xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.159 2003/05/10 23:12:46 thorpej Exp $	*/
2 /*-
3  * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Greg Oster; Jason R. Thorpe.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *        This product includes software developed by the NetBSD
20  *        Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 1988 University of Utah.
40  * Copyright (c) 1990, 1993
41  *      The Regents of the University of California.  All rights reserved.
42  *
43  * This code is derived from software contributed to Berkeley by
44  * the Systems Programming Group of the University of Utah Computer
45  * Science Department.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. All advertising materials mentioning features or use of this software
56  *    must display the following acknowledgement:
57  *      This product includes software developed by the University of
58  *      California, Berkeley and its contributors.
59  * 4. Neither the name of the University nor the names of its contributors
60  *    may be used to endorse or promote products derived from this software
61  *    without specific prior written permission.
62  *
63  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
65  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
66  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
67  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
68  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
69  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
70  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
71  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
72  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
73  * SUCH DAMAGE.
74  *
75  * from: Utah $Hdr: cd.c 1.6 90/11/28$
76  *
77  *      @(#)cd.c        8.2 (Berkeley) 11/16/93
78  */
79 
80 /*
81  * Copyright (c) 1995 Carnegie-Mellon University.
82  * All rights reserved.
83  *
84  * Authors: Mark Holland, Jim Zelenka
85  *
86  * Permission to use, copy, modify and distribute this software and
87  * its documentation is hereby granted, provided that both the copyright
88  * notice and this permission notice appear in all copies of the
89  * software, derivative works or modified versions, and any portions
90  * thereof, and that both notices appear in supporting documentation.
91  *
92  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
93  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
94  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
95  *
96  * Carnegie Mellon requests users of this software to return to
97  *
98  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
99  *  School of Computer Science
100  *  Carnegie Mellon University
101  *  Pittsburgh PA 15213-3890
102  *
103  * any improvements or extensions that they make and grant Carnegie the
104  * rights to redistribute these changes.
105  */
106 
107 /***********************************************************
108  *
109  * rf_kintf.c -- the kernel interface routines for RAIDframe
110  *
111  ***********************************************************/
112 
113 #include <sys/cdefs.h>
114 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.159 2003/05/10 23:12:46 thorpej Exp $");
115 
116 #include <sys/param.h>
117 #include <sys/errno.h>
118 #include <sys/pool.h>
119 #include <sys/proc.h>
120 #include <sys/queue.h>
121 #include <sys/disk.h>
122 #include <sys/device.h>
123 #include <sys/stat.h>
124 #include <sys/ioctl.h>
125 #include <sys/fcntl.h>
126 #include <sys/systm.h>
127 #include <sys/namei.h>
128 #include <sys/vnode.h>
129 #include <sys/disklabel.h>
130 #include <sys/conf.h>
131 #include <sys/lock.h>
132 #include <sys/buf.h>
133 #include <sys/user.h>
134 #include <sys/reboot.h>
135 
136 #include <dev/raidframe/raidframevar.h>
137 #include <dev/raidframe/raidframeio.h>
138 #include "raid.h"
139 #include "opt_raid_autoconfig.h"
140 #include "rf_raid.h"
141 #include "rf_copyback.h"
142 #include "rf_dag.h"
143 #include "rf_dagflags.h"
144 #include "rf_desc.h"
145 #include "rf_diskqueue.h"
146 #include "rf_etimer.h"
147 #include "rf_general.h"
148 #include "rf_kintf.h"
149 #include "rf_options.h"
150 #include "rf_driver.h"
151 #include "rf_parityscan.h"
152 #include "rf_threadstuff.h"
153 
154 #ifdef DEBUG
155 int     rf_kdebug_level = 0;
156 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
157 #else				/* DEBUG */
158 #define db1_printf(a) { }
159 #endif				/* DEBUG */
160 
161 static RF_Raid_t **raidPtrs;	/* global raid device descriptors */
162 
163 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
164 
165 static RF_SparetWait_t *rf_sparet_wait_queue;	/* requests to install a
166 						 * spare table */
167 static RF_SparetWait_t *rf_sparet_resp_queue;	/* responses from
168 						 * installation process */
169 
170 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures");
171 
172 /* prototypes */
173 static void KernelWakeupFunc(struct buf * bp);
174 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag,
175 		   dev_t dev, RF_SectorNum_t startSect,
176 		   RF_SectorCount_t numSect, caddr_t buf,
177 		   void (*cbFunc) (struct buf *), void *cbArg,
178 		   int logBytesPerSector, struct proc * b_proc);
179 static void raidinit(RF_Raid_t *);
180 
181 void raidattach(int);
182 
183 dev_type_open(raidopen);
184 dev_type_close(raidclose);
185 dev_type_read(raidread);
186 dev_type_write(raidwrite);
187 dev_type_ioctl(raidioctl);
188 dev_type_strategy(raidstrategy);
189 dev_type_dump(raiddump);
190 dev_type_size(raidsize);
191 
192 const struct bdevsw raid_bdevsw = {
193 	raidopen, raidclose, raidstrategy, raidioctl,
194 	raiddump, raidsize, D_DISK
195 };
196 
197 const struct cdevsw raid_cdevsw = {
198 	raidopen, raidclose, raidread, raidwrite, raidioctl,
199 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
200 };
201 
202 /*
203  * Pilfered from ccd.c
204  */
205 
206 struct raidbuf {
207 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
208 	struct buf *rf_obp;	/* ptr. to original I/O buf */
209 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
210 };
211 
212 /* component buffer pool */
213 struct pool raidframe_cbufpool;
214 
215 /* XXX Not sure if the following should be replacing the raidPtrs above,
216    or if it should be used in conjunction with that...
217 */
218 
219 struct raid_softc {
220 	int     sc_flags;	/* flags */
221 	int     sc_cflags;	/* configuration flags */
222 	size_t  sc_size;        /* size of the raid device */
223 	char    sc_xname[20];	/* XXX external name */
224 	struct disk sc_dkdev;	/* generic disk device info */
225 	struct bufq_state buf_queue;	/* used for the device queue */
226 };
227 /* sc_flags */
228 #define RAIDF_INITED	0x01	/* unit has been initialized */
229 #define RAIDF_WLABEL	0x02	/* label area is writable */
230 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
231 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
232 #define RAIDF_LOCKED	0x80	/* unit is locked */
233 
234 #define	raidunit(x)	DISKUNIT(x)
235 int numraid = 0;
236 
237 /*
238  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
239  * Be aware that large numbers can allow the driver to consume a lot of
240  * kernel memory, especially on writes, and in degraded mode reads.
241  *
242  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
243  * a single 64K write will typically require 64K for the old data,
244  * 64K for the old parity, and 64K for the new parity, for a total
245  * of 192K (if the parity buffer is not re-used immediately).
246  * Even it if is used immediately, that's still 128K, which when multiplied
247  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
248  *
249  * Now in degraded mode, for example, a 64K read on the above setup may
250  * require data reconstruction, which will require *all* of the 4 remaining
251  * disks to participate -- 4 * 32K/disk == 128K again.
252  */
253 
254 #ifndef RAIDOUTSTANDING
255 #define RAIDOUTSTANDING   6
256 #endif
257 
258 #define RAIDLABELDEV(dev)	\
259 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
260 
261 /* declared here, and made public, for the benefit of KVM stuff.. */
262 struct raid_softc *raid_softc;
263 
264 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
265 				     struct disklabel *);
266 static void raidgetdisklabel(dev_t);
267 static void raidmakedisklabel(struct raid_softc *);
268 
269 static int raidlock(struct raid_softc *);
270 static void raidunlock(struct raid_softc *);
271 
272 static void rf_markalldirty(RF_Raid_t *);
273 
274 struct device *raidrootdev;
275 
276 void rf_ReconThread(struct rf_recon_req *);
277 /* XXX what I want is: */
278 /*void rf_ReconThread(RF_Raid_t *raidPtr);  */
279 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
280 void rf_CopybackThread(RF_Raid_t *raidPtr);
281 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
282 int rf_autoconfig(struct device *self);
283 void rf_buildroothack(RF_ConfigSet_t *);
284 
285 RF_AutoConfig_t *rf_find_raid_components(void);
286 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
287 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
288 static int rf_reasonable_label(RF_ComponentLabel_t *);
289 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
290 int rf_set_autoconfig(RF_Raid_t *, int);
291 int rf_set_rootpartition(RF_Raid_t *, int);
292 void rf_release_all_vps(RF_ConfigSet_t *);
293 void rf_cleanup_config_set(RF_ConfigSet_t *);
294 int rf_have_enough_components(RF_ConfigSet_t *);
295 int rf_auto_config_set(RF_ConfigSet_t *, int *);
296 
297 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
298 				  allow autoconfig to take place.
299 			          Note that this is overridden by having
300 			          RAID_AUTOCONFIG as an option in the
301 			          kernel config file.  */
302 
303 void
304 raidattach(num)
305 	int     num;
306 {
307 	int raidID;
308 	int i, rc;
309 
310 #ifdef DEBUG
311 	printf("raidattach: Asked for %d units\n", num);
312 #endif
313 
314 	if (num <= 0) {
315 #ifdef DIAGNOSTIC
316 		panic("raidattach: count <= 0");
317 #endif
318 		return;
319 	}
320 	/* This is where all the initialization stuff gets done. */
321 
322 	numraid = num;
323 
324 	/* Make some space for requested number of units... */
325 
326 	RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **));
327 	if (raidPtrs == NULL) {
328 		panic("raidPtrs is NULL!!");
329 	}
330 
331 	/* Initialize the component buffer pool. */
332 	pool_init(&raidframe_cbufpool, sizeof(struct raidbuf), 0,
333 	    0, 0, "raidpl", NULL);
334 
335 	rc = rf_mutex_init(&rf_sparet_wait_mutex);
336 	if (rc) {
337 		RF_PANIC();
338 	}
339 
340 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
341 
342 	for (i = 0; i < num; i++)
343 		raidPtrs[i] = NULL;
344 	rc = rf_BootRaidframe();
345 	if (rc == 0)
346 		printf("Kernelized RAIDframe activated\n");
347 	else
348 		panic("Serious error booting RAID!!");
349 
350 	/* put together some datastructures like the CCD device does.. This
351 	 * lets us lock the device and what-not when it gets opened. */
352 
353 	raid_softc = (struct raid_softc *)
354 		malloc(num * sizeof(struct raid_softc),
355 		       M_RAIDFRAME, M_NOWAIT);
356 	if (raid_softc == NULL) {
357 		printf("WARNING: no memory for RAIDframe driver\n");
358 		return;
359 	}
360 
361 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
362 
363 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
364 					      M_RAIDFRAME, M_NOWAIT);
365 	if (raidrootdev == NULL) {
366 		panic("No memory for RAIDframe driver!!?!?!");
367 	}
368 
369 	for (raidID = 0; raidID < num; raidID++) {
370 		bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_FCFS);
371 
372 		raidrootdev[raidID].dv_class  = DV_DISK;
373 		raidrootdev[raidID].dv_cfdata = NULL;
374 		raidrootdev[raidID].dv_unit   = raidID;
375 		raidrootdev[raidID].dv_parent = NULL;
376 		raidrootdev[raidID].dv_flags  = 0;
377 		sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID);
378 
379 		RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t),
380 			  (RF_Raid_t *));
381 		if (raidPtrs[raidID] == NULL) {
382 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
383 			numraid = raidID;
384 			return;
385 		}
386 	}
387 
388 #ifdef RAID_AUTOCONFIG
389 	raidautoconfig = 1;
390 #endif
391 
392 	/*
393 	 * Register a finalizer which will be used to auto-config RAID
394 	 * sets once all real hardware devices have been found.
395 	 */
396 	if (config_finalize_register(NULL, rf_autoconfig) != 0)
397 		printf("WARNING: unable to register RAIDframe finalizer\n");
398 }
399 
400 int
401 rf_autoconfig(struct device *self)
402 {
403 	RF_AutoConfig_t *ac_list;
404 	RF_ConfigSet_t *config_sets;
405 
406 	if (raidautoconfig == 0)
407 		return (0);
408 
409 	/* XXX This code can only be run once. */
410 	raidautoconfig = 0;
411 
412 	/* 1. locate all RAID components on the system */
413 #ifdef DEBUG
414 	printf("Searching for RAID components...\n");
415 #endif
416 	ac_list = rf_find_raid_components();
417 
418 	/* 2. Sort them into their respective sets. */
419 	config_sets = rf_create_auto_sets(ac_list);
420 
421 	/*
422 	 * 3. Evaluate each set andconfigure the valid ones.
423 	 * This gets done in rf_buildroothack().
424 	 */
425 	rf_buildroothack(config_sets);
426 
427 	return (1);
428 }
429 
430 void
431 rf_buildroothack(RF_ConfigSet_t *config_sets)
432 {
433 	RF_ConfigSet_t *cset;
434 	RF_ConfigSet_t *next_cset;
435 	int retcode;
436 	int raidID;
437 	int rootID;
438 	int num_root;
439 
440 	rootID = 0;
441 	num_root = 0;
442 	cset = config_sets;
443 	while(cset != NULL ) {
444 		next_cset = cset->next;
445 		if (rf_have_enough_components(cset) &&
446 		    cset->ac->clabel->autoconfigure==1) {
447 			retcode = rf_auto_config_set(cset,&raidID);
448 			if (!retcode) {
449 				if (cset->rootable) {
450 					rootID = raidID;
451 					num_root++;
452 				}
453 			} else {
454 				/* The autoconfig didn't work :( */
455 #if DEBUG
456 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
457 #endif
458 				rf_release_all_vps(cset);
459 			}
460 		} else {
461 			/* we're not autoconfiguring this set...
462 			   release the associated resources */
463 			rf_release_all_vps(cset);
464 		}
465 		/* cleanup */
466 		rf_cleanup_config_set(cset);
467 		cset = next_cset;
468 	}
469 
470 	/* we found something bootable... */
471 
472 	if (num_root == 1) {
473 		booted_device = &raidrootdev[rootID];
474 	} else if (num_root > 1) {
475 		/* we can't guess.. require the user to answer... */
476 		boothowto |= RB_ASKNAME;
477 	}
478 }
479 
480 
481 int
482 raidsize(dev)
483 	dev_t   dev;
484 {
485 	struct raid_softc *rs;
486 	struct disklabel *lp;
487 	int     part, unit, omask, size;
488 
489 	unit = raidunit(dev);
490 	if (unit >= numraid)
491 		return (-1);
492 	rs = &raid_softc[unit];
493 
494 	if ((rs->sc_flags & RAIDF_INITED) == 0)
495 		return (-1);
496 
497 	part = DISKPART(dev);
498 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
499 	lp = rs->sc_dkdev.dk_label;
500 
501 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
502 		return (-1);
503 
504 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
505 		size = -1;
506 	else
507 		size = lp->d_partitions[part].p_size *
508 		    (lp->d_secsize / DEV_BSIZE);
509 
510 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
511 		return (-1);
512 
513 	return (size);
514 
515 }
516 
517 int
518 raiddump(dev, blkno, va, size)
519 	dev_t   dev;
520 	daddr_t blkno;
521 	caddr_t va;
522 	size_t  size;
523 {
524 	/* Not implemented. */
525 	return ENXIO;
526 }
527 /* ARGSUSED */
528 int
529 raidopen(dev, flags, fmt, p)
530 	dev_t   dev;
531 	int     flags, fmt;
532 	struct proc *p;
533 {
534 	int     unit = raidunit(dev);
535 	struct raid_softc *rs;
536 	struct disklabel *lp;
537 	int     part, pmask;
538 	int     error = 0;
539 
540 	if (unit >= numraid)
541 		return (ENXIO);
542 	rs = &raid_softc[unit];
543 
544 	if ((error = raidlock(rs)) != 0)
545 		return (error);
546 	lp = rs->sc_dkdev.dk_label;
547 
548 	part = DISKPART(dev);
549 	pmask = (1 << part);
550 
551 	if ((rs->sc_flags & RAIDF_INITED) &&
552 	    (rs->sc_dkdev.dk_openmask == 0))
553 		raidgetdisklabel(dev);
554 
555 	/* make sure that this partition exists */
556 
557 	if (part != RAW_PART) {
558 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
559 		    ((part >= lp->d_npartitions) ||
560 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
561 			error = ENXIO;
562 			raidunlock(rs);
563 			return (error);
564 		}
565 	}
566 	/* Prevent this unit from being unconfigured while open. */
567 	switch (fmt) {
568 	case S_IFCHR:
569 		rs->sc_dkdev.dk_copenmask |= pmask;
570 		break;
571 
572 	case S_IFBLK:
573 		rs->sc_dkdev.dk_bopenmask |= pmask;
574 		break;
575 	}
576 
577 	if ((rs->sc_dkdev.dk_openmask == 0) &&
578 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
579 		/* First one... mark things as dirty... Note that we *MUST*
580 		 have done a configure before this.  I DO NOT WANT TO BE
581 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
582 		 THAT THEY BELONG TOGETHER!!!!! */
583 		/* XXX should check to see if we're only open for reading
584 		   here... If so, we needn't do this, but then need some
585 		   other way of keeping track of what's happened.. */
586 
587 		rf_markalldirty( raidPtrs[unit] );
588 	}
589 
590 
591 	rs->sc_dkdev.dk_openmask =
592 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
593 
594 	raidunlock(rs);
595 
596 	return (error);
597 
598 
599 }
600 /* ARGSUSED */
601 int
602 raidclose(dev, flags, fmt, p)
603 	dev_t   dev;
604 	int     flags, fmt;
605 	struct proc *p;
606 {
607 	int     unit = raidunit(dev);
608 	struct raid_softc *rs;
609 	int     error = 0;
610 	int     part;
611 
612 	if (unit >= numraid)
613 		return (ENXIO);
614 	rs = &raid_softc[unit];
615 
616 	if ((error = raidlock(rs)) != 0)
617 		return (error);
618 
619 	part = DISKPART(dev);
620 
621 	/* ...that much closer to allowing unconfiguration... */
622 	switch (fmt) {
623 	case S_IFCHR:
624 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
625 		break;
626 
627 	case S_IFBLK:
628 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
629 		break;
630 	}
631 	rs->sc_dkdev.dk_openmask =
632 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
633 
634 	if ((rs->sc_dkdev.dk_openmask == 0) &&
635 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
636 		/* Last one... device is not unconfigured yet.
637 		   Device shutdown has taken care of setting the
638 		   clean bits if RAIDF_INITED is not set
639 		   mark things as clean... */
640 
641 		rf_update_component_labels(raidPtrs[unit],
642 						 RF_FINAL_COMPONENT_UPDATE);
643 		if (doing_shutdown) {
644 			/* last one, and we're going down, so
645 			   lights out for this RAID set too. */
646 			error = rf_Shutdown(raidPtrs[unit]);
647 
648 			/* It's no longer initialized... */
649 			rs->sc_flags &= ~RAIDF_INITED;
650 
651 			/* Detach the disk. */
652 			disk_detach(&rs->sc_dkdev);
653 		}
654 	}
655 
656 	raidunlock(rs);
657 	return (0);
658 
659 }
660 
661 void
662 raidstrategy(bp)
663 	struct buf *bp;
664 {
665 	int s;
666 
667 	unsigned int raidID = raidunit(bp->b_dev);
668 	RF_Raid_t *raidPtr;
669 	struct raid_softc *rs = &raid_softc[raidID];
670 	struct disklabel *lp;
671 	int     wlabel;
672 
673 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
674 		bp->b_error = ENXIO;
675 		bp->b_flags |= B_ERROR;
676 		bp->b_resid = bp->b_bcount;
677 		biodone(bp);
678 		return;
679 	}
680 	if (raidID >= numraid || !raidPtrs[raidID]) {
681 		bp->b_error = ENODEV;
682 		bp->b_flags |= B_ERROR;
683 		bp->b_resid = bp->b_bcount;
684 		biodone(bp);
685 		return;
686 	}
687 	raidPtr = raidPtrs[raidID];
688 	if (!raidPtr->valid) {
689 		bp->b_error = ENODEV;
690 		bp->b_flags |= B_ERROR;
691 		bp->b_resid = bp->b_bcount;
692 		biodone(bp);
693 		return;
694 	}
695 	if (bp->b_bcount == 0) {
696 		db1_printf(("b_bcount is zero..\n"));
697 		biodone(bp);
698 		return;
699 	}
700 	lp = rs->sc_dkdev.dk_label;
701 
702 	/*
703 	 * Do bounds checking and adjust transfer.  If there's an
704 	 * error, the bounds check will flag that for us.
705 	 */
706 
707 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
708 	if (DISKPART(bp->b_dev) != RAW_PART)
709 		if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) {
710 			db1_printf(("Bounds check failed!!:%d %d\n",
711 				(int) bp->b_blkno, (int) wlabel));
712 			biodone(bp);
713 			return;
714 		}
715 	s = splbio();
716 
717 	bp->b_resid = 0;
718 
719 	/* stuff it onto our queue */
720 	BUFQ_PUT(&rs->buf_queue, bp);
721 
722 	raidstart(raidPtrs[raidID]);
723 
724 	splx(s);
725 }
726 /* ARGSUSED */
727 int
728 raidread(dev, uio, flags)
729 	dev_t   dev;
730 	struct uio *uio;
731 	int     flags;
732 {
733 	int     unit = raidunit(dev);
734 	struct raid_softc *rs;
735 	int     part;
736 
737 	if (unit >= numraid)
738 		return (ENXIO);
739 	rs = &raid_softc[unit];
740 
741 	if ((rs->sc_flags & RAIDF_INITED) == 0)
742 		return (ENXIO);
743 	part = DISKPART(dev);
744 
745 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
746 
747 }
748 /* ARGSUSED */
749 int
750 raidwrite(dev, uio, flags)
751 	dev_t   dev;
752 	struct uio *uio;
753 	int     flags;
754 {
755 	int     unit = raidunit(dev);
756 	struct raid_softc *rs;
757 
758 	if (unit >= numraid)
759 		return (ENXIO);
760 	rs = &raid_softc[unit];
761 
762 	if ((rs->sc_flags & RAIDF_INITED) == 0)
763 		return (ENXIO);
764 
765 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
766 
767 }
768 
769 int
770 raidioctl(dev, cmd, data, flag, p)
771 	dev_t   dev;
772 	u_long  cmd;
773 	caddr_t data;
774 	int     flag;
775 	struct proc *p;
776 {
777 	int     unit = raidunit(dev);
778 	int     error = 0;
779 	int     part, pmask;
780 	struct raid_softc *rs;
781 	RF_Config_t *k_cfg, *u_cfg;
782 	RF_Raid_t *raidPtr;
783 	RF_RaidDisk_t *diskPtr;
784 	RF_AccTotals_t *totals;
785 	RF_DeviceConfig_t *d_cfg, **ucfgp;
786 	u_char *specific_buf;
787 	int retcode = 0;
788 	int row;
789 	int column;
790 	int raidid;
791 	struct rf_recon_req *rrcopy, *rr;
792 	RF_ComponentLabel_t *clabel;
793 	RF_ComponentLabel_t ci_label;
794 	RF_ComponentLabel_t **clabel_ptr;
795 	RF_SingleComponent_t *sparePtr,*componentPtr;
796 	RF_SingleComponent_t hot_spare;
797 	RF_SingleComponent_t component;
798 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
799 	int i, j, d;
800 #ifdef __HAVE_OLD_DISKLABEL
801 	struct disklabel newlabel;
802 #endif
803 
804 	if (unit >= numraid)
805 		return (ENXIO);
806 	rs = &raid_softc[unit];
807 	raidPtr = raidPtrs[unit];
808 
809 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
810 		(int) DISKPART(dev), (int) unit, (int) cmd));
811 
812 	/* Must be open for writes for these commands... */
813 	switch (cmd) {
814 	case DIOCSDINFO:
815 	case DIOCWDINFO:
816 #ifdef __HAVE_OLD_DISKLABEL
817 	case ODIOCWDINFO:
818 	case ODIOCSDINFO:
819 #endif
820 	case DIOCWLABEL:
821 		if ((flag & FWRITE) == 0)
822 			return (EBADF);
823 	}
824 
825 	/* Must be initialized for these... */
826 	switch (cmd) {
827 	case DIOCGDINFO:
828 	case DIOCSDINFO:
829 	case DIOCWDINFO:
830 #ifdef __HAVE_OLD_DISKLABEL
831 	case ODIOCGDINFO:
832 	case ODIOCWDINFO:
833 	case ODIOCSDINFO:
834 	case ODIOCGDEFLABEL:
835 #endif
836 	case DIOCGPART:
837 	case DIOCWLABEL:
838 	case DIOCGDEFLABEL:
839 	case RAIDFRAME_SHUTDOWN:
840 	case RAIDFRAME_REWRITEPARITY:
841 	case RAIDFRAME_GET_INFO:
842 	case RAIDFRAME_RESET_ACCTOTALS:
843 	case RAIDFRAME_GET_ACCTOTALS:
844 	case RAIDFRAME_KEEP_ACCTOTALS:
845 	case RAIDFRAME_GET_SIZE:
846 	case RAIDFRAME_FAIL_DISK:
847 	case RAIDFRAME_COPYBACK:
848 	case RAIDFRAME_CHECK_RECON_STATUS:
849 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
850 	case RAIDFRAME_GET_COMPONENT_LABEL:
851 	case RAIDFRAME_SET_COMPONENT_LABEL:
852 	case RAIDFRAME_ADD_HOT_SPARE:
853 	case RAIDFRAME_REMOVE_HOT_SPARE:
854 	case RAIDFRAME_INIT_LABELS:
855 	case RAIDFRAME_REBUILD_IN_PLACE:
856 	case RAIDFRAME_CHECK_PARITY:
857 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
858 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
859 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
860 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
861 	case RAIDFRAME_SET_AUTOCONFIG:
862 	case RAIDFRAME_SET_ROOT:
863 	case RAIDFRAME_DELETE_COMPONENT:
864 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
865 		if ((rs->sc_flags & RAIDF_INITED) == 0)
866 			return (ENXIO);
867 	}
868 
869 	switch (cmd) {
870 
871 		/* configure the system */
872 	case RAIDFRAME_CONFIGURE:
873 
874 		if (raidPtr->valid) {
875 			/* There is a valid RAID set running on this unit! */
876 			printf("raid%d: Device already configured!\n",unit);
877 			return(EINVAL);
878 		}
879 
880 		/* copy-in the configuration information */
881 		/* data points to a pointer to the configuration structure */
882 
883 		u_cfg = *((RF_Config_t **) data);
884 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
885 		if (k_cfg == NULL) {
886 			return (ENOMEM);
887 		}
888 		retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t));
889 		if (retcode) {
890 			RF_Free(k_cfg, sizeof(RF_Config_t));
891 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
892 				retcode));
893 			return (retcode);
894 		}
895 		/* allocate a buffer for the layout-specific data, and copy it
896 		 * in */
897 		if (k_cfg->layoutSpecificSize) {
898 			if (k_cfg->layoutSpecificSize > 10000) {
899 				/* sanity check */
900 				RF_Free(k_cfg, sizeof(RF_Config_t));
901 				return (EINVAL);
902 			}
903 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
904 			    (u_char *));
905 			if (specific_buf == NULL) {
906 				RF_Free(k_cfg, sizeof(RF_Config_t));
907 				return (ENOMEM);
908 			}
909 			retcode = copyin(k_cfg->layoutSpecific, specific_buf,
910 			    k_cfg->layoutSpecificSize);
911 			if (retcode) {
912 				RF_Free(k_cfg, sizeof(RF_Config_t));
913 				RF_Free(specific_buf,
914 					k_cfg->layoutSpecificSize);
915 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
916 					retcode));
917 				return (retcode);
918 			}
919 		} else
920 			specific_buf = NULL;
921 		k_cfg->layoutSpecific = specific_buf;
922 
923 		/* should do some kind of sanity check on the configuration.
924 		 * Store the sum of all the bytes in the last byte? */
925 
926 		/* configure the system */
927 
928 		/*
929 		 * Clear the entire RAID descriptor, just to make sure
930 		 *  there is no stale data left in the case of a
931 		 *  reconfiguration
932 		 */
933 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
934 		raidPtr->raidid = unit;
935 
936 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
937 
938 		if (retcode == 0) {
939 
940 			/* allow this many simultaneous IO's to
941 			   this RAID device */
942 			raidPtr->openings = RAIDOUTSTANDING;
943 
944 			raidinit(raidPtr);
945 			rf_markalldirty(raidPtr);
946 		}
947 		/* free the buffers.  No return code here. */
948 		if (k_cfg->layoutSpecificSize) {
949 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
950 		}
951 		RF_Free(k_cfg, sizeof(RF_Config_t));
952 
953 		return (retcode);
954 
955 		/* shutdown the system */
956 	case RAIDFRAME_SHUTDOWN:
957 
958 		if ((error = raidlock(rs)) != 0)
959 			return (error);
960 
961 		/*
962 		 * If somebody has a partition mounted, we shouldn't
963 		 * shutdown.
964 		 */
965 
966 		part = DISKPART(dev);
967 		pmask = (1 << part);
968 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
969 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
970 			(rs->sc_dkdev.dk_copenmask & pmask))) {
971 			raidunlock(rs);
972 			return (EBUSY);
973 		}
974 
975 		retcode = rf_Shutdown(raidPtr);
976 
977 		/* It's no longer initialized... */
978 		rs->sc_flags &= ~RAIDF_INITED;
979 
980 		/* Detach the disk. */
981 		disk_detach(&rs->sc_dkdev);
982 
983 		raidunlock(rs);
984 
985 		return (retcode);
986 	case RAIDFRAME_GET_COMPONENT_LABEL:
987 		clabel_ptr = (RF_ComponentLabel_t **) data;
988 		/* need to read the component label for the disk indicated
989 		   by row,column in clabel */
990 
991 		/* For practice, let's get it directly fromdisk, rather
992 		   than from the in-core copy */
993 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
994 			   (RF_ComponentLabel_t *));
995 		if (clabel == NULL)
996 			return (ENOMEM);
997 
998 		memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t));
999 
1000 		retcode = copyin( *clabel_ptr, clabel,
1001 				  sizeof(RF_ComponentLabel_t));
1002 
1003 		if (retcode) {
1004 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1005 			return(retcode);
1006 		}
1007 
1008 		row = clabel->row;
1009 		column = clabel->column;
1010 
1011 		if ((row < 0) || (row >= raidPtr->numRow) ||
1012 		    (column < 0) || (column >= raidPtr->numCol +
1013 				     raidPtr->numSpare)) {
1014 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1015 			return(EINVAL);
1016 		}
1017 
1018 		raidread_component_label(raidPtr->Disks[row][column].dev,
1019 				raidPtr->raid_cinfo[row][column].ci_vp,
1020 				clabel );
1021 
1022 		retcode = copyout(clabel, *clabel_ptr,
1023 				  sizeof(RF_ComponentLabel_t));
1024 		RF_Free(clabel, sizeof(RF_ComponentLabel_t));
1025 		return (retcode);
1026 
1027 	case RAIDFRAME_SET_COMPONENT_LABEL:
1028 		clabel = (RF_ComponentLabel_t *) data;
1029 
1030 		/* XXX check the label for valid stuff... */
1031 		/* Note that some things *should not* get modified --
1032 		   the user should be re-initing the labels instead of
1033 		   trying to patch things.
1034 		   */
1035 
1036 		raidid = raidPtr->raidid;
1037 		printf("raid%d: Got component label:\n", raidid);
1038 		printf("raid%d: Version: %d\n", raidid, clabel->version);
1039 		printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number);
1040 		printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter);
1041 		printf("raid%d: Row: %d\n", raidid, clabel->row);
1042 		printf("raid%d: Column: %d\n", raidid, clabel->column);
1043 		printf("raid%d: Num Rows: %d\n", raidid, clabel->num_rows);
1044 		printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns);
1045 		printf("raid%d: Clean: %d\n", raidid, clabel->clean);
1046 		printf("raid%d: Status: %d\n", raidid, clabel->status);
1047 
1048 		row = clabel->row;
1049 		column = clabel->column;
1050 
1051 		if ((row < 0) || (row >= raidPtr->numRow) ||
1052 		    (column < 0) || (column >= raidPtr->numCol)) {
1053 			return(EINVAL);
1054 		}
1055 
1056 		/* XXX this isn't allowed to do anything for now :-) */
1057 
1058 		/* XXX and before it is, we need to fill in the rest
1059 		   of the fields!?!?!?! */
1060 #if 0
1061 		raidwrite_component_label(
1062                             raidPtr->Disks[row][column].dev,
1063 			    raidPtr->raid_cinfo[row][column].ci_vp,
1064 			    clabel );
1065 #endif
1066 		return (0);
1067 
1068 	case RAIDFRAME_INIT_LABELS:
1069 		clabel = (RF_ComponentLabel_t *) data;
1070 		/*
1071 		   we only want the serial number from
1072 		   the above.  We get all the rest of the information
1073 		   from the config that was used to create this RAID
1074 		   set.
1075 		   */
1076 
1077 		raidPtr->serial_number = clabel->serial_number;
1078 
1079 		raid_init_component_label(raidPtr, &ci_label);
1080 		ci_label.serial_number = clabel->serial_number;
1081 
1082 		for(row=0;row<raidPtr->numRow;row++) {
1083 			ci_label.row = row;
1084 			for(column=0;column<raidPtr->numCol;column++) {
1085 				diskPtr = &raidPtr->Disks[row][column];
1086 				if (!RF_DEAD_DISK(diskPtr->status)) {
1087 					ci_label.partitionSize = diskPtr->partitionSize;
1088 					ci_label.column = column;
1089 					raidwrite_component_label(
1090 					  raidPtr->Disks[row][column].dev,
1091 					  raidPtr->raid_cinfo[row][column].ci_vp,
1092 					  &ci_label );
1093 				}
1094 			}
1095 		}
1096 
1097 		return (retcode);
1098 	case RAIDFRAME_SET_AUTOCONFIG:
1099 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1100 		printf("raid%d: New autoconfig value is: %d\n",
1101 		       raidPtr->raidid, d);
1102 		*(int *) data = d;
1103 		return (retcode);
1104 
1105 	case RAIDFRAME_SET_ROOT:
1106 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1107 		printf("raid%d: New rootpartition value is: %d\n",
1108 		       raidPtr->raidid, d);
1109 		*(int *) data = d;
1110 		return (retcode);
1111 
1112 		/* initialize all parity */
1113 	case RAIDFRAME_REWRITEPARITY:
1114 
1115 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1116 			/* Parity for RAID 0 is trivially correct */
1117 			raidPtr->parity_good = RF_RAID_CLEAN;
1118 			return(0);
1119 		}
1120 
1121 		if (raidPtr->parity_rewrite_in_progress == 1) {
1122 			/* Re-write is already in progress! */
1123 			return(EINVAL);
1124 		}
1125 
1126 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1127 					   rf_RewriteParityThread,
1128 					   raidPtr,"raid_parity");
1129 		return (retcode);
1130 
1131 
1132 	case RAIDFRAME_ADD_HOT_SPARE:
1133 		sparePtr = (RF_SingleComponent_t *) data;
1134 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1135 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1136 		return(retcode);
1137 
1138 	case RAIDFRAME_REMOVE_HOT_SPARE:
1139 		return(retcode);
1140 
1141 	case RAIDFRAME_DELETE_COMPONENT:
1142 		componentPtr = (RF_SingleComponent_t *)data;
1143 		memcpy( &component, componentPtr,
1144 			sizeof(RF_SingleComponent_t));
1145 		retcode = rf_delete_component(raidPtr, &component);
1146 		return(retcode);
1147 
1148 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1149 		componentPtr = (RF_SingleComponent_t *)data;
1150 		memcpy( &component, componentPtr,
1151 			sizeof(RF_SingleComponent_t));
1152 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1153 		return(retcode);
1154 
1155 	case RAIDFRAME_REBUILD_IN_PLACE:
1156 
1157 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1158 			/* Can't do this on a RAID 0!! */
1159 			return(EINVAL);
1160 		}
1161 
1162 		if (raidPtr->recon_in_progress == 1) {
1163 			/* a reconstruct is already in progress! */
1164 			return(EINVAL);
1165 		}
1166 
1167 		componentPtr = (RF_SingleComponent_t *) data;
1168 		memcpy( &component, componentPtr,
1169 			sizeof(RF_SingleComponent_t));
1170 		row = component.row;
1171 		column = component.column;
1172 
1173 		if ((row < 0) || (row >= raidPtr->numRow) ||
1174 		    (column < 0) || (column >= raidPtr->numCol)) {
1175 			return(EINVAL);
1176 		}
1177 
1178 		RF_LOCK_MUTEX(raidPtr->mutex);
1179 		if ((raidPtr->Disks[row][column].status == rf_ds_optimal) &&
1180 		    (raidPtr->numFailures > 0)) {
1181 			/* XXX 0 above shouldn't be constant!!! */
1182 			/* some component other than this has failed.
1183 			   Let's not make things worse than they already
1184 			   are... */
1185 			printf("raid%d: Unable to reconstruct to disk at:\n",
1186 			       raidPtr->raidid);
1187 			printf("raid%d:     Row: %d Col: %d   Too many failures.\n",
1188 			       raidPtr->raidid, row, column);
1189 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1190 			return (EINVAL);
1191 		}
1192 		if (raidPtr->Disks[row][column].status ==
1193 		    rf_ds_reconstructing) {
1194 			printf("raid%d: Unable to reconstruct to disk at:\n",
1195 			       raidPtr->raidid);
1196 			printf("raid%d:    Row: %d Col: %d   Reconstruction already occuring!\n", raidPtr->raidid, row, column);
1197 
1198 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1199 			return (EINVAL);
1200 		}
1201 		if (raidPtr->Disks[row][column].status == rf_ds_spared) {
1202 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1203 			return (EINVAL);
1204 		}
1205 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1206 
1207 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1208 		if (rrcopy == NULL)
1209 			return(ENOMEM);
1210 
1211 		rrcopy->raidPtr = (void *) raidPtr;
1212 		rrcopy->row = row;
1213 		rrcopy->col = column;
1214 
1215 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1216 					   rf_ReconstructInPlaceThread,
1217 					   rrcopy,"raid_reconip");
1218 		return(retcode);
1219 
1220 	case RAIDFRAME_GET_INFO:
1221 		if (!raidPtr->valid)
1222 			return (ENODEV);
1223 		ucfgp = (RF_DeviceConfig_t **) data;
1224 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1225 			  (RF_DeviceConfig_t *));
1226 		if (d_cfg == NULL)
1227 			return (ENOMEM);
1228 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1229 		d_cfg->rows = raidPtr->numRow;
1230 		d_cfg->cols = raidPtr->numCol;
1231 		d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol;
1232 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1233 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1234 			return (ENOMEM);
1235 		}
1236 		d_cfg->nspares = raidPtr->numSpare;
1237 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1238 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1239 			return (ENOMEM);
1240 		}
1241 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1242 		d = 0;
1243 		for (i = 0; i < d_cfg->rows; i++) {
1244 			for (j = 0; j < d_cfg->cols; j++) {
1245 				d_cfg->devs[d] = raidPtr->Disks[i][j];
1246 				d++;
1247 			}
1248 		}
1249 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1250 			d_cfg->spares[i] = raidPtr->Disks[0][j];
1251 		}
1252 		retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t));
1253 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1254 
1255 		return (retcode);
1256 
1257 	case RAIDFRAME_CHECK_PARITY:
1258 		*(int *) data = raidPtr->parity_good;
1259 		return (0);
1260 
1261 	case RAIDFRAME_RESET_ACCTOTALS:
1262 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1263 		return (0);
1264 
1265 	case RAIDFRAME_GET_ACCTOTALS:
1266 		totals = (RF_AccTotals_t *) data;
1267 		*totals = raidPtr->acc_totals;
1268 		return (0);
1269 
1270 	case RAIDFRAME_KEEP_ACCTOTALS:
1271 		raidPtr->keep_acc_totals = *(int *)data;
1272 		return (0);
1273 
1274 	case RAIDFRAME_GET_SIZE:
1275 		*(int *) data = raidPtr->totalSectors;
1276 		return (0);
1277 
1278 		/* fail a disk & optionally start reconstruction */
1279 	case RAIDFRAME_FAIL_DISK:
1280 
1281 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1282 			/* Can't do this on a RAID 0!! */
1283 			return(EINVAL);
1284 		}
1285 
1286 		rr = (struct rf_recon_req *) data;
1287 
1288 		if (rr->row < 0 || rr->row >= raidPtr->numRow
1289 		    || rr->col < 0 || rr->col >= raidPtr->numCol)
1290 			return (EINVAL);
1291 
1292 
1293 		RF_LOCK_MUTEX(raidPtr->mutex);
1294 		if ((raidPtr->Disks[rr->row][rr->col].status ==
1295 		     rf_ds_optimal) && (raidPtr->numFailures > 0)) {
1296 			/* some other component has failed.  Let's not make
1297 			   things worse. XXX wrong for RAID6 */
1298 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1299 			return (EINVAL);
1300 		}
1301 		if (raidPtr->Disks[rr->row][rr->col].status == rf_ds_spared) {
1302 			/* Can't fail a spared disk! */
1303 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1304 			return (EINVAL);
1305 		}
1306 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1307 
1308 		/* make a copy of the recon request so that we don't rely on
1309 		 * the user's buffer */
1310 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1311 		if (rrcopy == NULL)
1312 			return(ENOMEM);
1313 		memcpy(rrcopy, rr, sizeof(*rr));
1314 		rrcopy->raidPtr = (void *) raidPtr;
1315 
1316 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1317 					   rf_ReconThread,
1318 					   rrcopy,"raid_recon");
1319 		return (0);
1320 
1321 		/* invoke a copyback operation after recon on whatever disk
1322 		 * needs it, if any */
1323 	case RAIDFRAME_COPYBACK:
1324 
1325 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1326 			/* This makes no sense on a RAID 0!! */
1327 			return(EINVAL);
1328 		}
1329 
1330 		if (raidPtr->copyback_in_progress == 1) {
1331 			/* Copyback is already in progress! */
1332 			return(EINVAL);
1333 		}
1334 
1335 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1336 					   rf_CopybackThread,
1337 					   raidPtr,"raid_copyback");
1338 		return (retcode);
1339 
1340 		/* return the percentage completion of reconstruction */
1341 	case RAIDFRAME_CHECK_RECON_STATUS:
1342 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1343 			/* This makes no sense on a RAID 0, so tell the
1344 			   user it's done. */
1345 			*(int *) data = 100;
1346 			return(0);
1347 		}
1348 		row = 0; /* XXX we only consider a single row... */
1349 		if (raidPtr->status[row] != rf_rs_reconstructing)
1350 			*(int *) data = 100;
1351 		else
1352 			*(int *) data = raidPtr->reconControl[row]->percentComplete;
1353 		return (0);
1354 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1355 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1356 		row = 0; /* XXX we only consider a single row... */
1357 		if (raidPtr->status[row] != rf_rs_reconstructing) {
1358 			progressInfo.remaining = 0;
1359 			progressInfo.completed = 100;
1360 			progressInfo.total = 100;
1361 		} else {
1362 			progressInfo.total =
1363 				raidPtr->reconControl[row]->numRUsTotal;
1364 			progressInfo.completed =
1365 				raidPtr->reconControl[row]->numRUsComplete;
1366 			progressInfo.remaining = progressInfo.total -
1367 				progressInfo.completed;
1368 		}
1369 		retcode = copyout(&progressInfo, *progressInfoPtr,
1370 				  sizeof(RF_ProgressInfo_t));
1371 		return (retcode);
1372 
1373 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1374 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1375 			/* This makes no sense on a RAID 0, so tell the
1376 			   user it's done. */
1377 			*(int *) data = 100;
1378 			return(0);
1379 		}
1380 		if (raidPtr->parity_rewrite_in_progress == 1) {
1381 			*(int *) data = 100 *
1382 				raidPtr->parity_rewrite_stripes_done /
1383 				raidPtr->Layout.numStripe;
1384 		} else {
1385 			*(int *) data = 100;
1386 		}
1387 		return (0);
1388 
1389 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1390 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1391 		if (raidPtr->parity_rewrite_in_progress == 1) {
1392 			progressInfo.total = raidPtr->Layout.numStripe;
1393 			progressInfo.completed =
1394 				raidPtr->parity_rewrite_stripes_done;
1395 			progressInfo.remaining = progressInfo.total -
1396 				progressInfo.completed;
1397 		} else {
1398 			progressInfo.remaining = 0;
1399 			progressInfo.completed = 100;
1400 			progressInfo.total = 100;
1401 		}
1402 		retcode = copyout(&progressInfo, *progressInfoPtr,
1403 				  sizeof(RF_ProgressInfo_t));
1404 		return (retcode);
1405 
1406 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1407 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1408 			/* This makes no sense on a RAID 0 */
1409 			*(int *) data = 100;
1410 			return(0);
1411 		}
1412 		if (raidPtr->copyback_in_progress == 1) {
1413 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1414 				raidPtr->Layout.numStripe;
1415 		} else {
1416 			*(int *) data = 100;
1417 		}
1418 		return (0);
1419 
1420 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1421 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1422 		if (raidPtr->copyback_in_progress == 1) {
1423 			progressInfo.total = raidPtr->Layout.numStripe;
1424 			progressInfo.completed =
1425 				raidPtr->copyback_stripes_done;
1426 			progressInfo.remaining = progressInfo.total -
1427 				progressInfo.completed;
1428 		} else {
1429 			progressInfo.remaining = 0;
1430 			progressInfo.completed = 100;
1431 			progressInfo.total = 100;
1432 		}
1433 		retcode = copyout(&progressInfo, *progressInfoPtr,
1434 				  sizeof(RF_ProgressInfo_t));
1435 		return (retcode);
1436 
1437 		/* the sparetable daemon calls this to wait for the kernel to
1438 		 * need a spare table. this ioctl does not return until a
1439 		 * spare table is needed. XXX -- calling mpsleep here in the
1440 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1441 		 * -- I should either compute the spare table in the kernel,
1442 		 * or have a different -- XXX XXX -- interface (a different
1443 		 * character device) for delivering the table     -- XXX */
1444 #if 0
1445 	case RAIDFRAME_SPARET_WAIT:
1446 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1447 		while (!rf_sparet_wait_queue)
1448 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1449 		waitreq = rf_sparet_wait_queue;
1450 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1451 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1452 
1453 		/* structure assignment */
1454 		*((RF_SparetWait_t *) data) = *waitreq;
1455 
1456 		RF_Free(waitreq, sizeof(*waitreq));
1457 		return (0);
1458 
1459 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1460 		 * code in it that will cause the dameon to exit */
1461 	case RAIDFRAME_ABORT_SPARET_WAIT:
1462 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1463 		waitreq->fcol = -1;
1464 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1465 		waitreq->next = rf_sparet_wait_queue;
1466 		rf_sparet_wait_queue = waitreq;
1467 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1468 		wakeup(&rf_sparet_wait_queue);
1469 		return (0);
1470 
1471 		/* used by the spare table daemon to deliver a spare table
1472 		 * into the kernel */
1473 	case RAIDFRAME_SEND_SPARET:
1474 
1475 		/* install the spare table */
1476 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1477 
1478 		/* respond to the requestor.  the return status of the spare
1479 		 * table installation is passed in the "fcol" field */
1480 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1481 		waitreq->fcol = retcode;
1482 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1483 		waitreq->next = rf_sparet_resp_queue;
1484 		rf_sparet_resp_queue = waitreq;
1485 		wakeup(&rf_sparet_resp_queue);
1486 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1487 
1488 		return (retcode);
1489 #endif
1490 
1491 	default:
1492 		break; /* fall through to the os-specific code below */
1493 
1494 	}
1495 
1496 	if (!raidPtr->valid)
1497 		return (EINVAL);
1498 
1499 	/*
1500 	 * Add support for "regular" device ioctls here.
1501 	 */
1502 
1503 	switch (cmd) {
1504 	case DIOCGDINFO:
1505 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1506 		break;
1507 #ifdef __HAVE_OLD_DISKLABEL
1508 	case ODIOCGDINFO:
1509 		newlabel = *(rs->sc_dkdev.dk_label);
1510 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1511 			return ENOTTY;
1512 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1513 		break;
1514 #endif
1515 
1516 	case DIOCGPART:
1517 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1518 		((struct partinfo *) data)->part =
1519 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1520 		break;
1521 
1522 	case DIOCWDINFO:
1523 	case DIOCSDINFO:
1524 #ifdef __HAVE_OLD_DISKLABEL
1525 	case ODIOCWDINFO:
1526 	case ODIOCSDINFO:
1527 #endif
1528 	{
1529 		struct disklabel *lp;
1530 #ifdef __HAVE_OLD_DISKLABEL
1531 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1532 			memset(&newlabel, 0, sizeof newlabel);
1533 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1534 			lp = &newlabel;
1535 		} else
1536 #endif
1537 		lp = (struct disklabel *)data;
1538 
1539 		if ((error = raidlock(rs)) != 0)
1540 			return (error);
1541 
1542 		rs->sc_flags |= RAIDF_LABELLING;
1543 
1544 		error = setdisklabel(rs->sc_dkdev.dk_label,
1545 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1546 		if (error == 0) {
1547 			if (cmd == DIOCWDINFO
1548 #ifdef __HAVE_OLD_DISKLABEL
1549 			    || cmd == ODIOCWDINFO
1550 #endif
1551 			   )
1552 				error = writedisklabel(RAIDLABELDEV(dev),
1553 				    raidstrategy, rs->sc_dkdev.dk_label,
1554 				    rs->sc_dkdev.dk_cpulabel);
1555 		}
1556 		rs->sc_flags &= ~RAIDF_LABELLING;
1557 
1558 		raidunlock(rs);
1559 
1560 		if (error)
1561 			return (error);
1562 		break;
1563 	}
1564 
1565 	case DIOCWLABEL:
1566 		if (*(int *) data != 0)
1567 			rs->sc_flags |= RAIDF_WLABEL;
1568 		else
1569 			rs->sc_flags &= ~RAIDF_WLABEL;
1570 		break;
1571 
1572 	case DIOCGDEFLABEL:
1573 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1574 		break;
1575 
1576 #ifdef __HAVE_OLD_DISKLABEL
1577 	case ODIOCGDEFLABEL:
1578 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1579 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1580 			return ENOTTY;
1581 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1582 		break;
1583 #endif
1584 
1585 	default:
1586 		retcode = ENOTTY;
1587 	}
1588 	return (retcode);
1589 
1590 }
1591 
1592 
1593 /* raidinit -- complete the rest of the initialization for the
1594    RAIDframe device.  */
1595 
1596 
1597 static void
1598 raidinit(raidPtr)
1599 	RF_Raid_t *raidPtr;
1600 {
1601 	struct raid_softc *rs;
1602 	int     unit;
1603 
1604 	unit = raidPtr->raidid;
1605 
1606 	rs = &raid_softc[unit];
1607 
1608 	/* XXX should check return code first... */
1609 	rs->sc_flags |= RAIDF_INITED;
1610 
1611 	sprintf(rs->sc_xname, "raid%d", unit);	/* XXX doesn't check bounds. */
1612 
1613 	rs->sc_dkdev.dk_name = rs->sc_xname;
1614 
1615 	/* disk_attach actually creates space for the CPU disklabel, among
1616 	 * other things, so it's critical to call this *BEFORE* we try putzing
1617 	 * with disklabels. */
1618 
1619 	disk_attach(&rs->sc_dkdev);
1620 
1621 	/* XXX There may be a weird interaction here between this, and
1622 	 * protectedSectors, as used in RAIDframe.  */
1623 
1624 	rs->sc_size = raidPtr->totalSectors;
1625 
1626 }
1627 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0)
1628 /* wake up the daemon & tell it to get us a spare table
1629  * XXX
1630  * the entries in the queues should be tagged with the raidPtr
1631  * so that in the extremely rare case that two recons happen at once,
1632  * we know for which device were requesting a spare table
1633  * XXX
1634  *
1635  * XXX This code is not currently used. GO
1636  */
1637 int
1638 rf_GetSpareTableFromDaemon(req)
1639 	RF_SparetWait_t *req;
1640 {
1641 	int     retcode;
1642 
1643 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1644 	req->next = rf_sparet_wait_queue;
1645 	rf_sparet_wait_queue = req;
1646 	wakeup(&rf_sparet_wait_queue);
1647 
1648 	/* mpsleep unlocks the mutex */
1649 	while (!rf_sparet_resp_queue) {
1650 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1651 		    "raidframe getsparetable", 0);
1652 	}
1653 	req = rf_sparet_resp_queue;
1654 	rf_sparet_resp_queue = req->next;
1655 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1656 
1657 	retcode = req->fcol;
1658 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1659 					 * alloc'd */
1660 	return (retcode);
1661 }
1662 #endif
1663 
1664 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1665  * bp & passes it down.
1666  * any calls originating in the kernel must use non-blocking I/O
1667  * do some extra sanity checking to return "appropriate" error values for
1668  * certain conditions (to make some standard utilities work)
1669  *
1670  * Formerly known as: rf_DoAccessKernel
1671  */
1672 void
1673 raidstart(raidPtr)
1674 	RF_Raid_t *raidPtr;
1675 {
1676 	RF_SectorCount_t num_blocks, pb, sum;
1677 	RF_RaidAddr_t raid_addr;
1678 	struct partition *pp;
1679 	daddr_t blocknum;
1680 	int     unit;
1681 	struct raid_softc *rs;
1682 	int     do_async;
1683 	struct buf *bp;
1684 
1685 	unit = raidPtr->raidid;
1686 	rs = &raid_softc[unit];
1687 
1688 	/* quick check to see if anything has died recently */
1689 	RF_LOCK_MUTEX(raidPtr->mutex);
1690 	if (raidPtr->numNewFailures > 0) {
1691 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1692 		rf_update_component_labels(raidPtr,
1693 					   RF_NORMAL_COMPONENT_UPDATE);
1694 		RF_LOCK_MUTEX(raidPtr->mutex);
1695 		raidPtr->numNewFailures--;
1696 	}
1697 
1698 	/* Check to see if we're at the limit... */
1699 	while (raidPtr->openings > 0) {
1700 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1701 
1702 		/* get the next item, if any, from the queue */
1703 		if ((bp = BUFQ_GET(&rs->buf_queue)) == NULL) {
1704 			/* nothing more to do */
1705 			return;
1706 		}
1707 
1708 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1709 		 * partition.. Need to make it absolute to the underlying
1710 		 * device.. */
1711 
1712 		blocknum = bp->b_blkno;
1713 		if (DISKPART(bp->b_dev) != RAW_PART) {
1714 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1715 			blocknum += pp->p_offset;
1716 		}
1717 
1718 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1719 			    (int) blocknum));
1720 
1721 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1722 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1723 
1724 		/* *THIS* is where we adjust what block we're going to...
1725 		 * but DO NOT TOUCH bp->b_blkno!!! */
1726 		raid_addr = blocknum;
1727 
1728 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1729 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1730 		sum = raid_addr + num_blocks + pb;
1731 		if (1 || rf_debugKernelAccess) {
1732 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1733 				    (int) raid_addr, (int) sum, (int) num_blocks,
1734 				    (int) pb, (int) bp->b_resid));
1735 		}
1736 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1737 		    || (sum < num_blocks) || (sum < pb)) {
1738 			bp->b_error = ENOSPC;
1739 			bp->b_flags |= B_ERROR;
1740 			bp->b_resid = bp->b_bcount;
1741 			biodone(bp);
1742 			RF_LOCK_MUTEX(raidPtr->mutex);
1743 			continue;
1744 		}
1745 		/*
1746 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1747 		 */
1748 
1749 		if (bp->b_bcount & raidPtr->sectorMask) {
1750 			bp->b_error = EINVAL;
1751 			bp->b_flags |= B_ERROR;
1752 			bp->b_resid = bp->b_bcount;
1753 			biodone(bp);
1754 			RF_LOCK_MUTEX(raidPtr->mutex);
1755 			continue;
1756 
1757 		}
1758 		db1_printf(("Calling DoAccess..\n"));
1759 
1760 
1761 		RF_LOCK_MUTEX(raidPtr->mutex);
1762 		raidPtr->openings--;
1763 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1764 
1765 		/*
1766 		 * Everything is async.
1767 		 */
1768 		do_async = 1;
1769 
1770 		disk_busy(&rs->sc_dkdev);
1771 
1772 		/* XXX we're still at splbio() here... do we *really*
1773 		   need to be? */
1774 
1775 		/* don't ever condition on bp->b_flags & B_WRITE.
1776 		 * always condition on B_READ instead */
1777 
1778 		bp->b_error = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1779 				      RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1780 				      do_async, raid_addr, num_blocks,
1781 				      bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1782 
1783 		if (bp->b_error) {
1784 			bp->b_flags |= B_ERROR;
1785 		}
1786 
1787 		RF_LOCK_MUTEX(raidPtr->mutex);
1788 	}
1789 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1790 }
1791 
1792 
1793 
1794 
1795 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1796 
1797 int
1798 rf_DispatchKernelIO(queue, req)
1799 	RF_DiskQueue_t *queue;
1800 	RF_DiskQueueData_t *req;
1801 {
1802 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1803 	struct buf *bp;
1804 	struct raidbuf *raidbp = NULL;
1805 
1806 	req->queue = queue;
1807 
1808 #if DIAGNOSTIC
1809 	if (queue->raidPtr->raidid >= numraid) {
1810 		printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid,
1811 		    numraid);
1812 		panic("Invalid Unit number in rf_DispatchKernelIO");
1813 	}
1814 #endif
1815 
1816 	bp = req->bp;
1817 #if 1
1818 	/* XXX when there is a physical disk failure, someone is passing us a
1819 	 * buffer that contains old stuff!!  Attempt to deal with this problem
1820 	 * without taking a performance hit... (not sure where the real bug
1821 	 * is.  It's buried in RAIDframe somewhere) :-(  GO ) */
1822 
1823 	if (bp->b_flags & B_ERROR) {
1824 		bp->b_flags &= ~B_ERROR;
1825 	}
1826 	if (bp->b_error != 0) {
1827 		bp->b_error = 0;
1828 	}
1829 #endif
1830 	raidbp = pool_get(&raidframe_cbufpool, PR_NOWAIT);
1831 	if (raidbp == NULL) {
1832 		bp->b_flags |= B_ERROR;
1833 		bp->b_error = ENOMEM;
1834 		return (ENOMEM);
1835 	}
1836 	BUF_INIT(&raidbp->rf_buf);
1837 
1838 	/*
1839 	 * context for raidiodone
1840 	 */
1841 	raidbp->rf_obp = bp;
1842 	raidbp->req = req;
1843 
1844 	switch (req->type) {
1845 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1846 		/* XXX need to do something extra here.. */
1847 		/* I'm leaving this in, as I've never actually seen it used,
1848 		 * and I'd like folks to report it... GO */
1849 		printf(("WAKEUP CALLED\n"));
1850 		queue->numOutstanding++;
1851 
1852 		/* XXX need to glue the original buffer into this??  */
1853 
1854 		KernelWakeupFunc(&raidbp->rf_buf);
1855 		break;
1856 
1857 	case RF_IO_TYPE_READ:
1858 	case RF_IO_TYPE_WRITE:
1859 
1860 		if (req->tracerec) {
1861 			RF_ETIMER_START(req->tracerec->timer);
1862 		}
1863 		InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1864 		    op | bp->b_flags, queue->rf_cinfo->ci_dev,
1865 		    req->sectorOffset, req->numSector,
1866 		    req->buf, KernelWakeupFunc, (void *) req,
1867 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1868 
1869 		if (rf_debugKernelAccess) {
1870 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1871 				(long) bp->b_blkno));
1872 		}
1873 		queue->numOutstanding++;
1874 		queue->last_deq_sector = req->sectorOffset;
1875 		/* acc wouldn't have been let in if there were any pending
1876 		 * reqs at any other priority */
1877 		queue->curPriority = req->priority;
1878 
1879 		db1_printf(("Going for %c to unit %d row %d col %d\n",
1880 			    req->type, queue->raidPtr->raidid,
1881 			    queue->row, queue->col));
1882 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1883 			(int) req->sectorOffset, (int) req->numSector,
1884 			(int) (req->numSector <<
1885 			    queue->raidPtr->logBytesPerSector),
1886 			(int) queue->raidPtr->logBytesPerSector));
1887 		if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1888 			raidbp->rf_buf.b_vp->v_numoutput++;
1889 		}
1890 		VOP_STRATEGY(&raidbp->rf_buf);
1891 
1892 		break;
1893 
1894 	default:
1895 		panic("bad req->type in rf_DispatchKernelIO");
1896 	}
1897 	db1_printf(("Exiting from DispatchKernelIO\n"));
1898 
1899 	return (0);
1900 }
1901 /* this is the callback function associated with a I/O invoked from
1902    kernel code.
1903  */
1904 static void
1905 KernelWakeupFunc(vbp)
1906 	struct buf *vbp;
1907 {
1908 	RF_DiskQueueData_t *req = NULL;
1909 	RF_DiskQueue_t *queue;
1910 	struct raidbuf *raidbp = (struct raidbuf *) vbp;
1911 	struct buf *bp;
1912 	int s;
1913 
1914 	s = splbio();
1915 	db1_printf(("recovering the request queue:\n"));
1916 	req = raidbp->req;
1917 
1918 	bp = raidbp->rf_obp;
1919 
1920 	queue = (RF_DiskQueue_t *) req->queue;
1921 
1922 	if (raidbp->rf_buf.b_flags & B_ERROR) {
1923 		bp->b_flags |= B_ERROR;
1924 		bp->b_error = raidbp->rf_buf.b_error ?
1925 		    raidbp->rf_buf.b_error : EIO;
1926 	}
1927 
1928 	/* XXX methinks this could be wrong... */
1929 #if 1
1930 	bp->b_resid = raidbp->rf_buf.b_resid;
1931 #endif
1932 
1933 	if (req->tracerec) {
1934 		RF_ETIMER_STOP(req->tracerec->timer);
1935 		RF_ETIMER_EVAL(req->tracerec->timer);
1936 		RF_LOCK_MUTEX(rf_tracing_mutex);
1937 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1938 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1939 		req->tracerec->num_phys_ios++;
1940 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1941 	}
1942 	bp->b_bcount = raidbp->rf_buf.b_bcount;	/* XXXX ?? */
1943 
1944 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1945 	 * ballistic, and mark the component as hosed... */
1946 
1947 	if (bp->b_flags & B_ERROR) {
1948 		/* Mark the disk as dead */
1949 		/* but only mark it once... */
1950 		if (queue->raidPtr->Disks[queue->row][queue->col].status ==
1951 		    rf_ds_optimal) {
1952 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1953 			       queue->raidPtr->raidid,
1954 			       queue->raidPtr->Disks[queue->row][queue->col].devname);
1955 			queue->raidPtr->Disks[queue->row][queue->col].status =
1956 			    rf_ds_failed;
1957 			queue->raidPtr->status[queue->row] = rf_rs_degraded;
1958 			queue->raidPtr->numFailures++;
1959 			queue->raidPtr->numNewFailures++;
1960 		} else {	/* Disk is already dead... */
1961 			/* printf("Disk already marked as dead!\n"); */
1962 		}
1963 
1964 	}
1965 
1966 	pool_put(&raidframe_cbufpool, raidbp);
1967 
1968 	/* Fill in the error value */
1969 
1970 	req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0;
1971 
1972 	simple_lock(&queue->raidPtr->iodone_lock);
1973 
1974 	/* Drop this one on the "finished" queue... */
1975 	TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries);
1976 
1977 	/* Let the raidio thread know there is work to be done. */
1978 	wakeup(&(queue->raidPtr->iodone));
1979 
1980 	simple_unlock(&queue->raidPtr->iodone_lock);
1981 
1982 	splx(s);
1983 }
1984 
1985 
1986 
1987 /*
1988  * initialize a buf structure for doing an I/O in the kernel.
1989  */
1990 static void
1991 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg,
1992        logBytesPerSector, b_proc)
1993 	struct buf *bp;
1994 	struct vnode *b_vp;
1995 	unsigned rw_flag;
1996 	dev_t dev;
1997 	RF_SectorNum_t startSect;
1998 	RF_SectorCount_t numSect;
1999 	caddr_t buf;
2000 	void (*cbFunc) (struct buf *);
2001 	void *cbArg;
2002 	int logBytesPerSector;
2003 	struct proc *b_proc;
2004 {
2005 	/* bp->b_flags       = B_PHYS | rw_flag; */
2006 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
2007 	bp->b_bcount = numSect << logBytesPerSector;
2008 	bp->b_bufsize = bp->b_bcount;
2009 	bp->b_error = 0;
2010 	bp->b_dev = dev;
2011 	bp->b_data = buf;
2012 	bp->b_blkno = startSect;
2013 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
2014 	if (bp->b_bcount == 0) {
2015 		panic("bp->b_bcount is zero in InitBP!!");
2016 	}
2017 	bp->b_proc = b_proc;
2018 	bp->b_iodone = cbFunc;
2019 	bp->b_vp = b_vp;
2020 
2021 }
2022 
2023 static void
2024 raidgetdefaultlabel(raidPtr, rs, lp)
2025 	RF_Raid_t *raidPtr;
2026 	struct raid_softc *rs;
2027 	struct disklabel *lp;
2028 {
2029 	memset(lp, 0, sizeof(*lp));
2030 
2031 	/* fabricate a label... */
2032 	lp->d_secperunit = raidPtr->totalSectors;
2033 	lp->d_secsize = raidPtr->bytesPerSector;
2034 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
2035 	lp->d_ntracks = 4 * raidPtr->numCol;
2036 	lp->d_ncylinders = raidPtr->totalSectors /
2037 		(lp->d_nsectors * lp->d_ntracks);
2038 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
2039 
2040 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
2041 	lp->d_type = DTYPE_RAID;
2042 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
2043 	lp->d_rpm = 3600;
2044 	lp->d_interleave = 1;
2045 	lp->d_flags = 0;
2046 
2047 	lp->d_partitions[RAW_PART].p_offset = 0;
2048 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2049 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2050 	lp->d_npartitions = RAW_PART + 1;
2051 
2052 	lp->d_magic = DISKMAGIC;
2053 	lp->d_magic2 = DISKMAGIC;
2054 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2055 
2056 }
2057 /*
2058  * Read the disklabel from the raid device.  If one is not present, fake one
2059  * up.
2060  */
2061 static void
2062 raidgetdisklabel(dev)
2063 	dev_t   dev;
2064 {
2065 	int     unit = raidunit(dev);
2066 	struct raid_softc *rs = &raid_softc[unit];
2067 	const char   *errstring;
2068 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2069 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2070 	RF_Raid_t *raidPtr;
2071 
2072 	db1_printf(("Getting the disklabel...\n"));
2073 
2074 	memset(clp, 0, sizeof(*clp));
2075 
2076 	raidPtr = raidPtrs[unit];
2077 
2078 	raidgetdefaultlabel(raidPtr, rs, lp);
2079 
2080 	/*
2081 	 * Call the generic disklabel extraction routine.
2082 	 */
2083 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2084 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2085 	if (errstring)
2086 		raidmakedisklabel(rs);
2087 	else {
2088 		int     i;
2089 		struct partition *pp;
2090 
2091 		/*
2092 		 * Sanity check whether the found disklabel is valid.
2093 		 *
2094 		 * This is necessary since total size of the raid device
2095 		 * may vary when an interleave is changed even though exactly
2096 		 * same componets are used, and old disklabel may used
2097 		 * if that is found.
2098 		 */
2099 		if (lp->d_secperunit != rs->sc_size)
2100 			printf("raid%d: WARNING: %s: "
2101 			    "total sector size in disklabel (%d) != "
2102 			    "the size of raid (%ld)\n", unit, rs->sc_xname,
2103 			    lp->d_secperunit, (long) rs->sc_size);
2104 		for (i = 0; i < lp->d_npartitions; i++) {
2105 			pp = &lp->d_partitions[i];
2106 			if (pp->p_offset + pp->p_size > rs->sc_size)
2107 				printf("raid%d: WARNING: %s: end of partition `%c' "
2108 				       "exceeds the size of raid (%ld)\n",
2109 				       unit, rs->sc_xname, 'a' + i, (long) rs->sc_size);
2110 		}
2111 	}
2112 
2113 }
2114 /*
2115  * Take care of things one might want to take care of in the event
2116  * that a disklabel isn't present.
2117  */
2118 static void
2119 raidmakedisklabel(rs)
2120 	struct raid_softc *rs;
2121 {
2122 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2123 	db1_printf(("Making a label..\n"));
2124 
2125 	/*
2126 	 * For historical reasons, if there's no disklabel present
2127 	 * the raw partition must be marked FS_BSDFFS.
2128 	 */
2129 
2130 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2131 
2132 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2133 
2134 	lp->d_checksum = dkcksum(lp);
2135 }
2136 /*
2137  * Lookup the provided name in the filesystem.  If the file exists,
2138  * is a valid block device, and isn't being used by anyone else,
2139  * set *vpp to the file's vnode.
2140  * You'll find the original of this in ccd.c
2141  */
2142 int
2143 raidlookup(path, p, vpp)
2144 	char   *path;
2145 	struct proc *p;
2146 	struct vnode **vpp;	/* result */
2147 {
2148 	struct nameidata nd;
2149 	struct vnode *vp;
2150 	struct vattr va;
2151 	int     error;
2152 
2153 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2154 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2155 		return (error);
2156 	}
2157 	vp = nd.ni_vp;
2158 	if (vp->v_usecount > 1) {
2159 		VOP_UNLOCK(vp, 0);
2160 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2161 		return (EBUSY);
2162 	}
2163 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2164 		VOP_UNLOCK(vp, 0);
2165 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2166 		return (error);
2167 	}
2168 	/* XXX: eventually we should handle VREG, too. */
2169 	if (va.va_type != VBLK) {
2170 		VOP_UNLOCK(vp, 0);
2171 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2172 		return (ENOTBLK);
2173 	}
2174 	VOP_UNLOCK(vp, 0);
2175 	*vpp = vp;
2176 	return (0);
2177 }
2178 /*
2179  * Wait interruptibly for an exclusive lock.
2180  *
2181  * XXX
2182  * Several drivers do this; it should be abstracted and made MP-safe.
2183  * (Hmm... where have we seen this warning before :->  GO )
2184  */
2185 static int
2186 raidlock(rs)
2187 	struct raid_softc *rs;
2188 {
2189 	int     error;
2190 
2191 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2192 		rs->sc_flags |= RAIDF_WANTED;
2193 		if ((error =
2194 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2195 			return (error);
2196 	}
2197 	rs->sc_flags |= RAIDF_LOCKED;
2198 	return (0);
2199 }
2200 /*
2201  * Unlock and wake up any waiters.
2202  */
2203 static void
2204 raidunlock(rs)
2205 	struct raid_softc *rs;
2206 {
2207 
2208 	rs->sc_flags &= ~RAIDF_LOCKED;
2209 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2210 		rs->sc_flags &= ~RAIDF_WANTED;
2211 		wakeup(rs);
2212 	}
2213 }
2214 
2215 
2216 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2217 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2218 
2219 int
2220 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2221 {
2222 	RF_ComponentLabel_t clabel;
2223 	raidread_component_label(dev, b_vp, &clabel);
2224 	clabel.mod_counter = mod_counter;
2225 	clabel.clean = RF_RAID_CLEAN;
2226 	raidwrite_component_label(dev, b_vp, &clabel);
2227 	return(0);
2228 }
2229 
2230 
2231 int
2232 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2233 {
2234 	RF_ComponentLabel_t clabel;
2235 	raidread_component_label(dev, b_vp, &clabel);
2236 	clabel.mod_counter = mod_counter;
2237 	clabel.clean = RF_RAID_DIRTY;
2238 	raidwrite_component_label(dev, b_vp, &clabel);
2239 	return(0);
2240 }
2241 
2242 /* ARGSUSED */
2243 int
2244 raidread_component_label(dev, b_vp, clabel)
2245 	dev_t dev;
2246 	struct vnode *b_vp;
2247 	RF_ComponentLabel_t *clabel;
2248 {
2249 	struct buf *bp;
2250 	const struct bdevsw *bdev;
2251 	int error;
2252 
2253 	/* XXX should probably ensure that we don't try to do this if
2254 	   someone has changed rf_protected_sectors. */
2255 
2256 	if (b_vp == NULL) {
2257 		/* For whatever reason, this component is not valid.
2258 		   Don't try to read a component label from it. */
2259 		return(EINVAL);
2260 	}
2261 
2262 	/* get a block of the appropriate size... */
2263 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2264 	bp->b_dev = dev;
2265 
2266 	/* get our ducks in a row for the read */
2267 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2268 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2269 	bp->b_flags |= B_READ;
2270  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2271 
2272 	bdev = bdevsw_lookup(bp->b_dev);
2273 	if (bdev == NULL)
2274 		return (ENXIO);
2275 	(*bdev->d_strategy)(bp);
2276 
2277 	error = biowait(bp);
2278 
2279 	if (!error) {
2280 		memcpy(clabel, bp->b_data,
2281 		       sizeof(RF_ComponentLabel_t));
2282         }
2283 
2284 	brelse(bp);
2285 	return(error);
2286 }
2287 /* ARGSUSED */
2288 int
2289 raidwrite_component_label(dev, b_vp, clabel)
2290 	dev_t dev;
2291 	struct vnode *b_vp;
2292 	RF_ComponentLabel_t *clabel;
2293 {
2294 	struct buf *bp;
2295 	const struct bdevsw *bdev;
2296 	int error;
2297 
2298 	/* get a block of the appropriate size... */
2299 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2300 	bp->b_dev = dev;
2301 
2302 	/* get our ducks in a row for the write */
2303 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2304 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2305 	bp->b_flags |= B_WRITE;
2306  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2307 
2308 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2309 
2310 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2311 
2312 	bdev = bdevsw_lookup(bp->b_dev);
2313 	if (bdev == NULL)
2314 		return (ENXIO);
2315 	(*bdev->d_strategy)(bp);
2316 	error = biowait(bp);
2317 	brelse(bp);
2318 	if (error) {
2319 #if 1
2320 		printf("Failed to write RAID component info!\n");
2321 #endif
2322 	}
2323 
2324 	return(error);
2325 }
2326 
2327 void
2328 rf_markalldirty(raidPtr)
2329 	RF_Raid_t *raidPtr;
2330 {
2331 	RF_ComponentLabel_t clabel;
2332 	int sparecol;
2333 	int r,c;
2334 	int i,j;
2335 	int srow, scol;
2336 
2337 	raidPtr->mod_counter++;
2338 	for (r = 0; r < raidPtr->numRow; r++) {
2339 		for (c = 0; c < raidPtr->numCol; c++) {
2340 			/* we don't want to touch (at all) a disk that has
2341 			   failed */
2342 			if (!RF_DEAD_DISK(raidPtr->Disks[r][c].status)) {
2343 				raidread_component_label(
2344 					raidPtr->Disks[r][c].dev,
2345 					raidPtr->raid_cinfo[r][c].ci_vp,
2346 					&clabel);
2347 				if (clabel.status == rf_ds_spared) {
2348 					/* XXX do something special...
2349 					 but whatever you do, don't
2350 					 try to access it!! */
2351 				} else {
2352 					raidmarkdirty(
2353 					      raidPtr->Disks[r][c].dev,
2354 					      raidPtr->raid_cinfo[r][c].ci_vp,
2355 					      raidPtr->mod_counter);
2356 				}
2357 			}
2358 		}
2359 	}
2360 
2361 	for( c = 0; c < raidPtr->numSpare ; c++) {
2362 		sparecol = raidPtr->numCol + c;
2363 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
2364 			/*
2365 
2366 			   we claim this disk is "optimal" if it's
2367 			   rf_ds_used_spare, as that means it should be
2368 			   directly substitutable for the disk it replaced.
2369 			   We note that too...
2370 
2371 			 */
2372 
2373 			for(i=0;i<raidPtr->numRow;i++) {
2374 				for(j=0;j<raidPtr->numCol;j++) {
2375 					if ((raidPtr->Disks[i][j].spareRow ==
2376 					     0) &&
2377 					    (raidPtr->Disks[i][j].spareCol ==
2378 					     sparecol)) {
2379 						srow = i;
2380 						scol = j;
2381 						break;
2382 					}
2383 				}
2384 			}
2385 
2386 			raidread_component_label(
2387 				 raidPtr->Disks[0][sparecol].dev,
2388 				 raidPtr->raid_cinfo[0][sparecol].ci_vp,
2389 				 &clabel);
2390 			/* make sure status is noted */
2391 
2392 			raid_init_component_label(raidPtr, &clabel);
2393 
2394 			clabel.row = srow;
2395 			clabel.column = scol;
2396 			/* Note: we *don't* change status from rf_ds_used_spare
2397 			   to rf_ds_optimal */
2398 			/* clabel.status = rf_ds_optimal; */
2399 
2400 			raidmarkdirty(raidPtr->Disks[0][sparecol].dev,
2401 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2402 				      raidPtr->mod_counter);
2403 		}
2404 	}
2405 }
2406 
2407 
2408 void
2409 rf_update_component_labels(raidPtr, final)
2410 	RF_Raid_t *raidPtr;
2411 	int final;
2412 {
2413 	RF_ComponentLabel_t clabel;
2414 	int sparecol;
2415 	int r,c;
2416 	int i,j;
2417 	int srow, scol;
2418 
2419 	srow = -1;
2420 	scol = -1;
2421 
2422 	/* XXX should do extra checks to make sure things really are clean,
2423 	   rather than blindly setting the clean bit... */
2424 
2425 	raidPtr->mod_counter++;
2426 
2427 	for (r = 0; r < raidPtr->numRow; r++) {
2428 		for (c = 0; c < raidPtr->numCol; c++) {
2429 			if (raidPtr->Disks[r][c].status == rf_ds_optimal) {
2430 				raidread_component_label(
2431 					raidPtr->Disks[r][c].dev,
2432 					raidPtr->raid_cinfo[r][c].ci_vp,
2433 					&clabel);
2434 				/* make sure status is noted */
2435 				clabel.status = rf_ds_optimal;
2436 				/* bump the counter */
2437 				clabel.mod_counter = raidPtr->mod_counter;
2438 
2439 				raidwrite_component_label(
2440 					raidPtr->Disks[r][c].dev,
2441 					raidPtr->raid_cinfo[r][c].ci_vp,
2442 					&clabel);
2443 				if (final == RF_FINAL_COMPONENT_UPDATE) {
2444 					if (raidPtr->parity_good == RF_RAID_CLEAN) {
2445 						raidmarkclean(
2446 							      raidPtr->Disks[r][c].dev,
2447 							      raidPtr->raid_cinfo[r][c].ci_vp,
2448 							      raidPtr->mod_counter);
2449 					}
2450 				}
2451 			}
2452 			/* else we don't touch it.. */
2453 		}
2454 	}
2455 
2456 	for( c = 0; c < raidPtr->numSpare ; c++) {
2457 		sparecol = raidPtr->numCol + c;
2458 		/* Need to ensure that the reconstruct actually completed! */
2459 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
2460 			/*
2461 
2462 			   we claim this disk is "optimal" if it's
2463 			   rf_ds_used_spare, as that means it should be
2464 			   directly substitutable for the disk it replaced.
2465 			   We note that too...
2466 
2467 			 */
2468 
2469 			for(i=0;i<raidPtr->numRow;i++) {
2470 				for(j=0;j<raidPtr->numCol;j++) {
2471 					if ((raidPtr->Disks[i][j].spareRow ==
2472 					     0) &&
2473 					    (raidPtr->Disks[i][j].spareCol ==
2474 					     sparecol)) {
2475 						srow = i;
2476 						scol = j;
2477 						break;
2478 					}
2479 				}
2480 			}
2481 
2482 			/* XXX shouldn't *really* need this... */
2483 			raidread_component_label(
2484 				      raidPtr->Disks[0][sparecol].dev,
2485 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2486 				      &clabel);
2487 			/* make sure status is noted */
2488 
2489 			raid_init_component_label(raidPtr, &clabel);
2490 
2491 			clabel.mod_counter = raidPtr->mod_counter;
2492 			clabel.row = srow;
2493 			clabel.column = scol;
2494 			clabel.status = rf_ds_optimal;
2495 
2496 			raidwrite_component_label(
2497 				      raidPtr->Disks[0][sparecol].dev,
2498 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2499 				      &clabel);
2500 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2501 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2502 					raidmarkclean( raidPtr->Disks[0][sparecol].dev,
2503 						       raidPtr->raid_cinfo[0][sparecol].ci_vp,
2504 						       raidPtr->mod_counter);
2505 				}
2506 			}
2507 		}
2508 	}
2509 }
2510 
2511 void
2512 rf_close_component(raidPtr, vp, auto_configured)
2513 	RF_Raid_t *raidPtr;
2514 	struct vnode *vp;
2515 	int auto_configured;
2516 {
2517 	struct proc *p;
2518 
2519 	p = raidPtr->engine_thread;
2520 
2521 	if (vp != NULL) {
2522 		if (auto_configured == 1) {
2523 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2524 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2525 			vput(vp);
2526 
2527 		} else {
2528 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2529 		}
2530 	}
2531 }
2532 
2533 
2534 void
2535 rf_UnconfigureVnodes(raidPtr)
2536 	RF_Raid_t *raidPtr;
2537 {
2538 	int r,c;
2539 	struct vnode *vp;
2540 	int acd;
2541 
2542 
2543 	/* We take this opportunity to close the vnodes like we should.. */
2544 
2545 	for (r = 0; r < raidPtr->numRow; r++) {
2546 		for (c = 0; c < raidPtr->numCol; c++) {
2547 			vp = raidPtr->raid_cinfo[r][c].ci_vp;
2548 			acd = raidPtr->Disks[r][c].auto_configured;
2549 			rf_close_component(raidPtr, vp, acd);
2550 			raidPtr->raid_cinfo[r][c].ci_vp = NULL;
2551 			raidPtr->Disks[r][c].auto_configured = 0;
2552 		}
2553 	}
2554 	for (r = 0; r < raidPtr->numSpare; r++) {
2555 		vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp;
2556 		acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured;
2557 		rf_close_component(raidPtr, vp, acd);
2558 		raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL;
2559 		raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0;
2560 	}
2561 }
2562 
2563 
2564 void
2565 rf_ReconThread(req)
2566 	struct rf_recon_req *req;
2567 {
2568 	int     s;
2569 	RF_Raid_t *raidPtr;
2570 
2571 	s = splbio();
2572 	raidPtr = (RF_Raid_t *) req->raidPtr;
2573 	raidPtr->recon_in_progress = 1;
2574 
2575 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col,
2576 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2577 
2578 	/* XXX get rid of this! we don't need it at all.. */
2579 	RF_Free(req, sizeof(*req));
2580 
2581 	raidPtr->recon_in_progress = 0;
2582 	splx(s);
2583 
2584 	/* That's all... */
2585 	kthread_exit(0);        /* does not return */
2586 }
2587 
2588 void
2589 rf_RewriteParityThread(raidPtr)
2590 	RF_Raid_t *raidPtr;
2591 {
2592 	int retcode;
2593 	int s;
2594 
2595 	raidPtr->parity_rewrite_in_progress = 1;
2596 	s = splbio();
2597 	retcode = rf_RewriteParity(raidPtr);
2598 	splx(s);
2599 	if (retcode) {
2600 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2601 	} else {
2602 		/* set the clean bit!  If we shutdown correctly,
2603 		   the clean bit on each component label will get
2604 		   set */
2605 		raidPtr->parity_good = RF_RAID_CLEAN;
2606 	}
2607 	raidPtr->parity_rewrite_in_progress = 0;
2608 
2609 	/* Anyone waiting for us to stop?  If so, inform them... */
2610 	if (raidPtr->waitShutdown) {
2611 		wakeup(&raidPtr->parity_rewrite_in_progress);
2612 	}
2613 
2614 	/* That's all... */
2615 	kthread_exit(0);        /* does not return */
2616 }
2617 
2618 
2619 void
2620 rf_CopybackThread(raidPtr)
2621 	RF_Raid_t *raidPtr;
2622 {
2623 	int s;
2624 
2625 	raidPtr->copyback_in_progress = 1;
2626 	s = splbio();
2627 	rf_CopybackReconstructedData(raidPtr);
2628 	splx(s);
2629 	raidPtr->copyback_in_progress = 0;
2630 
2631 	/* That's all... */
2632 	kthread_exit(0);        /* does not return */
2633 }
2634 
2635 
2636 void
2637 rf_ReconstructInPlaceThread(req)
2638 	struct rf_recon_req *req;
2639 {
2640 	int s;
2641 	RF_Raid_t *raidPtr;
2642 
2643 	s = splbio();
2644 	raidPtr = req->raidPtr;
2645 	raidPtr->recon_in_progress = 1;
2646 	rf_ReconstructInPlace(raidPtr, req->row, req->col);
2647 	RF_Free(req, sizeof(*req));
2648 	raidPtr->recon_in_progress = 0;
2649 	splx(s);
2650 
2651 	/* That's all... */
2652 	kthread_exit(0);        /* does not return */
2653 }
2654 
2655 RF_AutoConfig_t *
2656 rf_find_raid_components()
2657 {
2658 	struct vnode *vp;
2659 	struct disklabel label;
2660 	struct device *dv;
2661 	dev_t dev;
2662 	int bmajor;
2663 	int error;
2664 	int i;
2665 	int good_one;
2666 	RF_ComponentLabel_t *clabel;
2667 	RF_AutoConfig_t *ac_list;
2668 	RF_AutoConfig_t *ac;
2669 
2670 
2671 	/* initialize the AutoConfig list */
2672 	ac_list = NULL;
2673 
2674 	/* we begin by trolling through *all* the devices on the system */
2675 
2676 	for (dv = alldevs.tqh_first; dv != NULL;
2677 	     dv = dv->dv_list.tqe_next) {
2678 
2679 		/* we are only interested in disks... */
2680 		if (dv->dv_class != DV_DISK)
2681 			continue;
2682 
2683 		/* we don't care about floppies... */
2684 		if (!strcmp(dv->dv_cfdata->cf_name,"fd")) {
2685 			continue;
2686 		}
2687 
2688 		/* we don't care about CD's... */
2689 		if (!strcmp(dv->dv_cfdata->cf_name,"cd")) {
2690 			continue;
2691 		}
2692 
2693 		/* hdfd is the Atari/Hades floppy driver */
2694 		if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) {
2695 			continue;
2696 		}
2697 		/* fdisa is the Atari/Milan floppy driver */
2698 		if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) {
2699 			continue;
2700 		}
2701 
2702 		/* need to find the device_name_to_block_device_major stuff */
2703 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2704 
2705 		/* get a vnode for the raw partition of this disk */
2706 
2707 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2708 		if (bdevvp(dev, &vp))
2709 			panic("RAID can't alloc vnode");
2710 
2711 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2712 
2713 		if (error) {
2714 			/* "Who cares."  Continue looking
2715 			   for something that exists*/
2716 			vput(vp);
2717 			continue;
2718 		}
2719 
2720 		/* Ok, the disk exists.  Go get the disklabel. */
2721 		error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0);
2722 		if (error) {
2723 			/*
2724 			 * XXX can't happen - open() would
2725 			 * have errored out (or faked up one)
2726 			 */
2727 			printf("can't get label for dev %s%c (%d)!?!?\n",
2728 			       dv->dv_xname, 'a' + RAW_PART, error);
2729 		}
2730 
2731 		/* don't need this any more.  We'll allocate it again
2732 		   a little later if we really do... */
2733 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2734 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2735 		vput(vp);
2736 
2737 		for (i=0; i < label.d_npartitions; i++) {
2738 			/* We only support partitions marked as RAID */
2739 			if (label.d_partitions[i].p_fstype != FS_RAID)
2740 				continue;
2741 
2742 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2743 			if (bdevvp(dev, &vp))
2744 				panic("RAID can't alloc vnode");
2745 
2746 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2747 			if (error) {
2748 				/* Whatever... */
2749 				vput(vp);
2750 				continue;
2751 			}
2752 
2753 			good_one = 0;
2754 
2755 			clabel = (RF_ComponentLabel_t *)
2756 				malloc(sizeof(RF_ComponentLabel_t),
2757 				       M_RAIDFRAME, M_NOWAIT);
2758 			if (clabel == NULL) {
2759 				/* XXX CLEANUP HERE */
2760 				printf("RAID auto config: out of memory!\n");
2761 				return(NULL); /* XXX probably should panic? */
2762 			}
2763 
2764 			if (!raidread_component_label(dev, vp, clabel)) {
2765 				/* Got the label.  Does it look reasonable? */
2766 				if (rf_reasonable_label(clabel) &&
2767 				    (clabel->partitionSize <=
2768 				     label.d_partitions[i].p_size)) {
2769 #if DEBUG
2770 					printf("Component on: %s%c: %d\n",
2771 					       dv->dv_xname, 'a'+i,
2772 					       label.d_partitions[i].p_size);
2773 					rf_print_component_label(clabel);
2774 #endif
2775 					/* if it's reasonable, add it,
2776 					   else ignore it. */
2777 					ac = (RF_AutoConfig_t *)
2778 						malloc(sizeof(RF_AutoConfig_t),
2779 						       M_RAIDFRAME,
2780 						       M_NOWAIT);
2781 					if (ac == NULL) {
2782 						/* XXX should panic?? */
2783 						return(NULL);
2784 					}
2785 
2786 					sprintf(ac->devname, "%s%c",
2787 						dv->dv_xname, 'a'+i);
2788 					ac->dev = dev;
2789 					ac->vp = vp;
2790 					ac->clabel = clabel;
2791 					ac->next = ac_list;
2792 					ac_list = ac;
2793 					good_one = 1;
2794 				}
2795 			}
2796 			if (!good_one) {
2797 				/* cleanup */
2798 				free(clabel, M_RAIDFRAME);
2799 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2800 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2801 				vput(vp);
2802 			}
2803 		}
2804 	}
2805 	return(ac_list);
2806 }
2807 
2808 static int
2809 rf_reasonable_label(clabel)
2810 	RF_ComponentLabel_t *clabel;
2811 {
2812 
2813 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2814 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2815 	    ((clabel->clean == RF_RAID_CLEAN) ||
2816 	     (clabel->clean == RF_RAID_DIRTY)) &&
2817 	    clabel->row >=0 &&
2818 	    clabel->column >= 0 &&
2819 	    clabel->num_rows > 0 &&
2820 	    clabel->num_columns > 0 &&
2821 	    clabel->row < clabel->num_rows &&
2822 	    clabel->column < clabel->num_columns &&
2823 	    clabel->blockSize > 0 &&
2824 	    clabel->numBlocks > 0) {
2825 		/* label looks reasonable enough... */
2826 		return(1);
2827 	}
2828 	return(0);
2829 }
2830 
2831 
2832 #if DEBUG
2833 void
2834 rf_print_component_label(clabel)
2835 	RF_ComponentLabel_t *clabel;
2836 {
2837 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2838 	       clabel->row, clabel->column,
2839 	       clabel->num_rows, clabel->num_columns);
2840 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2841 	       clabel->version, clabel->serial_number,
2842 	       clabel->mod_counter);
2843 	printf("   Clean: %s Status: %d\n",
2844 	       clabel->clean ? "Yes" : "No", clabel->status );
2845 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2846 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2847 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2848 	       (char) clabel->parityConfig, clabel->blockSize,
2849 	       clabel->numBlocks);
2850 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2851 	printf("   Contains root partition: %s\n",
2852 	       clabel->root_partition ? "Yes" : "No" );
2853 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2854 #if 0
2855 	   printf("   Config order: %d\n", clabel->config_order);
2856 #endif
2857 
2858 }
2859 #endif
2860 
2861 RF_ConfigSet_t *
2862 rf_create_auto_sets(ac_list)
2863 	RF_AutoConfig_t *ac_list;
2864 {
2865 	RF_AutoConfig_t *ac;
2866 	RF_ConfigSet_t *config_sets;
2867 	RF_ConfigSet_t *cset;
2868 	RF_AutoConfig_t *ac_next;
2869 
2870 
2871 	config_sets = NULL;
2872 
2873 	/* Go through the AutoConfig list, and figure out which components
2874 	   belong to what sets.  */
2875 	ac = ac_list;
2876 	while(ac!=NULL) {
2877 		/* we're going to putz with ac->next, so save it here
2878 		   for use at the end of the loop */
2879 		ac_next = ac->next;
2880 
2881 		if (config_sets == NULL) {
2882 			/* will need at least this one... */
2883 			config_sets = (RF_ConfigSet_t *)
2884 				malloc(sizeof(RF_ConfigSet_t),
2885 				       M_RAIDFRAME, M_NOWAIT);
2886 			if (config_sets == NULL) {
2887 				panic("rf_create_auto_sets: No memory!");
2888 			}
2889 			/* this one is easy :) */
2890 			config_sets->ac = ac;
2891 			config_sets->next = NULL;
2892 			config_sets->rootable = 0;
2893 			ac->next = NULL;
2894 		} else {
2895 			/* which set does this component fit into? */
2896 			cset = config_sets;
2897 			while(cset!=NULL) {
2898 				if (rf_does_it_fit(cset, ac)) {
2899 					/* looks like it matches... */
2900 					ac->next = cset->ac;
2901 					cset->ac = ac;
2902 					break;
2903 				}
2904 				cset = cset->next;
2905 			}
2906 			if (cset==NULL) {
2907 				/* didn't find a match above... new set..*/
2908 				cset = (RF_ConfigSet_t *)
2909 					malloc(sizeof(RF_ConfigSet_t),
2910 					       M_RAIDFRAME, M_NOWAIT);
2911 				if (cset == NULL) {
2912 					panic("rf_create_auto_sets: No memory!");
2913 				}
2914 				cset->ac = ac;
2915 				ac->next = NULL;
2916 				cset->next = config_sets;
2917 				cset->rootable = 0;
2918 				config_sets = cset;
2919 			}
2920 		}
2921 		ac = ac_next;
2922 	}
2923 
2924 
2925 	return(config_sets);
2926 }
2927 
2928 static int
2929 rf_does_it_fit(cset, ac)
2930 	RF_ConfigSet_t *cset;
2931 	RF_AutoConfig_t *ac;
2932 {
2933 	RF_ComponentLabel_t *clabel1, *clabel2;
2934 
2935 	/* If this one matches the *first* one in the set, that's good
2936 	   enough, since the other members of the set would have been
2937 	   through here too... */
2938 	/* note that we are not checking partitionSize here..
2939 
2940 	   Note that we are also not checking the mod_counters here.
2941 	   If everything else matches execpt the mod_counter, that's
2942 	   good enough for this test.  We will deal with the mod_counters
2943 	   a little later in the autoconfiguration process.
2944 
2945 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2946 
2947 	   The reason we don't check for this is that failed disks
2948 	   will have lower modification counts.  If those disks are
2949 	   not added to the set they used to belong to, then they will
2950 	   form their own set, which may result in 2 different sets,
2951 	   for example, competing to be configured at raid0, and
2952 	   perhaps competing to be the root filesystem set.  If the
2953 	   wrong ones get configured, or both attempt to become /,
2954 	   weird behaviour and or serious lossage will occur.  Thus we
2955 	   need to bring them into the fold here, and kick them out at
2956 	   a later point.
2957 
2958 	*/
2959 
2960 	clabel1 = cset->ac->clabel;
2961 	clabel2 = ac->clabel;
2962 	if ((clabel1->version == clabel2->version) &&
2963 	    (clabel1->serial_number == clabel2->serial_number) &&
2964 	    (clabel1->num_rows == clabel2->num_rows) &&
2965 	    (clabel1->num_columns == clabel2->num_columns) &&
2966 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2967 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2968 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2969 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2970 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2971 	    (clabel1->blockSize == clabel2->blockSize) &&
2972 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2973 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2974 	    (clabel1->root_partition == clabel2->root_partition) &&
2975 	    (clabel1->last_unit == clabel2->last_unit) &&
2976 	    (clabel1->config_order == clabel2->config_order)) {
2977 		/* if it get's here, it almost *has* to be a match */
2978 	} else {
2979 		/* it's not consistent with somebody in the set..
2980 		   punt */
2981 		return(0);
2982 	}
2983 	/* all was fine.. it must fit... */
2984 	return(1);
2985 }
2986 
2987 int
2988 rf_have_enough_components(cset)
2989 	RF_ConfigSet_t *cset;
2990 {
2991 	RF_AutoConfig_t *ac;
2992 	RF_AutoConfig_t *auto_config;
2993 	RF_ComponentLabel_t *clabel;
2994 	int r,c;
2995 	int num_rows;
2996 	int num_cols;
2997 	int num_missing;
2998 	int mod_counter;
2999 	int mod_counter_found;
3000 	int even_pair_failed;
3001 	char parity_type;
3002 
3003 
3004 	/* check to see that we have enough 'live' components
3005 	   of this set.  If so, we can configure it if necessary */
3006 
3007 	num_rows = cset->ac->clabel->num_rows;
3008 	num_cols = cset->ac->clabel->num_columns;
3009 	parity_type = cset->ac->clabel->parityConfig;
3010 
3011 	/* XXX Check for duplicate components!?!?!? */
3012 
3013 	/* Determine what the mod_counter is supposed to be for this set. */
3014 
3015 	mod_counter_found = 0;
3016 	mod_counter = 0;
3017 	ac = cset->ac;
3018 	while(ac!=NULL) {
3019 		if (mod_counter_found==0) {
3020 			mod_counter = ac->clabel->mod_counter;
3021 			mod_counter_found = 1;
3022 		} else {
3023 			if (ac->clabel->mod_counter > mod_counter) {
3024 				mod_counter = ac->clabel->mod_counter;
3025 			}
3026 		}
3027 		ac = ac->next;
3028 	}
3029 
3030 	num_missing = 0;
3031 	auto_config = cset->ac;
3032 
3033 	for(r=0; r<num_rows; r++) {
3034 		even_pair_failed = 0;
3035 		for(c=0; c<num_cols; c++) {
3036 			ac = auto_config;
3037 			while(ac!=NULL) {
3038 				if ((ac->clabel->row == r) &&
3039 				    (ac->clabel->column == c) &&
3040 				    (ac->clabel->mod_counter == mod_counter)) {
3041 					/* it's this one... */
3042 #if DEBUG
3043 					printf("Found: %s at %d,%d\n",
3044 					       ac->devname,r,c);
3045 #endif
3046 					break;
3047 				}
3048 				ac=ac->next;
3049 			}
3050 			if (ac==NULL) {
3051 				/* Didn't find one here! */
3052 				/* special case for RAID 1, especially
3053 				   where there are more than 2
3054 				   components (where RAIDframe treats
3055 				   things a little differently :( ) */
3056 				if (parity_type == '1') {
3057 					if (c%2 == 0) { /* even component */
3058 						even_pair_failed = 1;
3059 					} else { /* odd component.  If
3060                                                     we're failed, and
3061                                                     so is the even
3062                                                     component, it's
3063                                                     "Good Night, Charlie" */
3064 						if (even_pair_failed == 1) {
3065 							return(0);
3066 						}
3067 					}
3068 				} else {
3069 					/* normal accounting */
3070 					num_missing++;
3071 				}
3072 			}
3073 			if ((parity_type == '1') && (c%2 == 1)) {
3074 				/* Just did an even component, and we didn't
3075 				   bail.. reset the even_pair_failed flag,
3076 				   and go on to the next component.... */
3077 				even_pair_failed = 0;
3078 			}
3079 		}
3080 	}
3081 
3082 	clabel = cset->ac->clabel;
3083 
3084 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3085 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3086 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3087 		/* XXX this needs to be made *much* more general */
3088 		/* Too many failures */
3089 		return(0);
3090 	}
3091 	/* otherwise, all is well, and we've got enough to take a kick
3092 	   at autoconfiguring this set */
3093 	return(1);
3094 }
3095 
3096 void
3097 rf_create_configuration(ac,config,raidPtr)
3098 	RF_AutoConfig_t *ac;
3099 	RF_Config_t *config;
3100 	RF_Raid_t *raidPtr;
3101 {
3102 	RF_ComponentLabel_t *clabel;
3103 	int i;
3104 
3105 	clabel = ac->clabel;
3106 
3107 	/* 1. Fill in the common stuff */
3108 	config->numRow = clabel->num_rows;
3109 	config->numCol = clabel->num_columns;
3110 	config->numSpare = 0; /* XXX should this be set here? */
3111 	config->sectPerSU = clabel->sectPerSU;
3112 	config->SUsPerPU = clabel->SUsPerPU;
3113 	config->SUsPerRU = clabel->SUsPerRU;
3114 	config->parityConfig = clabel->parityConfig;
3115 	/* XXX... */
3116 	strcpy(config->diskQueueType,"fifo");
3117 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3118 	config->layoutSpecificSize = 0; /* XXX ?? */
3119 
3120 	while(ac!=NULL) {
3121 		/* row/col values will be in range due to the checks
3122 		   in reasonable_label() */
3123 		strcpy(config->devnames[ac->clabel->row][ac->clabel->column],
3124 		       ac->devname);
3125 		ac = ac->next;
3126 	}
3127 
3128 	for(i=0;i<RF_MAXDBGV;i++) {
3129 		config->debugVars[i][0] = NULL;
3130 	}
3131 }
3132 
3133 int
3134 rf_set_autoconfig(raidPtr, new_value)
3135 	RF_Raid_t *raidPtr;
3136 	int new_value;
3137 {
3138 	RF_ComponentLabel_t clabel;
3139 	struct vnode *vp;
3140 	dev_t dev;
3141 	int row, column;
3142 	int sparecol;
3143 
3144 	raidPtr->autoconfigure = new_value;
3145 	for(row=0; row<raidPtr->numRow; row++) {
3146 		for(column=0; column<raidPtr->numCol; column++) {
3147 			if (raidPtr->Disks[row][column].status ==
3148 			    rf_ds_optimal) {
3149 				dev = raidPtr->Disks[row][column].dev;
3150 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3151 				raidread_component_label(dev, vp, &clabel);
3152 				clabel.autoconfigure = new_value;
3153 				raidwrite_component_label(dev, vp, &clabel);
3154 			}
3155 		}
3156 	}
3157 	for(column = 0; column < raidPtr->numSpare ; column++) {
3158 		sparecol = raidPtr->numCol + column;
3159 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
3160 			dev = raidPtr->Disks[0][sparecol].dev;
3161 			vp = raidPtr->raid_cinfo[0][sparecol].ci_vp;
3162 			raidread_component_label(dev, vp, &clabel);
3163 			clabel.autoconfigure = new_value;
3164 			raidwrite_component_label(dev, vp, &clabel);
3165 		}
3166 	}
3167 	return(new_value);
3168 }
3169 
3170 int
3171 rf_set_rootpartition(raidPtr, new_value)
3172 	RF_Raid_t *raidPtr;
3173 	int new_value;
3174 {
3175 	RF_ComponentLabel_t clabel;
3176 	struct vnode *vp;
3177 	dev_t dev;
3178 	int row, column;
3179 	int sparecol;
3180 
3181 	raidPtr->root_partition = new_value;
3182 	for(row=0; row<raidPtr->numRow; row++) {
3183 		for(column=0; column<raidPtr->numCol; column++) {
3184 			if (raidPtr->Disks[row][column].status ==
3185 			    rf_ds_optimal) {
3186 				dev = raidPtr->Disks[row][column].dev;
3187 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3188 				raidread_component_label(dev, vp, &clabel);
3189 				clabel.root_partition = new_value;
3190 				raidwrite_component_label(dev, vp, &clabel);
3191 			}
3192 		}
3193 	}
3194 	for(column = 0; column < raidPtr->numSpare ; column++) {
3195 		sparecol = raidPtr->numCol + column;
3196 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
3197 			dev = raidPtr->Disks[0][sparecol].dev;
3198 			vp = raidPtr->raid_cinfo[0][sparecol].ci_vp;
3199 			raidread_component_label(dev, vp, &clabel);
3200 			clabel.root_partition = new_value;
3201 			raidwrite_component_label(dev, vp, &clabel);
3202 		}
3203 	}
3204 	return(new_value);
3205 }
3206 
3207 void
3208 rf_release_all_vps(cset)
3209 	RF_ConfigSet_t *cset;
3210 {
3211 	RF_AutoConfig_t *ac;
3212 
3213 	ac = cset->ac;
3214 	while(ac!=NULL) {
3215 		/* Close the vp, and give it back */
3216 		if (ac->vp) {
3217 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3218 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3219 			vput(ac->vp);
3220 			ac->vp = NULL;
3221 		}
3222 		ac = ac->next;
3223 	}
3224 }
3225 
3226 
3227 void
3228 rf_cleanup_config_set(cset)
3229 	RF_ConfigSet_t *cset;
3230 {
3231 	RF_AutoConfig_t *ac;
3232 	RF_AutoConfig_t *next_ac;
3233 
3234 	ac = cset->ac;
3235 	while(ac!=NULL) {
3236 		next_ac = ac->next;
3237 		/* nuke the label */
3238 		free(ac->clabel, M_RAIDFRAME);
3239 		/* cleanup the config structure */
3240 		free(ac, M_RAIDFRAME);
3241 		/* "next.." */
3242 		ac = next_ac;
3243 	}
3244 	/* and, finally, nuke the config set */
3245 	free(cset, M_RAIDFRAME);
3246 }
3247 
3248 
3249 void
3250 raid_init_component_label(raidPtr, clabel)
3251 	RF_Raid_t *raidPtr;
3252 	RF_ComponentLabel_t *clabel;
3253 {
3254 	/* current version number */
3255 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3256 	clabel->serial_number = raidPtr->serial_number;
3257 	clabel->mod_counter = raidPtr->mod_counter;
3258 	clabel->num_rows = raidPtr->numRow;
3259 	clabel->num_columns = raidPtr->numCol;
3260 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3261 	clabel->status = rf_ds_optimal; /* "It's good!" */
3262 
3263 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3264 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3265 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3266 
3267 	clabel->blockSize = raidPtr->bytesPerSector;
3268 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3269 
3270 	/* XXX not portable */
3271 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3272 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3273 	clabel->autoconfigure = raidPtr->autoconfigure;
3274 	clabel->root_partition = raidPtr->root_partition;
3275 	clabel->last_unit = raidPtr->raidid;
3276 	clabel->config_order = raidPtr->config_order;
3277 }
3278 
3279 int
3280 rf_auto_config_set(cset,unit)
3281 	RF_ConfigSet_t *cset;
3282 	int *unit;
3283 {
3284 	RF_Raid_t *raidPtr;
3285 	RF_Config_t *config;
3286 	int raidID;
3287 	int retcode;
3288 
3289 #if DEBUG
3290 	printf("RAID autoconfigure\n");
3291 #endif
3292 
3293 	retcode = 0;
3294 	*unit = -1;
3295 
3296 	/* 1. Create a config structure */
3297 
3298 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3299 				       M_RAIDFRAME,
3300 				       M_NOWAIT);
3301 	if (config==NULL) {
3302 		printf("Out of mem!?!?\n");
3303 				/* XXX do something more intelligent here. */
3304 		return(1);
3305 	}
3306 
3307 	memset(config, 0, sizeof(RF_Config_t));
3308 
3309 	/*
3310 	   2. Figure out what RAID ID this one is supposed to live at
3311 	   See if we can get the same RAID dev that it was configured
3312 	   on last time..
3313 	*/
3314 
3315 	raidID = cset->ac->clabel->last_unit;
3316 	if ((raidID < 0) || (raidID >= numraid)) {
3317 		/* let's not wander off into lala land. */
3318 		raidID = numraid - 1;
3319 	}
3320 	if (raidPtrs[raidID]->valid != 0) {
3321 
3322 		/*
3323 		   Nope... Go looking for an alternative...
3324 		   Start high so we don't immediately use raid0 if that's
3325 		   not taken.
3326 		*/
3327 
3328 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3329 			if (raidPtrs[raidID]->valid == 0) {
3330 				/* can use this one! */
3331 				break;
3332 			}
3333 		}
3334 	}
3335 
3336 	if (raidID < 0) {
3337 		/* punt... */
3338 		printf("Unable to auto configure this set!\n");
3339 		printf("(Out of RAID devs!)\n");
3340 		return(1);
3341 	}
3342 
3343 #if DEBUG
3344 	printf("Configuring raid%d:\n",raidID);
3345 #endif
3346 
3347 	raidPtr = raidPtrs[raidID];
3348 
3349 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3350 	raidPtr->raidid = raidID;
3351 	raidPtr->openings = RAIDOUTSTANDING;
3352 
3353 	/* 3. Build the configuration structure */
3354 	rf_create_configuration(cset->ac, config, raidPtr);
3355 
3356 	/* 4. Do the configuration */
3357 	retcode = rf_Configure(raidPtr, config, cset->ac);
3358 
3359 	if (retcode == 0) {
3360 
3361 		raidinit(raidPtrs[raidID]);
3362 
3363 		rf_markalldirty(raidPtrs[raidID]);
3364 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3365 		if (cset->ac->clabel->root_partition==1) {
3366 			/* everything configured just fine.  Make a note
3367 			   that this set is eligible to be root. */
3368 			cset->rootable = 1;
3369 			/* XXX do this here? */
3370 			raidPtrs[raidID]->root_partition = 1;
3371 		}
3372 	}
3373 
3374 	/* 5. Cleanup */
3375 	free(config, M_RAIDFRAME);
3376 
3377 	*unit = raidID;
3378 	return(retcode);
3379 }
3380 
3381 void
3382 rf_disk_unbusy(desc)
3383 	RF_RaidAccessDesc_t *desc;
3384 {
3385 	struct buf *bp;
3386 
3387 	bp = (struct buf *)desc->bp;
3388 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3389 	    (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ));
3390 }
3391