xref: /freebsd/sys/geom/geom_ccd.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * SPDX-License-Identifier: (BSD-2-Clause AND BSD-3-Clause)
3  *
4  * Copyright (c) 2003 Poul-Henning Kamp.
5  * Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
6  * All rights reserved.
7  *
8  * This code is derived from software contributed to The NetBSD Foundation
9  * by Jason R. Thorpe.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  *
32  * $NetBSD: ccd.c,v 1.22 1995/12/08 19:13:26 thorpej Exp $
33  */
34 
35 /*-
36  * Copyright (c) 1988 University of Utah.
37  * Copyright (c) 1990, 1993
38  *	The Regents of the University of California.  All rights reserved.
39  *
40  * This code is derived from software contributed to Berkeley by
41  * the Systems Programming Group of the University of Utah Computer
42  * Science Department.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. Neither the name of the University nor the names of its contributors
53  *    may be used to endorse or promote products derived from this software
54  *    without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66  * SUCH DAMAGE.
67  *
68  * from: Utah $Hdr: cd.c 1.6 90/11/28$
69  *
70  *	@(#)cd.c	8.2 (Berkeley) 11/16/93
71  */
72 
73 /*
74  * Dynamic configuration and disklabel support by:
75  *	Jason R. Thorpe <thorpej@nas.nasa.gov>
76  *	Numerical Aerodynamic Simulation Facility
77  *	Mail Stop 258-6
78  *	NASA Ames Research Center
79  *	Moffett Field, CA 94035
80  */
81 
82 #include <sys/cdefs.h>
83 #include <sys/param.h>
84 #include <sys/systm.h>
85 #include <sys/kernel.h>
86 #include <sys/module.h>
87 #include <sys/bio.h>
88 #include <sys/malloc.h>
89 #include <sys/sbuf.h>
90 #include <geom/geom.h>
91 
92 /*
93  * Number of blocks to untouched in front of a component partition.
94  * This is to avoid violating its disklabel area when it starts at the
95  * beginning of the slice.
96  */
97 #if !defined(CCD_OFFSET)
98 #define CCD_OFFSET 16
99 #endif
100 
101 /* sc_flags */
102 #define CCDF_UNIFORM	0x02	/* use LCCD of sizes for uniform interleave */
103 #define CCDF_MIRROR	0x04	/* use mirroring */
104 #define CCDF_NO_OFFSET	0x08	/* do not leave space in front */
105 #define CCDF_LINUX	0x10	/* use Linux compatibility mode */
106 
107 /* Mask of user-settable ccd flags. */
108 #define CCDF_USERMASK	(CCDF_UNIFORM|CCDF_MIRROR)
109 
110 /*
111  * Interleave description table.
112  * Computed at boot time to speed irregular-interleave lookups.
113  * The idea is that we interleave in "groups".  First we interleave
114  * evenly over all component disks up to the size of the smallest
115  * component (the first group), then we interleave evenly over all
116  * remaining disks up to the size of the next-smallest (second group),
117  * and so on.
118  *
119  * Each table entry describes the interleave characteristics of one
120  * of these groups.  For example if a concatenated disk consisted of
121  * three components of 5, 3, and 7 DEV_BSIZE blocks interleaved at
122  * DEV_BSIZE (1), the table would have three entries:
123  *
124  *	ndisk	startblk	startoff	dev
125  *	3	0		0		0, 1, 2
126  *	2	9		3		0, 2
127  *	1	13		5		2
128  *	0	-		-		-
129  *
130  * which says that the first nine blocks (0-8) are interleaved over
131  * 3 disks (0, 1, 2) starting at block offset 0 on any component disk,
132  * the next 4 blocks (9-12) are interleaved over 2 disks (0, 2) starting
133  * at component block 3, and the remaining blocks (13-14) are on disk
134  * 2 starting at offset 5.
135  */
136 struct ccdiinfo {
137 	int	ii_ndisk;	/* # of disks range is interleaved over */
138 	daddr_t	ii_startblk;	/* starting scaled block # for range */
139 	daddr_t	ii_startoff;	/* starting component offset (block #) */
140 	int	*ii_index;	/* ordered list of components in range */
141 };
142 
143 /*
144  * Component info table.
145  * Describes a single component of a concatenated disk.
146  */
147 struct ccdcinfo {
148 	daddr_t		ci_size; 		/* size */
149 	struct g_provider *ci_provider;		/* provider */
150 	struct g_consumer *ci_consumer;		/* consumer */
151 };
152 
153 /*
154  * A concatenated disk is described by this structure.
155  */
156 
157 struct ccd_s {
158 	LIST_ENTRY(ccd_s) list;
159 
160 	int		 sc_unit;		/* logical unit number */
161 	int		 sc_flags;		/* flags */
162 	daddr_t		 sc_size;		/* size of ccd */
163 	int		 sc_ileave;		/* interleave */
164 	u_int		 sc_ndisks;		/* number of components */
165 	struct ccdcinfo	 *sc_cinfo;		/* component info */
166 	struct ccdiinfo	 *sc_itable;		/* interleave table */
167 	uint32_t	 sc_secsize;		/* # bytes per sector */
168 	int		 sc_pick;		/* side of mirror picked */
169 	daddr_t		 sc_blk[2];		/* mirror localization */
170 	uint32_t	 sc_offset;		/* actual offset used */
171 };
172 
173 static g_start_t g_ccd_start;
174 static void ccdiodone(struct bio *bp);
175 static void ccdinterleave(struct ccd_s *);
176 static int ccdinit(struct gctl_req *req, struct ccd_s *);
177 static int ccdbuffer(struct bio **ret, struct ccd_s *,
178 		      struct bio *, daddr_t, caddr_t, long);
179 
180 static void
181 g_ccd_orphan(struct g_consumer *cp)
182 {
183 	/*
184 	 * XXX: We don't do anything here.  It is not obvious
185 	 * XXX: what DTRT would be, so we do what the previous
186 	 * XXX: code did: ignore it and let the user cope.
187 	 */
188 }
189 
190 static int
191 g_ccd_access(struct g_provider *pp, int dr, int dw, int de)
192 {
193 	struct g_geom *gp;
194 	struct g_consumer *cp1, *cp2;
195 	int error;
196 
197 	de += dr;
198 	de += dw;
199 
200 	gp = pp->geom;
201 	error = ENXIO;
202 	LIST_FOREACH(cp1, &gp->consumer, consumer) {
203 		error = g_access(cp1, dr, dw, de);
204 		if (error) {
205 			LIST_FOREACH(cp2, &gp->consumer, consumer) {
206 				if (cp1 == cp2)
207 					break;
208 				g_access(cp2, -dr, -dw, -de);
209 			}
210 			break;
211 		}
212 	}
213 	return (error);
214 }
215 
216 /*
217  * Free the softc and its substructures.
218  */
219 static void
220 g_ccd_freesc(struct ccd_s *sc)
221 {
222 	struct ccdiinfo *ii;
223 
224 	g_free(sc->sc_cinfo);
225 	if (sc->sc_itable != NULL) {
226 		for (ii = sc->sc_itable; ii->ii_ndisk > 0; ii++)
227 			g_free(ii->ii_index);
228 		g_free(sc->sc_itable);
229 	}
230 	g_free(sc);
231 }
232 
233 static int
234 ccdinit(struct gctl_req *req, struct ccd_s *cs)
235 {
236 	struct ccdcinfo *ci;
237 	daddr_t size;
238 	int ix;
239 	daddr_t minsize;
240 	int maxsecsize;
241 	off_t mediasize;
242 	u_int sectorsize;
243 
244 	cs->sc_size = 0;
245 
246 	maxsecsize = 0;
247 	minsize = 0;
248 
249 	if (cs->sc_flags & CCDF_LINUX) {
250 		cs->sc_offset = 0;
251 		cs->sc_ileave *= 2;
252 		if (cs->sc_flags & CCDF_MIRROR && cs->sc_ndisks != 2)
253 			gctl_error(req, "Mirror mode for Linux raids is "
254 			                "only supported with 2 devices");
255 	} else {
256 		if (cs->sc_flags & CCDF_NO_OFFSET)
257 			cs->sc_offset = 0;
258 		else
259 			cs->sc_offset = CCD_OFFSET;
260 	}
261 	for (ix = 0; ix < cs->sc_ndisks; ix++) {
262 		ci = &cs->sc_cinfo[ix];
263 
264 		mediasize = ci->ci_provider->mediasize;
265 		sectorsize = ci->ci_provider->sectorsize;
266 		if (sectorsize > maxsecsize)
267 			maxsecsize = sectorsize;
268 		size = mediasize / DEV_BSIZE - cs->sc_offset;
269 
270 		/* Truncate to interleave boundary */
271 
272 		if (cs->sc_ileave > 1)
273 			size -= size % cs->sc_ileave;
274 
275 		if (size == 0) {
276 			gctl_error(req, "Component %s has effective size zero",
277 			    ci->ci_provider->name);
278 			return(ENODEV);
279 		}
280 
281 		if (minsize == 0 || size < minsize)
282 			minsize = size;
283 		ci->ci_size = size;
284 		cs->sc_size += size;
285 	}
286 
287 	/*
288 	 * Don't allow the interleave to be smaller than
289 	 * the biggest component sector.
290 	 */
291 	if ((cs->sc_ileave > 0) &&
292 	    (cs->sc_ileave < (maxsecsize / DEV_BSIZE))) {
293 		gctl_error(req, "Interleave to small for sector size");
294 		return(EINVAL);
295 	}
296 
297 	/*
298 	 * If uniform interleave is desired set all sizes to that of
299 	 * the smallest component.  This will guarantee that a single
300 	 * interleave table is generated.
301 	 *
302 	 * Lost space must be taken into account when calculating the
303 	 * overall size.  Half the space is lost when CCDF_MIRROR is
304 	 * specified.
305 	 */
306 	if (cs->sc_flags & CCDF_UNIFORM) {
307 		for (ix = 0; ix < cs->sc_ndisks; ix++) {
308 			ci = &cs->sc_cinfo[ix];
309 			ci->ci_size = minsize;
310 		}
311 		cs->sc_size = cs->sc_ndisks * minsize;
312 	}
313 
314 	if (cs->sc_flags & CCDF_MIRROR) {
315 		/*
316 		 * Check to see if an even number of components
317 		 * have been specified.  The interleave must also
318 		 * be non-zero in order for us to be able to
319 		 * guarantee the topology.
320 		 */
321 		if (cs->sc_ndisks % 2) {
322 			gctl_error(req,
323 			      "Mirroring requires an even number of disks");
324 			return(EINVAL);
325 		}
326 		if (cs->sc_ileave == 0) {
327 			gctl_error(req,
328 			     "An interleave must be specified when mirroring");
329 			return(EINVAL);
330 		}
331 		cs->sc_size = (cs->sc_ndisks/2) * minsize;
332 	}
333 
334 	/*
335 	 * Construct the interleave table.
336 	 */
337 	ccdinterleave(cs);
338 
339 	/*
340 	 * Create pseudo-geometry based on 1MB cylinders.  It's
341 	 * pretty close.
342 	 */
343 	cs->sc_secsize = maxsecsize;
344 
345 	return (0);
346 }
347 
348 static void
349 ccdinterleave(struct ccd_s *cs)
350 {
351 	struct ccdcinfo *ci, *smallci;
352 	struct ccdiinfo *ii;
353 	daddr_t bn, lbn;
354 	int ix;
355 	daddr_t size;
356 
357 	/*
358 	 * Allocate an interleave table.  The worst case occurs when each
359 	 * of N disks is of a different size, resulting in N interleave
360 	 * tables.
361 	 *
362 	 * Chances are this is too big, but we don't care.
363 	 */
364 	size = (cs->sc_ndisks + 1) * sizeof(struct ccdiinfo);
365 	cs->sc_itable = g_malloc(size, M_WAITOK | M_ZERO);
366 
367 	/*
368 	 * Trivial case: no interleave (actually interleave of disk size).
369 	 * Each table entry represents a single component in its entirety.
370 	 *
371 	 * An interleave of 0 may not be used with a mirror setup.
372 	 */
373 	if (cs->sc_ileave == 0) {
374 		bn = 0;
375 		ii = cs->sc_itable;
376 
377 		for (ix = 0; ix < cs->sc_ndisks; ix++) {
378 			/* Allocate space for ii_index. */
379 			ii->ii_index = g_malloc(sizeof(int), M_WAITOK);
380 			ii->ii_ndisk = 1;
381 			ii->ii_startblk = bn;
382 			ii->ii_startoff = 0;
383 			ii->ii_index[0] = ix;
384 			bn += cs->sc_cinfo[ix].ci_size;
385 			ii++;
386 		}
387 		ii->ii_ndisk = 0;
388 		return;
389 	}
390 
391 	/*
392 	 * The following isn't fast or pretty; it doesn't have to be.
393 	 */
394 	size = 0;
395 	bn = lbn = 0;
396 	for (ii = cs->sc_itable; ; ii++) {
397 		/*
398 		 * Allocate space for ii_index.  We might allocate more then
399 		 * we use.
400 		 */
401 		ii->ii_index = g_malloc((sizeof(int) * cs->sc_ndisks),
402 		    M_WAITOK);
403 
404 		/*
405 		 * Locate the smallest of the remaining components
406 		 */
407 		smallci = NULL;
408 		for (ci = cs->sc_cinfo; ci < &cs->sc_cinfo[cs->sc_ndisks];
409 		    ci++) {
410 			if (ci->ci_size > size &&
411 			    (smallci == NULL ||
412 			     ci->ci_size < smallci->ci_size)) {
413 				smallci = ci;
414 			}
415 		}
416 
417 		/*
418 		 * Nobody left, all done
419 		 */
420 		if (smallci == NULL) {
421 			ii->ii_ndisk = 0;
422 			g_free(ii->ii_index);
423 			ii->ii_index = NULL;
424 			break;
425 		}
426 
427 		/*
428 		 * Record starting logical block using an sc_ileave blocksize.
429 		 */
430 		ii->ii_startblk = bn / cs->sc_ileave;
431 
432 		/*
433 		 * Record starting component block using an sc_ileave
434 		 * blocksize.  This value is relative to the beginning of
435 		 * a component disk.
436 		 */
437 		ii->ii_startoff = lbn;
438 
439 		/*
440 		 * Determine how many disks take part in this interleave
441 		 * and record their indices.
442 		 */
443 		ix = 0;
444 		for (ci = cs->sc_cinfo;
445 		    ci < &cs->sc_cinfo[cs->sc_ndisks]; ci++) {
446 			if (ci->ci_size >= smallci->ci_size) {
447 				ii->ii_index[ix++] = ci - cs->sc_cinfo;
448 			}
449 		}
450 		ii->ii_ndisk = ix;
451 		bn += ix * (smallci->ci_size - size);
452 		lbn = smallci->ci_size / cs->sc_ileave;
453 		size = smallci->ci_size;
454 	}
455 }
456 
457 static void
458 g_ccd_start(struct bio *bp)
459 {
460 	long bcount, rcount;
461 	struct bio *cbp[2];
462 	caddr_t addr;
463 	daddr_t bn;
464 	int err;
465 	struct ccd_s *cs;
466 
467 	cs = bp->bio_to->geom->softc;
468 
469 	/*
470 	 * Block all GETATTR requests, we wouldn't know which of our
471 	 * subdevices we should ship it off to.
472 	 * XXX: this may not be the right policy.
473 	 */
474 	if(bp->bio_cmd == BIO_GETATTR) {
475 		g_io_deliver(bp, EINVAL);
476 		return;
477 	}
478 
479 	/*
480 	 * Translate the partition-relative block number to an absolute.
481 	 */
482 	bn = bp->bio_offset / cs->sc_secsize;
483 
484 	/*
485 	 * Allocate component buffers and fire off the requests
486 	 */
487 	addr = bp->bio_data;
488 	for (bcount = bp->bio_length; bcount > 0; bcount -= rcount) {
489 		err = ccdbuffer(cbp, cs, bp, bn, addr, bcount);
490 		if (err) {
491 			bp->bio_completed += bcount;
492 			if (bp->bio_error == 0)
493 				bp->bio_error = err;
494 			if (bp->bio_completed == bp->bio_length)
495 				g_io_deliver(bp, bp->bio_error);
496 			return;
497 		}
498 		rcount = cbp[0]->bio_length;
499 
500 		if (cs->sc_flags & CCDF_MIRROR) {
501 			/*
502 			 * Mirroring.  Writes go to both disks, reads are
503 			 * taken from whichever disk seems most appropriate.
504 			 *
505 			 * We attempt to localize reads to the disk whos arm
506 			 * is nearest the read request.  We ignore seeks due
507 			 * to writes when making this determination and we
508 			 * also try to avoid hogging.
509 			 */
510 			if (cbp[0]->bio_cmd != BIO_READ) {
511 				g_io_request(cbp[0], cbp[0]->bio_from);
512 				g_io_request(cbp[1], cbp[1]->bio_from);
513 			} else {
514 				int pick = cs->sc_pick;
515 				daddr_t range = cs->sc_size / 16;
516 
517 				if (bn < cs->sc_blk[pick] - range ||
518 				    bn > cs->sc_blk[pick] + range
519 				) {
520 					cs->sc_pick = pick = 1 - pick;
521 				}
522 				cs->sc_blk[pick] = bn + btodb(rcount);
523 				g_io_request(cbp[pick], cbp[pick]->bio_from);
524 			}
525 		} else {
526 			/*
527 			 * Not mirroring
528 			 */
529 			g_io_request(cbp[0], cbp[0]->bio_from);
530 		}
531 		bn += btodb(rcount);
532 		addr += rcount;
533 	}
534 }
535 
536 /*
537  * Build a component buffer header.
538  */
539 static int
540 ccdbuffer(struct bio **cb, struct ccd_s *cs, struct bio *bp, daddr_t bn, caddr_t addr, long bcount)
541 {
542 	struct ccdcinfo *ci, *ci2 = NULL;
543 	struct bio *cbp;
544 	daddr_t cbn, cboff;
545 	off_t cbc;
546 
547 	/*
548 	 * Determine which component bn falls in.
549 	 */
550 	cbn = bn;
551 	cboff = 0;
552 
553 	if (cs->sc_ileave == 0) {
554 		/*
555 		 * Serially concatenated and neither a mirror nor a parity
556 		 * config.  This is a special case.
557 		 */
558 		daddr_t sblk;
559 
560 		sblk = 0;
561 		for (ci = cs->sc_cinfo; cbn >= sblk + ci->ci_size; ci++)
562 			sblk += ci->ci_size;
563 		cbn -= sblk;
564 	} else {
565 		struct ccdiinfo *ii;
566 		int ccdisk, off;
567 
568 		/*
569 		 * Calculate cbn, the logical superblock (sc_ileave chunks),
570 		 * and cboff, a normal block offset (DEV_BSIZE chunks) relative
571 		 * to cbn.
572 		 */
573 		cboff = cbn % cs->sc_ileave;	/* DEV_BSIZE gran */
574 		cbn = cbn / cs->sc_ileave;	/* DEV_BSIZE * ileave gran */
575 
576 		/*
577 		 * Figure out which interleave table to use.
578 		 */
579 		for (ii = cs->sc_itable; ii->ii_ndisk; ii++) {
580 			if (ii->ii_startblk > cbn)
581 				break;
582 		}
583 		ii--;
584 
585 		/*
586 		 * off is the logical superblock relative to the beginning
587 		 * of this interleave block.
588 		 */
589 		off = cbn - ii->ii_startblk;
590 
591 		/*
592 		 * We must calculate which disk component to use (ccdisk),
593 		 * and recalculate cbn to be the superblock relative to
594 		 * the beginning of the component.  This is typically done by
595 		 * adding 'off' and ii->ii_startoff together.  However, 'off'
596 		 * must typically be divided by the number of components in
597 		 * this interleave array to be properly convert it from a
598 		 * CCD-relative logical superblock number to a
599 		 * component-relative superblock number.
600 		 */
601 		if (ii->ii_ndisk == 1) {
602 			/*
603 			 * When we have just one disk, it can't be a mirror
604 			 * or a parity config.
605 			 */
606 			ccdisk = ii->ii_index[0];
607 			cbn = ii->ii_startoff + off;
608 		} else {
609 			if (cs->sc_flags & CCDF_MIRROR) {
610 				/*
611 				 * We have forced a uniform mapping, resulting
612 				 * in a single interleave array.  We double
613 				 * up on the first half of the available
614 				 * components and our mirror is in the second
615 				 * half.  This only works with a single
616 				 * interleave array because doubling up
617 				 * doubles the number of sectors, so there
618 				 * cannot be another interleave array because
619 				 * the next interleave array's calculations
620 				 * would be off.
621 				 */
622 				int ndisk2 = ii->ii_ndisk / 2;
623 				ccdisk = ii->ii_index[off % ndisk2];
624 				cbn = ii->ii_startoff + off / ndisk2;
625 				ci2 = &cs->sc_cinfo[ccdisk + ndisk2];
626 			} else {
627 				ccdisk = ii->ii_index[off % ii->ii_ndisk];
628 				cbn = ii->ii_startoff + off / ii->ii_ndisk;
629 			}
630 		}
631 
632 		ci = &cs->sc_cinfo[ccdisk];
633 
634 		/*
635 		 * Convert cbn from a superblock to a normal block so it
636 		 * can be used to calculate (along with cboff) the normal
637 		 * block index into this particular disk.
638 		 */
639 		cbn *= cs->sc_ileave;
640 	}
641 
642 	/*
643 	 * Fill in the component buf structure.
644 	 */
645 	cbp = g_clone_bio(bp);
646 	if (cbp == NULL)
647 		return (ENOMEM);
648 	cbp->bio_done = g_std_done;
649 	cbp->bio_offset = dbtob(cbn + cboff + cs->sc_offset);
650 	cbp->bio_data = addr;
651 	if (cs->sc_ileave == 0)
652               cbc = dbtob((off_t)(ci->ci_size - cbn));
653 	else
654               cbc = dbtob((off_t)(cs->sc_ileave - cboff));
655 	cbp->bio_length = (cbc < bcount) ? cbc : bcount;
656 
657 	cbp->bio_from = ci->ci_consumer;
658 	cb[0] = cbp;
659 
660 	if (cs->sc_flags & CCDF_MIRROR) {
661 		cbp = g_clone_bio(bp);
662 		if (cbp == NULL)
663 			return (ENOMEM);
664 		cbp->bio_done = cb[0]->bio_done = ccdiodone;
665 		cbp->bio_offset = cb[0]->bio_offset;
666 		cbp->bio_data = cb[0]->bio_data;
667 		cbp->bio_length = cb[0]->bio_length;
668 		cbp->bio_from = ci2->ci_consumer;
669 		cbp->bio_caller1 = cb[0];
670 		cb[0]->bio_caller1 = cbp;
671 		cb[1] = cbp;
672 	}
673 	return (0);
674 }
675 
676 /*
677  * Called only for mirrored operations.
678  */
679 static void
680 ccdiodone(struct bio *cbp)
681 {
682 	struct bio *mbp, *pbp;
683 
684 	mbp = cbp->bio_caller1;
685 	pbp = cbp->bio_parent;
686 
687 	if (pbp->bio_cmd == BIO_READ) {
688 		if (cbp->bio_error == 0) {
689 			/* We will not be needing the partner bio */
690 			if (mbp != NULL) {
691 				pbp->bio_inbed++;
692 				g_destroy_bio(mbp);
693 			}
694 			g_std_done(cbp);
695 			return;
696 		}
697 		if (mbp != NULL) {
698 			/* Try partner the bio instead */
699 			mbp->bio_caller1 = NULL;
700 			pbp->bio_inbed++;
701 			g_destroy_bio(cbp);
702 			g_io_request(mbp, mbp->bio_from);
703 			/*
704 			 * XXX: If this comes back OK, we should actually
705 			 * try to write the good data on the failed mirror
706 			 */
707 			return;
708 		}
709 		g_std_done(cbp);
710 		return;
711 	}
712 	if (mbp != NULL) {
713 		mbp->bio_caller1 = NULL;
714 		pbp->bio_inbed++;
715 		if (cbp->bio_error != 0 && pbp->bio_error == 0)
716 			pbp->bio_error = cbp->bio_error;
717 		g_destroy_bio(cbp);
718 		return;
719 	}
720 	g_std_done(cbp);
721 }
722 
723 static void
724 g_ccd_create(struct gctl_req *req, struct g_class *mp)
725 {
726 	int *unit, *ileave, *nprovider;
727 	struct g_geom *gp;
728 	struct g_consumer *cp;
729 	struct g_provider *pp;
730 	struct ccd_s *sc;
731 	struct sbuf *sb;
732 	char buf[20];
733 	int i, error;
734 
735 	g_topology_assert();
736 	unit = gctl_get_paraml(req, "unit", sizeof (*unit));
737 	if (unit == NULL) {
738 		gctl_error(req, "unit parameter not given");
739 		return;
740 	}
741 	ileave = gctl_get_paraml(req, "ileave", sizeof (*ileave));
742 	if (ileave == NULL) {
743 		gctl_error(req, "ileave parameter not given");
744 		return;
745 	}
746 	nprovider = gctl_get_paraml(req, "nprovider", sizeof (*nprovider));
747 	if (nprovider == NULL) {
748 		gctl_error(req, "nprovider parameter not given");
749 		return;
750 	}
751 
752 	/* Check for duplicate unit */
753 	LIST_FOREACH(gp, &mp->geom, geom) {
754 		sc = gp->softc;
755 		if (sc != NULL && sc->sc_unit == *unit) {
756 			gctl_error(req, "Unit %d already configured", *unit);
757 			return;
758 		}
759 	}
760 
761 	if (*nprovider <= 0) {
762 		gctl_error(req, "Bogus nprovider argument (= %d)", *nprovider);
763 		return;
764 	}
765 
766 	/* Check all providers are valid */
767 	for (i = 0; i < *nprovider; i++) {
768 		snprintf(buf, sizeof(buf), "provider%d", i);
769 		pp = gctl_get_provider(req, buf);
770 		if (pp == NULL)
771 			return;
772 	}
773 
774 	gp = g_new_geomf(mp, "ccd%d", *unit);
775 	sc = g_malloc(sizeof *sc, M_WAITOK | M_ZERO);
776 	gp->softc = sc;
777 	sc->sc_ndisks = *nprovider;
778 
779 	/* Allocate space for the component info. */
780 	sc->sc_cinfo = g_malloc(sc->sc_ndisks * sizeof(struct ccdcinfo),
781 	    M_WAITOK | M_ZERO);
782 
783 	/* Create consumers and attach to all providers */
784 	for (i = 0; i < *nprovider; i++) {
785 		snprintf(buf, sizeof(buf), "provider%d", i);
786 		pp = gctl_get_provider(req, buf);
787 		cp = g_new_consumer(gp);
788 		error = g_attach(cp, pp);
789 		KASSERT(error == 0, ("attach to %s failed", pp->name));
790 		sc->sc_cinfo[i].ci_consumer = cp;
791 		sc->sc_cinfo[i].ci_provider = pp;
792 	}
793 
794 	sc->sc_unit = *unit;
795 	sc->sc_ileave = *ileave;
796 
797 	if (gctl_get_param(req, "no_offset", NULL))
798 		sc->sc_flags |= CCDF_NO_OFFSET;
799 	if (gctl_get_param(req, "linux", NULL))
800 		sc->sc_flags |= CCDF_LINUX;
801 
802 	if (gctl_get_param(req, "uniform", NULL))
803 		sc->sc_flags |= CCDF_UNIFORM;
804 	if (gctl_get_param(req, "mirror", NULL))
805 		sc->sc_flags |= CCDF_MIRROR;
806 
807 	if (sc->sc_ileave == 0 && (sc->sc_flags & CCDF_MIRROR)) {
808 		printf("%s: disabling mirror, interleave is 0\n", gp->name);
809 		sc->sc_flags &= ~(CCDF_MIRROR);
810 	}
811 
812 	if ((sc->sc_flags & CCDF_MIRROR) && !(sc->sc_flags & CCDF_UNIFORM)) {
813 		printf("%s: mirror/parity forces uniform flag\n", gp->name);
814 		sc->sc_flags |= CCDF_UNIFORM;
815 	}
816 
817 	error = ccdinit(req, sc);
818 	if (error != 0) {
819 		g_ccd_freesc(sc);
820 		gp->softc = NULL;
821 		g_wither_geom(gp, ENXIO);
822 		return;
823 	}
824 
825 	pp = g_new_providerf(gp, "%s", gp->name);
826 	pp->mediasize = sc->sc_size * (off_t)sc->sc_secsize;
827 	pp->sectorsize = sc->sc_secsize;
828 	g_error_provider(pp, 0);
829 
830 	sb = sbuf_new_auto();
831 	sbuf_printf(sb, "ccd%d: %d components ", sc->sc_unit, *nprovider);
832 	for (i = 0; i < *nprovider; i++) {
833 		sbuf_printf(sb, "%s%s",
834 		    i == 0 ? "(" : ", ",
835 		    sc->sc_cinfo[i].ci_provider->name);
836 	}
837 	sbuf_printf(sb, "), %jd blocks ", (off_t)pp->mediasize / DEV_BSIZE);
838 	if (sc->sc_ileave != 0)
839 		sbuf_printf(sb, "interleaved at %d blocks\n",
840 			sc->sc_ileave);
841 	else
842 		sbuf_printf(sb, "concatenated\n");
843 	sbuf_finish(sb);
844 	gctl_set_param_err(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
845 	sbuf_delete(sb);
846 }
847 
848 static int
849 g_ccd_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
850 {
851 	struct g_provider *pp;
852 	struct ccd_s *sc;
853 
854 	g_topology_assert();
855 	sc = gp->softc;
856 	pp = LIST_FIRST(&gp->provider);
857 	if (sc == NULL || pp == NULL)
858 		return (EBUSY);
859 	if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0) {
860 		gctl_error(req, "%s is open(r%dw%de%d)", gp->name,
861 		    pp->acr, pp->acw, pp->ace);
862 		return (EBUSY);
863 	}
864 	g_ccd_freesc(sc);
865 	gp->softc = NULL;
866 	g_wither_geom(gp, ENXIO);
867 	return (0);
868 }
869 
870 static void
871 g_ccd_list(struct gctl_req *req, struct g_class *mp)
872 {
873 	struct sbuf *sb;
874 	struct ccd_s *cs;
875 	struct g_geom *gp;
876 	int i, unit, *up;
877 
878 	up = gctl_get_paraml(req, "unit", sizeof (*up));
879 	if (up == NULL) {
880 		gctl_error(req, "unit parameter not given");
881 		return;
882 	}
883 	unit = *up;
884 	sb = sbuf_new_auto();
885 	LIST_FOREACH(gp, &mp->geom, geom) {
886 		cs = gp->softc;
887 		if (cs == NULL || (unit >= 0 && unit != cs->sc_unit))
888 			continue;
889 		sbuf_printf(sb, "ccd%d\t\t%d\t%d\t",
890 		    cs->sc_unit, cs->sc_ileave, cs->sc_flags & CCDF_USERMASK);
891 
892 		for (i = 0; i < cs->sc_ndisks; ++i) {
893 			sbuf_printf(sb, "%s/dev/%s", i == 0 ? "" : " ",
894 			    cs->sc_cinfo[i].ci_provider->name);
895 		}
896 		sbuf_printf(sb, "\n");
897 	}
898 	sbuf_finish(sb);
899 	gctl_set_param_err(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
900 	sbuf_delete(sb);
901 }
902 
903 static void
904 g_ccd_config(struct gctl_req *req, struct g_class *mp, char const *verb)
905 {
906 	struct g_geom *gp;
907 
908 	g_topology_assert();
909 	if (!strcmp(verb, "create geom")) {
910 		g_ccd_create(req, mp);
911 	} else if (!strcmp(verb, "destroy geom")) {
912 		gp = gctl_get_geom(req, mp, "geom");
913 		if (gp != NULL)
914 			g_ccd_destroy_geom(req, mp, gp);
915 	} else if (!strcmp(verb, "list")) {
916 		g_ccd_list(req, mp);
917 	} else {
918 		gctl_error(req, "unknown verb");
919 	}
920 }
921 
922 static struct g_class g_ccd_class = {
923 	.name = "CCD",
924 	.version = G_VERSION,
925 	.ctlreq = g_ccd_config,
926 	.destroy_geom = g_ccd_destroy_geom,
927 	.start = g_ccd_start,
928 	.orphan = g_ccd_orphan,
929 	.access = g_ccd_access,
930 };
931 
932 DECLARE_GEOM_CLASS(g_ccd_class, g_ccd);
933 MODULE_VERSION(geom_ccd, 0);
934