev);
1008 	struct ccd_softc *cs;
1009 
1010 #ifdef DEBUG
1011 	if (ccddebug & CCDB_FOLLOW)
1012 		printf("ccdwrite(0x%"PRIx64", %p)\n", dev, uio);
1013 #endif
1014 	if (unit >= numccd)
1015 		return (ENXIO);
1016 	cs = &ccd_softc[unit];
1017 
1018 	/* Unlocked advisory check, ccdstrategy check is synchronous. */
1019 	if ((cs->sc_flags & CCDF_INITED) == 0)
1020 		return (ENXIO);
1021 
1022 	return (physio(ccdstrategy, NULL, dev, B_WRITE, minphys, uio));
1023 }
1024 
1025 static int
1026 ccdioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
1027 {
1028 	int unit = ccdunit(dev);
1029 	int i, j, lookedup = 0, error = 0;
1030 	int part, pmask;
1031 	struct ccd_softc *cs;
1032 	struct ccd_ioctl *ccio = (struct ccd_ioctl *)data;
1033 	kauth_cred_t uc;
1034 	char **cpp;
1035 	struct pathbuf *pb;
1036 	struct vnode **vpp;
1037 #ifdef __HAVE_OLD_DISKLABEL
1038 	struct disklabel newlabel;
1039 #endif
1040 
1041 	if (unit >= numccd)
1042 		return (ENXIO);
1043 	cs = &ccd_softc[unit];
1044 	uc = kauth_cred_get();
1045 
1046 	/* Must be open for writes for these commands... */
1047 	switch (cmd) {
1048 	case CCDIOCSET:
1049 	case CCDIOCCLR:
1050 	case DIOCSDINFO:
1051 	case DIOCWDINFO:
1052 #ifdef __HAVE_OLD_DISKLABEL
1053 	case ODIOCSDINFO:
1054 	case ODIOCWDINFO:
1055 #endif
1056 	case DIOCKLABEL:
1057 	case DIOCWLABEL:
1058 		if ((flag & FWRITE) == 0)
1059 			return (EBADF);
1060 	}
1061 
1062 	mutex_enter(&cs->sc_dvlock);
1063 
1064 	/* Must be initialized for these... */
1065 	switch (cmd) {
1066 	case CCDIOCCLR:
1067 	case DIOCGDINFO:
1068 	case DIOCCACHESYNC:
1069 	case DIOCSDINFO:
1070 	case DIOCWDINFO:
1071 	case DIOCGPART:
1072 	case DIOCWLABEL:
1073 	case DIOCKLABEL:
1074 	case DIOCGDEFLABEL:
1075 #ifdef __HAVE_OLD_DISKLABEL
1076 	case ODIOCGDINFO:
1077 	case ODIOCSDINFO:
1078 	case ODIOCWDINFO:
1079 	case ODIOCGDEFLABEL:
1080 #endif
1081 		if ((cs->sc_flags & CCDF_INITED) == 0) {
1082 			error = ENXIO;
1083 			goto out;
1084 		}
1085 	}
1086 
1087 	switch (cmd) {
1088 	case CCDIOCSET:
1089 		if (cs->sc_flags & CCDF_INITED) {
1090 			error = EBUSY;
1091 			goto out;
1092 		}
1093 
1094 		/* Validate the flags. */
1095 		if ((ccio->ccio_flags & CCDF_USERMASK) != ccio->ccio_flags) {
1096 			error = EINVAL;
1097 			goto out;
1098 		}
1099 
1100 		if (ccio->ccio_ndisks > CCD_MAXNDISKS ||
1101 		    ccio->ccio_ndisks == 0) {
1102 			error = EINVAL;
1103 			goto out;
1104 		}
1105 
1106 		/* Fill in some important bits. */
1107 		cs->sc_ileave = ccio->ccio_ileave;
1108 		cs->sc_nccdisks = ccio->ccio_ndisks;
1109 		cs->sc_flags = ccio->ccio_flags & CCDF_USERMASK;
1110 
1111 		/*
1112 		 * Allocate space for and copy in the array of
1113 		 * componet pathnames and device numbers.
1114 		 */
1115 		cpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*cpp), KM_SLEEP);
1116 		vpp = kmem_alloc(ccio->ccio_ndisks * sizeof(*vpp), KM_SLEEP);
1117 		error = copyin(ccio->ccio_disks, cpp,
1118 		    ccio->ccio_ndisks * sizeof(*cpp));
1119 		if (error) {
1120 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1121 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1122 			goto out;
1123 		}
1124 
1125 #ifdef DEBUG
1126 		if (ccddebug & CCDB_INIT)
1127 			for (i = 0; i < ccio->ccio_ndisks; ++i)
1128 				printf("ccdioctl: component %d: %p\n",
1129 				    i, cpp[i]);
1130 #endif
1131 
1132 		for (i = 0; i < ccio->ccio_ndisks; ++i) {
1133 #ifdef DEBUG
1134 			if (ccddebug & CCDB_INIT)
1135 				printf("ccdioctl: lookedup = %d\n", lookedup);
1136 #endif
1137 			error = pathbuf_copyin(cpp[i], &pb);
1138 			if (error == 0) {
1139 				error = dk_lookup(pb, l, &vpp[i]);
1140 			}
1141 			pathbuf_destroy(pb);
1142 			if (error != 0) {
1143 				for (j = 0; j < lookedup; ++j)
1144 					(void)vn_close(vpp[j], FREAD|FWRITE,
1145 					    uc);
1146 				kmem_free(vpp, ccio->ccio_ndisks *
1147 				    sizeof(*vpp));
1148 				kmem_free(cpp, ccio->ccio_ndisks *
1149 				    sizeof(*cpp));
1150 				goto out;
1151 			}
1152 			++lookedup;
1153 		}
1154 
1155 		/* Attach the disk. */
1156 		disk_attach(&cs->sc_dkdev);
1157 		bufq_alloc(&cs->sc_bufq, "fcfs", 0);
1158 
1159 		/*
1160 		 * Initialize the ccd.  Fills in the softc for us.
1161 		 */
1162 		if ((error = ccdinit(cs, cpp, vpp, l)) != 0) {
1163 			for (j = 0; j < lookedup; ++j)
1164 				(void)vn_close(vpp[j], FREAD|FWRITE,
1165 				    uc);
1166 			kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1167 			kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1168 			disk_detach(&cs->sc_dkdev);
1169 			bufq_free(cs->sc_bufq);
1170 			goto out;
1171 		}
1172 
1173 		/* We can free the temporary variables now. */
1174 		kmem_free(vpp, ccio->ccio_ndisks * sizeof(*vpp));
1175 		kmem_free(cpp, ccio->ccio_ndisks * sizeof(*cpp));
1176 
1177 		/*
1178 		 * The ccd has been successfully initialized, so
1179 		 * we can place it into the array.  Don't try to
1180 		 * read the disklabel until the disk has been attached,
1181 		 * because space for the disklabel is allocated
1182 		 * in disk_attach();
1183 		 */
1184 		ccio->ccio_unit = unit;
1185 		ccio->ccio_size = cs->sc_size;
1186 
1187 		/* Try and read the disklabel. */
1188 		ccdgetdisklabel(dev);
1189 		break;
1190 
1191 	case CCDIOCCLR:
1192 		/*
1193 		 * Don't unconfigure if any other partitions are open
1194 		 * or if both the character and block flavors of this
1195 		 * partition are open.
1196 		 */
1197 		part = DISKPART(dev);
1198 		pmask = (1 << part);
1199 		if ((cs->sc_dkdev.dk_openmask & ~pmask) ||
1200 		    ((cs->sc_dkdev.dk_bopenmask & pmask) &&
1201 		    (cs->sc_dkdev.dk_copenmask & pmask))) {
1202 			error = EBUSY;
1203 			goto out;
1204 		}
1205 
1206 		/* Stop new I/O, wait for in-flight I/O to complete. */
1207 		mutex_enter(cs->sc_iolock);
1208 		cs->sc_flags &= ~(CCDF_INITED|CCDF_VLABEL);
1209 		cs->sc_zap = true;
1210 		while (disk_isbusy(&cs->sc_dkdev) ||
1211 		    bufq_peek(cs->sc_bufq) != NULL ||
1212 		    cs->sc_thread != NULL) {
1213 			cv_broadcast(&cs->sc_push);
1214 			(void)cv_timedwait(&cs->sc_stop, cs->sc_iolock, hz);
1215 		}
1216 		mutex_exit(cs->sc_iolock);
1217 
1218 		/*
1219 		 * Free ccd_softc information and clear entry.
1220 		 */
1221 
1222 		/* Close the components and free their pathnames. */
1223 		for (i = 0; i < cs->sc_nccdisks; ++i) {
1224 			/*
1225 			 * XXX: this close could potentially fail and
1226 			 * cause Bad Things.  Maybe we need to force
1227 			 * the close to happen?
1228 			 */
1229 #ifdef DEBUG
1230 			if (ccddebug & CCDB_VNODE)
1231 				vprint("CCDIOCCLR: vnode info",
1232 				    cs->sc_cinfo[i].ci_vp);
1233 #endif
1234 			(void)vn_close(cs->sc_cinfo[i].ci_vp, FREAD|FWRITE,
1235 			    uc);
1236 			kmem_free(cs->sc_cinfo[i].ci_path,
1237 			    cs->sc_cinfo[i].ci_pathlen);
1238 		}
1239 
1240 		/* Free interleave index. */
1241 		for (i = 0; cs->sc_itable[i].ii_ndisk; ++i) {
1242 			kmem_free(cs->sc_itable[i].ii_index,
1243 			    cs->sc_itable[i].ii_indexsz);
1244 		}
1245 
1246 		/* Free component info and interleave table. */
1247 		kmem_free(cs->sc_cinfo, cs->sc_nccdisks *
1248 		    sizeof(struct ccdcinfo));
1249 		kmem_free(cs->sc_itable, (cs->sc_nccdisks + 1) *
1250 		    sizeof(struct ccdiinfo));
1251 
1252 		/* Detatch the disk. */
1253 		disk_detach(&cs->sc_dkdev);
1254 		bufq_free(cs->sc_bufq);
1255 		break;
1256 
1257 	case DIOCGDINFO:
1258 		*(struct disklabel *)data = *(cs->sc_dkdev.dk_label);
1259 		break;
1260 
1261 #ifdef __HAVE_OLD_DISKLABEL
1262 	case ODIOCGDINFO:
1263 		newlabel = *(cs->sc_dkdev.dk_label);
1264 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1265 			return ENOTTY;
1266 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1267 		break;
1268 #endif
1269 
1270 	case DIOCGPART:
1271 		((struct partinfo *)data)->disklab = cs->sc_dkdev.dk_label;
1272 		((struct partinfo *)data)->part =
1273 		    &cs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1274 		break;
1275 
1276 	case DIOCCACHESYNC:
1277 		/*
1278 		 * XXX Do we really need to care about having a writable
1279 		 * file descriptor here?
1280 		 */
1281 		if ((flag & FWRITE) == 0)
1282 			return (EBADF);
1283 
1284 		/*
1285 		 * We pass this call down to all components and report
1286 		 * the first error we encounter.
1287 		 */
1288 		for (error = 0, i = 0; i < cs->sc_nccdisks; i++) {
1289 			j = VOP_IOCTL(cs->sc_cinfo[i].ci_vp, cmd, data,
1290 				      flag, uc);
1291 			if (j != 0 && error == 0)
1292 				error = j;
1293 		}
1294 		break;
1295 
1296 	case DIOCWDINFO:
1297 	case DIOCSDINFO:
1298 #ifdef __HAVE_OLD_DISKLABEL
1299 	case ODIOCWDINFO:
1300 	case ODIOCSDINFO:
1301 #endif
1302 	{
1303 		struct disklabel *lp;
1304 #ifdef __HAVE_OLD_DISKLABEL
1305 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1306 			memset(&newlabel, 0, sizeof newlabel);
1307 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1308 			lp = &newlabel;
1309 		} else
1310 #endif
1311 		lp = (struct disklabel *)data;
1312 
1313 		cs->sc_flags |= CCDF_LABELLING;
1314 
1315 		error = setdisklabel(cs->sc_dkdev.dk_label,
1316 		    lp, 0, cs->sc_dkdev.dk_cpulabel);
1317 		if (error == 0) {
1318 			if (cmd == DIOCWDINFO
1319 #ifdef __HAVE_OLD_DISKLABEL
1320 			    || cmd == ODIOCWDINFO
1321 #endif
1322 			   )
1323 				error = writedisklabel(CCDLABELDEV(dev),
1324 				    ccdstrategy, cs->sc_dkdev.dk_label,
1325 				    cs->sc_dkdev.dk_cpulabel);
1326 		}
1327 
1328 		cs->sc_flags &= ~CCDF_LABELLING;
1329 		break;
1330 	}
1331 
1332 	case DIOCKLABEL:
1333 		if (*(int *)data != 0)
1334 			cs->sc_flags |= CCDF_KLABEL;
1335 		else
1336 			cs->sc_flags &= ~CCDF_KLABEL;
1337 		break;
1338 
1339 	case DIOCWLABEL:
1340 		if (*(int *)data != 0)
1341 			cs->sc_flags |= CCDF_WLABEL;
1342 		else
1343 			cs->sc_flags &= ~CCDF_WLABEL;
1344 		break;
1345 
1346 	case DIOCGDEFLABEL:
1347 		ccdgetdefaultlabel(cs, (struct disklabel *)data);
1348 		break;
1349 
1350 #ifdef __HAVE_OLD_DISKLABEL
1351 	case ODIOCGDEFLABEL:
1352 		ccdgetdefaultlabel(cs, &newlabel);
1353 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1354 			return ENOTTY;
1355 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1356 		break;
1357 #endif
1358 
1359 	default:
1360 		error = ENOTTY;
1361 	}
1362 
1363  out:
1364 	mutex_exit(&cs->sc_dvlock);
1365 	return (error);
1366 }
1367 
1368 static int
1369 ccdsize(dev_t dev)
1370 {
1371 	struct ccd_softc *cs;
1372 	struct disklabel *lp;
1373 	int part, unit, omask, size;
1374 
1375 	unit = ccdunit(dev);
1376 	if (unit >= numccd)
1377 		return (-1);
1378 	cs = &ccd_softc[unit];
1379 
1380 	if ((cs->sc_flags & CCDF_INITED) == 0)
1381 		return (-1);
1382 
1383 	part = DISKPART(dev);
1384 	omask = cs->sc_dkdev.dk_openmask & (1 << part);
1385 	lp = cs->sc_dkdev.dk_label;
1386 
1387 	if (omask == 0 && ccdopen(dev, 0, S_IFBLK, curlwp))
1388 		return (-1);
1389 
1390 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
1391 		size = -1;
1392 	else
1393 		size = lp->d_partitions[part].p_size *
1394 		    (lp->d_secsize / DEV_BSIZE);
1395 
1396 	if (omask == 0 && ccdclose(dev, 0, S_IFBLK, curlwp))
1397 		return (-1);
1398 
1399 	return (size);
1400 }
1401 
1402 static void
1403 ccdgetdefaultlabel(struct ccd_softc *cs, struct disklabel *lp)
1404 {
1405 	struct ccdgeom *ccg = &cs->sc_geom;
1406 
1407 	memset(lp, 0, sizeof(*lp));
1408 
1409 	lp->d_secperunit = cs->sc_size;
1410 	lp->d_secsize = ccg->ccg_secsize;
1411 	lp->d_nsectors = ccg->ccg_nsectors;
1412 	lp->d_ntracks = ccg->ccg_ntracks;
1413 	lp->d_ncylinders = ccg->ccg_ncylinders;
1414 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1415 
1416 	strncpy(lp->d_typename, "ccd", sizeof(lp->d_typename));
1417 	lp->d_type = DTYPE_CCD;
1418 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1419 	lp->d_rpm = 3600;
1420 	lp->d_interleave = 1;
1421 	lp->d_flags = 0;
1422 
1423 	lp->d_partitions[RAW_PART].p_offset = 0;
1424 	lp->d_partitions[RAW_PART].p_size = cs->sc_size;
1425 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
1426 	lp->d_npartitions = RAW_PART + 1;
1427 
1428 	lp->d_magic = DISKMAGIC;
1429 	lp->d_magic2 = DISKMAGIC;
1430 	lp->d_checksum = dkcksum(cs->sc_dkdev.dk_label);
1431 }
1432 
1433 /*
1434  * Read the disklabel from the ccd.  If one is not present, fake one
1435  * up.
1436  */
1437 static void
1438 ccdgetdisklabel(dev_t dev)
1439 {
1440 	int unit = ccdunit(dev);
1441 	struct ccd_softc *cs = &ccd_softc[unit];
1442 	const char *errstring;
1443 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1444 	struct cpu_disklabel *clp = cs->sc_dkdev.dk_cpulabel;
1445 
1446 	KASSERT(mutex_owned(&cs->sc_dvlock));
1447 
1448 	memset(clp, 0, sizeof(*clp));
1449 
1450 	ccdgetdefaultlabel(cs, lp);
1451 
1452 	/*
1453 	 * Call the generic disklabel extraction routine.
1454 	 */
1455 	cs->sc_flags |= CCDF_RLABEL;
1456 	if ((cs->sc_flags & CCDF_NOLABEL) != 0)
1457 		errstring = "CCDF_NOLABEL set; ignoring on-disk label";
1458 	else
1459 		errstring = readdisklabel(CCDLABELDEV(dev), ccdstrategy,
1460 		    cs->sc_dkdev.dk_label, cs->sc_dkdev.dk_cpulabel);
1461 	if (errstring)
1462 		ccdmakedisklabel(cs);
1463 	else {
1464 		int i;
1465 		struct partition *pp;
1466 
1467 		/*
1468 		 * Sanity check whether the found disklabel is valid.
1469 		 *
1470 		 * This is necessary since total size of ccd may vary
1471 		 * when an interleave is changed even though exactly
1472 		 * same componets are used, and old disklabel may used
1473 		 * if that is found.
1474 		 */
1475 		if (lp->d_secperunit != cs->sc_size)
1476 			printf("WARNING: %s: "
1477 			    "total sector size in disklabel (%d) != "
1478 			    "the size of ccd (%lu)\n", cs->sc_xname,
1479 			    lp->d_secperunit, (u_long)cs->sc_size);
1480 		for (i = 0; i < lp->d_npartitions; i++) {
1481 			pp = &lp->d_partitions[i];
1482 			if (pp->p_offset + pp->p_size > cs->sc_size)
1483 				printf("WARNING: %s: end of partition `%c' "
1484 				    "exceeds the size of ccd (%lu)\n",
1485 				    cs->sc_xname, 'a' + i, (u_long)cs->sc_size);
1486 		}
1487 	}
1488 
1489 #ifdef DEBUG
1490 	/* It's actually extremely common to have unlabeled ccds. */
1491 	if (ccddebug & CCDB_LABEL)
1492 		if (errstring != NULL)
1493 			printf("%s: %s\n", cs->sc_xname, errstring);
1494 #endif
1495 
1496 	/* In-core label now valid. */
1497 	cs->sc_flags = (cs->sc_flags | CCDF_VLABEL) & ~CCDF_RLABEL;
1498 }
1499 
1500 /*
1501  * Take care of things one might want to take care of in the event
1502  * that a disklabel isn't present.
1503  */
1504 static void
1505 ccdmakedisklabel(struct ccd_softc *cs)
1506 {
1507 	struct disklabel *lp = cs->sc_dkdev.dk_label;
1508 
1509 	/*
1510 	 * For historical reasons, if there's no disklabel present
1511 	 * the raw partition must be marked FS_BSDFFS.
1512 	 */
1513 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
1514 
1515 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
1516 
1517 	lp->d_checksum = dkcksum(lp);
1518 }
1519 
1520 #ifdef DEBUG
1521 static void
1522 printiinfo(struct ccdiinfo *ii)
1523 {
1524 	int ix, i;
1525 
1526 	for (ix = 0; ii->ii_ndisk; ix++, ii++) {
1527 		printf(" itab[%d]: #dk %d sblk %" PRId64 " soff %" PRId64,
1528 		    ix, ii->ii_ndisk, ii->ii_startblk, ii->ii_startoff);
1529 		for (i = 0; i < ii->ii_ndisk; i++)
1530 			printf(" %d", ii->ii_index[i]);
1531 		printf("\n");
1532 	}
1533 }
1534 #endif
1535 
1536 MODULE(MODULE_CLASS_DRIVER, ccd, NULL);
1537 
1538 static int
1539 ccd_modcmd(modcmd_t cmd, void *arg)
1540 {
1541 	int bmajor, cmajor, error = 0;
1542 
1543 	bmajor = cmajor = -1;
1544 
1545 	switch (cmd) {
1546 	case MODULE_CMD_INIT:
1547 #ifdef _MODULE
1548 		ccdattach(4);
1549 
1550 		return devsw_attach("ccd", &ccd_bdevsw, &bmajor,
1551 		    &ccd_cdevsw, &cmajor);
1552 #endif
1553 		break;
1554 
1555 	case MODULE_CMD_FINI:
1556 #ifdef _MODULE
1557 		return devsw_detach(&ccd_bdevsw, &ccd_cdevsw);
1558 #endif
1559 		break;
1560 
1561 	case MODULE_CMD_STAT:
1562 		return ENOTTY;
1563 
1564 	default:
1565 		return ENOTTY;
1566 	}
1567 
1568 	return error;
1569 }
1570