xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.130 2002/09/06 13:18:43 gehenna 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 
82 
83 /*
84  * Copyright (c) 1995 Carnegie-Mellon University.
85  * All rights reserved.
86  *
87  * Authors: Mark Holland, Jim Zelenka
88  *
89  * Permission to use, copy, modify and distribute this software and
90  * its documentation is hereby granted, provided that both the copyright
91  * notice and this permission notice appear in all copies of the
92  * software, derivative works or modified versions, and any portions
93  * thereof, and that both notices appear in supporting documentation.
94  *
95  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
96  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
97  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
98  *
99  * Carnegie Mellon requests users of this software to return to
100  *
101  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
102  *  School of Computer Science
103  *  Carnegie Mellon University
104  *  Pittsburgh PA 15213-3890
105  *
106  * any improvements or extensions that they make and grant Carnegie the
107  * rights to redistribute these changes.
108  */
109 
110 /***********************************************************
111  *
112  * rf_kintf.c -- the kernel interface routines for RAIDframe
113  *
114  ***********************************************************/
115 
116 #include <sys/cdefs.h>
117 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.130 2002/09/06 13:18:43 gehenna Exp $");
118 
119 #include <sys/param.h>
120 #include <sys/errno.h>
121 #include <sys/pool.h>
122 #include <sys/queue.h>
123 #include <sys/disk.h>
124 #include <sys/device.h>
125 #include <sys/stat.h>
126 #include <sys/ioctl.h>
127 #include <sys/fcntl.h>
128 #include <sys/systm.h>
129 #include <sys/namei.h>
130 #include <sys/vnode.h>
131 #include <sys/disklabel.h>
132 #include <sys/conf.h>
133 #include <sys/lock.h>
134 #include <sys/buf.h>
135 #include <sys/user.h>
136 #include <sys/reboot.h>
137 
138 #include <dev/raidframe/raidframevar.h>
139 #include <dev/raidframe/raidframeio.h>
140 #include "raid.h"
141 #include "opt_raid_autoconfig.h"
142 #include "rf_raid.h"
143 #include "rf_copyback.h"
144 #include "rf_dag.h"
145 #include "rf_dagflags.h"
146 #include "rf_desc.h"
147 #include "rf_diskqueue.h"
148 #include "rf_etimer.h"
149 #include "rf_general.h"
150 #include "rf_kintf.h"
151 #include "rf_options.h"
152 #include "rf_driver.h"
153 #include "rf_parityscan.h"
154 #include "rf_threadstuff.h"
155 
156 int     rf_kdebug_level = 0;
157 
158 #ifdef DEBUG
159 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
160 #else				/* DEBUG */
161 #define db1_printf(a) { }
162 #endif				/* DEBUG */
163 
164 static RF_Raid_t **raidPtrs;	/* global raid device descriptors */
165 
166 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
167 
168 static RF_SparetWait_t *rf_sparet_wait_queue;	/* requests to install a
169 						 * spare table */
170 static RF_SparetWait_t *rf_sparet_resp_queue;	/* responses from
171 						 * installation process */
172 
173 /* prototypes */
174 static void KernelWakeupFunc(struct buf * bp);
175 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag,
176 		   dev_t dev, RF_SectorNum_t startSect,
177 		   RF_SectorCount_t numSect, caddr_t buf,
178 		   void (*cbFunc) (struct buf *), void *cbArg,
179 		   int logBytesPerSector, struct proc * b_proc);
180 static void raidinit(RF_Raid_t *);
181 
182 void raidattach(int);
183 
184 dev_type_open(raidopen);
185 dev_type_close(raidclose);
186 dev_type_read(raidread);
187 dev_type_write(raidwrite);
188 dev_type_ioctl(raidioctl);
189 dev_type_strategy(raidstrategy);
190 dev_type_dump(raiddump);
191 dev_type_size(raidsize);
192 
193 const struct bdevsw raid_bdevsw = {
194 	raidopen, raidclose, raidstrategy, raidioctl,
195 	raiddump, raidsize, D_DISK
196 };
197 
198 const struct cdevsw raid_cdevsw = {
199 	raidopen, raidclose, raidread, raidwrite, raidioctl,
200 	nostop, notty, nopoll, nommap, D_DISK
201 };
202 
203 /*
204  * Pilfered from ccd.c
205  */
206 
207 struct raidbuf {
208 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
209 	struct buf *rf_obp;	/* ptr. to original I/O buf */
210 	int     rf_flags;	/* misc. flags */
211 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
212 };
213 
214 /* component buffer pool */
215 struct pool raidframe_cbufpool;
216 
217 #define RAIDGETBUF(rs) pool_get(&raidframe_cbufpool, PR_NOWAIT)
218 #define	RAIDPUTBUF(rs, cbp) pool_put(&raidframe_cbufpool, cbp)
219 
220 /* XXX Not sure if the following should be replacing the raidPtrs above,
221    or if it should be used in conjunction with that...
222 */
223 
224 struct raid_softc {
225 	int     sc_flags;	/* flags */
226 	int     sc_cflags;	/* configuration flags */
227 	size_t  sc_size;        /* size of the raid device */
228 	char    sc_xname[20];	/* XXX external name */
229 	struct disk sc_dkdev;	/* generic disk device info */
230 	struct bufq_state buf_queue;	/* used for the device queue */
231 };
232 /* sc_flags */
233 #define RAIDF_INITED	0x01	/* unit has been initialized */
234 #define RAIDF_WLABEL	0x02	/* label area is writable */
235 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
236 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
237 #define RAIDF_LOCKED	0x80	/* unit is locked */
238 
239 #define	raidunit(x)	DISKUNIT(x)
240 int numraid = 0;
241 
242 /*
243  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
244  * Be aware that large numbers can allow the driver to consume a lot of
245  * kernel memory, especially on writes, and in degraded mode reads.
246  *
247  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
248  * a single 64K write will typically require 64K for the old data,
249  * 64K for the old parity, and 64K for the new parity, for a total
250  * of 192K (if the parity buffer is not re-used immediately).
251  * Even it if is used immediately, that's still 128K, which when multiplied
252  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
253  *
254  * Now in degraded mode, for example, a 64K read on the above setup may
255  * require data reconstruction, which will require *all* of the 4 remaining
256  * disks to participate -- 4 * 32K/disk == 128K again.
257  */
258 
259 #ifndef RAIDOUTSTANDING
260 #define RAIDOUTSTANDING   6
261 #endif
262 
263 #define RAIDLABELDEV(dev)	\
264 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
265 
266 /* declared here, and made public, for the benefit of KVM stuff.. */
267 struct raid_softc *raid_softc;
268 
269 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
270 				     struct disklabel *);
271 static void raidgetdisklabel(dev_t);
272 static void raidmakedisklabel(struct raid_softc *);
273 
274 static int raidlock(struct raid_softc *);
275 static void raidunlock(struct raid_softc *);
276 
277 static void rf_markalldirty(RF_Raid_t *);
278 void rf_mountroot_hook(struct device *);
279 
280 struct device *raidrootdev;
281 
282 void rf_ReconThread(struct rf_recon_req *);
283 /* XXX what I want is: */
284 /*void rf_ReconThread(RF_Raid_t *raidPtr);  */
285 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
286 void rf_CopybackThread(RF_Raid_t *raidPtr);
287 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
288 void rf_buildroothack(void *);
289 
290 RF_AutoConfig_t *rf_find_raid_components(void);
291 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
292 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
293 static int rf_reasonable_label(RF_ComponentLabel_t *);
294 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
295 int rf_set_autoconfig(RF_Raid_t *, int);
296 int rf_set_rootpartition(RF_Raid_t *, int);
297 void rf_release_all_vps(RF_ConfigSet_t *);
298 void rf_cleanup_config_set(RF_ConfigSet_t *);
299 int rf_have_enough_components(RF_ConfigSet_t *);
300 int rf_auto_config_set(RF_ConfigSet_t *, int *);
301 
302 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
303 				  allow autoconfig to take place.
304 			          Note that this is overridden by having
305 			          RAID_AUTOCONFIG as an option in the
306 			          kernel config file.  */
307 
308 void
309 raidattach(num)
310 	int     num;
311 {
312 	int raidID;
313 	int i, rc;
314 	RF_AutoConfig_t *ac_list; /* autoconfig list */
315 	RF_ConfigSet_t *config_sets;
316 
317 #ifdef DEBUG
318 	printf("raidattach: Asked for %d units\n", num);
319 #endif
320 
321 	if (num <= 0) {
322 #ifdef DIAGNOSTIC
323 		panic("raidattach: count <= 0");
324 #endif
325 		return;
326 	}
327 	/* This is where all the initialization stuff gets done. */
328 
329 	numraid = num;
330 
331 	/* Make some space for requested number of units... */
332 
333 	RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **));
334 	if (raidPtrs == NULL) {
335 		panic("raidPtrs is NULL!!\n");
336 	}
337 
338 	/* Initialize the component buffer pool. */
339 	pool_init(&raidframe_cbufpool, sizeof(struct raidbuf), 0,
340 	    0, 0, "raidpl", NULL);
341 
342 	rc = rf_mutex_init(&rf_sparet_wait_mutex);
343 	if (rc) {
344 		RF_PANIC();
345 	}
346 
347 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
348 
349 	for (i = 0; i < num; i++)
350 		raidPtrs[i] = NULL;
351 	rc = rf_BootRaidframe();
352 	if (rc == 0)
353 		printf("Kernelized RAIDframe activated\n");
354 	else
355 		panic("Serious error booting RAID!!\n");
356 
357 	/* put together some datastructures like the CCD device does.. This
358 	 * lets us lock the device and what-not when it gets opened. */
359 
360 	raid_softc = (struct raid_softc *)
361 		malloc(num * sizeof(struct raid_softc),
362 		       M_RAIDFRAME, M_NOWAIT);
363 	if (raid_softc == NULL) {
364 		printf("WARNING: no memory for RAIDframe driver\n");
365 		return;
366 	}
367 
368 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
369 
370 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
371 					      M_RAIDFRAME, M_NOWAIT);
372 	if (raidrootdev == NULL) {
373 		panic("No memory for RAIDframe driver!!?!?!\n");
374 	}
375 
376 	for (raidID = 0; raidID < num; raidID++) {
377 		bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_FCFS);
378 
379 		raidrootdev[raidID].dv_class  = DV_DISK;
380 		raidrootdev[raidID].dv_cfdata = NULL;
381 		raidrootdev[raidID].dv_unit   = raidID;
382 		raidrootdev[raidID].dv_parent = NULL;
383 		raidrootdev[raidID].dv_flags  = 0;
384 		sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID);
385 
386 		RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t),
387 			  (RF_Raid_t *));
388 		if (raidPtrs[raidID] == NULL) {
389 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
390 			numraid = raidID;
391 			return;
392 		}
393 	}
394 
395 #ifdef RAID_AUTOCONFIG
396 	raidautoconfig = 1;
397 #endif
398 
399 if (raidautoconfig) {
400 	/* 1. locate all RAID components on the system */
401 
402 #if DEBUG
403 	printf("Searching for raid components...\n");
404 #endif
405 	ac_list = rf_find_raid_components();
406 
407 	/* 2. sort them into their respective sets */
408 
409 	config_sets = rf_create_auto_sets(ac_list);
410 
411 	/* 3. evaluate each set and configure the valid ones
412 	   This gets done in rf_buildroothack() */
413 
414 	/* schedule the creation of the thread to do the
415 	   "/ on RAID" stuff */
416 
417 	kthread_create(rf_buildroothack,config_sets);
418 
419 #if 0
420 	mountroothook_establish(rf_mountroot_hook, &raidrootdev[0]);
421 #endif
422 }
423 
424 }
425 
426 void
427 rf_buildroothack(arg)
428 	void *arg;
429 {
430 	RF_ConfigSet_t *config_sets = arg;
431 	RF_ConfigSet_t *cset;
432 	RF_ConfigSet_t *next_cset;
433 	int retcode;
434 	int raidID;
435 	int rootID;
436 	int num_root;
437 
438 	rootID = 0;
439 	num_root = 0;
440 	cset = config_sets;
441 	while(cset != NULL ) {
442 		next_cset = cset->next;
443 		if (rf_have_enough_components(cset) &&
444 		    cset->ac->clabel->autoconfigure==1) {
445 			retcode = rf_auto_config_set(cset,&raidID);
446 			if (!retcode) {
447 				if (cset->rootable) {
448 					rootID = raidID;
449 					num_root++;
450 				}
451 			} else {
452 				/* The autoconfig didn't work :( */
453 #if DEBUG
454 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
455 #endif
456 				rf_release_all_vps(cset);
457 			}
458 		} else {
459 			/* we're not autoconfiguring this set...
460 			   release the associated resources */
461 			rf_release_all_vps(cset);
462 		}
463 		/* cleanup */
464 		rf_cleanup_config_set(cset);
465 		cset = next_cset;
466 	}
467 
468 	/* we found something bootable... */
469 
470 	if (num_root == 1) {
471 		booted_device = &raidrootdev[rootID];
472 	} else if (num_root > 1) {
473 		/* we can't guess.. require the user to answer... */
474 		boothowto |= RB_ASKNAME;
475 	}
476 }
477 
478 
479 int
480 raidsize(dev)
481 	dev_t   dev;
482 {
483 	struct raid_softc *rs;
484 	struct disklabel *lp;
485 	int     part, unit, omask, size;
486 
487 	unit = raidunit(dev);
488 	if (unit >= numraid)
489 		return (-1);
490 	rs = &raid_softc[unit];
491 
492 	if ((rs->sc_flags & RAIDF_INITED) == 0)
493 		return (-1);
494 
495 	part = DISKPART(dev);
496 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
497 	lp = rs->sc_dkdev.dk_label;
498 
499 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
500 		return (-1);
501 
502 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
503 		size = -1;
504 	else
505 		size = lp->d_partitions[part].p_size *
506 		    (lp->d_secsize / DEV_BSIZE);
507 
508 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
509 		return (-1);
510 
511 	return (size);
512 
513 }
514 
515 int
516 raiddump(dev, blkno, va, size)
517 	dev_t   dev;
518 	daddr_t blkno;
519 	caddr_t va;
520 	size_t  size;
521 {
522 	/* Not implemented. */
523 	return ENXIO;
524 }
525 /* ARGSUSED */
526 int
527 raidopen(dev, flags, fmt, p)
528 	dev_t   dev;
529 	int     flags, fmt;
530 	struct proc *p;
531 {
532 	int     unit = raidunit(dev);
533 	struct raid_softc *rs;
534 	struct disklabel *lp;
535 	int     part, pmask;
536 	int     error = 0;
537 
538 	if (unit >= numraid)
539 		return (ENXIO);
540 	rs = &raid_softc[unit];
541 
542 	if ((error = raidlock(rs)) != 0)
543 		return (error);
544 	lp = rs->sc_dkdev.dk_label;
545 
546 	part = DISKPART(dev);
547 	pmask = (1 << part);
548 
549 	db1_printf(("Opening raid device number: %d partition: %d\n",
550 		unit, part));
551 
552 
553 	if ((rs->sc_flags & RAIDF_INITED) &&
554 	    (rs->sc_dkdev.dk_openmask == 0))
555 		raidgetdisklabel(dev);
556 
557 	/* make sure that this partition exists */
558 
559 	if (part != RAW_PART) {
560 		db1_printf(("Not a raw partition..\n"));
561 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
562 		    ((part >= lp->d_npartitions) ||
563 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
564 			error = ENXIO;
565 			raidunlock(rs);
566 			db1_printf(("Bailing out...\n"));
567 			return (error);
568 		}
569 	}
570 	/* Prevent this unit from being unconfigured while open. */
571 	switch (fmt) {
572 	case S_IFCHR:
573 		rs->sc_dkdev.dk_copenmask |= pmask;
574 		break;
575 
576 	case S_IFBLK:
577 		rs->sc_dkdev.dk_bopenmask |= pmask;
578 		break;
579 	}
580 
581 	if ((rs->sc_dkdev.dk_openmask == 0) &&
582 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
583 		/* First one... mark things as dirty... Note that we *MUST*
584 		 have done a configure before this.  I DO NOT WANT TO BE
585 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
586 		 THAT THEY BELONG TOGETHER!!!!! */
587 		/* XXX should check to see if we're only open for reading
588 		   here... If so, we needn't do this, but then need some
589 		   other way of keeping track of what's happened.. */
590 
591 		rf_markalldirty( raidPtrs[unit] );
592 	}
593 
594 
595 	rs->sc_dkdev.dk_openmask =
596 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
597 
598 	raidunlock(rs);
599 
600 	return (error);
601 
602 
603 }
604 /* ARGSUSED */
605 int
606 raidclose(dev, flags, fmt, p)
607 	dev_t   dev;
608 	int     flags, fmt;
609 	struct proc *p;
610 {
611 	int     unit = raidunit(dev);
612 	struct raid_softc *rs;
613 	int     error = 0;
614 	int     part;
615 
616 	if (unit >= numraid)
617 		return (ENXIO);
618 	rs = &raid_softc[unit];
619 
620 	if ((error = raidlock(rs)) != 0)
621 		return (error);
622 
623 	part = DISKPART(dev);
624 
625 	/* ...that much closer to allowing unconfiguration... */
626 	switch (fmt) {
627 	case S_IFCHR:
628 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
629 		break;
630 
631 	case S_IFBLK:
632 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
633 		break;
634 	}
635 	rs->sc_dkdev.dk_openmask =
636 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
637 
638 	if ((rs->sc_dkdev.dk_openmask == 0) &&
639 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
640 		/* Last one... device is not unconfigured yet.
641 		   Device shutdown has taken care of setting the
642 		   clean bits if RAIDF_INITED is not set
643 		   mark things as clean... */
644 #if 0
645 		printf("Last one on raid%d.  Updating status.\n",unit);
646 #endif
647 		rf_update_component_labels(raidPtrs[unit],
648 						 RF_FINAL_COMPONENT_UPDATE);
649 		if (doing_shutdown) {
650 			/* last one, and we're going down, so
651 			   lights out for this RAID set too. */
652 			error = rf_Shutdown(raidPtrs[unit]);
653 
654 			/* It's no longer initialized... */
655 			rs->sc_flags &= ~RAIDF_INITED;
656 
657 			/* Detach the disk. */
658 			disk_detach(&rs->sc_dkdev);
659 		}
660 	}
661 
662 	raidunlock(rs);
663 	return (0);
664 
665 }
666 
667 void
668 raidstrategy(bp)
669 	struct buf *bp;
670 {
671 	int s;
672 
673 	unsigned int raidID = raidunit(bp->b_dev);
674 	RF_Raid_t *raidPtr;
675 	struct raid_softc *rs = &raid_softc[raidID];
676 	struct disklabel *lp;
677 	int     wlabel;
678 
679 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
680 		bp->b_error = ENXIO;
681 		bp->b_flags |= B_ERROR;
682 		bp->b_resid = bp->b_bcount;
683 		biodone(bp);
684 		return;
685 	}
686 	if (raidID >= numraid || !raidPtrs[raidID]) {
687 		bp->b_error = ENODEV;
688 		bp->b_flags |= B_ERROR;
689 		bp->b_resid = bp->b_bcount;
690 		biodone(bp);
691 		return;
692 	}
693 	raidPtr = raidPtrs[raidID];
694 	if (!raidPtr->valid) {
695 		bp->b_error = ENODEV;
696 		bp->b_flags |= B_ERROR;
697 		bp->b_resid = bp->b_bcount;
698 		biodone(bp);
699 		return;
700 	}
701 	if (bp->b_bcount == 0) {
702 		db1_printf(("b_bcount is zero..\n"));
703 		biodone(bp);
704 		return;
705 	}
706 	lp = rs->sc_dkdev.dk_label;
707 
708 	/*
709 	 * Do bounds checking and adjust transfer.  If there's an
710 	 * error, the bounds check will flag that for us.
711 	 */
712 
713 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
714 	if (DISKPART(bp->b_dev) != RAW_PART)
715 		if (bounds_check_with_label(bp, lp, wlabel) <= 0) {
716 			db1_printf(("Bounds check failed!!:%d %d\n",
717 				(int) bp->b_blkno, (int) wlabel));
718 			biodone(bp);
719 			return;
720 		}
721 	s = splbio();
722 
723 	bp->b_resid = 0;
724 
725 	/* stuff it onto our queue */
726 	BUFQ_PUT(&rs->buf_queue, bp);
727 
728 	raidstart(raidPtrs[raidID]);
729 
730 	splx(s);
731 }
732 /* ARGSUSED */
733 int
734 raidread(dev, uio, flags)
735 	dev_t   dev;
736 	struct uio *uio;
737 	int     flags;
738 {
739 	int     unit = raidunit(dev);
740 	struct raid_softc *rs;
741 	int     part;
742 
743 	if (unit >= numraid)
744 		return (ENXIO);
745 	rs = &raid_softc[unit];
746 
747 	if ((rs->sc_flags & RAIDF_INITED) == 0)
748 		return (ENXIO);
749 	part = DISKPART(dev);
750 
751 	db1_printf(("raidread: unit: %d partition: %d\n", unit, part));
752 
753 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
754 
755 }
756 /* ARGSUSED */
757 int
758 raidwrite(dev, uio, flags)
759 	dev_t   dev;
760 	struct uio *uio;
761 	int     flags;
762 {
763 	int     unit = raidunit(dev);
764 	struct raid_softc *rs;
765 
766 	if (unit >= numraid)
767 		return (ENXIO);
768 	rs = &raid_softc[unit];
769 
770 	if ((rs->sc_flags & RAIDF_INITED) == 0)
771 		return (ENXIO);
772 	db1_printf(("raidwrite\n"));
773 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
774 
775 }
776 
777 int
778 raidioctl(dev, cmd, data, flag, p)
779 	dev_t   dev;
780 	u_long  cmd;
781 	caddr_t data;
782 	int     flag;
783 	struct proc *p;
784 {
785 	int     unit = raidunit(dev);
786 	int     error = 0;
787 	int     part, pmask;
788 	struct raid_softc *rs;
789 	RF_Config_t *k_cfg, *u_cfg;
790 	RF_Raid_t *raidPtr;
791 	RF_RaidDisk_t *diskPtr;
792 	RF_AccTotals_t *totals;
793 	RF_DeviceConfig_t *d_cfg, **ucfgp;
794 	u_char *specific_buf;
795 	int retcode = 0;
796 	int row;
797 	int column;
798 	int raidid;
799 	struct rf_recon_req *rrcopy, *rr;
800 	RF_ComponentLabel_t *clabel;
801 	RF_ComponentLabel_t ci_label;
802 	RF_ComponentLabel_t **clabel_ptr;
803 	RF_SingleComponent_t *sparePtr,*componentPtr;
804 	RF_SingleComponent_t hot_spare;
805 	RF_SingleComponent_t component;
806 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
807 	int i, j, d;
808 #ifdef __HAVE_OLD_DISKLABEL
809 	struct disklabel newlabel;
810 #endif
811 
812 	if (unit >= numraid)
813 		return (ENXIO);
814 	rs = &raid_softc[unit];
815 	raidPtr = raidPtrs[unit];
816 
817 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
818 		(int) DISKPART(dev), (int) unit, (int) cmd));
819 
820 	/* Must be open for writes for these commands... */
821 	switch (cmd) {
822 	case DIOCSDINFO:
823 	case DIOCWDINFO:
824 #ifdef __HAVE_OLD_DISKLABEL
825 	case ODIOCWDINFO:
826 	case ODIOCSDINFO:
827 #endif
828 	case DIOCWLABEL:
829 		if ((flag & FWRITE) == 0)
830 			return (EBADF);
831 	}
832 
833 	/* Must be initialized for these... */
834 	switch (cmd) {
835 	case DIOCGDINFO:
836 	case DIOCSDINFO:
837 	case DIOCWDINFO:
838 #ifdef __HAVE_OLD_DISKLABEL
839 	case ODIOCGDINFO:
840 	case ODIOCWDINFO:
841 	case ODIOCSDINFO:
842 	case ODIOCGDEFLABEL:
843 #endif
844 	case DIOCGPART:
845 	case DIOCWLABEL:
846 	case DIOCGDEFLABEL:
847 	case RAIDFRAME_SHUTDOWN:
848 	case RAIDFRAME_REWRITEPARITY:
849 	case RAIDFRAME_GET_INFO:
850 	case RAIDFRAME_RESET_ACCTOTALS:
851 	case RAIDFRAME_GET_ACCTOTALS:
852 	case RAIDFRAME_KEEP_ACCTOTALS:
853 	case RAIDFRAME_GET_SIZE:
854 	case RAIDFRAME_FAIL_DISK:
855 	case RAIDFRAME_COPYBACK:
856 	case RAIDFRAME_CHECK_RECON_STATUS:
857 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
858 	case RAIDFRAME_GET_COMPONENT_LABEL:
859 	case RAIDFRAME_SET_COMPONENT_LABEL:
860 	case RAIDFRAME_ADD_HOT_SPARE:
861 	case RAIDFRAME_REMOVE_HOT_SPARE:
862 	case RAIDFRAME_INIT_LABELS:
863 	case RAIDFRAME_REBUILD_IN_PLACE:
864 	case RAIDFRAME_CHECK_PARITY:
865 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
866 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
867 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
868 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
869 	case RAIDFRAME_SET_AUTOCONFIG:
870 	case RAIDFRAME_SET_ROOT:
871 	case RAIDFRAME_DELETE_COMPONENT:
872 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
873 		if ((rs->sc_flags & RAIDF_INITED) == 0)
874 			return (ENXIO);
875 	}
876 
877 	switch (cmd) {
878 
879 		/* configure the system */
880 	case RAIDFRAME_CONFIGURE:
881 
882 		if (raidPtr->valid) {
883 			/* There is a valid RAID set running on this unit! */
884 			printf("raid%d: Device already configured!\n",unit);
885 			return(EINVAL);
886 		}
887 
888 		/* copy-in the configuration information */
889 		/* data points to a pointer to the configuration structure */
890 
891 		u_cfg = *((RF_Config_t **) data);
892 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
893 		if (k_cfg == NULL) {
894 			return (ENOMEM);
895 		}
896 		retcode = copyin((caddr_t) u_cfg, (caddr_t) k_cfg,
897 		    sizeof(RF_Config_t));
898 		if (retcode) {
899 			RF_Free(k_cfg, sizeof(RF_Config_t));
900 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
901 				retcode));
902 			return (retcode);
903 		}
904 		/* allocate a buffer for the layout-specific data, and copy it
905 		 * in */
906 		if (k_cfg->layoutSpecificSize) {
907 			if (k_cfg->layoutSpecificSize > 10000) {
908 				/* sanity check */
909 				RF_Free(k_cfg, sizeof(RF_Config_t));
910 				return (EINVAL);
911 			}
912 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
913 			    (u_char *));
914 			if (specific_buf == NULL) {
915 				RF_Free(k_cfg, sizeof(RF_Config_t));
916 				return (ENOMEM);
917 			}
918 			retcode = copyin(k_cfg->layoutSpecific,
919 			    (caddr_t) specific_buf,
920 			    k_cfg->layoutSpecificSize);
921 			if (retcode) {
922 				RF_Free(k_cfg, sizeof(RF_Config_t));
923 				RF_Free(specific_buf,
924 					k_cfg->layoutSpecificSize);
925 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
926 					retcode));
927 				return (retcode);
928 			}
929 		} else
930 			specific_buf = NULL;
931 		k_cfg->layoutSpecific = specific_buf;
932 
933 		/* should do some kind of sanity check on the configuration.
934 		 * Store the sum of all the bytes in the last byte? */
935 
936 		/* configure the system */
937 
938 		/*
939 		 * Clear the entire RAID descriptor, just to make sure
940 		 *  there is no stale data left in the case of a
941 		 *  reconfiguration
942 		 */
943 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
944 		raidPtr->raidid = unit;
945 
946 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
947 
948 		if (retcode == 0) {
949 
950 			/* allow this many simultaneous IO's to
951 			   this RAID device */
952 			raidPtr->openings = RAIDOUTSTANDING;
953 
954 			raidinit(raidPtr);
955 			rf_markalldirty(raidPtr);
956 		}
957 		/* free the buffers.  No return code here. */
958 		if (k_cfg->layoutSpecificSize) {
959 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
960 		}
961 		RF_Free(k_cfg, sizeof(RF_Config_t));
962 
963 		return (retcode);
964 
965 		/* shutdown the system */
966 	case RAIDFRAME_SHUTDOWN:
967 
968 		if ((error = raidlock(rs)) != 0)
969 			return (error);
970 
971 		/*
972 		 * If somebody has a partition mounted, we shouldn't
973 		 * shutdown.
974 		 */
975 
976 		part = DISKPART(dev);
977 		pmask = (1 << part);
978 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
979 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
980 			(rs->sc_dkdev.dk_copenmask & pmask))) {
981 			raidunlock(rs);
982 			return (EBUSY);
983 		}
984 
985 		retcode = rf_Shutdown(raidPtr);
986 
987 		/* It's no longer initialized... */
988 		rs->sc_flags &= ~RAIDF_INITED;
989 
990 		/* Detach the disk. */
991 		disk_detach(&rs->sc_dkdev);
992 
993 		raidunlock(rs);
994 
995 		return (retcode);
996 	case RAIDFRAME_GET_COMPONENT_LABEL:
997 		clabel_ptr = (RF_ComponentLabel_t **) data;
998 		/* need to read the component label for the disk indicated
999 		   by row,column in clabel */
1000 
1001 		/* For practice, let's get it directly fromdisk, rather
1002 		   than from the in-core copy */
1003 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
1004 			   (RF_ComponentLabel_t *));
1005 		if (clabel == NULL)
1006 			return (ENOMEM);
1007 
1008 		memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t));
1009 
1010 		retcode = copyin( *clabel_ptr, clabel,
1011 				  sizeof(RF_ComponentLabel_t));
1012 
1013 		if (retcode) {
1014 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1015 			return(retcode);
1016 		}
1017 
1018 		row = clabel->row;
1019 		column = clabel->column;
1020 
1021 		if ((row < 0) || (row >= raidPtr->numRow) ||
1022 		    (column < 0) || (column >= raidPtr->numCol +
1023 				     raidPtr->numSpare)) {
1024 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1025 			return(EINVAL);
1026 		}
1027 
1028 		raidread_component_label(raidPtr->Disks[row][column].dev,
1029 				raidPtr->raid_cinfo[row][column].ci_vp,
1030 				clabel );
1031 
1032 		retcode = copyout((caddr_t) clabel,
1033 				  (caddr_t) *clabel_ptr,
1034 				  sizeof(RF_ComponentLabel_t));
1035 		RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1036 		return (retcode);
1037 
1038 	case RAIDFRAME_SET_COMPONENT_LABEL:
1039 		clabel = (RF_ComponentLabel_t *) data;
1040 
1041 		/* XXX check the label for valid stuff... */
1042 		/* Note that some things *should not* get modified --
1043 		   the user should be re-initing the labels instead of
1044 		   trying to patch things.
1045 		   */
1046 
1047 		raidid = raidPtr->raidid;
1048 		printf("raid%d: Got component label:\n", raidid);
1049 		printf("raid%d: Version: %d\n", raidid, clabel->version);
1050 		printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number);
1051 		printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter);
1052 		printf("raid%d: Row: %d\n", raidid, clabel->row);
1053 		printf("raid%d: Column: %d\n", raidid, clabel->column);
1054 		printf("raid%d: Num Rows: %d\n", raidid, clabel->num_rows);
1055 		printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns);
1056 		printf("raid%d: Clean: %d\n", raidid, clabel->clean);
1057 		printf("raid%d: Status: %d\n", raidid, clabel->status);
1058 
1059 		row = clabel->row;
1060 		column = clabel->column;
1061 
1062 		if ((row < 0) || (row >= raidPtr->numRow) ||
1063 		    (column < 0) || (column >= raidPtr->numCol)) {
1064 			return(EINVAL);
1065 		}
1066 
1067 		/* XXX this isn't allowed to do anything for now :-) */
1068 
1069 		/* XXX and before it is, we need to fill in the rest
1070 		   of the fields!?!?!?! */
1071 #if 0
1072 		raidwrite_component_label(
1073                             raidPtr->Disks[row][column].dev,
1074 			    raidPtr->raid_cinfo[row][column].ci_vp,
1075 			    clabel );
1076 #endif
1077 		return (0);
1078 
1079 	case RAIDFRAME_INIT_LABELS:
1080 		clabel = (RF_ComponentLabel_t *) data;
1081 		/*
1082 		   we only want the serial number from
1083 		   the above.  We get all the rest of the information
1084 		   from the config that was used to create this RAID
1085 		   set.
1086 		   */
1087 
1088 		raidPtr->serial_number = clabel->serial_number;
1089 
1090 		raid_init_component_label(raidPtr, &ci_label);
1091 		ci_label.serial_number = clabel->serial_number;
1092 
1093 		for(row=0;row<raidPtr->numRow;row++) {
1094 			ci_label.row = row;
1095 			for(column=0;column<raidPtr->numCol;column++) {
1096 				diskPtr = &raidPtr->Disks[row][column];
1097 				if (!RF_DEAD_DISK(diskPtr->status)) {
1098 					ci_label.partitionSize = diskPtr->partitionSize;
1099 					ci_label.column = column;
1100 					raidwrite_component_label(
1101 					  raidPtr->Disks[row][column].dev,
1102 					  raidPtr->raid_cinfo[row][column].ci_vp,
1103 					  &ci_label );
1104 				}
1105 			}
1106 		}
1107 
1108 		return (retcode);
1109 	case RAIDFRAME_SET_AUTOCONFIG:
1110 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1111 		printf("raid%d: New autoconfig value is: %d\n",
1112 		       raidPtr->raidid, d);
1113 		*(int *) data = d;
1114 		return (retcode);
1115 
1116 	case RAIDFRAME_SET_ROOT:
1117 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1118 		printf("raid%d: New rootpartition value is: %d\n",
1119 		       raidPtr->raidid, d);
1120 		*(int *) data = d;
1121 		return (retcode);
1122 
1123 		/* initialize all parity */
1124 	case RAIDFRAME_REWRITEPARITY:
1125 
1126 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1127 			/* Parity for RAID 0 is trivially correct */
1128 			raidPtr->parity_good = RF_RAID_CLEAN;
1129 			return(0);
1130 		}
1131 
1132 		if (raidPtr->parity_rewrite_in_progress == 1) {
1133 			/* Re-write is already in progress! */
1134 			return(EINVAL);
1135 		}
1136 
1137 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1138 					   rf_RewriteParityThread,
1139 					   raidPtr,"raid_parity");
1140 		return (retcode);
1141 
1142 
1143 	case RAIDFRAME_ADD_HOT_SPARE:
1144 		sparePtr = (RF_SingleComponent_t *) data;
1145 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1146 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1147 		return(retcode);
1148 
1149 	case RAIDFRAME_REMOVE_HOT_SPARE:
1150 		return(retcode);
1151 
1152 	case RAIDFRAME_DELETE_COMPONENT:
1153 		componentPtr = (RF_SingleComponent_t *)data;
1154 		memcpy( &component, componentPtr,
1155 			sizeof(RF_SingleComponent_t));
1156 		retcode = rf_delete_component(raidPtr, &component);
1157 		return(retcode);
1158 
1159 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1160 		componentPtr = (RF_SingleComponent_t *)data;
1161 		memcpy( &component, componentPtr,
1162 			sizeof(RF_SingleComponent_t));
1163 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1164 		return(retcode);
1165 
1166 	case RAIDFRAME_REBUILD_IN_PLACE:
1167 
1168 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1169 			/* Can't do this on a RAID 0!! */
1170 			return(EINVAL);
1171 		}
1172 
1173 		if (raidPtr->recon_in_progress == 1) {
1174 			/* a reconstruct is already in progress! */
1175 			return(EINVAL);
1176 		}
1177 
1178 		componentPtr = (RF_SingleComponent_t *) data;
1179 		memcpy( &component, componentPtr,
1180 			sizeof(RF_SingleComponent_t));
1181 		row = component.row;
1182 		column = component.column;
1183 		printf("raid%d: Rebuild: %d %d\n", raidPtr->raidid,
1184 		       row, column);
1185 		if ((row < 0) || (row >= raidPtr->numRow) ||
1186 		    (column < 0) || (column >= raidPtr->numCol)) {
1187 			return(EINVAL);
1188 		}
1189 
1190 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1191 		if (rrcopy == NULL)
1192 			return(ENOMEM);
1193 
1194 		rrcopy->raidPtr = (void *) raidPtr;
1195 		rrcopy->row = row;
1196 		rrcopy->col = column;
1197 
1198 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1199 					   rf_ReconstructInPlaceThread,
1200 					   rrcopy,"raid_reconip");
1201 		return(retcode);
1202 
1203 	case RAIDFRAME_GET_INFO:
1204 		if (!raidPtr->valid)
1205 			return (ENODEV);
1206 		ucfgp = (RF_DeviceConfig_t **) data;
1207 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1208 			  (RF_DeviceConfig_t *));
1209 		if (d_cfg == NULL)
1210 			return (ENOMEM);
1211 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1212 		d_cfg->rows = raidPtr->numRow;
1213 		d_cfg->cols = raidPtr->numCol;
1214 		d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol;
1215 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1216 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1217 			return (ENOMEM);
1218 		}
1219 		d_cfg->nspares = raidPtr->numSpare;
1220 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1221 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1222 			return (ENOMEM);
1223 		}
1224 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1225 		d = 0;
1226 		for (i = 0; i < d_cfg->rows; i++) {
1227 			for (j = 0; j < d_cfg->cols; j++) {
1228 				d_cfg->devs[d] = raidPtr->Disks[i][j];
1229 				d++;
1230 			}
1231 		}
1232 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1233 			d_cfg->spares[i] = raidPtr->Disks[0][j];
1234 		}
1235 		retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp,
1236 				  sizeof(RF_DeviceConfig_t));
1237 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1238 
1239 		return (retcode);
1240 
1241 	case RAIDFRAME_CHECK_PARITY:
1242 		*(int *) data = raidPtr->parity_good;
1243 		return (0);
1244 
1245 	case RAIDFRAME_RESET_ACCTOTALS:
1246 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1247 		return (0);
1248 
1249 	case RAIDFRAME_GET_ACCTOTALS:
1250 		totals = (RF_AccTotals_t *) data;
1251 		*totals = raidPtr->acc_totals;
1252 		return (0);
1253 
1254 	case RAIDFRAME_KEEP_ACCTOTALS:
1255 		raidPtr->keep_acc_totals = *(int *)data;
1256 		return (0);
1257 
1258 	case RAIDFRAME_GET_SIZE:
1259 		*(int *) data = raidPtr->totalSectors;
1260 		return (0);
1261 
1262 		/* fail a disk & optionally start reconstruction */
1263 	case RAIDFRAME_FAIL_DISK:
1264 
1265 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1266 			/* Can't do this on a RAID 0!! */
1267 			return(EINVAL);
1268 		}
1269 
1270 		rr = (struct rf_recon_req *) data;
1271 
1272 		if (rr->row < 0 || rr->row >= raidPtr->numRow
1273 		    || rr->col < 0 || rr->col >= raidPtr->numCol)
1274 			return (EINVAL);
1275 
1276 		printf("raid%d: Failing the disk: row: %d col: %d\n",
1277 		       unit, rr->row, rr->col);
1278 
1279 		/* make a copy of the recon request so that we don't rely on
1280 		 * the user's buffer */
1281 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1282 		if (rrcopy == NULL)
1283 			return(ENOMEM);
1284 		memcpy(rrcopy, rr, sizeof(*rr));
1285 		rrcopy->raidPtr = (void *) raidPtr;
1286 
1287 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1288 					   rf_ReconThread,
1289 					   rrcopy,"raid_recon");
1290 		return (0);
1291 
1292 		/* invoke a copyback operation after recon on whatever disk
1293 		 * needs it, if any */
1294 	case RAIDFRAME_COPYBACK:
1295 
1296 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1297 			/* This makes no sense on a RAID 0!! */
1298 			return(EINVAL);
1299 		}
1300 
1301 		if (raidPtr->copyback_in_progress == 1) {
1302 			/* Copyback is already in progress! */
1303 			return(EINVAL);
1304 		}
1305 
1306 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1307 					   rf_CopybackThread,
1308 					   raidPtr,"raid_copyback");
1309 		return (retcode);
1310 
1311 		/* return the percentage completion of reconstruction */
1312 	case RAIDFRAME_CHECK_RECON_STATUS:
1313 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1314 			/* This makes no sense on a RAID 0, so tell the
1315 			   user it's done. */
1316 			*(int *) data = 100;
1317 			return(0);
1318 		}
1319 		row = 0; /* XXX we only consider a single row... */
1320 		if (raidPtr->status[row] != rf_rs_reconstructing)
1321 			*(int *) data = 100;
1322 		else
1323 			*(int *) data = raidPtr->reconControl[row]->percentComplete;
1324 		return (0);
1325 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1326 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1327 		row = 0; /* XXX we only consider a single row... */
1328 		if (raidPtr->status[row] != rf_rs_reconstructing) {
1329 			progressInfo.remaining = 0;
1330 			progressInfo.completed = 100;
1331 			progressInfo.total = 100;
1332 		} else {
1333 			progressInfo.total =
1334 				raidPtr->reconControl[row]->numRUsTotal;
1335 			progressInfo.completed =
1336 				raidPtr->reconControl[row]->numRUsComplete;
1337 			progressInfo.remaining = progressInfo.total -
1338 				progressInfo.completed;
1339 		}
1340 		retcode = copyout((caddr_t) &progressInfo,
1341 				  (caddr_t) *progressInfoPtr,
1342 				  sizeof(RF_ProgressInfo_t));
1343 		return (retcode);
1344 
1345 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1346 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1347 			/* This makes no sense on a RAID 0, so tell the
1348 			   user it's done. */
1349 			*(int *) data = 100;
1350 			return(0);
1351 		}
1352 		if (raidPtr->parity_rewrite_in_progress == 1) {
1353 			*(int *) data = 100 *
1354 				raidPtr->parity_rewrite_stripes_done /
1355 				raidPtr->Layout.numStripe;
1356 		} else {
1357 			*(int *) data = 100;
1358 		}
1359 		return (0);
1360 
1361 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1362 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1363 		if (raidPtr->parity_rewrite_in_progress == 1) {
1364 			progressInfo.total = raidPtr->Layout.numStripe;
1365 			progressInfo.completed =
1366 				raidPtr->parity_rewrite_stripes_done;
1367 			progressInfo.remaining = progressInfo.total -
1368 				progressInfo.completed;
1369 		} else {
1370 			progressInfo.remaining = 0;
1371 			progressInfo.completed = 100;
1372 			progressInfo.total = 100;
1373 		}
1374 		retcode = copyout((caddr_t) &progressInfo,
1375 				  (caddr_t) *progressInfoPtr,
1376 				  sizeof(RF_ProgressInfo_t));
1377 		return (retcode);
1378 
1379 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1380 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1381 			/* This makes no sense on a RAID 0 */
1382 			*(int *) data = 100;
1383 			return(0);
1384 		}
1385 		if (raidPtr->copyback_in_progress == 1) {
1386 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1387 				raidPtr->Layout.numStripe;
1388 		} else {
1389 			*(int *) data = 100;
1390 		}
1391 		return (0);
1392 
1393 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1394 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1395 		if (raidPtr->copyback_in_progress == 1) {
1396 			progressInfo.total = raidPtr->Layout.numStripe;
1397 			progressInfo.completed =
1398 				raidPtr->copyback_stripes_done;
1399 			progressInfo.remaining = progressInfo.total -
1400 				progressInfo.completed;
1401 		} else {
1402 			progressInfo.remaining = 0;
1403 			progressInfo.completed = 100;
1404 			progressInfo.total = 100;
1405 		}
1406 		retcode = copyout((caddr_t) &progressInfo,
1407 				  (caddr_t) *progressInfoPtr,
1408 				  sizeof(RF_ProgressInfo_t));
1409 		return (retcode);
1410 
1411 		/* the sparetable daemon calls this to wait for the kernel to
1412 		 * need a spare table. this ioctl does not return until a
1413 		 * spare table is needed. XXX -- calling mpsleep here in the
1414 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1415 		 * -- I should either compute the spare table in the kernel,
1416 		 * or have a different -- XXX XXX -- interface (a different
1417 		 * character device) for delivering the table     -- XXX */
1418 #if 0
1419 	case RAIDFRAME_SPARET_WAIT:
1420 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1421 		while (!rf_sparet_wait_queue)
1422 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1423 		waitreq = rf_sparet_wait_queue;
1424 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1425 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1426 
1427 		/* structure assignment */
1428 		*((RF_SparetWait_t *) data) = *waitreq;
1429 
1430 		RF_Free(waitreq, sizeof(*waitreq));
1431 		return (0);
1432 
1433 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1434 		 * code in it that will cause the dameon to exit */
1435 	case RAIDFRAME_ABORT_SPARET_WAIT:
1436 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1437 		waitreq->fcol = -1;
1438 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1439 		waitreq->next = rf_sparet_wait_queue;
1440 		rf_sparet_wait_queue = waitreq;
1441 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1442 		wakeup(&rf_sparet_wait_queue);
1443 		return (0);
1444 
1445 		/* used by the spare table daemon to deliver a spare table
1446 		 * into the kernel */
1447 	case RAIDFRAME_SEND_SPARET:
1448 
1449 		/* install the spare table */
1450 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1451 
1452 		/* respond to the requestor.  the return status of the spare
1453 		 * table installation is passed in the "fcol" field */
1454 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1455 		waitreq->fcol = retcode;
1456 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1457 		waitreq->next = rf_sparet_resp_queue;
1458 		rf_sparet_resp_queue = waitreq;
1459 		wakeup(&rf_sparet_resp_queue);
1460 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1461 
1462 		return (retcode);
1463 #endif
1464 
1465 	default:
1466 		break; /* fall through to the os-specific code below */
1467 
1468 	}
1469 
1470 	if (!raidPtr->valid)
1471 		return (EINVAL);
1472 
1473 	/*
1474 	 * Add support for "regular" device ioctls here.
1475 	 */
1476 
1477 	switch (cmd) {
1478 	case DIOCGDINFO:
1479 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1480 		break;
1481 #ifdef __HAVE_OLD_DISKLABEL
1482 	case ODIOCGDINFO:
1483 		newlabel = *(rs->sc_dkdev.dk_label);
1484 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1485 			return ENOTTY;
1486 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1487 		break;
1488 #endif
1489 
1490 	case DIOCGPART:
1491 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1492 		((struct partinfo *) data)->part =
1493 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1494 		break;
1495 
1496 	case DIOCWDINFO:
1497 	case DIOCSDINFO:
1498 #ifdef __HAVE_OLD_DISKLABEL
1499 	case ODIOCWDINFO:
1500 	case ODIOCSDINFO:
1501 #endif
1502 	{
1503 		struct disklabel *lp;
1504 #ifdef __HAVE_OLD_DISKLABEL
1505 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1506 			memset(&newlabel, 0, sizeof newlabel);
1507 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1508 			lp = &newlabel;
1509 		} else
1510 #endif
1511 		lp = (struct disklabel *)data;
1512 
1513 		if ((error = raidlock(rs)) != 0)
1514 			return (error);
1515 
1516 		rs->sc_flags |= RAIDF_LABELLING;
1517 
1518 		error = setdisklabel(rs->sc_dkdev.dk_label,
1519 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1520 		if (error == 0) {
1521 			if (cmd == DIOCWDINFO
1522 #ifdef __HAVE_OLD_DISKLABEL
1523 			    || cmd == ODIOCWDINFO
1524 #endif
1525 			   )
1526 				error = writedisklabel(RAIDLABELDEV(dev),
1527 				    raidstrategy, rs->sc_dkdev.dk_label,
1528 				    rs->sc_dkdev.dk_cpulabel);
1529 		}
1530 		rs->sc_flags &= ~RAIDF_LABELLING;
1531 
1532 		raidunlock(rs);
1533 
1534 		if (error)
1535 			return (error);
1536 		break;
1537 	}
1538 
1539 	case DIOCWLABEL:
1540 		if (*(int *) data != 0)
1541 			rs->sc_flags |= RAIDF_WLABEL;
1542 		else
1543 			rs->sc_flags &= ~RAIDF_WLABEL;
1544 		break;
1545 
1546 	case DIOCGDEFLABEL:
1547 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1548 		break;
1549 
1550 #ifdef __HAVE_OLD_DISKLABEL
1551 	case ODIOCGDEFLABEL:
1552 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1553 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1554 			return ENOTTY;
1555 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1556 		break;
1557 #endif
1558 
1559 	default:
1560 		retcode = ENOTTY;
1561 	}
1562 	return (retcode);
1563 
1564 }
1565 
1566 
1567 /* raidinit -- complete the rest of the initialization for the
1568    RAIDframe device.  */
1569 
1570 
1571 static void
1572 raidinit(raidPtr)
1573 	RF_Raid_t *raidPtr;
1574 {
1575 	struct raid_softc *rs;
1576 	int     unit;
1577 
1578 	unit = raidPtr->raidid;
1579 
1580 	rs = &raid_softc[unit];
1581 
1582 	/* XXX should check return code first... */
1583 	rs->sc_flags |= RAIDF_INITED;
1584 
1585 	sprintf(rs->sc_xname, "raid%d", unit);	/* XXX doesn't check bounds. */
1586 
1587 	rs->sc_dkdev.dk_name = rs->sc_xname;
1588 
1589 	/* disk_attach actually creates space for the CPU disklabel, among
1590 	 * other things, so it's critical to call this *BEFORE* we try putzing
1591 	 * with disklabels. */
1592 
1593 	disk_attach(&rs->sc_dkdev);
1594 
1595 	/* XXX There may be a weird interaction here between this, and
1596 	 * protectedSectors, as used in RAIDframe.  */
1597 
1598 	rs->sc_size = raidPtr->totalSectors;
1599 
1600 }
1601 
1602 /* wake up the daemon & tell it to get us a spare table
1603  * XXX
1604  * the entries in the queues should be tagged with the raidPtr
1605  * so that in the extremely rare case that two recons happen at once,
1606  * we know for which device were requesting a spare table
1607  * XXX
1608  *
1609  * XXX This code is not currently used. GO
1610  */
1611 int
1612 rf_GetSpareTableFromDaemon(req)
1613 	RF_SparetWait_t *req;
1614 {
1615 	int     retcode;
1616 
1617 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1618 	req->next = rf_sparet_wait_queue;
1619 	rf_sparet_wait_queue = req;
1620 	wakeup(&rf_sparet_wait_queue);
1621 
1622 	/* mpsleep unlocks the mutex */
1623 	while (!rf_sparet_resp_queue) {
1624 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1625 		    "raidframe getsparetable", 0);
1626 	}
1627 	req = rf_sparet_resp_queue;
1628 	rf_sparet_resp_queue = req->next;
1629 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1630 
1631 	retcode = req->fcol;
1632 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1633 					 * alloc'd */
1634 	return (retcode);
1635 }
1636 
1637 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1638  * bp & passes it down.
1639  * any calls originating in the kernel must use non-blocking I/O
1640  * do some extra sanity checking to return "appropriate" error values for
1641  * certain conditions (to make some standard utilities work)
1642  *
1643  * Formerly known as: rf_DoAccessKernel
1644  */
1645 void
1646 raidstart(raidPtr)
1647 	RF_Raid_t *raidPtr;
1648 {
1649 	RF_SectorCount_t num_blocks, pb, sum;
1650 	RF_RaidAddr_t raid_addr;
1651 	int     retcode;
1652 	struct partition *pp;
1653 	daddr_t blocknum;
1654 	int     unit;
1655 	struct raid_softc *rs;
1656 	int     do_async;
1657 	struct buf *bp;
1658 
1659 	unit = raidPtr->raidid;
1660 	rs = &raid_softc[unit];
1661 
1662 	/* quick check to see if anything has died recently */
1663 	RF_LOCK_MUTEX(raidPtr->mutex);
1664 	if (raidPtr->numNewFailures > 0) {
1665 		rf_update_component_labels(raidPtr,
1666 					   RF_NORMAL_COMPONENT_UPDATE);
1667 		raidPtr->numNewFailures--;
1668 	}
1669 
1670 	/* Check to see if we're at the limit... */
1671 	while (raidPtr->openings > 0) {
1672 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1673 
1674 		/* get the next item, if any, from the queue */
1675 		if ((bp = BUFQ_GET(&rs->buf_queue)) == NULL) {
1676 			/* nothing more to do */
1677 			return;
1678 		}
1679 
1680 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1681 		 * partition.. Need to make it absolute to the underlying
1682 		 * device.. */
1683 
1684 		blocknum = bp->b_blkno;
1685 		if (DISKPART(bp->b_dev) != RAW_PART) {
1686 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1687 			blocknum += pp->p_offset;
1688 		}
1689 
1690 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1691 			    (int) blocknum));
1692 
1693 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1694 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1695 
1696 		/* *THIS* is where we adjust what block we're going to...
1697 		 * but DO NOT TOUCH bp->b_blkno!!! */
1698 		raid_addr = blocknum;
1699 
1700 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1701 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1702 		sum = raid_addr + num_blocks + pb;
1703 		if (1 || rf_debugKernelAccess) {
1704 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1705 				    (int) raid_addr, (int) sum, (int) num_blocks,
1706 				    (int) pb, (int) bp->b_resid));
1707 		}
1708 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1709 		    || (sum < num_blocks) || (sum < pb)) {
1710 			bp->b_error = ENOSPC;
1711 			bp->b_flags |= B_ERROR;
1712 			bp->b_resid = bp->b_bcount;
1713 			biodone(bp);
1714 			RF_LOCK_MUTEX(raidPtr->mutex);
1715 			continue;
1716 		}
1717 		/*
1718 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1719 		 */
1720 
1721 		if (bp->b_bcount & raidPtr->sectorMask) {
1722 			bp->b_error = EINVAL;
1723 			bp->b_flags |= B_ERROR;
1724 			bp->b_resid = bp->b_bcount;
1725 			biodone(bp);
1726 			RF_LOCK_MUTEX(raidPtr->mutex);
1727 			continue;
1728 
1729 		}
1730 		db1_printf(("Calling DoAccess..\n"));
1731 
1732 
1733 		RF_LOCK_MUTEX(raidPtr->mutex);
1734 		raidPtr->openings--;
1735 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1736 
1737 		/*
1738 		 * Everything is async.
1739 		 */
1740 		do_async = 1;
1741 
1742 		disk_busy(&rs->sc_dkdev);
1743 
1744 		/* XXX we're still at splbio() here... do we *really*
1745 		   need to be? */
1746 
1747 		/* don't ever condition on bp->b_flags & B_WRITE.
1748 		 * always condition on B_READ instead */
1749 
1750 		retcode = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1751 				      RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1752 				      do_async, raid_addr, num_blocks,
1753 				      bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1754 
1755 		RF_LOCK_MUTEX(raidPtr->mutex);
1756 	}
1757 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1758 }
1759 
1760 
1761 
1762 
1763 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1764 
1765 int
1766 rf_DispatchKernelIO(queue, req)
1767 	RF_DiskQueue_t *queue;
1768 	RF_DiskQueueData_t *req;
1769 {
1770 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1771 	struct buf *bp;
1772 	struct raidbuf *raidbp = NULL;
1773 	struct raid_softc *rs;
1774 	int     unit;
1775 	int s;
1776 
1777 	s=0;
1778 	/* s = splbio();*/ /* want to test this */
1779 	/* XXX along with the vnode, we also need the softc associated with
1780 	 * this device.. */
1781 
1782 	req->queue = queue;
1783 
1784 	unit = queue->raidPtr->raidid;
1785 
1786 	db1_printf(("DispatchKernelIO unit: %d\n", unit));
1787 
1788 	if (unit >= numraid) {
1789 		printf("Invalid unit number: %d %d\n", unit, numraid);
1790 		panic("Invalid Unit number in rf_DispatchKernelIO\n");
1791 	}
1792 	rs = &raid_softc[unit];
1793 
1794 	bp = req->bp;
1795 #if 1
1796 	/* XXX when there is a physical disk failure, someone is passing us a
1797 	 * buffer that contains old stuff!!  Attempt to deal with this problem
1798 	 * without taking a performance hit... (not sure where the real bug
1799 	 * is.  It's buried in RAIDframe somewhere) :-(  GO ) */
1800 
1801 	if (bp->b_flags & B_ERROR) {
1802 		bp->b_flags &= ~B_ERROR;
1803 	}
1804 	if (bp->b_error != 0) {
1805 		bp->b_error = 0;
1806 	}
1807 #endif
1808 	raidbp = RAIDGETBUF(rs);
1809 
1810 	raidbp->rf_flags = 0;	/* XXX not really used anywhere... */
1811 
1812 	/*
1813 	 * context for raidiodone
1814 	 */
1815 	raidbp->rf_obp = bp;
1816 	raidbp->req = req;
1817 
1818 	LIST_INIT(&raidbp->rf_buf.b_dep);
1819 
1820 	switch (req->type) {
1821 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1822 		/* XXX need to do something extra here.. */
1823 		/* I'm leaving this in, as I've never actually seen it used,
1824 		 * and I'd like folks to report it... GO */
1825 		printf(("WAKEUP CALLED\n"));
1826 		queue->numOutstanding++;
1827 
1828 		/* XXX need to glue the original buffer into this??  */
1829 
1830 		KernelWakeupFunc(&raidbp->rf_buf);
1831 		break;
1832 
1833 	case RF_IO_TYPE_READ:
1834 	case RF_IO_TYPE_WRITE:
1835 
1836 		if (req->tracerec) {
1837 			RF_ETIMER_START(req->tracerec->timer);
1838 		}
1839 		InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1840 		    op | bp->b_flags, queue->rf_cinfo->ci_dev,
1841 		    req->sectorOffset, req->numSector,
1842 		    req->buf, KernelWakeupFunc, (void *) req,
1843 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1844 
1845 		if (rf_debugKernelAccess) {
1846 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1847 				(long) bp->b_blkno));
1848 		}
1849 		queue->numOutstanding++;
1850 		queue->last_deq_sector = req->sectorOffset;
1851 		/* acc wouldn't have been let in if there were any pending
1852 		 * reqs at any other priority */
1853 		queue->curPriority = req->priority;
1854 
1855 		db1_printf(("Going for %c to unit %d row %d col %d\n",
1856 			req->type, unit, queue->row, queue->col));
1857 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1858 			(int) req->sectorOffset, (int) req->numSector,
1859 			(int) (req->numSector <<
1860 			    queue->raidPtr->logBytesPerSector),
1861 			(int) queue->raidPtr->logBytesPerSector));
1862 		if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1863 			raidbp->rf_buf.b_vp->v_numoutput++;
1864 		}
1865 		VOP_STRATEGY(&raidbp->rf_buf);
1866 
1867 		break;
1868 
1869 	default:
1870 		panic("bad req->type in rf_DispatchKernelIO");
1871 	}
1872 	db1_printf(("Exiting from DispatchKernelIO\n"));
1873 	/* splx(s); */ /* want to test this */
1874 	return (0);
1875 }
1876 /* this is the callback function associated with a I/O invoked from
1877    kernel code.
1878  */
1879 static void
1880 KernelWakeupFunc(vbp)
1881 	struct buf *vbp;
1882 {
1883 	RF_DiskQueueData_t *req = NULL;
1884 	RF_DiskQueue_t *queue;
1885 	struct raidbuf *raidbp = (struct raidbuf *) vbp;
1886 	struct buf *bp;
1887 	struct raid_softc *rs;
1888 	int     unit;
1889 	int s;
1890 
1891 	s = splbio();
1892 	db1_printf(("recovering the request queue:\n"));
1893 	req = raidbp->req;
1894 
1895 	bp = raidbp->rf_obp;
1896 
1897 	queue = (RF_DiskQueue_t *) req->queue;
1898 
1899 	if (raidbp->rf_buf.b_flags & B_ERROR) {
1900 		bp->b_flags |= B_ERROR;
1901 		bp->b_error = raidbp->rf_buf.b_error ?
1902 		    raidbp->rf_buf.b_error : EIO;
1903 	}
1904 
1905 	/* XXX methinks this could be wrong... */
1906 #if 1
1907 	bp->b_resid = raidbp->rf_buf.b_resid;
1908 #endif
1909 
1910 	if (req->tracerec) {
1911 		RF_ETIMER_STOP(req->tracerec->timer);
1912 		RF_ETIMER_EVAL(req->tracerec->timer);
1913 		RF_LOCK_MUTEX(rf_tracing_mutex);
1914 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1915 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1916 		req->tracerec->num_phys_ios++;
1917 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1918 	}
1919 	bp->b_bcount = raidbp->rf_buf.b_bcount;	/* XXXX ?? */
1920 
1921 	unit = queue->raidPtr->raidid;	/* *Much* simpler :-> */
1922 
1923 
1924 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1925 	 * ballistic, and mark the component as hosed... */
1926 
1927 	if (bp->b_flags & B_ERROR) {
1928 		/* Mark the disk as dead */
1929 		/* but only mark it once... */
1930 		if (queue->raidPtr->Disks[queue->row][queue->col].status ==
1931 		    rf_ds_optimal) {
1932 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1933 			    unit, queue->raidPtr->Disks[queue->row][queue->col].devname);
1934 			queue->raidPtr->Disks[queue->row][queue->col].status =
1935 			    rf_ds_failed;
1936 			queue->raidPtr->status[queue->row] = rf_rs_degraded;
1937 			queue->raidPtr->numFailures++;
1938 			queue->raidPtr->numNewFailures++;
1939 		} else {	/* Disk is already dead... */
1940 			/* printf("Disk already marked as dead!\n"); */
1941 		}
1942 
1943 	}
1944 
1945 	rs = &raid_softc[unit];
1946 	RAIDPUTBUF(rs, raidbp);
1947 
1948 	rf_DiskIOComplete(queue, req, (bp->b_flags & B_ERROR) ? 1 : 0);
1949 	(req->CompleteFunc) (req->argument, (bp->b_flags & B_ERROR) ? 1 : 0);
1950 
1951 	splx(s);
1952 }
1953 
1954 
1955 
1956 /*
1957  * initialize a buf structure for doing an I/O in the kernel.
1958  */
1959 static void
1960 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg,
1961        logBytesPerSector, b_proc)
1962 	struct buf *bp;
1963 	struct vnode *b_vp;
1964 	unsigned rw_flag;
1965 	dev_t dev;
1966 	RF_SectorNum_t startSect;
1967 	RF_SectorCount_t numSect;
1968 	caddr_t buf;
1969 	void (*cbFunc) (struct buf *);
1970 	void *cbArg;
1971 	int logBytesPerSector;
1972 	struct proc *b_proc;
1973 {
1974 	/* bp->b_flags       = B_PHYS | rw_flag; */
1975 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
1976 	bp->b_bcount = numSect << logBytesPerSector;
1977 	bp->b_bufsize = bp->b_bcount;
1978 	bp->b_error = 0;
1979 	bp->b_dev = dev;
1980 	bp->b_data = buf;
1981 	bp->b_blkno = startSect;
1982 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
1983 	if (bp->b_bcount == 0) {
1984 		panic("bp->b_bcount is zero in InitBP!!\n");
1985 	}
1986 	bp->b_proc = b_proc;
1987 	bp->b_iodone = cbFunc;
1988 	bp->b_vp = b_vp;
1989 
1990 }
1991 
1992 static void
1993 raidgetdefaultlabel(raidPtr, rs, lp)
1994 	RF_Raid_t *raidPtr;
1995 	struct raid_softc *rs;
1996 	struct disklabel *lp;
1997 {
1998 	db1_printf(("Building a default label...\n"));
1999 	memset(lp, 0, sizeof(*lp));
2000 
2001 	/* fabricate a label... */
2002 	lp->d_secperunit = raidPtr->totalSectors;
2003 	lp->d_secsize = raidPtr->bytesPerSector;
2004 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
2005 	lp->d_ntracks = 4 * raidPtr->numCol;
2006 	lp->d_ncylinders = raidPtr->totalSectors /
2007 		(lp->d_nsectors * lp->d_ntracks);
2008 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
2009 
2010 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
2011 	lp->d_type = DTYPE_RAID;
2012 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
2013 	lp->d_rpm = 3600;
2014 	lp->d_interleave = 1;
2015 	lp->d_flags = 0;
2016 
2017 	lp->d_partitions[RAW_PART].p_offset = 0;
2018 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2019 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2020 	lp->d_npartitions = RAW_PART + 1;
2021 
2022 	lp->d_magic = DISKMAGIC;
2023 	lp->d_magic2 = DISKMAGIC;
2024 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2025 
2026 }
2027 /*
2028  * Read the disklabel from the raid device.  If one is not present, fake one
2029  * up.
2030  */
2031 static void
2032 raidgetdisklabel(dev)
2033 	dev_t   dev;
2034 {
2035 	int     unit = raidunit(dev);
2036 	struct raid_softc *rs = &raid_softc[unit];
2037 	char   *errstring;
2038 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2039 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2040 	RF_Raid_t *raidPtr;
2041 
2042 	db1_printf(("Getting the disklabel...\n"));
2043 
2044 	memset(clp, 0, sizeof(*clp));
2045 
2046 	raidPtr = raidPtrs[unit];
2047 
2048 	raidgetdefaultlabel(raidPtr, rs, lp);
2049 
2050 	/*
2051 	 * Call the generic disklabel extraction routine.
2052 	 */
2053 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2054 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2055 	if (errstring)
2056 		raidmakedisklabel(rs);
2057 	else {
2058 		int     i;
2059 		struct partition *pp;
2060 
2061 		/*
2062 		 * Sanity check whether the found disklabel is valid.
2063 		 *
2064 		 * This is necessary since total size of the raid device
2065 		 * may vary when an interleave is changed even though exactly
2066 		 * same componets are used, and old disklabel may used
2067 		 * if that is found.
2068 		 */
2069 		if (lp->d_secperunit != rs->sc_size)
2070 			printf("raid%d: WARNING: %s: "
2071 			    "total sector size in disklabel (%d) != "
2072 			    "the size of raid (%ld)\n", unit, rs->sc_xname,
2073 			    lp->d_secperunit, (long) rs->sc_size);
2074 		for (i = 0; i < lp->d_npartitions; i++) {
2075 			pp = &lp->d_partitions[i];
2076 			if (pp->p_offset + pp->p_size > rs->sc_size)
2077 				printf("raid%d: WARNING: %s: end of partition `%c' "
2078 				       "exceeds the size of raid (%ld)\n",
2079 				       unit, rs->sc_xname, 'a' + i, (long) rs->sc_size);
2080 		}
2081 	}
2082 
2083 }
2084 /*
2085  * Take care of things one might want to take care of in the event
2086  * that a disklabel isn't present.
2087  */
2088 static void
2089 raidmakedisklabel(rs)
2090 	struct raid_softc *rs;
2091 {
2092 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2093 	db1_printf(("Making a label..\n"));
2094 
2095 	/*
2096 	 * For historical reasons, if there's no disklabel present
2097 	 * the raw partition must be marked FS_BSDFFS.
2098 	 */
2099 
2100 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2101 
2102 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2103 
2104 	lp->d_checksum = dkcksum(lp);
2105 }
2106 /*
2107  * Lookup the provided name in the filesystem.  If the file exists,
2108  * is a valid block device, and isn't being used by anyone else,
2109  * set *vpp to the file's vnode.
2110  * You'll find the original of this in ccd.c
2111  */
2112 int
2113 raidlookup(path, p, vpp)
2114 	char   *path;
2115 	struct proc *p;
2116 	struct vnode **vpp;	/* result */
2117 {
2118 	struct nameidata nd;
2119 	struct vnode *vp;
2120 	struct vattr va;
2121 	int     error;
2122 
2123 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2124 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2125 #if 0
2126 		printf("RAIDframe: vn_open returned %d\n", error);
2127 #endif
2128 		return (error);
2129 	}
2130 	vp = nd.ni_vp;
2131 	if (vp->v_usecount > 1) {
2132 		VOP_UNLOCK(vp, 0);
2133 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2134 		return (EBUSY);
2135 	}
2136 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2137 		VOP_UNLOCK(vp, 0);
2138 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2139 		return (error);
2140 	}
2141 	/* XXX: eventually we should handle VREG, too. */
2142 	if (va.va_type != VBLK) {
2143 		VOP_UNLOCK(vp, 0);
2144 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2145 		return (ENOTBLK);
2146 	}
2147 	VOP_UNLOCK(vp, 0);
2148 	*vpp = vp;
2149 	return (0);
2150 }
2151 /*
2152  * Wait interruptibly for an exclusive lock.
2153  *
2154  * XXX
2155  * Several drivers do this; it should be abstracted and made MP-safe.
2156  * (Hmm... where have we seen this warning before :->  GO )
2157  */
2158 static int
2159 raidlock(rs)
2160 	struct raid_softc *rs;
2161 {
2162 	int     error;
2163 
2164 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2165 		rs->sc_flags |= RAIDF_WANTED;
2166 		if ((error =
2167 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2168 			return (error);
2169 	}
2170 	rs->sc_flags |= RAIDF_LOCKED;
2171 	return (0);
2172 }
2173 /*
2174  * Unlock and wake up any waiters.
2175  */
2176 static void
2177 raidunlock(rs)
2178 	struct raid_softc *rs;
2179 {
2180 
2181 	rs->sc_flags &= ~RAIDF_LOCKED;
2182 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2183 		rs->sc_flags &= ~RAIDF_WANTED;
2184 		wakeup(rs);
2185 	}
2186 }
2187 
2188 
2189 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2190 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2191 
2192 int
2193 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2194 {
2195 	RF_ComponentLabel_t clabel;
2196 	raidread_component_label(dev, b_vp, &clabel);
2197 	clabel.mod_counter = mod_counter;
2198 	clabel.clean = RF_RAID_CLEAN;
2199 	raidwrite_component_label(dev, b_vp, &clabel);
2200 	return(0);
2201 }
2202 
2203 
2204 int
2205 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2206 {
2207 	RF_ComponentLabel_t clabel;
2208 	raidread_component_label(dev, b_vp, &clabel);
2209 	clabel.mod_counter = mod_counter;
2210 	clabel.clean = RF_RAID_DIRTY;
2211 	raidwrite_component_label(dev, b_vp, &clabel);
2212 	return(0);
2213 }
2214 
2215 /* ARGSUSED */
2216 int
2217 raidread_component_label(dev, b_vp, clabel)
2218 	dev_t dev;
2219 	struct vnode *b_vp;
2220 	RF_ComponentLabel_t *clabel;
2221 {
2222 	struct buf *bp;
2223 	const struct bdevsw *bdev;
2224 	int error;
2225 
2226 	/* XXX should probably ensure that we don't try to do this if
2227 	   someone has changed rf_protected_sectors. */
2228 
2229 	if (b_vp == NULL) {
2230 		/* For whatever reason, this component is not valid.
2231 		   Don't try to read a component label from it. */
2232 		return(EINVAL);
2233 	}
2234 
2235 	/* get a block of the appropriate size... */
2236 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2237 	bp->b_dev = dev;
2238 
2239 	/* get our ducks in a row for the read */
2240 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2241 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2242 	bp->b_flags |= B_READ;
2243  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2244 
2245 	bdev = bdevsw_lookup(bp->b_dev);
2246 	if (bdev == NULL)
2247 		return (ENXIO);
2248 	(*bdev->d_strategy)(bp);
2249 
2250 	error = biowait(bp);
2251 
2252 	if (!error) {
2253 		memcpy(clabel, bp->b_data,
2254 		       sizeof(RF_ComponentLabel_t));
2255 #if 0
2256 		rf_print_component_label( clabel );
2257 #endif
2258         } else {
2259 #if 0
2260 		printf("Failed to read RAID component label!\n");
2261 #endif
2262 	}
2263 
2264 	brelse(bp);
2265 	return(error);
2266 }
2267 /* ARGSUSED */
2268 int
2269 raidwrite_component_label(dev, b_vp, clabel)
2270 	dev_t dev;
2271 	struct vnode *b_vp;
2272 	RF_ComponentLabel_t *clabel;
2273 {
2274 	struct buf *bp;
2275 	const struct bdevsw *bdev;
2276 	int error;
2277 
2278 	/* get a block of the appropriate size... */
2279 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2280 	bp->b_dev = dev;
2281 
2282 	/* get our ducks in a row for the write */
2283 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2284 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2285 	bp->b_flags |= B_WRITE;
2286  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2287 
2288 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2289 
2290 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2291 
2292 	bdev = bdevsw_lookup(bp->b_dev);
2293 	if (bdev == NULL)
2294 		return (ENXIO);
2295 	(*bdev->d_strategy)(bp);
2296 	error = biowait(bp);
2297 	brelse(bp);
2298 	if (error) {
2299 #if 1
2300 		printf("Failed to write RAID component info!\n");
2301 #endif
2302 	}
2303 
2304 	return(error);
2305 }
2306 
2307 void
2308 rf_markalldirty(raidPtr)
2309 	RF_Raid_t *raidPtr;
2310 {
2311 	RF_ComponentLabel_t clabel;
2312 	int r,c;
2313 
2314 	raidPtr->mod_counter++;
2315 	for (r = 0; r < raidPtr->numRow; r++) {
2316 		for (c = 0; c < raidPtr->numCol; c++) {
2317 			/* we don't want to touch (at all) a disk that has
2318 			   failed */
2319 			if (!RF_DEAD_DISK(raidPtr->Disks[r][c].status)) {
2320 				raidread_component_label(
2321 					raidPtr->Disks[r][c].dev,
2322 					raidPtr->raid_cinfo[r][c].ci_vp,
2323 					&clabel);
2324 				if (clabel.status == rf_ds_spared) {
2325 					/* XXX do something special...
2326 					 but whatever you do, don't
2327 					 try to access it!! */
2328 				} else {
2329 #if 0
2330 				clabel.status =
2331 					raidPtr->Disks[r][c].status;
2332 				raidwrite_component_label(
2333 					raidPtr->Disks[r][c].dev,
2334 					raidPtr->raid_cinfo[r][c].ci_vp,
2335 					&clabel);
2336 #endif
2337 				raidmarkdirty(
2338 				       raidPtr->Disks[r][c].dev,
2339 				       raidPtr->raid_cinfo[r][c].ci_vp,
2340 				       raidPtr->mod_counter);
2341 				}
2342 			}
2343 		}
2344 	}
2345 	/* printf("Component labels marked dirty.\n"); */
2346 #if 0
2347 	for( c = 0; c < raidPtr->numSpare ; c++) {
2348 		sparecol = raidPtr->numCol + c;
2349 		if (raidPtr->Disks[r][sparecol].status == rf_ds_used_spare) {
2350 			/*
2351 
2352 			   XXX this is where we get fancy and map this spare
2353 			   into it's correct spot in the array.
2354 
2355 			 */
2356 			/*
2357 
2358 			   we claim this disk is "optimal" if it's
2359 			   rf_ds_used_spare, as that means it should be
2360 			   directly substitutable for the disk it replaced.
2361 			   We note that too...
2362 
2363 			 */
2364 
2365 			for(i=0;i<raidPtr->numRow;i++) {
2366 				for(j=0;j<raidPtr->numCol;j++) {
2367 					if ((raidPtr->Disks[i][j].spareRow ==
2368 					     r) &&
2369 					    (raidPtr->Disks[i][j].spareCol ==
2370 					     sparecol)) {
2371 						srow = r;
2372 						scol = sparecol;
2373 						break;
2374 					}
2375 				}
2376 			}
2377 
2378 			raidread_component_label(
2379 				      raidPtr->Disks[r][sparecol].dev,
2380 				      raidPtr->raid_cinfo[r][sparecol].ci_vp,
2381 				      &clabel);
2382 			/* make sure status is noted */
2383 			clabel.version = RF_COMPONENT_LABEL_VERSION;
2384 			clabel.mod_counter = raidPtr->mod_counter;
2385 			clabel.serial_number = raidPtr->serial_number;
2386 			clabel.row = srow;
2387 			clabel.column = scol;
2388 			clabel.num_rows = raidPtr->numRow;
2389 			clabel.num_columns = raidPtr->numCol;
2390 			clabel.clean = RF_RAID_DIRTY; /* changed in a bit*/
2391 			clabel.status = rf_ds_optimal;
2392 			raidwrite_component_label(
2393 				      raidPtr->Disks[r][sparecol].dev,
2394 				      raidPtr->raid_cinfo[r][sparecol].ci_vp,
2395 				      &clabel);
2396 			raidmarkclean( raidPtr->Disks[r][sparecol].dev,
2397 			              raidPtr->raid_cinfo[r][sparecol].ci_vp);
2398 		}
2399 	}
2400 
2401 #endif
2402 }
2403 
2404 
2405 void
2406 rf_update_component_labels(raidPtr, final)
2407 	RF_Raid_t *raidPtr;
2408 	int final;
2409 {
2410 	RF_ComponentLabel_t clabel;
2411 	int sparecol;
2412 	int r,c;
2413 	int i,j;
2414 	int srow, scol;
2415 
2416 	srow = -1;
2417 	scol = -1;
2418 
2419 	/* XXX should do extra checks to make sure things really are clean,
2420 	   rather than blindly setting the clean bit... */
2421 
2422 	raidPtr->mod_counter++;
2423 
2424 	for (r = 0; r < raidPtr->numRow; r++) {
2425 		for (c = 0; c < raidPtr->numCol; c++) {
2426 			if (raidPtr->Disks[r][c].status == rf_ds_optimal) {
2427 				raidread_component_label(
2428 					raidPtr->Disks[r][c].dev,
2429 					raidPtr->raid_cinfo[r][c].ci_vp,
2430 					&clabel);
2431 				/* make sure status is noted */
2432 				clabel.status = rf_ds_optimal;
2433 				/* bump the counter */
2434 				clabel.mod_counter = raidPtr->mod_counter;
2435 
2436 				raidwrite_component_label(
2437 					raidPtr->Disks[r][c].dev,
2438 					raidPtr->raid_cinfo[r][c].ci_vp,
2439 					&clabel);
2440 				if (final == RF_FINAL_COMPONENT_UPDATE) {
2441 					if (raidPtr->parity_good == RF_RAID_CLEAN) {
2442 						raidmarkclean(
2443 							      raidPtr->Disks[r][c].dev,
2444 							      raidPtr->raid_cinfo[r][c].ci_vp,
2445 							      raidPtr->mod_counter);
2446 					}
2447 				}
2448 			}
2449 			/* else we don't touch it.. */
2450 		}
2451 	}
2452 
2453 	for( c = 0; c < raidPtr->numSpare ; c++) {
2454 		sparecol = raidPtr->numCol + c;
2455 		/* Need to ensure that the reconstruct actually completed! */
2456 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
2457 			/*
2458 
2459 			   we claim this disk is "optimal" if it's
2460 			   rf_ds_used_spare, as that means it should be
2461 			   directly substitutable for the disk it replaced.
2462 			   We note that too...
2463 
2464 			 */
2465 
2466 			for(i=0;i<raidPtr->numRow;i++) {
2467 				for(j=0;j<raidPtr->numCol;j++) {
2468 					if ((raidPtr->Disks[i][j].spareRow ==
2469 					     0) &&
2470 					    (raidPtr->Disks[i][j].spareCol ==
2471 					     sparecol)) {
2472 						srow = i;
2473 						scol = j;
2474 						break;
2475 					}
2476 				}
2477 			}
2478 
2479 			/* XXX shouldn't *really* need this... */
2480 			raidread_component_label(
2481 				      raidPtr->Disks[0][sparecol].dev,
2482 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2483 				      &clabel);
2484 			/* make sure status is noted */
2485 
2486 			raid_init_component_label(raidPtr, &clabel);
2487 
2488 			clabel.mod_counter = raidPtr->mod_counter;
2489 			clabel.row = srow;
2490 			clabel.column = scol;
2491 			clabel.status = rf_ds_optimal;
2492 
2493 			raidwrite_component_label(
2494 				      raidPtr->Disks[0][sparecol].dev,
2495 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2496 				      &clabel);
2497 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2498 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2499 					raidmarkclean( raidPtr->Disks[0][sparecol].dev,
2500 						       raidPtr->raid_cinfo[0][sparecol].ci_vp,
2501 						       raidPtr->mod_counter);
2502 				}
2503 			}
2504 		}
2505 	}
2506 	/* 	printf("Component labels updated\n"); */
2507 }
2508 
2509 void
2510 rf_close_component(raidPtr, vp, auto_configured)
2511 	RF_Raid_t *raidPtr;
2512 	struct vnode *vp;
2513 	int auto_configured;
2514 {
2515 	struct proc *p;
2516 
2517 	p = raidPtr->engine_thread;
2518 
2519 	if (vp != NULL) {
2520 		if (auto_configured == 1) {
2521 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2522 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2523 			vput(vp);
2524 
2525 		} else {
2526 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2527 		}
2528 	} else {
2529 #if 0
2530 		printf("vnode was NULL\n");
2531 #endif
2532 	}
2533 }
2534 
2535 
2536 void
2537 rf_UnconfigureVnodes(raidPtr)
2538 	RF_Raid_t *raidPtr;
2539 {
2540 	int r,c;
2541 	struct proc *p;
2542 	struct vnode *vp;
2543 	int acd;
2544 
2545 
2546 	/* We take this opportunity to close the vnodes like we should.. */
2547 
2548 	p = raidPtr->engine_thread;
2549 
2550 	for (r = 0; r < raidPtr->numRow; r++) {
2551 		for (c = 0; c < raidPtr->numCol; c++) {
2552 #if 0
2553 			printf("raid%d: Closing vnode for row: %d col: %d\n",
2554 			       raidPtr->raidid, r, c);
2555 #endif
2556 			vp = raidPtr->raid_cinfo[r][c].ci_vp;
2557 			acd = raidPtr->Disks[r][c].auto_configured;
2558 			rf_close_component(raidPtr, vp, acd);
2559 			raidPtr->raid_cinfo[r][c].ci_vp = NULL;
2560 			raidPtr->Disks[r][c].auto_configured = 0;
2561 		}
2562 	}
2563 	for (r = 0; r < raidPtr->numSpare; r++) {
2564 #if 0
2565 		printf("raid%d: Closing vnode for spare: %d\n",
2566 		       raidPtr->raidid, r);
2567 #endif
2568 		vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp;
2569 		acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured;
2570 		rf_close_component(raidPtr, vp, acd);
2571 		raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL;
2572 		raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0;
2573 	}
2574 }
2575 
2576 
2577 void
2578 rf_ReconThread(req)
2579 	struct rf_recon_req *req;
2580 {
2581 	int     s;
2582 	RF_Raid_t *raidPtr;
2583 
2584 	s = splbio();
2585 	raidPtr = (RF_Raid_t *) req->raidPtr;
2586 	raidPtr->recon_in_progress = 1;
2587 
2588 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col,
2589 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2590 
2591 	/* XXX get rid of this! we don't need it at all.. */
2592 	RF_Free(req, sizeof(*req));
2593 
2594 	raidPtr->recon_in_progress = 0;
2595 	splx(s);
2596 
2597 	/* That's all... */
2598 	kthread_exit(0);        /* does not return */
2599 }
2600 
2601 void
2602 rf_RewriteParityThread(raidPtr)
2603 	RF_Raid_t *raidPtr;
2604 {
2605 	int retcode;
2606 	int s;
2607 
2608 	raidPtr->parity_rewrite_in_progress = 1;
2609 	s = splbio();
2610 	retcode = rf_RewriteParity(raidPtr);
2611 	splx(s);
2612 	if (retcode) {
2613 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2614 	} else {
2615 		/* set the clean bit!  If we shutdown correctly,
2616 		   the clean bit on each component label will get
2617 		   set */
2618 		raidPtr->parity_good = RF_RAID_CLEAN;
2619 	}
2620 	raidPtr->parity_rewrite_in_progress = 0;
2621 
2622 	/* Anyone waiting for us to stop?  If so, inform them... */
2623 	if (raidPtr->waitShutdown) {
2624 		wakeup(&raidPtr->parity_rewrite_in_progress);
2625 	}
2626 
2627 	/* That's all... */
2628 	kthread_exit(0);        /* does not return */
2629 }
2630 
2631 
2632 void
2633 rf_CopybackThread(raidPtr)
2634 	RF_Raid_t *raidPtr;
2635 {
2636 	int s;
2637 
2638 	raidPtr->copyback_in_progress = 1;
2639 	s = splbio();
2640 	rf_CopybackReconstructedData(raidPtr);
2641 	splx(s);
2642 	raidPtr->copyback_in_progress = 0;
2643 
2644 	/* That's all... */
2645 	kthread_exit(0);        /* does not return */
2646 }
2647 
2648 
2649 void
2650 rf_ReconstructInPlaceThread(req)
2651 	struct rf_recon_req *req;
2652 {
2653 	int retcode;
2654 	int s;
2655 	RF_Raid_t *raidPtr;
2656 
2657 	s = splbio();
2658 	raidPtr = req->raidPtr;
2659 	raidPtr->recon_in_progress = 1;
2660 	retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col);
2661 	RF_Free(req, sizeof(*req));
2662 	raidPtr->recon_in_progress = 0;
2663 	splx(s);
2664 
2665 	/* That's all... */
2666 	kthread_exit(0);        /* does not return */
2667 }
2668 
2669 void
2670 rf_mountroot_hook(dev)
2671 	struct device *dev;
2672 {
2673 
2674 }
2675 
2676 
2677 RF_AutoConfig_t *
2678 rf_find_raid_components()
2679 {
2680 	struct vnode *vp;
2681 	struct disklabel label;
2682 	struct device *dv;
2683 	dev_t dev;
2684 	int bmajor;
2685 	int error;
2686 	int i;
2687 	int good_one;
2688 	RF_ComponentLabel_t *clabel;
2689 	RF_AutoConfig_t *ac_list;
2690 	RF_AutoConfig_t *ac;
2691 
2692 
2693 	/* initialize the AutoConfig list */
2694 	ac_list = NULL;
2695 
2696 	/* we begin by trolling through *all* the devices on the system */
2697 
2698 	for (dv = alldevs.tqh_first; dv != NULL;
2699 	     dv = dv->dv_list.tqe_next) {
2700 
2701 		/* we are only interested in disks... */
2702 		if (dv->dv_class != DV_DISK)
2703 			continue;
2704 
2705 		/* we don't care about floppies... */
2706 		if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fd")) {
2707 			continue;
2708 		}
2709 
2710 		/* we don't care about CD's... */
2711 		if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"cd")) {
2712 			continue;
2713 		}
2714 
2715 		/* hdfd is the Atari/Hades floppy driver */
2716 		if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"hdfd")) {
2717 			continue;
2718 		}
2719 		/* fdisa is the Atari/Milan floppy driver */
2720 		if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fdisa")) {
2721 			continue;
2722 		}
2723 
2724 		/* need to find the device_name_to_block_device_major stuff */
2725 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2726 
2727 		/* get a vnode for the raw partition of this disk */
2728 
2729 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2730 		if (bdevvp(dev, &vp))
2731 			panic("RAID can't alloc vnode");
2732 
2733 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2734 
2735 		if (error) {
2736 			/* "Who cares."  Continue looking
2737 			   for something that exists*/
2738 			vput(vp);
2739 			continue;
2740 		}
2741 
2742 		/* Ok, the disk exists.  Go get the disklabel. */
2743 		error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label,
2744 				  FREAD, NOCRED, 0);
2745 		if (error) {
2746 			/*
2747 			 * XXX can't happen - open() would
2748 			 * have errored out (or faked up one)
2749 			 */
2750 			printf("can't get label for dev %s%c (%d)!?!?\n",
2751 			       dv->dv_xname, 'a' + RAW_PART, error);
2752 		}
2753 
2754 		/* don't need this any more.  We'll allocate it again
2755 		   a little later if we really do... */
2756 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2757 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2758 		vput(vp);
2759 
2760 		for (i=0; i < label.d_npartitions; i++) {
2761 			/* We only support partitions marked as RAID */
2762 			if (label.d_partitions[i].p_fstype != FS_RAID)
2763 				continue;
2764 
2765 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2766 			if (bdevvp(dev, &vp))
2767 				panic("RAID can't alloc vnode");
2768 
2769 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2770 			if (error) {
2771 				/* Whatever... */
2772 				vput(vp);
2773 				continue;
2774 			}
2775 
2776 			good_one = 0;
2777 
2778 			clabel = (RF_ComponentLabel_t *)
2779 				malloc(sizeof(RF_ComponentLabel_t),
2780 				       M_RAIDFRAME, M_NOWAIT);
2781 			if (clabel == NULL) {
2782 				/* XXX CLEANUP HERE */
2783 				printf("RAID auto config: out of memory!\n");
2784 				return(NULL); /* XXX probably should panic? */
2785 			}
2786 
2787 			if (!raidread_component_label(dev, vp, clabel)) {
2788 				/* Got the label.  Does it look reasonable? */
2789 				if (rf_reasonable_label(clabel) &&
2790 				    (clabel->partitionSize <=
2791 				     label.d_partitions[i].p_size)) {
2792 #if DEBUG
2793 					printf("Component on: %s%c: %d\n",
2794 					       dv->dv_xname, 'a'+i,
2795 					       label.d_partitions[i].p_size);
2796 					rf_print_component_label(clabel);
2797 #endif
2798 					/* if it's reasonable, add it,
2799 					   else ignore it. */
2800 					ac = (RF_AutoConfig_t *)
2801 						malloc(sizeof(RF_AutoConfig_t),
2802 						       M_RAIDFRAME,
2803 						       M_NOWAIT);
2804 					if (ac == NULL) {
2805 						/* XXX should panic?? */
2806 						return(NULL);
2807 					}
2808 
2809 					sprintf(ac->devname, "%s%c",
2810 						dv->dv_xname, 'a'+i);
2811 					ac->dev = dev;
2812 					ac->vp = vp;
2813 					ac->clabel = clabel;
2814 					ac->next = ac_list;
2815 					ac_list = ac;
2816 					good_one = 1;
2817 				}
2818 			}
2819 			if (!good_one) {
2820 				/* cleanup */
2821 				free(clabel, M_RAIDFRAME);
2822 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2823 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2824 				vput(vp);
2825 			}
2826 		}
2827 	}
2828 	return(ac_list);
2829 }
2830 
2831 static int
2832 rf_reasonable_label(clabel)
2833 	RF_ComponentLabel_t *clabel;
2834 {
2835 
2836 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2837 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2838 	    ((clabel->clean == RF_RAID_CLEAN) ||
2839 	     (clabel->clean == RF_RAID_DIRTY)) &&
2840 	    clabel->row >=0 &&
2841 	    clabel->column >= 0 &&
2842 	    clabel->num_rows > 0 &&
2843 	    clabel->num_columns > 0 &&
2844 	    clabel->row < clabel->num_rows &&
2845 	    clabel->column < clabel->num_columns &&
2846 	    clabel->blockSize > 0 &&
2847 	    clabel->numBlocks > 0) {
2848 		/* label looks reasonable enough... */
2849 		return(1);
2850 	}
2851 	return(0);
2852 }
2853 
2854 
2855 void
2856 rf_print_component_label(clabel)
2857 	RF_ComponentLabel_t *clabel;
2858 {
2859 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2860 	       clabel->row, clabel->column,
2861 	       clabel->num_rows, clabel->num_columns);
2862 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2863 	       clabel->version, clabel->serial_number,
2864 	       clabel->mod_counter);
2865 	printf("   Clean: %s Status: %d\n",
2866 	       clabel->clean ? "Yes" : "No", clabel->status );
2867 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2868 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2869 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2870 	       (char) clabel->parityConfig, clabel->blockSize,
2871 	       clabel->numBlocks);
2872 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2873 	printf("   Contains root partition: %s\n",
2874 	       clabel->root_partition ? "Yes" : "No" );
2875 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2876 #if 0
2877 	   printf("   Config order: %d\n", clabel->config_order);
2878 #endif
2879 
2880 }
2881 
2882 RF_ConfigSet_t *
2883 rf_create_auto_sets(ac_list)
2884 	RF_AutoConfig_t *ac_list;
2885 {
2886 	RF_AutoConfig_t *ac;
2887 	RF_ConfigSet_t *config_sets;
2888 	RF_ConfigSet_t *cset;
2889 	RF_AutoConfig_t *ac_next;
2890 
2891 
2892 	config_sets = NULL;
2893 
2894 	/* Go through the AutoConfig list, and figure out which components
2895 	   belong to what sets.  */
2896 	ac = ac_list;
2897 	while(ac!=NULL) {
2898 		/* we're going to putz with ac->next, so save it here
2899 		   for use at the end of the loop */
2900 		ac_next = ac->next;
2901 
2902 		if (config_sets == NULL) {
2903 			/* will need at least this one... */
2904 			config_sets = (RF_ConfigSet_t *)
2905 				malloc(sizeof(RF_ConfigSet_t),
2906 				       M_RAIDFRAME, M_NOWAIT);
2907 			if (config_sets == NULL) {
2908 				panic("rf_create_auto_sets: No memory!\n");
2909 			}
2910 			/* this one is easy :) */
2911 			config_sets->ac = ac;
2912 			config_sets->next = NULL;
2913 			config_sets->rootable = 0;
2914 			ac->next = NULL;
2915 		} else {
2916 			/* which set does this component fit into? */
2917 			cset = config_sets;
2918 			while(cset!=NULL) {
2919 				if (rf_does_it_fit(cset, ac)) {
2920 					/* looks like it matches... */
2921 					ac->next = cset->ac;
2922 					cset->ac = ac;
2923 					break;
2924 				}
2925 				cset = cset->next;
2926 			}
2927 			if (cset==NULL) {
2928 				/* didn't find a match above... new set..*/
2929 				cset = (RF_ConfigSet_t *)
2930 					malloc(sizeof(RF_ConfigSet_t),
2931 					       M_RAIDFRAME, M_NOWAIT);
2932 				if (cset == NULL) {
2933 					panic("rf_create_auto_sets: No memory!\n");
2934 				}
2935 				cset->ac = ac;
2936 				ac->next = NULL;
2937 				cset->next = config_sets;
2938 				cset->rootable = 0;
2939 				config_sets = cset;
2940 			}
2941 		}
2942 		ac = ac_next;
2943 	}
2944 
2945 
2946 	return(config_sets);
2947 }
2948 
2949 static int
2950 rf_does_it_fit(cset, ac)
2951 	RF_ConfigSet_t *cset;
2952 	RF_AutoConfig_t *ac;
2953 {
2954 	RF_ComponentLabel_t *clabel1, *clabel2;
2955 
2956 	/* If this one matches the *first* one in the set, that's good
2957 	   enough, since the other members of the set would have been
2958 	   through here too... */
2959 	/* note that we are not checking partitionSize here..
2960 
2961 	   Note that we are also not checking the mod_counters here.
2962 	   If everything else matches execpt the mod_counter, that's
2963 	   good enough for this test.  We will deal with the mod_counters
2964 	   a little later in the autoconfiguration process.
2965 
2966 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2967 
2968 	   The reason we don't check for this is that failed disks
2969 	   will have lower modification counts.  If those disks are
2970 	   not added to the set they used to belong to, then they will
2971 	   form their own set, which may result in 2 different sets,
2972 	   for example, competing to be configured at raid0, and
2973 	   perhaps competing to be the root filesystem set.  If the
2974 	   wrong ones get configured, or both attempt to become /,
2975 	   weird behaviour and or serious lossage will occur.  Thus we
2976 	   need to bring them into the fold here, and kick them out at
2977 	   a later point.
2978 
2979 	*/
2980 
2981 	clabel1 = cset->ac->clabel;
2982 	clabel2 = ac->clabel;
2983 	if ((clabel1->version == clabel2->version) &&
2984 	    (clabel1->serial_number == clabel2->serial_number) &&
2985 	    (clabel1->num_rows == clabel2->num_rows) &&
2986 	    (clabel1->num_columns == clabel2->num_columns) &&
2987 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2988 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2989 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2990 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2991 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2992 	    (clabel1->blockSize == clabel2->blockSize) &&
2993 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2994 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2995 	    (clabel1->root_partition == clabel2->root_partition) &&
2996 	    (clabel1->last_unit == clabel2->last_unit) &&
2997 	    (clabel1->config_order == clabel2->config_order)) {
2998 		/* if it get's here, it almost *has* to be a match */
2999 	} else {
3000 		/* it's not consistent with somebody in the set..
3001 		   punt */
3002 		return(0);
3003 	}
3004 	/* all was fine.. it must fit... */
3005 	return(1);
3006 }
3007 
3008 int
3009 rf_have_enough_components(cset)
3010 	RF_ConfigSet_t *cset;
3011 {
3012 	RF_AutoConfig_t *ac;
3013 	RF_AutoConfig_t *auto_config;
3014 	RF_ComponentLabel_t *clabel;
3015 	int r,c;
3016 	int num_rows;
3017 	int num_cols;
3018 	int num_missing;
3019 	int mod_counter;
3020 	int mod_counter_found;
3021 	int even_pair_failed;
3022 	char parity_type;
3023 
3024 
3025 	/* check to see that we have enough 'live' components
3026 	   of this set.  If so, we can configure it if necessary */
3027 
3028 	num_rows = cset->ac->clabel->num_rows;
3029 	num_cols = cset->ac->clabel->num_columns;
3030 	parity_type = cset->ac->clabel->parityConfig;
3031 
3032 	/* XXX Check for duplicate components!?!?!? */
3033 
3034 	/* Determine what the mod_counter is supposed to be for this set. */
3035 
3036 	mod_counter_found = 0;
3037 	mod_counter = 0;
3038 	ac = cset->ac;
3039 	while(ac!=NULL) {
3040 		if (mod_counter_found==0) {
3041 			mod_counter = ac->clabel->mod_counter;
3042 			mod_counter_found = 1;
3043 		} else {
3044 			if (ac->clabel->mod_counter > mod_counter) {
3045 				mod_counter = ac->clabel->mod_counter;
3046 			}
3047 		}
3048 		ac = ac->next;
3049 	}
3050 
3051 	num_missing = 0;
3052 	auto_config = cset->ac;
3053 
3054 	for(r=0; r<num_rows; r++) {
3055 		even_pair_failed = 0;
3056 		for(c=0; c<num_cols; c++) {
3057 			ac = auto_config;
3058 			while(ac!=NULL) {
3059 				if ((ac->clabel->row == r) &&
3060 				    (ac->clabel->column == c) &&
3061 				    (ac->clabel->mod_counter == mod_counter)) {
3062 					/* it's this one... */
3063 #if DEBUG
3064 					printf("Found: %s at %d,%d\n",
3065 					       ac->devname,r,c);
3066 #endif
3067 					break;
3068 				}
3069 				ac=ac->next;
3070 			}
3071 			if (ac==NULL) {
3072 				/* Didn't find one here! */
3073 				/* special case for RAID 1, especially
3074 				   where there are more than 2
3075 				   components (where RAIDframe treats
3076 				   things a little differently :( ) */
3077 				if (parity_type == '1') {
3078 					if (c%2 == 0) { /* even component */
3079 						even_pair_failed = 1;
3080 					} else { /* odd component.  If
3081                                                     we're failed, and
3082                                                     so is the even
3083                                                     component, it's
3084                                                     "Good Night, Charlie" */
3085 						if (even_pair_failed == 1) {
3086 							return(0);
3087 						}
3088 					}
3089 				} else {
3090 					/* normal accounting */
3091 					num_missing++;
3092 				}
3093 			}
3094 			if ((parity_type == '1') && (c%2 == 1)) {
3095 				/* Just did an even component, and we didn't
3096 				   bail.. reset the even_pair_failed flag,
3097 				   and go on to the next component.... */
3098 				even_pair_failed = 0;
3099 			}
3100 		}
3101 	}
3102 
3103 	clabel = cset->ac->clabel;
3104 
3105 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3106 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3107 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3108 		/* XXX this needs to be made *much* more general */
3109 		/* Too many failures */
3110 		return(0);
3111 	}
3112 	/* otherwise, all is well, and we've got enough to take a kick
3113 	   at autoconfiguring this set */
3114 	return(1);
3115 }
3116 
3117 void
3118 rf_create_configuration(ac,config,raidPtr)
3119 	RF_AutoConfig_t *ac;
3120 	RF_Config_t *config;
3121 	RF_Raid_t *raidPtr;
3122 {
3123 	RF_ComponentLabel_t *clabel;
3124 	int i;
3125 
3126 	clabel = ac->clabel;
3127 
3128 	/* 1. Fill in the common stuff */
3129 	config->numRow = clabel->num_rows;
3130 	config->numCol = clabel->num_columns;
3131 	config->numSpare = 0; /* XXX should this be set here? */
3132 	config->sectPerSU = clabel->sectPerSU;
3133 	config->SUsPerPU = clabel->SUsPerPU;
3134 	config->SUsPerRU = clabel->SUsPerRU;
3135 	config->parityConfig = clabel->parityConfig;
3136 	/* XXX... */
3137 	strcpy(config->diskQueueType,"fifo");
3138 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3139 	config->layoutSpecificSize = 0; /* XXX ?? */
3140 
3141 	while(ac!=NULL) {
3142 		/* row/col values will be in range due to the checks
3143 		   in reasonable_label() */
3144 		strcpy(config->devnames[ac->clabel->row][ac->clabel->column],
3145 		       ac->devname);
3146 		ac = ac->next;
3147 	}
3148 
3149 	for(i=0;i<RF_MAXDBGV;i++) {
3150 		config->debugVars[i][0] = NULL;
3151 	}
3152 }
3153 
3154 int
3155 rf_set_autoconfig(raidPtr, new_value)
3156 	RF_Raid_t *raidPtr;
3157 	int new_value;
3158 {
3159 	RF_ComponentLabel_t clabel;
3160 	struct vnode *vp;
3161 	dev_t dev;
3162 	int row, column;
3163 
3164 	raidPtr->autoconfigure = new_value;
3165 	for(row=0; row<raidPtr->numRow; row++) {
3166 		for(column=0; column<raidPtr->numCol; column++) {
3167 			if (raidPtr->Disks[row][column].status ==
3168 			    rf_ds_optimal) {
3169 				dev = raidPtr->Disks[row][column].dev;
3170 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3171 				raidread_component_label(dev, vp, &clabel);
3172 				clabel.autoconfigure = new_value;
3173 				raidwrite_component_label(dev, vp, &clabel);
3174 			}
3175 		}
3176 	}
3177 	return(new_value);
3178 }
3179 
3180 int
3181 rf_set_rootpartition(raidPtr, new_value)
3182 	RF_Raid_t *raidPtr;
3183 	int new_value;
3184 {
3185 	RF_ComponentLabel_t clabel;
3186 	struct vnode *vp;
3187 	dev_t dev;
3188 	int row, column;
3189 
3190 	raidPtr->root_partition = new_value;
3191 	for(row=0; row<raidPtr->numRow; row++) {
3192 		for(column=0; column<raidPtr->numCol; column++) {
3193 			if (raidPtr->Disks[row][column].status ==
3194 			    rf_ds_optimal) {
3195 				dev = raidPtr->Disks[row][column].dev;
3196 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3197 				raidread_component_label(dev, vp, &clabel);
3198 				clabel.root_partition = new_value;
3199 				raidwrite_component_label(dev, vp, &clabel);
3200 			}
3201 		}
3202 	}
3203 	return(new_value);
3204 }
3205 
3206 void
3207 rf_release_all_vps(cset)
3208 	RF_ConfigSet_t *cset;
3209 {
3210 	RF_AutoConfig_t *ac;
3211 
3212 	ac = cset->ac;
3213 	while(ac!=NULL) {
3214 		/* Close the vp, and give it back */
3215 		if (ac->vp) {
3216 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3217 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3218 			vput(ac->vp);
3219 			ac->vp = NULL;
3220 		}
3221 		ac = ac->next;
3222 	}
3223 }
3224 
3225 
3226 void
3227 rf_cleanup_config_set(cset)
3228 	RF_ConfigSet_t *cset;
3229 {
3230 	RF_AutoConfig_t *ac;
3231 	RF_AutoConfig_t *next_ac;
3232 
3233 	ac = cset->ac;
3234 	while(ac!=NULL) {
3235 		next_ac = ac->next;
3236 		/* nuke the label */
3237 		free(ac->clabel, M_RAIDFRAME);
3238 		/* cleanup the config structure */
3239 		free(ac, M_RAIDFRAME);
3240 		/* "next.." */
3241 		ac = next_ac;
3242 	}
3243 	/* and, finally, nuke the config set */
3244 	free(cset, M_RAIDFRAME);
3245 }
3246 
3247 
3248 void
3249 raid_init_component_label(raidPtr, clabel)
3250 	RF_Raid_t *raidPtr;
3251 	RF_ComponentLabel_t *clabel;
3252 {
3253 	/* current version number */
3254 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3255 	clabel->serial_number = raidPtr->serial_number;
3256 	clabel->mod_counter = raidPtr->mod_counter;
3257 	clabel->num_rows = raidPtr->numRow;
3258 	clabel->num_columns = raidPtr->numCol;
3259 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3260 	clabel->status = rf_ds_optimal; /* "It's good!" */
3261 
3262 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3263 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3264 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3265 
3266 	clabel->blockSize = raidPtr->bytesPerSector;
3267 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3268 
3269 	/* XXX not portable */
3270 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3271 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3272 	clabel->autoconfigure = raidPtr->autoconfigure;
3273 	clabel->root_partition = raidPtr->root_partition;
3274 	clabel->last_unit = raidPtr->raidid;
3275 	clabel->config_order = raidPtr->config_order;
3276 }
3277 
3278 int
3279 rf_auto_config_set(cset,unit)
3280 	RF_ConfigSet_t *cset;
3281 	int *unit;
3282 {
3283 	RF_Raid_t *raidPtr;
3284 	RF_Config_t *config;
3285 	int raidID;
3286 	int retcode;
3287 
3288 #if DEBUG
3289 	printf("RAID autoconfigure\n");
3290 #endif
3291 
3292 	retcode = 0;
3293 	*unit = -1;
3294 
3295 	/* 1. Create a config structure */
3296 
3297 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3298 				       M_RAIDFRAME,
3299 				       M_NOWAIT);
3300 	if (config==NULL) {
3301 		printf("Out of mem!?!?\n");
3302 				/* XXX do something more intelligent here. */
3303 		return(1);
3304 	}
3305 
3306 	memset(config, 0, sizeof(RF_Config_t));
3307 
3308 	/* XXX raidID needs to be set correctly.. */
3309 
3310 	/*
3311 	   2. Figure out what RAID ID this one is supposed to live at
3312 	   See if we can get the same RAID dev that it was configured
3313 	   on last time..
3314 	*/
3315 
3316 	raidID = cset->ac->clabel->last_unit;
3317 	if ((raidID < 0) || (raidID >= numraid)) {
3318 		/* let's not wander off into lala land. */
3319 		raidID = numraid - 1;
3320 	}
3321 	if (raidPtrs[raidID]->valid != 0) {
3322 
3323 		/*
3324 		   Nope... Go looking for an alternative...
3325 		   Start high so we don't immediately use raid0 if that's
3326 		   not taken.
3327 		*/
3328 
3329 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3330 			if (raidPtrs[raidID]->valid == 0) {
3331 				/* can use this one! */
3332 				break;
3333 			}
3334 		}
3335 	}
3336 
3337 	if (raidID < 0) {
3338 		/* punt... */
3339 		printf("Unable to auto configure this set!\n");
3340 		printf("(Out of RAID devs!)\n");
3341 		return(1);
3342 	}
3343 
3344 #if DEBUG
3345 	printf("Configuring raid%d:\n",raidID);
3346 #endif
3347 
3348 	raidPtr = raidPtrs[raidID];
3349 
3350 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3351 	raidPtr->raidid = raidID;
3352 	raidPtr->openings = RAIDOUTSTANDING;
3353 
3354 	/* 3. Build the configuration structure */
3355 	rf_create_configuration(cset->ac, config, raidPtr);
3356 
3357 	/* 4. Do the configuration */
3358 	retcode = rf_Configure(raidPtr, config, cset->ac);
3359 
3360 	if (retcode == 0) {
3361 
3362 		raidinit(raidPtrs[raidID]);
3363 
3364 		rf_markalldirty(raidPtrs[raidID]);
3365 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3366 		if (cset->ac->clabel->root_partition==1) {
3367 			/* everything configured just fine.  Make a note
3368 			   that this set is eligible to be root. */
3369 			cset->rootable = 1;
3370 			/* XXX do this here? */
3371 			raidPtrs[raidID]->root_partition = 1;
3372 		}
3373 	}
3374 
3375 	/* 5. Cleanup */
3376 	free(config, M_RAIDFRAME);
3377 
3378 	*unit = raidID;
3379 	return(retcode);
3380 }
3381 
3382 void
3383 rf_disk_unbusy(desc)
3384 	RF_RaidAccessDesc_t *desc;
3385 {
3386 	struct buf *bp;
3387 
3388 	bp = (struct buf *)desc->bp;
3389 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3390 			    (bp->b_bcount - bp->b_resid));
3391 }
3392