xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision 220b5c059a84c51ea44107ea8951a57ffaecdc8c)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.114 2001/11/28 05:39:13 lukem 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 1.114 2001/11/28 05:39:13 lukem 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_acctrace.h"
149 #include "rf_etimer.h"
150 #include "rf_general.h"
151 #include "rf_debugMem.h"
152 #include "rf_kintf.h"
153 #include "rf_options.h"
154 #include "rf_driver.h"
155 #include "rf_parityscan.h"
156 #include "rf_debugprint.h"
157 #include "rf_threadstuff.h"
158 
159 int     rf_kdebug_level = 0;
160 
161 #ifdef DEBUG
162 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
163 #else				/* DEBUG */
164 #define db1_printf(a) { }
165 #endif				/* DEBUG */
166 
167 static RF_Raid_t **raidPtrs;	/* global raid device descriptors */
168 
169 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
170 
171 static RF_SparetWait_t *rf_sparet_wait_queue;	/* requests to install a
172 						 * spare table */
173 static RF_SparetWait_t *rf_sparet_resp_queue;	/* responses from
174 						 * installation process */
175 
176 /* prototypes */
177 static void KernelWakeupFunc(struct buf * bp);
178 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag,
179 		   dev_t dev, RF_SectorNum_t startSect,
180 		   RF_SectorCount_t numSect, caddr_t buf,
181 		   void (*cbFunc) (struct buf *), void *cbArg,
182 		   int logBytesPerSector, struct proc * b_proc);
183 static void raidinit(RF_Raid_t *);
184 
185 void raidattach(int);
186 int raidsize(dev_t);
187 int raidopen(dev_t, int, int, struct proc *);
188 int raidclose(dev_t, int, int, struct proc *);
189 int raidioctl(dev_t, u_long, caddr_t, int, struct proc *);
190 int raidwrite(dev_t, struct uio *, int);
191 int raidread(dev_t, struct uio *, int);
192 void raidstrategy(struct buf *);
193 int raiddump(dev_t, daddr_t, caddr_t, size_t);
194 
195 /*
196  * Pilfered from ccd.c
197  */
198 
199 struct raidbuf {
200 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
201 	struct buf *rf_obp;	/* ptr. to original I/O buf */
202 	int     rf_flags;	/* misc. flags */
203 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
204 };
205 
206 
207 #define RAIDGETBUF(rs) pool_get(&(rs)->sc_cbufpool, PR_NOWAIT)
208 #define	RAIDPUTBUF(rs, cbp) pool_put(&(rs)->sc_cbufpool, cbp)
209 
210 /* XXX Not sure if the following should be replacing the raidPtrs above,
211    or if it should be used in conjunction with that...
212 */
213 
214 struct raid_softc {
215 	int     sc_flags;	/* flags */
216 	int     sc_cflags;	/* configuration flags */
217 	size_t  sc_size;        /* size of the raid device */
218 	char    sc_xname[20];	/* XXX external name */
219 	struct disk sc_dkdev;	/* generic disk device info */
220 	struct pool sc_cbufpool;	/* component buffer pool */
221 	struct buf_queue buf_queue;	/* used for the device queue */
222 };
223 /* sc_flags */
224 #define RAIDF_INITED	0x01	/* unit has been initialized */
225 #define RAIDF_WLABEL	0x02	/* label area is writable */
226 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
227 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
228 #define RAIDF_LOCKED	0x80	/* unit is locked */
229 
230 #define	raidunit(x)	DISKUNIT(x)
231 int numraid = 0;
232 
233 /*
234  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
235  * Be aware that large numbers can allow the driver to consume a lot of
236  * kernel memory, especially on writes, and in degraded mode reads.
237  *
238  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
239  * a single 64K write will typically require 64K for the old data,
240  * 64K for the old parity, and 64K for the new parity, for a total
241  * of 192K (if the parity buffer is not re-used immediately).
242  * Even it if is used immediately, that's still 128K, which when multiplied
243  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
244  *
245  * Now in degraded mode, for example, a 64K read on the above setup may
246  * require data reconstruction, which will require *all* of the 4 remaining
247  * disks to participate -- 4 * 32K/disk == 128K again.
248  */
249 
250 #ifndef RAIDOUTSTANDING
251 #define RAIDOUTSTANDING   6
252 #endif
253 
254 #define RAIDLABELDEV(dev)	\
255 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
256 
257 /* declared here, and made public, for the benefit of KVM stuff.. */
258 struct raid_softc *raid_softc;
259 
260 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
261 				     struct disklabel *);
262 static void raidgetdisklabel(dev_t);
263 static void raidmakedisklabel(struct raid_softc *);
264 
265 static int raidlock(struct raid_softc *);
266 static void raidunlock(struct raid_softc *);
267 
268 static void rf_markalldirty(RF_Raid_t *);
269 void rf_mountroot_hook(struct device *);
270 
271 struct device *raidrootdev;
272 
273 void rf_ReconThread(struct rf_recon_req *);
274 /* XXX what I want is: */
275 /*void rf_ReconThread(RF_Raid_t *raidPtr);  */
276 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
277 void rf_CopybackThread(RF_Raid_t *raidPtr);
278 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
279 void rf_buildroothack(void *);
280 
281 RF_AutoConfig_t *rf_find_raid_components(void);
282 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
283 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
284 static int rf_reasonable_label(RF_ComponentLabel_t *);
285 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
286 int rf_set_autoconfig(RF_Raid_t *, int);
287 int rf_set_rootpartition(RF_Raid_t *, int);
288 void rf_release_all_vps(RF_ConfigSet_t *);
289 void rf_cleanup_config_set(RF_ConfigSet_t *);
290 int rf_have_enough_components(RF_ConfigSet_t *);
291 int rf_auto_config_set(RF_ConfigSet_t *, int *);
292 
293 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
294 				  allow autoconfig to take place.
295 			          Note that this is overridden by having
296 			          RAID_AUTOCONFIG as an option in the
297 			          kernel config file.  */
298 
299 void
300 raidattach(num)
301 	int     num;
302 {
303 	int raidID;
304 	int i, rc;
305 	RF_AutoConfig_t *ac_list; /* autoconfig list */
306 	RF_ConfigSet_t *config_sets;
307 
308 #ifdef DEBUG
309 	printf("raidattach: Asked for %d units\n", num);
310 #endif
311 
312 	if (num <= 0) {
313 #ifdef DIAGNOSTIC
314 		panic("raidattach: count <= 0");
315 #endif
316 		return;
317 	}
318 	/* This is where all the initialization stuff gets done. */
319 
320 	numraid = num;
321 
322 	/* Make some space for requested number of units... */
323 
324 	RF_Calloc(raidPtrs, num, sizeof(RF_Raid_t *), (RF_Raid_t **));
325 	if (raidPtrs == NULL) {
326 		panic("raidPtrs is NULL!!\n");
327 	}
328 
329 	rc = rf_mutex_init(&rf_sparet_wait_mutex);
330 	if (rc) {
331 		RF_PANIC();
332 	}
333 
334 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
335 
336 	for (i = 0; i < num; i++)
337 		raidPtrs[i] = NULL;
338 	rc = rf_BootRaidframe();
339 	if (rc == 0)
340 		printf("Kernelized RAIDframe activated\n");
341 	else
342 		panic("Serious error booting RAID!!\n");
343 
344 	/* put together some datastructures like the CCD device does.. This
345 	 * lets us lock the device and what-not when it gets opened. */
346 
347 	raid_softc = (struct raid_softc *)
348 		malloc(num * sizeof(struct raid_softc),
349 		       M_RAIDFRAME, M_NOWAIT);
350 	if (raid_softc == NULL) {
351 		printf("WARNING: no memory for RAIDframe driver\n");
352 		return;
353 	}
354 
355 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
356 
357 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
358 					      M_RAIDFRAME, M_NOWAIT);
359 	if (raidrootdev == NULL) {
360 		panic("No memory for RAIDframe driver!!?!?!\n");
361 	}
362 
363 	for (raidID = 0; raidID < num; raidID++) {
364 		BUFQ_INIT(&raid_softc[raidID].buf_queue);
365 
366 		raidrootdev[raidID].dv_class  = DV_DISK;
367 		raidrootdev[raidID].dv_cfdata = NULL;
368 		raidrootdev[raidID].dv_unit   = raidID;
369 		raidrootdev[raidID].dv_parent = NULL;
370 		raidrootdev[raidID].dv_flags  = 0;
371 		sprintf(raidrootdev[raidID].dv_xname,"raid%d",raidID);
372 
373 		RF_Calloc(raidPtrs[raidID], 1, sizeof(RF_Raid_t),
374 			  (RF_Raid_t *));
375 		if (raidPtrs[raidID] == NULL) {
376 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
377 			numraid = raidID;
378 			return;
379 		}
380 	}
381 
382 #ifdef RAID_AUTOCONFIG
383 	raidautoconfig = 1;
384 #endif
385 
386 if (raidautoconfig) {
387 	/* 1. locate all RAID components on the system */
388 
389 #if DEBUG
390 	printf("Searching for raid components...\n");
391 #endif
392 	ac_list = rf_find_raid_components();
393 
394 	/* 2. sort them into their respective sets */
395 
396 	config_sets = rf_create_auto_sets(ac_list);
397 
398 	/* 3. evaluate each set and configure the valid ones
399 	   This gets done in rf_buildroothack() */
400 
401 	/* schedule the creation of the thread to do the
402 	   "/ on RAID" stuff */
403 
404 	kthread_create(rf_buildroothack,config_sets);
405 
406 #if 0
407 	mountroothook_establish(rf_mountroot_hook, &raidrootdev[0]);
408 #endif
409 }
410 
411 }
412 
413 void
414 rf_buildroothack(arg)
415 	void *arg;
416 {
417 	RF_ConfigSet_t *config_sets = arg;
418 	RF_ConfigSet_t *cset;
419 	RF_ConfigSet_t *next_cset;
420 	int retcode;
421 	int raidID;
422 	int rootID;
423 	int num_root;
424 
425 	rootID = 0;
426 	num_root = 0;
427 	cset = config_sets;
428 	while(cset != NULL ) {
429 		next_cset = cset->next;
430 		if (rf_have_enough_components(cset) &&
431 		    cset->ac->clabel->autoconfigure==1) {
432 			retcode = rf_auto_config_set(cset,&raidID);
433 			if (!retcode) {
434 				if (cset->rootable) {
435 					rootID = raidID;
436 					num_root++;
437 				}
438 			} else {
439 				/* The autoconfig didn't work :( */
440 #if DEBUG
441 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
442 #endif
443 				rf_release_all_vps(cset);
444 			}
445 		} else {
446 			/* we're not autoconfiguring this set...
447 			   release the associated resources */
448 			rf_release_all_vps(cset);
449 		}
450 		/* cleanup */
451 		rf_cleanup_config_set(cset);
452 		cset = next_cset;
453 	}
454 	if (boothowto & RB_ASKNAME) {
455 		/* We don't auto-config... */
456 	} else {
457 		/* They didn't ask, and we found something bootable... */
458 
459 		if (num_root == 1) {
460 			booted_device = &raidrootdev[rootID];
461 		} else if (num_root > 1) {
462 			/* we can't guess.. require the user to answer... */
463 			boothowto |= RB_ASKNAME;
464 		}
465 	}
466 }
467 
468 
469 int
470 raidsize(dev)
471 	dev_t   dev;
472 {
473 	struct raid_softc *rs;
474 	struct disklabel *lp;
475 	int     part, unit, omask, size;
476 
477 	unit = raidunit(dev);
478 	if (unit >= numraid)
479 		return (-1);
480 	rs = &raid_softc[unit];
481 
482 	if ((rs->sc_flags & RAIDF_INITED) == 0)
483 		return (-1);
484 
485 	part = DISKPART(dev);
486 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
487 	lp = rs->sc_dkdev.dk_label;
488 
489 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
490 		return (-1);
491 
492 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
493 		size = -1;
494 	else
495 		size = lp->d_partitions[part].p_size *
496 		    (lp->d_secsize / DEV_BSIZE);
497 
498 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
499 		return (-1);
500 
501 	return (size);
502 
503 }
504 
505 int
506 raiddump(dev, blkno, va, size)
507 	dev_t   dev;
508 	daddr_t blkno;
509 	caddr_t va;
510 	size_t  size;
511 {
512 	/* Not implemented. */
513 	return ENXIO;
514 }
515 /* ARGSUSED */
516 int
517 raidopen(dev, flags, fmt, p)
518 	dev_t   dev;
519 	int     flags, fmt;
520 	struct proc *p;
521 {
522 	int     unit = raidunit(dev);
523 	struct raid_softc *rs;
524 	struct disklabel *lp;
525 	int     part, pmask;
526 	int     error = 0;
527 
528 	if (unit >= numraid)
529 		return (ENXIO);
530 	rs = &raid_softc[unit];
531 
532 	if ((error = raidlock(rs)) != 0)
533 		return (error);
534 	lp = rs->sc_dkdev.dk_label;
535 
536 	part = DISKPART(dev);
537 	pmask = (1 << part);
538 
539 	db1_printf(("Opening raid device number: %d partition: %d\n",
540 		unit, part));
541 
542 
543 	if ((rs->sc_flags & RAIDF_INITED) &&
544 	    (rs->sc_dkdev.dk_openmask == 0))
545 		raidgetdisklabel(dev);
546 
547 	/* make sure that this partition exists */
548 
549 	if (part != RAW_PART) {
550 		db1_printf(("Not a raw partition..\n"));
551 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
552 		    ((part >= lp->d_npartitions) ||
553 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
554 			error = ENXIO;
555 			raidunlock(rs);
556 			db1_printf(("Bailing out...\n"));
557 			return (error);
558 		}
559 	}
560 	/* Prevent this unit from being unconfigured while open. */
561 	switch (fmt) {
562 	case S_IFCHR:
563 		rs->sc_dkdev.dk_copenmask |= pmask;
564 		break;
565 
566 	case S_IFBLK:
567 		rs->sc_dkdev.dk_bopenmask |= pmask;
568 		break;
569 	}
570 
571 	if ((rs->sc_dkdev.dk_openmask == 0) &&
572 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
573 		/* First one... mark things as dirty... Note that we *MUST*
574 		 have done a configure before this.  I DO NOT WANT TO BE
575 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
576 		 THAT THEY BELONG TOGETHER!!!!! */
577 		/* XXX should check to see if we're only open for reading
578 		   here... If so, we needn't do this, but then need some
579 		   other way of keeping track of what's happened.. */
580 
581 		rf_markalldirty( raidPtrs[unit] );
582 	}
583 
584 
585 	rs->sc_dkdev.dk_openmask =
586 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
587 
588 	raidunlock(rs);
589 
590 	return (error);
591 
592 
593 }
594 /* ARGSUSED */
595 int
596 raidclose(dev, flags, fmt, p)
597 	dev_t   dev;
598 	int     flags, fmt;
599 	struct proc *p;
600 {
601 	int     unit = raidunit(dev);
602 	struct raid_softc *rs;
603 	int     error = 0;
604 	int     part;
605 
606 	if (unit >= numraid)
607 		return (ENXIO);
608 	rs = &raid_softc[unit];
609 
610 	if ((error = raidlock(rs)) != 0)
611 		return (error);
612 
613 	part = DISKPART(dev);
614 
615 	/* ...that much closer to allowing unconfiguration... */
616 	switch (fmt) {
617 	case S_IFCHR:
618 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
619 		break;
620 
621 	case S_IFBLK:
622 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
623 		break;
624 	}
625 	rs->sc_dkdev.dk_openmask =
626 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
627 
628 	if ((rs->sc_dkdev.dk_openmask == 0) &&
629 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
630 		/* Last one... device is not unconfigured yet.
631 		   Device shutdown has taken care of setting the
632 		   clean bits if RAIDF_INITED is not set
633 		   mark things as clean... */
634 #if 0
635 		printf("Last one on raid%d.  Updating status.\n",unit);
636 #endif
637 		rf_update_component_labels(raidPtrs[unit],
638 						 RF_FINAL_COMPONENT_UPDATE);
639 		if (doing_shutdown) {
640 			/* last one, and we're going down, so
641 			   lights out for this RAID set too. */
642 			error = rf_Shutdown(raidPtrs[unit]);
643 			pool_destroy(&rs->sc_cbufpool);
644 
645 			/* It's no longer initialized... */
646 			rs->sc_flags &= ~RAIDF_INITED;
647 
648 			/* Detach the disk. */
649 			disk_detach(&rs->sc_dkdev);
650 		}
651 	}
652 
653 	raidunlock(rs);
654 	return (0);
655 
656 }
657 
658 void
659 raidstrategy(bp)
660 	struct buf *bp;
661 {
662 	int s;
663 
664 	unsigned int raidID = raidunit(bp->b_dev);
665 	RF_Raid_t *raidPtr;
666 	struct raid_softc *rs = &raid_softc[raidID];
667 	struct disklabel *lp;
668 	int     wlabel;
669 
670 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
671 		bp->b_error = ENXIO;
672 		bp->b_flags |= B_ERROR;
673 		bp->b_resid = bp->b_bcount;
674 		biodone(bp);
675 		return;
676 	}
677 	if (raidID >= numraid || !raidPtrs[raidID]) {
678 		bp->b_error = ENODEV;
679 		bp->b_flags |= B_ERROR;
680 		bp->b_resid = bp->b_bcount;
681 		biodone(bp);
682 		return;
683 	}
684 	raidPtr = raidPtrs[raidID];
685 	if (!raidPtr->valid) {
686 		bp->b_error = ENODEV;
687 		bp->b_flags |= B_ERROR;
688 		bp->b_resid = bp->b_bcount;
689 		biodone(bp);
690 		return;
691 	}
692 	if (bp->b_bcount == 0) {
693 		db1_printf(("b_bcount is zero..\n"));
694 		biodone(bp);
695 		return;
696 	}
697 	lp = rs->sc_dkdev.dk_label;
698 
699 	/*
700 	 * Do bounds checking and adjust transfer.  If there's an
701 	 * error, the bounds check will flag that for us.
702 	 */
703 
704 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
705 	if (DISKPART(bp->b_dev) != RAW_PART)
706 		if (bounds_check_with_label(bp, lp, wlabel) <= 0) {
707 			db1_printf(("Bounds check failed!!:%d %d\n",
708 				(int) bp->b_blkno, (int) wlabel));
709 			biodone(bp);
710 			return;
711 		}
712 	s = splbio();
713 
714 	bp->b_resid = 0;
715 
716 	/* stuff it onto our queue */
717 	BUFQ_INSERT_TAIL(&rs->buf_queue, bp);
718 
719 	raidstart(raidPtrs[raidID]);
720 
721 	splx(s);
722 }
723 /* ARGSUSED */
724 int
725 raidread(dev, uio, flags)
726 	dev_t   dev;
727 	struct uio *uio;
728 	int     flags;
729 {
730 	int     unit = raidunit(dev);
731 	struct raid_softc *rs;
732 	int     part;
733 
734 	if (unit >= numraid)
735 		return (ENXIO);
736 	rs = &raid_softc[unit];
737 
738 	if ((rs->sc_flags & RAIDF_INITED) == 0)
739 		return (ENXIO);
740 	part = DISKPART(dev);
741 
742 	db1_printf(("raidread: unit: %d partition: %d\n", unit, part));
743 
744 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
745 
746 }
747 /* ARGSUSED */
748 int
749 raidwrite(dev, uio, flags)
750 	dev_t   dev;
751 	struct uio *uio;
752 	int     flags;
753 {
754 	int     unit = raidunit(dev);
755 	struct raid_softc *rs;
756 
757 	if (unit >= numraid)
758 		return (ENXIO);
759 	rs = &raid_softc[unit];
760 
761 	if ((rs->sc_flags & RAIDF_INITED) == 0)
762 		return (ENXIO);
763 	db1_printf(("raidwrite\n"));
764 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
765 
766 }
767 
768 int
769 raidioctl(dev, cmd, data, flag, p)
770 	dev_t   dev;
771 	u_long  cmd;
772 	caddr_t data;
773 	int     flag;
774 	struct proc *p;
775 {
776 	int     unit = raidunit(dev);
777 	int     error = 0;
778 	int     part, pmask;
779 	struct raid_softc *rs;
780 	RF_Config_t *k_cfg, *u_cfg;
781 	RF_Raid_t *raidPtr;
782 	RF_RaidDisk_t *diskPtr;
783 	RF_AccTotals_t *totals;
784 	RF_DeviceConfig_t *d_cfg, **ucfgp;
785 	u_char *specific_buf;
786 	int retcode = 0;
787 	int row;
788 	int column;
789 	struct rf_recon_req *rrcopy, *rr;
790 	RF_ComponentLabel_t *clabel;
791 	RF_ComponentLabel_t ci_label;
792 	RF_ComponentLabel_t **clabel_ptr;
793 	RF_SingleComponent_t *sparePtr,*componentPtr;
794 	RF_SingleComponent_t hot_spare;
795 	RF_SingleComponent_t component;
796 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
797 	int i, j, d;
798 #ifdef __HAVE_OLD_DISKLABEL
799 	struct disklabel newlabel;
800 #endif
801 
802 	if (unit >= numraid)
803 		return (ENXIO);
804 	rs = &raid_softc[unit];
805 	raidPtr = raidPtrs[unit];
806 
807 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
808 		(int) DISKPART(dev), (int) unit, (int) cmd));
809 
810 	/* Must be open for writes for these commands... */
811 	switch (cmd) {
812 	case DIOCSDINFO:
813 	case DIOCWDINFO:
814 #ifdef __HAVE_OLD_DISKLABEL
815 	case ODIOCWDINFO:
816 	case ODIOCSDINFO:
817 #endif
818 	case DIOCWLABEL:
819 		if ((flag & FWRITE) == 0)
820 			return (EBADF);
821 	}
822 
823 	/* Must be initialized for these... */
824 	switch (cmd) {
825 	case DIOCGDINFO:
826 	case DIOCSDINFO:
827 	case DIOCWDINFO:
828 #ifdef __HAVE_OLD_DISKLABEL
829 	case ODIOCGDINFO:
830 	case ODIOCWDINFO:
831 	case ODIOCSDINFO:
832 	case ODIOCGDEFLABEL:
833 #endif
834 	case DIOCGPART:
835 	case DIOCWLABEL:
836 	case DIOCGDEFLABEL:
837 	case RAIDFRAME_SHUTDOWN:
838 	case RAIDFRAME_REWRITEPARITY:
839 	case RAIDFRAME_GET_INFO:
840 	case RAIDFRAME_RESET_ACCTOTALS:
841 	case RAIDFRAME_GET_ACCTOTALS:
842 	case RAIDFRAME_KEEP_ACCTOTALS:
843 	case RAIDFRAME_GET_SIZE:
844 	case RAIDFRAME_FAIL_DISK:
845 	case RAIDFRAME_COPYBACK:
846 	case RAIDFRAME_CHECK_RECON_STATUS:
847 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
848 	case RAIDFRAME_GET_COMPONENT_LABEL:
849 	case RAIDFRAME_SET_COMPONENT_LABEL:
850 	case RAIDFRAME_ADD_HOT_SPARE:
851 	case RAIDFRAME_REMOVE_HOT_SPARE:
852 	case RAIDFRAME_INIT_LABELS:
853 	case RAIDFRAME_REBUILD_IN_PLACE:
854 	case RAIDFRAME_CHECK_PARITY:
855 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
856 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
857 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
858 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
859 	case RAIDFRAME_SET_AUTOCONFIG:
860 	case RAIDFRAME_SET_ROOT:
861 	case RAIDFRAME_DELETE_COMPONENT:
862 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
863 		if ((rs->sc_flags & RAIDF_INITED) == 0)
864 			return (ENXIO);
865 	}
866 
867 	switch (cmd) {
868 
869 		/* configure the system */
870 	case RAIDFRAME_CONFIGURE:
871 
872 		if (raidPtr->valid) {
873 			/* There is a valid RAID set running on this unit! */
874 			printf("raid%d: Device already configured!\n",unit);
875 			return(EINVAL);
876 		}
877 
878 		/* copy-in the configuration information */
879 		/* data points to a pointer to the configuration structure */
880 
881 		u_cfg = *((RF_Config_t **) data);
882 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
883 		if (k_cfg == NULL) {
884 			return (ENOMEM);
885 		}
886 		retcode = copyin((caddr_t) u_cfg, (caddr_t) k_cfg,
887 		    sizeof(RF_Config_t));
888 		if (retcode) {
889 			RF_Free(k_cfg, sizeof(RF_Config_t));
890 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
891 				retcode));
892 			return (retcode);
893 		}
894 		/* allocate a buffer for the layout-specific data, and copy it
895 		 * in */
896 		if (k_cfg->layoutSpecificSize) {
897 			if (k_cfg->layoutSpecificSize > 10000) {
898 				/* sanity check */
899 				RF_Free(k_cfg, sizeof(RF_Config_t));
900 				return (EINVAL);
901 			}
902 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
903 			    (u_char *));
904 			if (specific_buf == NULL) {
905 				RF_Free(k_cfg, sizeof(RF_Config_t));
906 				return (ENOMEM);
907 			}
908 			retcode = copyin(k_cfg->layoutSpecific,
909 			    (caddr_t) specific_buf,
910 			    k_cfg->layoutSpecificSize);
911 			if (retcode) {
912 				RF_Free(k_cfg, sizeof(RF_Config_t));
913 				RF_Free(specific_buf,
914 					k_cfg->layoutSpecificSize);
915 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
916 					retcode));
917 				return (retcode);
918 			}
919 		} else
920 			specific_buf = NULL;
921 		k_cfg->layoutSpecific = specific_buf;
922 
923 		/* should do some kind of sanity check on the configuration.
924 		 * Store the sum of all the bytes in the last byte? */
925 
926 		/* configure the system */
927 
928 		/*
929 		 * Clear the entire RAID descriptor, just to make sure
930 		 *  there is no stale data left in the case of a
931 		 *  reconfiguration
932 		 */
933 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
934 		raidPtr->raidid = unit;
935 
936 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
937 
938 		if (retcode == 0) {
939 
940 			/* allow this many simultaneous IO's to
941 			   this RAID device */
942 			raidPtr->openings = RAIDOUTSTANDING;
943 
944 			raidinit(raidPtr);
945 			rf_markalldirty(raidPtr);
946 		}
947 		/* free the buffers.  No return code here. */
948 		if (k_cfg->layoutSpecificSize) {
949 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
950 		}
951 		RF_Free(k_cfg, sizeof(RF_Config_t));
952 
953 		return (retcode);
954 
955 		/* shutdown the system */
956 	case RAIDFRAME_SHUTDOWN:
957 
958 		if ((error = raidlock(rs)) != 0)
959 			return (error);
960 
961 		/*
962 		 * If somebody has a partition mounted, we shouldn't
963 		 * shutdown.
964 		 */
965 
966 		part = DISKPART(dev);
967 		pmask = (1 << part);
968 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
969 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
970 			(rs->sc_dkdev.dk_copenmask & pmask))) {
971 			raidunlock(rs);
972 			return (EBUSY);
973 		}
974 
975 		retcode = rf_Shutdown(raidPtr);
976 
977 		pool_destroy(&rs->sc_cbufpool);
978 
979 		/* It's no longer initialized... */
980 		rs->sc_flags &= ~RAIDF_INITED;
981 
982 		/* Detach the disk. */
983 		disk_detach(&rs->sc_dkdev);
984 
985 		raidunlock(rs);
986 
987 		return (retcode);
988 	case RAIDFRAME_GET_COMPONENT_LABEL:
989 		clabel_ptr = (RF_ComponentLabel_t **) data;
990 		/* need to read the component label for the disk indicated
991 		   by row,column in clabel */
992 
993 		/* For practice, let's get it directly fromdisk, rather
994 		   than from the in-core copy */
995 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
996 			   (RF_ComponentLabel_t *));
997 		if (clabel == NULL)
998 			return (ENOMEM);
999 
1000 		memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t));
1001 
1002 		retcode = copyin( *clabel_ptr, clabel,
1003 				  sizeof(RF_ComponentLabel_t));
1004 
1005 		if (retcode) {
1006 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1007 			return(retcode);
1008 		}
1009 
1010 		row = clabel->row;
1011 		column = clabel->column;
1012 
1013 		if ((row < 0) || (row >= raidPtr->numRow) ||
1014 		    (column < 0) || (column >= raidPtr->numCol +
1015 				     raidPtr->numSpare)) {
1016 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1017 			return(EINVAL);
1018 		}
1019 
1020 		raidread_component_label(raidPtr->Disks[row][column].dev,
1021 				raidPtr->raid_cinfo[row][column].ci_vp,
1022 				clabel );
1023 
1024 		retcode = copyout((caddr_t) clabel,
1025 				  (caddr_t) *clabel_ptr,
1026 				  sizeof(RF_ComponentLabel_t));
1027 		RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1028 		return (retcode);
1029 
1030 	case RAIDFRAME_SET_COMPONENT_LABEL:
1031 		clabel = (RF_ComponentLabel_t *) data;
1032 
1033 		/* XXX check the label for valid stuff... */
1034 		/* Note that some things *should not* get modified --
1035 		   the user should be re-initing the labels instead of
1036 		   trying to patch things.
1037 		   */
1038 
1039 		printf("Got component label:\n");
1040 		printf("Version: %d\n",clabel->version);
1041 		printf("Serial Number: %d\n",clabel->serial_number);
1042 		printf("Mod counter: %d\n",clabel->mod_counter);
1043 		printf("Row: %d\n", clabel->row);
1044 		printf("Column: %d\n", clabel->column);
1045 		printf("Num Rows: %d\n", clabel->num_rows);
1046 		printf("Num Columns: %d\n", clabel->num_columns);
1047 		printf("Clean: %d\n", clabel->clean);
1048 		printf("Status: %d\n", clabel->status);
1049 
1050 		row = clabel->row;
1051 		column = clabel->column;
1052 
1053 		if ((row < 0) || (row >= raidPtr->numRow) ||
1054 		    (column < 0) || (column >= raidPtr->numCol)) {
1055 			return(EINVAL);
1056 		}
1057 
1058 		/* XXX this isn't allowed to do anything for now :-) */
1059 
1060 		/* XXX and before it is, we need to fill in the rest
1061 		   of the fields!?!?!?! */
1062 #if 0
1063 		raidwrite_component_label(
1064                             raidPtr->Disks[row][column].dev,
1065 			    raidPtr->raid_cinfo[row][column].ci_vp,
1066 			    clabel );
1067 #endif
1068 		return (0);
1069 
1070 	case RAIDFRAME_INIT_LABELS:
1071 		clabel = (RF_ComponentLabel_t *) data;
1072 		/*
1073 		   we only want the serial number from
1074 		   the above.  We get all the rest of the information
1075 		   from the config that was used to create this RAID
1076 		   set.
1077 		   */
1078 
1079 		raidPtr->serial_number = clabel->serial_number;
1080 
1081 		raid_init_component_label(raidPtr, &ci_label);
1082 		ci_label.serial_number = clabel->serial_number;
1083 
1084 		for(row=0;row<raidPtr->numRow;row++) {
1085 			ci_label.row = row;
1086 			for(column=0;column<raidPtr->numCol;column++) {
1087 				diskPtr = &raidPtr->Disks[row][column];
1088 				if (!RF_DEAD_DISK(diskPtr->status)) {
1089 					ci_label.partitionSize = diskPtr->partitionSize;
1090 					ci_label.column = column;
1091 					raidwrite_component_label(
1092 					  raidPtr->Disks[row][column].dev,
1093 					  raidPtr->raid_cinfo[row][column].ci_vp,
1094 					  &ci_label );
1095 				}
1096 			}
1097 		}
1098 
1099 		return (retcode);
1100 	case RAIDFRAME_SET_AUTOCONFIG:
1101 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1102 		printf("New autoconfig value is: %d\n", d);
1103 		*(int *) data = d;
1104 		return (retcode);
1105 
1106 	case RAIDFRAME_SET_ROOT:
1107 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1108 		printf("New rootpartition value is: %d\n", d);
1109 		*(int *) data = d;
1110 		return (retcode);
1111 
1112 		/* initialize all parity */
1113 	case RAIDFRAME_REWRITEPARITY:
1114 
1115 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1116 			/* Parity for RAID 0 is trivially correct */
1117 			raidPtr->parity_good = RF_RAID_CLEAN;
1118 			return(0);
1119 		}
1120 
1121 		if (raidPtr->parity_rewrite_in_progress == 1) {
1122 			/* Re-write is already in progress! */
1123 			return(EINVAL);
1124 		}
1125 
1126 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1127 					   rf_RewriteParityThread,
1128 					   raidPtr,"raid_parity");
1129 		return (retcode);
1130 
1131 
1132 	case RAIDFRAME_ADD_HOT_SPARE:
1133 		sparePtr = (RF_SingleComponent_t *) data;
1134 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1135 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1136 		return(retcode);
1137 
1138 	case RAIDFRAME_REMOVE_HOT_SPARE:
1139 		return(retcode);
1140 
1141 	case RAIDFRAME_DELETE_COMPONENT:
1142 		componentPtr = (RF_SingleComponent_t *)data;
1143 		memcpy( &component, componentPtr,
1144 			sizeof(RF_SingleComponent_t));
1145 		retcode = rf_delete_component(raidPtr, &component);
1146 		return(retcode);
1147 
1148 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1149 		componentPtr = (RF_SingleComponent_t *)data;
1150 		memcpy( &component, componentPtr,
1151 			sizeof(RF_SingleComponent_t));
1152 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1153 		return(retcode);
1154 
1155 	case RAIDFRAME_REBUILD_IN_PLACE:
1156 
1157 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1158 			/* Can't do this on a RAID 0!! */
1159 			return(EINVAL);
1160 		}
1161 
1162 		if (raidPtr->recon_in_progress == 1) {
1163 			/* a reconstruct is already in progress! */
1164 			return(EINVAL);
1165 		}
1166 
1167 		componentPtr = (RF_SingleComponent_t *) data;
1168 		memcpy( &component, componentPtr,
1169 			sizeof(RF_SingleComponent_t));
1170 		row = component.row;
1171 		column = component.column;
1172 		printf("Rebuild: %d %d\n",row, column);
1173 		if ((row < 0) || (row >= raidPtr->numRow) ||
1174 		    (column < 0) || (column >= raidPtr->numCol)) {
1175 			return(EINVAL);
1176 		}
1177 
1178 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1179 		if (rrcopy == NULL)
1180 			return(ENOMEM);
1181 
1182 		rrcopy->raidPtr = (void *) raidPtr;
1183 		rrcopy->row = row;
1184 		rrcopy->col = column;
1185 
1186 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1187 					   rf_ReconstructInPlaceThread,
1188 					   rrcopy,"raid_reconip");
1189 		return(retcode);
1190 
1191 	case RAIDFRAME_GET_INFO:
1192 		if (!raidPtr->valid)
1193 			return (ENODEV);
1194 		ucfgp = (RF_DeviceConfig_t **) data;
1195 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1196 			  (RF_DeviceConfig_t *));
1197 		if (d_cfg == NULL)
1198 			return (ENOMEM);
1199 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1200 		d_cfg->rows = raidPtr->numRow;
1201 		d_cfg->cols = raidPtr->numCol;
1202 		d_cfg->ndevs = raidPtr->numRow * raidPtr->numCol;
1203 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1204 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1205 			return (ENOMEM);
1206 		}
1207 		d_cfg->nspares = raidPtr->numSpare;
1208 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1209 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1210 			return (ENOMEM);
1211 		}
1212 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1213 		d = 0;
1214 		for (i = 0; i < d_cfg->rows; i++) {
1215 			for (j = 0; j < d_cfg->cols; j++) {
1216 				d_cfg->devs[d] = raidPtr->Disks[i][j];
1217 				d++;
1218 			}
1219 		}
1220 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1221 			d_cfg->spares[i] = raidPtr->Disks[0][j];
1222 		}
1223 		retcode = copyout((caddr_t) d_cfg, (caddr_t) * ucfgp,
1224 				  sizeof(RF_DeviceConfig_t));
1225 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1226 
1227 		return (retcode);
1228 
1229 	case RAIDFRAME_CHECK_PARITY:
1230 		*(int *) data = raidPtr->parity_good;
1231 		return (0);
1232 
1233 	case RAIDFRAME_RESET_ACCTOTALS:
1234 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1235 		return (0);
1236 
1237 	case RAIDFRAME_GET_ACCTOTALS:
1238 		totals = (RF_AccTotals_t *) data;
1239 		*totals = raidPtr->acc_totals;
1240 		return (0);
1241 
1242 	case RAIDFRAME_KEEP_ACCTOTALS:
1243 		raidPtr->keep_acc_totals = *(int *)data;
1244 		return (0);
1245 
1246 	case RAIDFRAME_GET_SIZE:
1247 		*(int *) data = raidPtr->totalSectors;
1248 		return (0);
1249 
1250 		/* fail a disk & optionally start reconstruction */
1251 	case RAIDFRAME_FAIL_DISK:
1252 
1253 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1254 			/* Can't do this on a RAID 0!! */
1255 			return(EINVAL);
1256 		}
1257 
1258 		rr = (struct rf_recon_req *) data;
1259 
1260 		if (rr->row < 0 || rr->row >= raidPtr->numRow
1261 		    || rr->col < 0 || rr->col >= raidPtr->numCol)
1262 			return (EINVAL);
1263 
1264 		printf("raid%d: Failing the disk: row: %d col: %d\n",
1265 		       unit, rr->row, rr->col);
1266 
1267 		/* make a copy of the recon request so that we don't rely on
1268 		 * the user's buffer */
1269 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1270 		if (rrcopy == NULL)
1271 			return(ENOMEM);
1272 		bcopy(rr, rrcopy, sizeof(*rr));
1273 		rrcopy->raidPtr = (void *) raidPtr;
1274 
1275 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1276 					   rf_ReconThread,
1277 					   rrcopy,"raid_recon");
1278 		return (0);
1279 
1280 		/* invoke a copyback operation after recon on whatever disk
1281 		 * needs it, if any */
1282 	case RAIDFRAME_COPYBACK:
1283 
1284 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1285 			/* This makes no sense on a RAID 0!! */
1286 			return(EINVAL);
1287 		}
1288 
1289 		if (raidPtr->copyback_in_progress == 1) {
1290 			/* Copyback is already in progress! */
1291 			return(EINVAL);
1292 		}
1293 
1294 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1295 					   rf_CopybackThread,
1296 					   raidPtr,"raid_copyback");
1297 		return (retcode);
1298 
1299 		/* return the percentage completion of reconstruction */
1300 	case RAIDFRAME_CHECK_RECON_STATUS:
1301 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1302 			/* This makes no sense on a RAID 0, so tell the
1303 			   user it's done. */
1304 			*(int *) data = 100;
1305 			return(0);
1306 		}
1307 		row = 0; /* XXX we only consider a single row... */
1308 		if (raidPtr->status[row] != rf_rs_reconstructing)
1309 			*(int *) data = 100;
1310 		else
1311 			*(int *) data = raidPtr->reconControl[row]->percentComplete;
1312 		return (0);
1313 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1314 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1315 		row = 0; /* XXX we only consider a single row... */
1316 		if (raidPtr->status[row] != rf_rs_reconstructing) {
1317 			progressInfo.remaining = 0;
1318 			progressInfo.completed = 100;
1319 			progressInfo.total = 100;
1320 		} else {
1321 			progressInfo.total =
1322 				raidPtr->reconControl[row]->numRUsTotal;
1323 			progressInfo.completed =
1324 				raidPtr->reconControl[row]->numRUsComplete;
1325 			progressInfo.remaining = progressInfo.total -
1326 				progressInfo.completed;
1327 		}
1328 		retcode = copyout((caddr_t) &progressInfo,
1329 				  (caddr_t) *progressInfoPtr,
1330 				  sizeof(RF_ProgressInfo_t));
1331 		return (retcode);
1332 
1333 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1334 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1335 			/* This makes no sense on a RAID 0, so tell the
1336 			   user it's done. */
1337 			*(int *) data = 100;
1338 			return(0);
1339 		}
1340 		if (raidPtr->parity_rewrite_in_progress == 1) {
1341 			*(int *) data = 100 *
1342 				raidPtr->parity_rewrite_stripes_done /
1343 				raidPtr->Layout.numStripe;
1344 		} else {
1345 			*(int *) data = 100;
1346 		}
1347 		return (0);
1348 
1349 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1350 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1351 		if (raidPtr->parity_rewrite_in_progress == 1) {
1352 			progressInfo.total = raidPtr->Layout.numStripe;
1353 			progressInfo.completed =
1354 				raidPtr->parity_rewrite_stripes_done;
1355 			progressInfo.remaining = progressInfo.total -
1356 				progressInfo.completed;
1357 		} else {
1358 			progressInfo.remaining = 0;
1359 			progressInfo.completed = 100;
1360 			progressInfo.total = 100;
1361 		}
1362 		retcode = copyout((caddr_t) &progressInfo,
1363 				  (caddr_t) *progressInfoPtr,
1364 				  sizeof(RF_ProgressInfo_t));
1365 		return (retcode);
1366 
1367 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1368 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1369 			/* This makes no sense on a RAID 0 */
1370 			*(int *) data = 100;
1371 			return(0);
1372 		}
1373 		if (raidPtr->copyback_in_progress == 1) {
1374 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1375 				raidPtr->Layout.numStripe;
1376 		} else {
1377 			*(int *) data = 100;
1378 		}
1379 		return (0);
1380 
1381 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1382 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1383 		if (raidPtr->copyback_in_progress == 1) {
1384 			progressInfo.total = raidPtr->Layout.numStripe;
1385 			progressInfo.completed =
1386 				raidPtr->copyback_stripes_done;
1387 			progressInfo.remaining = progressInfo.total -
1388 				progressInfo.completed;
1389 		} else {
1390 			progressInfo.remaining = 0;
1391 			progressInfo.completed = 100;
1392 			progressInfo.total = 100;
1393 		}
1394 		retcode = copyout((caddr_t) &progressInfo,
1395 				  (caddr_t) *progressInfoPtr,
1396 				  sizeof(RF_ProgressInfo_t));
1397 		return (retcode);
1398 
1399 		/* the sparetable daemon calls this to wait for the kernel to
1400 		 * need a spare table. this ioctl does not return until a
1401 		 * spare table is needed. XXX -- calling mpsleep here in the
1402 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1403 		 * -- I should either compute the spare table in the kernel,
1404 		 * or have a different -- XXX XXX -- interface (a different
1405 		 * character device) for delivering the table     -- XXX */
1406 #if 0
1407 	case RAIDFRAME_SPARET_WAIT:
1408 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1409 		while (!rf_sparet_wait_queue)
1410 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1411 		waitreq = rf_sparet_wait_queue;
1412 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1413 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1414 
1415 		/* structure assignment */
1416 		*((RF_SparetWait_t *) data) = *waitreq;
1417 
1418 		RF_Free(waitreq, sizeof(*waitreq));
1419 		return (0);
1420 
1421 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1422 		 * code in it that will cause the dameon to exit */
1423 	case RAIDFRAME_ABORT_SPARET_WAIT:
1424 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1425 		waitreq->fcol = -1;
1426 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1427 		waitreq->next = rf_sparet_wait_queue;
1428 		rf_sparet_wait_queue = waitreq;
1429 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1430 		wakeup(&rf_sparet_wait_queue);
1431 		return (0);
1432 
1433 		/* used by the spare table daemon to deliver a spare table
1434 		 * into the kernel */
1435 	case RAIDFRAME_SEND_SPARET:
1436 
1437 		/* install the spare table */
1438 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1439 
1440 		/* respond to the requestor.  the return status of the spare
1441 		 * table installation is passed in the "fcol" field */
1442 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1443 		waitreq->fcol = retcode;
1444 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1445 		waitreq->next = rf_sparet_resp_queue;
1446 		rf_sparet_resp_queue = waitreq;
1447 		wakeup(&rf_sparet_resp_queue);
1448 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1449 
1450 		return (retcode);
1451 #endif
1452 
1453 	default:
1454 		break; /* fall through to the os-specific code below */
1455 
1456 	}
1457 
1458 	if (!raidPtr->valid)
1459 		return (EINVAL);
1460 
1461 	/*
1462 	 * Add support for "regular" device ioctls here.
1463 	 */
1464 
1465 	switch (cmd) {
1466 	case DIOCGDINFO:
1467 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1468 		break;
1469 #ifdef __HAVE_OLD_DISKLABEL
1470 	case ODIOCGDINFO:
1471 		newlabel = *(rs->sc_dkdev.dk_label);
1472 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1473 			return ENOTTY;
1474 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1475 		break;
1476 #endif
1477 
1478 	case DIOCGPART:
1479 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1480 		((struct partinfo *) data)->part =
1481 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1482 		break;
1483 
1484 	case DIOCWDINFO:
1485 	case DIOCSDINFO:
1486 #ifdef __HAVE_OLD_DISKLABEL
1487 	case ODIOCWDINFO:
1488 	case ODIOCSDINFO:
1489 #endif
1490 	{
1491 		struct disklabel *lp;
1492 #ifdef __HAVE_OLD_DISKLABEL
1493 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1494 			memset(&newlabel, 0, sizeof newlabel);
1495 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1496 			lp = &newlabel;
1497 		} else
1498 #endif
1499 		lp = (struct disklabel *)data;
1500 
1501 		if ((error = raidlock(rs)) != 0)
1502 			return (error);
1503 
1504 		rs->sc_flags |= RAIDF_LABELLING;
1505 
1506 		error = setdisklabel(rs->sc_dkdev.dk_label,
1507 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1508 		if (error == 0) {
1509 			if (cmd == DIOCWDINFO
1510 #ifdef __HAVE_OLD_DISKLABEL
1511 			    || cmd == ODIOCWDINFO
1512 #endif
1513 			   )
1514 				error = writedisklabel(RAIDLABELDEV(dev),
1515 				    raidstrategy, rs->sc_dkdev.dk_label,
1516 				    rs->sc_dkdev.dk_cpulabel);
1517 		}
1518 		rs->sc_flags &= ~RAIDF_LABELLING;
1519 
1520 		raidunlock(rs);
1521 
1522 		if (error)
1523 			return (error);
1524 		break;
1525 	}
1526 
1527 	case DIOCWLABEL:
1528 		if (*(int *) data != 0)
1529 			rs->sc_flags |= RAIDF_WLABEL;
1530 		else
1531 			rs->sc_flags &= ~RAIDF_WLABEL;
1532 		break;
1533 
1534 	case DIOCGDEFLABEL:
1535 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1536 		break;
1537 
1538 #ifdef __HAVE_OLD_DISKLABEL
1539 	case ODIOCGDEFLABEL:
1540 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1541 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1542 			return ENOTTY;
1543 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1544 		break;
1545 #endif
1546 
1547 	default:
1548 		retcode = ENOTTY;
1549 	}
1550 	return (retcode);
1551 
1552 }
1553 
1554 
1555 /* raidinit -- complete the rest of the initialization for the
1556    RAIDframe device.  */
1557 
1558 
1559 static void
1560 raidinit(raidPtr)
1561 	RF_Raid_t *raidPtr;
1562 {
1563 	struct raid_softc *rs;
1564 	int     unit;
1565 
1566 	unit = raidPtr->raidid;
1567 
1568 	rs = &raid_softc[unit];
1569 	pool_init(&rs->sc_cbufpool, sizeof(struct raidbuf), 0,
1570 		  0, 0, "raidpl", 0, NULL, NULL, M_RAIDFRAME);
1571 
1572 
1573 	/* XXX should check return code first... */
1574 	rs->sc_flags |= RAIDF_INITED;
1575 
1576 	sprintf(rs->sc_xname, "raid%d", unit);	/* XXX doesn't check bounds. */
1577 
1578 	rs->sc_dkdev.dk_name = rs->sc_xname;
1579 
1580 	/* disk_attach actually creates space for the CPU disklabel, among
1581 	 * other things, so it's critical to call this *BEFORE* we try putzing
1582 	 * with disklabels. */
1583 
1584 	disk_attach(&rs->sc_dkdev);
1585 
1586 	/* XXX There may be a weird interaction here between this, and
1587 	 * protectedSectors, as used in RAIDframe.  */
1588 
1589 	rs->sc_size = raidPtr->totalSectors;
1590 
1591 }
1592 
1593 /* wake up the daemon & tell it to get us a spare table
1594  * XXX
1595  * the entries in the queues should be tagged with the raidPtr
1596  * so that in the extremely rare case that two recons happen at once,
1597  * we know for which device were requesting a spare table
1598  * XXX
1599  *
1600  * XXX This code is not currently used. GO
1601  */
1602 int
1603 rf_GetSpareTableFromDaemon(req)
1604 	RF_SparetWait_t *req;
1605 {
1606 	int     retcode;
1607 
1608 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1609 	req->next = rf_sparet_wait_queue;
1610 	rf_sparet_wait_queue = req;
1611 	wakeup(&rf_sparet_wait_queue);
1612 
1613 	/* mpsleep unlocks the mutex */
1614 	while (!rf_sparet_resp_queue) {
1615 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1616 		    "raidframe getsparetable", 0);
1617 	}
1618 	req = rf_sparet_resp_queue;
1619 	rf_sparet_resp_queue = req->next;
1620 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1621 
1622 	retcode = req->fcol;
1623 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1624 					 * alloc'd */
1625 	return (retcode);
1626 }
1627 
1628 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1629  * bp & passes it down.
1630  * any calls originating in the kernel must use non-blocking I/O
1631  * do some extra sanity checking to return "appropriate" error values for
1632  * certain conditions (to make some standard utilities work)
1633  *
1634  * Formerly known as: rf_DoAccessKernel
1635  */
1636 void
1637 raidstart(raidPtr)
1638 	RF_Raid_t *raidPtr;
1639 {
1640 	RF_SectorCount_t num_blocks, pb, sum;
1641 	RF_RaidAddr_t raid_addr;
1642 	int     retcode;
1643 	struct partition *pp;
1644 	daddr_t blocknum;
1645 	int     unit;
1646 	struct raid_softc *rs;
1647 	int     do_async;
1648 	struct buf *bp;
1649 
1650 	unit = raidPtr->raidid;
1651 	rs = &raid_softc[unit];
1652 
1653 	/* quick check to see if anything has died recently */
1654 	RF_LOCK_MUTEX(raidPtr->mutex);
1655 	if (raidPtr->numNewFailures > 0) {
1656 		rf_update_component_labels(raidPtr,
1657 					   RF_NORMAL_COMPONENT_UPDATE);
1658 		raidPtr->numNewFailures--;
1659 	}
1660 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1661 
1662 	/* Check to see if we're at the limit... */
1663 	RF_LOCK_MUTEX(raidPtr->mutex);
1664 	while (raidPtr->openings > 0) {
1665 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1666 
1667 		/* get the next item, if any, from the queue */
1668 		if ((bp = BUFQ_FIRST(&rs->buf_queue)) == NULL) {
1669 			/* nothing more to do */
1670 			return;
1671 		}
1672 		BUFQ_REMOVE(&rs->buf_queue, bp);
1673 
1674 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1675 		 * partition.. Need to make it absolute to the underlying
1676 		 * device.. */
1677 
1678 		blocknum = bp->b_blkno;
1679 		if (DISKPART(bp->b_dev) != RAW_PART) {
1680 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1681 			blocknum += pp->p_offset;
1682 		}
1683 
1684 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1685 			    (int) blocknum));
1686 
1687 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1688 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1689 
1690 		/* *THIS* is where we adjust what block we're going to...
1691 		 * but DO NOT TOUCH bp->b_blkno!!! */
1692 		raid_addr = blocknum;
1693 
1694 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1695 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1696 		sum = raid_addr + num_blocks + pb;
1697 		if (1 || rf_debugKernelAccess) {
1698 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1699 				    (int) raid_addr, (int) sum, (int) num_blocks,
1700 				    (int) pb, (int) bp->b_resid));
1701 		}
1702 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1703 		    || (sum < num_blocks) || (sum < pb)) {
1704 			bp->b_error = ENOSPC;
1705 			bp->b_flags |= B_ERROR;
1706 			bp->b_resid = bp->b_bcount;
1707 			biodone(bp);
1708 			RF_LOCK_MUTEX(raidPtr->mutex);
1709 			continue;
1710 		}
1711 		/*
1712 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1713 		 */
1714 
1715 		if (bp->b_bcount & raidPtr->sectorMask) {
1716 			bp->b_error = EINVAL;
1717 			bp->b_flags |= B_ERROR;
1718 			bp->b_resid = bp->b_bcount;
1719 			biodone(bp);
1720 			RF_LOCK_MUTEX(raidPtr->mutex);
1721 			continue;
1722 
1723 		}
1724 		db1_printf(("Calling DoAccess..\n"));
1725 
1726 
1727 		RF_LOCK_MUTEX(raidPtr->mutex);
1728 		raidPtr->openings--;
1729 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1730 
1731 		/*
1732 		 * Everything is async.
1733 		 */
1734 		do_async = 1;
1735 
1736 		disk_busy(&rs->sc_dkdev);
1737 
1738 		/* XXX we're still at splbio() here... do we *really*
1739 		   need to be? */
1740 
1741 		/* don't ever condition on bp->b_flags & B_WRITE.
1742 		 * always condition on B_READ instead */
1743 
1744 		retcode = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1745 				      RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1746 				      do_async, raid_addr, num_blocks,
1747 				      bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1748 
1749 		RF_LOCK_MUTEX(raidPtr->mutex);
1750 	}
1751 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1752 }
1753 
1754 
1755 
1756 
1757 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1758 
1759 int
1760 rf_DispatchKernelIO(queue, req)
1761 	RF_DiskQueue_t *queue;
1762 	RF_DiskQueueData_t *req;
1763 {
1764 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1765 	struct buf *bp;
1766 	struct raidbuf *raidbp = NULL;
1767 	struct raid_softc *rs;
1768 	int     unit;
1769 	int s;
1770 
1771 	s=0;
1772 	/* s = splbio();*/ /* want to test this */
1773 	/* XXX along with the vnode, we also need the softc associated with
1774 	 * this device.. */
1775 
1776 	req->queue = queue;
1777 
1778 	unit = queue->raidPtr->raidid;
1779 
1780 	db1_printf(("DispatchKernelIO unit: %d\n", unit));
1781 
1782 	if (unit >= numraid) {
1783 		printf("Invalid unit number: %d %d\n", unit, numraid);
1784 		panic("Invalid Unit number in rf_DispatchKernelIO\n");
1785 	}
1786 	rs = &raid_softc[unit];
1787 
1788 	bp = req->bp;
1789 #if 1
1790 	/* XXX when there is a physical disk failure, someone is passing us a
1791 	 * buffer that contains old stuff!!  Attempt to deal with this problem
1792 	 * without taking a performance hit... (not sure where the real bug
1793 	 * is.  It's buried in RAIDframe somewhere) :-(  GO ) */
1794 
1795 	if (bp->b_flags & B_ERROR) {
1796 		bp->b_flags &= ~B_ERROR;
1797 	}
1798 	if (bp->b_error != 0) {
1799 		bp->b_error = 0;
1800 	}
1801 #endif
1802 	raidbp = RAIDGETBUF(rs);
1803 
1804 	raidbp->rf_flags = 0;	/* XXX not really used anywhere... */
1805 
1806 	/*
1807 	 * context for raidiodone
1808 	 */
1809 	raidbp->rf_obp = bp;
1810 	raidbp->req = req;
1811 
1812 	LIST_INIT(&raidbp->rf_buf.b_dep);
1813 
1814 	switch (req->type) {
1815 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1816 		/* XXX need to do something extra here.. */
1817 		/* I'm leaving this in, as I've never actually seen it used,
1818 		 * and I'd like folks to report it... GO */
1819 		printf(("WAKEUP CALLED\n"));
1820 		queue->numOutstanding++;
1821 
1822 		/* XXX need to glue the original buffer into this??  */
1823 
1824 		KernelWakeupFunc(&raidbp->rf_buf);
1825 		break;
1826 
1827 	case RF_IO_TYPE_READ:
1828 	case RF_IO_TYPE_WRITE:
1829 
1830 		if (req->tracerec) {
1831 			RF_ETIMER_START(req->tracerec->timer);
1832 		}
1833 		InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1834 		    op | bp->b_flags, queue->rf_cinfo->ci_dev,
1835 		    req->sectorOffset, req->numSector,
1836 		    req->buf, KernelWakeupFunc, (void *) req,
1837 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1838 
1839 		if (rf_debugKernelAccess) {
1840 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1841 				(long) bp->b_blkno));
1842 		}
1843 		queue->numOutstanding++;
1844 		queue->last_deq_sector = req->sectorOffset;
1845 		/* acc wouldn't have been let in if there were any pending
1846 		 * reqs at any other priority */
1847 		queue->curPriority = req->priority;
1848 
1849 		db1_printf(("Going for %c to unit %d row %d col %d\n",
1850 			req->type, unit, queue->row, queue->col));
1851 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1852 			(int) req->sectorOffset, (int) req->numSector,
1853 			(int) (req->numSector <<
1854 			    queue->raidPtr->logBytesPerSector),
1855 			(int) queue->raidPtr->logBytesPerSector));
1856 		if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1857 			raidbp->rf_buf.b_vp->v_numoutput++;
1858 		}
1859 		VOP_STRATEGY(&raidbp->rf_buf);
1860 
1861 		break;
1862 
1863 	default:
1864 		panic("bad req->type in rf_DispatchKernelIO");
1865 	}
1866 	db1_printf(("Exiting from DispatchKernelIO\n"));
1867 	/* splx(s); */ /* want to test this */
1868 	return (0);
1869 }
1870 /* this is the callback function associated with a I/O invoked from
1871    kernel code.
1872  */
1873 static void
1874 KernelWakeupFunc(vbp)
1875 	struct buf *vbp;
1876 {
1877 	RF_DiskQueueData_t *req = NULL;
1878 	RF_DiskQueue_t *queue;
1879 	struct raidbuf *raidbp = (struct raidbuf *) vbp;
1880 	struct buf *bp;
1881 	struct raid_softc *rs;
1882 	int     unit;
1883 	int s;
1884 
1885 	s = splbio();
1886 	db1_printf(("recovering the request queue:\n"));
1887 	req = raidbp->req;
1888 
1889 	bp = raidbp->rf_obp;
1890 
1891 	queue = (RF_DiskQueue_t *) req->queue;
1892 
1893 	if (raidbp->rf_buf.b_flags & B_ERROR) {
1894 		bp->b_flags |= B_ERROR;
1895 		bp->b_error = raidbp->rf_buf.b_error ?
1896 		    raidbp->rf_buf.b_error : EIO;
1897 	}
1898 
1899 	/* XXX methinks this could be wrong... */
1900 #if 1
1901 	bp->b_resid = raidbp->rf_buf.b_resid;
1902 #endif
1903 
1904 	if (req->tracerec) {
1905 		RF_ETIMER_STOP(req->tracerec->timer);
1906 		RF_ETIMER_EVAL(req->tracerec->timer);
1907 		RF_LOCK_MUTEX(rf_tracing_mutex);
1908 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1909 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1910 		req->tracerec->num_phys_ios++;
1911 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1912 	}
1913 	bp->b_bcount = raidbp->rf_buf.b_bcount;	/* XXXX ?? */
1914 
1915 	unit = queue->raidPtr->raidid;	/* *Much* simpler :-> */
1916 
1917 
1918 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1919 	 * ballistic, and mark the component as hosed... */
1920 
1921 	if (bp->b_flags & B_ERROR) {
1922 		/* Mark the disk as dead */
1923 		/* but only mark it once... */
1924 		if (queue->raidPtr->Disks[queue->row][queue->col].status ==
1925 		    rf_ds_optimal) {
1926 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1927 			    unit, queue->raidPtr->Disks[queue->row][queue->col].devname);
1928 			queue->raidPtr->Disks[queue->row][queue->col].status =
1929 			    rf_ds_failed;
1930 			queue->raidPtr->status[queue->row] = rf_rs_degraded;
1931 			queue->raidPtr->numFailures++;
1932 			queue->raidPtr->numNewFailures++;
1933 		} else {	/* Disk is already dead... */
1934 			/* printf("Disk already marked as dead!\n"); */
1935 		}
1936 
1937 	}
1938 
1939 	rs = &raid_softc[unit];
1940 	RAIDPUTBUF(rs, raidbp);
1941 
1942 	rf_DiskIOComplete(queue, req, (bp->b_flags & B_ERROR) ? 1 : 0);
1943 	(req->CompleteFunc) (req->argument, (bp->b_flags & B_ERROR) ? 1 : 0);
1944 
1945 	splx(s);
1946 }
1947 
1948 
1949 
1950 /*
1951  * initialize a buf structure for doing an I/O in the kernel.
1952  */
1953 static void
1954 InitBP(bp, b_vp, rw_flag, dev, startSect, numSect, buf, cbFunc, cbArg,
1955        logBytesPerSector, b_proc)
1956 	struct buf *bp;
1957 	struct vnode *b_vp;
1958 	unsigned rw_flag;
1959 	dev_t dev;
1960 	RF_SectorNum_t startSect;
1961 	RF_SectorCount_t numSect;
1962 	caddr_t buf;
1963 	void (*cbFunc) (struct buf *);
1964 	void *cbArg;
1965 	int logBytesPerSector;
1966 	struct proc *b_proc;
1967 {
1968 	/* bp->b_flags       = B_PHYS | rw_flag; */
1969 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
1970 	bp->b_bcount = numSect << logBytesPerSector;
1971 	bp->b_bufsize = bp->b_bcount;
1972 	bp->b_error = 0;
1973 	bp->b_dev = dev;
1974 	bp->b_data = buf;
1975 	bp->b_blkno = startSect;
1976 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
1977 	if (bp->b_bcount == 0) {
1978 		panic("bp->b_bcount is zero in InitBP!!\n");
1979 	}
1980 	bp->b_proc = b_proc;
1981 	bp->b_iodone = cbFunc;
1982 	bp->b_vp = b_vp;
1983 
1984 }
1985 
1986 static void
1987 raidgetdefaultlabel(raidPtr, rs, lp)
1988 	RF_Raid_t *raidPtr;
1989 	struct raid_softc *rs;
1990 	struct disklabel *lp;
1991 {
1992 	db1_printf(("Building a default label...\n"));
1993 	memset(lp, 0, sizeof(*lp));
1994 
1995 	/* fabricate a label... */
1996 	lp->d_secperunit = raidPtr->totalSectors;
1997 	lp->d_secsize = raidPtr->bytesPerSector;
1998 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
1999 	lp->d_ntracks = 4 * raidPtr->numCol;
2000 	lp->d_ncylinders = raidPtr->totalSectors /
2001 		(lp->d_nsectors * lp->d_ntracks);
2002 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
2003 
2004 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
2005 	lp->d_type = DTYPE_RAID;
2006 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
2007 	lp->d_rpm = 3600;
2008 	lp->d_interleave = 1;
2009 	lp->d_flags = 0;
2010 
2011 	lp->d_partitions[RAW_PART].p_offset = 0;
2012 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2013 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2014 	lp->d_npartitions = RAW_PART + 1;
2015 
2016 	lp->d_magic = DISKMAGIC;
2017 	lp->d_magic2 = DISKMAGIC;
2018 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2019 
2020 }
2021 /*
2022  * Read the disklabel from the raid device.  If one is not present, fake one
2023  * up.
2024  */
2025 static void
2026 raidgetdisklabel(dev)
2027 	dev_t   dev;
2028 {
2029 	int     unit = raidunit(dev);
2030 	struct raid_softc *rs = &raid_softc[unit];
2031 	char   *errstring;
2032 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2033 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2034 	RF_Raid_t *raidPtr;
2035 
2036 	db1_printf(("Getting the disklabel...\n"));
2037 
2038 	memset(clp, 0, sizeof(*clp));
2039 
2040 	raidPtr = raidPtrs[unit];
2041 
2042 	raidgetdefaultlabel(raidPtr, rs, lp);
2043 
2044 	/*
2045 	 * Call the generic disklabel extraction routine.
2046 	 */
2047 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2048 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2049 	if (errstring)
2050 		raidmakedisklabel(rs);
2051 	else {
2052 		int     i;
2053 		struct partition *pp;
2054 
2055 		/*
2056 		 * Sanity check whether the found disklabel is valid.
2057 		 *
2058 		 * This is necessary since total size of the raid device
2059 		 * may vary when an interleave is changed even though exactly
2060 		 * same componets are used, and old disklabel may used
2061 		 * if that is found.
2062 		 */
2063 		if (lp->d_secperunit != rs->sc_size)
2064 			printf("WARNING: %s: "
2065 			    "total sector size in disklabel (%d) != "
2066 			    "the size of raid (%ld)\n", rs->sc_xname,
2067 			    lp->d_secperunit, (long) rs->sc_size);
2068 		for (i = 0; i < lp->d_npartitions; i++) {
2069 			pp = &lp->d_partitions[i];
2070 			if (pp->p_offset + pp->p_size > rs->sc_size)
2071 				printf("WARNING: %s: end of partition `%c' "
2072 				    "exceeds the size of raid (%ld)\n",
2073 				    rs->sc_xname, 'a' + i, (long) rs->sc_size);
2074 		}
2075 	}
2076 
2077 }
2078 /*
2079  * Take care of things one might want to take care of in the event
2080  * that a disklabel isn't present.
2081  */
2082 static void
2083 raidmakedisklabel(rs)
2084 	struct raid_softc *rs;
2085 {
2086 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2087 	db1_printf(("Making a label..\n"));
2088 
2089 	/*
2090 	 * For historical reasons, if there's no disklabel present
2091 	 * the raw partition must be marked FS_BSDFFS.
2092 	 */
2093 
2094 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2095 
2096 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2097 
2098 	lp->d_checksum = dkcksum(lp);
2099 }
2100 /*
2101  * Lookup the provided name in the filesystem.  If the file exists,
2102  * is a valid block device, and isn't being used by anyone else,
2103  * set *vpp to the file's vnode.
2104  * You'll find the original of this in ccd.c
2105  */
2106 int
2107 raidlookup(path, p, vpp)
2108 	char   *path;
2109 	struct proc *p;
2110 	struct vnode **vpp;	/* result */
2111 {
2112 	struct nameidata nd;
2113 	struct vnode *vp;
2114 	struct vattr va;
2115 	int     error;
2116 
2117 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2118 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2119 #ifdef DEBUG
2120 		printf("RAIDframe: vn_open returned %d\n", error);
2121 #endif
2122 		return (error);
2123 	}
2124 	vp = nd.ni_vp;
2125 	if (vp->v_usecount > 1) {
2126 		VOP_UNLOCK(vp, 0);
2127 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2128 		return (EBUSY);
2129 	}
2130 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2131 		VOP_UNLOCK(vp, 0);
2132 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2133 		return (error);
2134 	}
2135 	/* XXX: eventually we should handle VREG, too. */
2136 	if (va.va_type != VBLK) {
2137 		VOP_UNLOCK(vp, 0);
2138 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2139 		return (ENOTBLK);
2140 	}
2141 	VOP_UNLOCK(vp, 0);
2142 	*vpp = vp;
2143 	return (0);
2144 }
2145 /*
2146  * Wait interruptibly for an exclusive lock.
2147  *
2148  * XXX
2149  * Several drivers do this; it should be abstracted and made MP-safe.
2150  * (Hmm... where have we seen this warning before :->  GO )
2151  */
2152 static int
2153 raidlock(rs)
2154 	struct raid_softc *rs;
2155 {
2156 	int     error;
2157 
2158 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2159 		rs->sc_flags |= RAIDF_WANTED;
2160 		if ((error =
2161 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2162 			return (error);
2163 	}
2164 	rs->sc_flags |= RAIDF_LOCKED;
2165 	return (0);
2166 }
2167 /*
2168  * Unlock and wake up any waiters.
2169  */
2170 static void
2171 raidunlock(rs)
2172 	struct raid_softc *rs;
2173 {
2174 
2175 	rs->sc_flags &= ~RAIDF_LOCKED;
2176 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2177 		rs->sc_flags &= ~RAIDF_WANTED;
2178 		wakeup(rs);
2179 	}
2180 }
2181 
2182 
2183 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2184 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2185 
2186 int
2187 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2188 {
2189 	RF_ComponentLabel_t clabel;
2190 	raidread_component_label(dev, b_vp, &clabel);
2191 	clabel.mod_counter = mod_counter;
2192 	clabel.clean = RF_RAID_CLEAN;
2193 	raidwrite_component_label(dev, b_vp, &clabel);
2194 	return(0);
2195 }
2196 
2197 
2198 int
2199 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2200 {
2201 	RF_ComponentLabel_t clabel;
2202 	raidread_component_label(dev, b_vp, &clabel);
2203 	clabel.mod_counter = mod_counter;
2204 	clabel.clean = RF_RAID_DIRTY;
2205 	raidwrite_component_label(dev, b_vp, &clabel);
2206 	return(0);
2207 }
2208 
2209 /* ARGSUSED */
2210 int
2211 raidread_component_label(dev, b_vp, clabel)
2212 	dev_t dev;
2213 	struct vnode *b_vp;
2214 	RF_ComponentLabel_t *clabel;
2215 {
2216 	struct buf *bp;
2217 	int error;
2218 
2219 	/* XXX should probably ensure that we don't try to do this if
2220 	   someone has changed rf_protected_sectors. */
2221 
2222 	if (b_vp == NULL) {
2223 		/* For whatever reason, this component is not valid.
2224 		   Don't try to read a component label from it. */
2225 		return(EINVAL);
2226 	}
2227 
2228 	/* get a block of the appropriate size... */
2229 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2230 	bp->b_dev = dev;
2231 
2232 	/* get our ducks in a row for the read */
2233 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2234 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2235 	bp->b_flags |= B_READ;
2236  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2237 
2238 	(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
2239 
2240 	error = biowait(bp);
2241 
2242 	if (!error) {
2243 		memcpy(clabel, bp->b_data,
2244 		       sizeof(RF_ComponentLabel_t));
2245 #if 0
2246 		rf_print_component_label( clabel );
2247 #endif
2248         } else {
2249 #if 0
2250 		printf("Failed to read RAID component label!\n");
2251 #endif
2252 	}
2253 
2254 	brelse(bp);
2255 	return(error);
2256 }
2257 /* ARGSUSED */
2258 int
2259 raidwrite_component_label(dev, b_vp, clabel)
2260 	dev_t dev;
2261 	struct vnode *b_vp;
2262 	RF_ComponentLabel_t *clabel;
2263 {
2264 	struct buf *bp;
2265 	int error;
2266 
2267 	/* get a block of the appropriate size... */
2268 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2269 	bp->b_dev = dev;
2270 
2271 	/* get our ducks in a row for the write */
2272 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2273 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2274 	bp->b_flags |= B_WRITE;
2275  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2276 
2277 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2278 
2279 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2280 
2281 	(*bdevsw[major(bp->b_dev)].d_strategy)(bp);
2282 	error = biowait(bp);
2283 	brelse(bp);
2284 	if (error) {
2285 #if 1
2286 		printf("Failed to write RAID component info!\n");
2287 #endif
2288 	}
2289 
2290 	return(error);
2291 }
2292 
2293 void
2294 rf_markalldirty(raidPtr)
2295 	RF_Raid_t *raidPtr;
2296 {
2297 	RF_ComponentLabel_t clabel;
2298 	int r,c;
2299 
2300 	raidPtr->mod_counter++;
2301 	for (r = 0; r < raidPtr->numRow; r++) {
2302 		for (c = 0; c < raidPtr->numCol; c++) {
2303 			/* we don't want to touch (at all) a disk that has
2304 			   failed */
2305 			if (!RF_DEAD_DISK(raidPtr->Disks[r][c].status)) {
2306 				raidread_component_label(
2307 					raidPtr->Disks[r][c].dev,
2308 					raidPtr->raid_cinfo[r][c].ci_vp,
2309 					&clabel);
2310 				if (clabel.status == rf_ds_spared) {
2311 					/* XXX do something special...
2312 					 but whatever you do, don't
2313 					 try to access it!! */
2314 				} else {
2315 #if 0
2316 				clabel.status =
2317 					raidPtr->Disks[r][c].status;
2318 				raidwrite_component_label(
2319 					raidPtr->Disks[r][c].dev,
2320 					raidPtr->raid_cinfo[r][c].ci_vp,
2321 					&clabel);
2322 #endif
2323 				raidmarkdirty(
2324 				       raidPtr->Disks[r][c].dev,
2325 				       raidPtr->raid_cinfo[r][c].ci_vp,
2326 				       raidPtr->mod_counter);
2327 				}
2328 			}
2329 		}
2330 	}
2331 	/* printf("Component labels marked dirty.\n"); */
2332 #if 0
2333 	for( c = 0; c < raidPtr->numSpare ; c++) {
2334 		sparecol = raidPtr->numCol + c;
2335 		if (raidPtr->Disks[r][sparecol].status == rf_ds_used_spare) {
2336 			/*
2337 
2338 			   XXX this is where we get fancy and map this spare
2339 			   into it's correct spot in the array.
2340 
2341 			 */
2342 			/*
2343 
2344 			   we claim this disk is "optimal" if it's
2345 			   rf_ds_used_spare, as that means it should be
2346 			   directly substitutable for the disk it replaced.
2347 			   We note that too...
2348 
2349 			 */
2350 
2351 			for(i=0;i<raidPtr->numRow;i++) {
2352 				for(j=0;j<raidPtr->numCol;j++) {
2353 					if ((raidPtr->Disks[i][j].spareRow ==
2354 					     r) &&
2355 					    (raidPtr->Disks[i][j].spareCol ==
2356 					     sparecol)) {
2357 						srow = r;
2358 						scol = sparecol;
2359 						break;
2360 					}
2361 				}
2362 			}
2363 
2364 			raidread_component_label(
2365 				      raidPtr->Disks[r][sparecol].dev,
2366 				      raidPtr->raid_cinfo[r][sparecol].ci_vp,
2367 				      &clabel);
2368 			/* make sure status is noted */
2369 			clabel.version = RF_COMPONENT_LABEL_VERSION;
2370 			clabel.mod_counter = raidPtr->mod_counter;
2371 			clabel.serial_number = raidPtr->serial_number;
2372 			clabel.row = srow;
2373 			clabel.column = scol;
2374 			clabel.num_rows = raidPtr->numRow;
2375 			clabel.num_columns = raidPtr->numCol;
2376 			clabel.clean = RF_RAID_DIRTY; /* changed in a bit*/
2377 			clabel.status = rf_ds_optimal;
2378 			raidwrite_component_label(
2379 				      raidPtr->Disks[r][sparecol].dev,
2380 				      raidPtr->raid_cinfo[r][sparecol].ci_vp,
2381 				      &clabel);
2382 			raidmarkclean( raidPtr->Disks[r][sparecol].dev,
2383 			              raidPtr->raid_cinfo[r][sparecol].ci_vp);
2384 		}
2385 	}
2386 
2387 #endif
2388 }
2389 
2390 
2391 void
2392 rf_update_component_labels(raidPtr, final)
2393 	RF_Raid_t *raidPtr;
2394 	int final;
2395 {
2396 	RF_ComponentLabel_t clabel;
2397 	int sparecol;
2398 	int r,c;
2399 	int i,j;
2400 	int srow, scol;
2401 
2402 	srow = -1;
2403 	scol = -1;
2404 
2405 	/* XXX should do extra checks to make sure things really are clean,
2406 	   rather than blindly setting the clean bit... */
2407 
2408 	raidPtr->mod_counter++;
2409 
2410 	for (r = 0; r < raidPtr->numRow; r++) {
2411 		for (c = 0; c < raidPtr->numCol; c++) {
2412 			if (raidPtr->Disks[r][c].status == rf_ds_optimal) {
2413 				raidread_component_label(
2414 					raidPtr->Disks[r][c].dev,
2415 					raidPtr->raid_cinfo[r][c].ci_vp,
2416 					&clabel);
2417 				/* make sure status is noted */
2418 				clabel.status = rf_ds_optimal;
2419 				/* bump the counter */
2420 				clabel.mod_counter = raidPtr->mod_counter;
2421 
2422 				raidwrite_component_label(
2423 					raidPtr->Disks[r][c].dev,
2424 					raidPtr->raid_cinfo[r][c].ci_vp,
2425 					&clabel);
2426 				if (final == RF_FINAL_COMPONENT_UPDATE) {
2427 					if (raidPtr->parity_good == RF_RAID_CLEAN) {
2428 						raidmarkclean(
2429 							      raidPtr->Disks[r][c].dev,
2430 							      raidPtr->raid_cinfo[r][c].ci_vp,
2431 							      raidPtr->mod_counter);
2432 					}
2433 				}
2434 			}
2435 			/* else we don't touch it.. */
2436 		}
2437 	}
2438 
2439 	for( c = 0; c < raidPtr->numSpare ; c++) {
2440 		sparecol = raidPtr->numCol + c;
2441 		/* Need to ensure that the reconstruct actually completed! */
2442 		if (raidPtr->Disks[0][sparecol].status == rf_ds_used_spare) {
2443 			/*
2444 
2445 			   we claim this disk is "optimal" if it's
2446 			   rf_ds_used_spare, as that means it should be
2447 			   directly substitutable for the disk it replaced.
2448 			   We note that too...
2449 
2450 			 */
2451 
2452 			for(i=0;i<raidPtr->numRow;i++) {
2453 				for(j=0;j<raidPtr->numCol;j++) {
2454 					if ((raidPtr->Disks[i][j].spareRow ==
2455 					     0) &&
2456 					    (raidPtr->Disks[i][j].spareCol ==
2457 					     sparecol)) {
2458 						srow = i;
2459 						scol = j;
2460 						break;
2461 					}
2462 				}
2463 			}
2464 
2465 			/* XXX shouldn't *really* need this... */
2466 			raidread_component_label(
2467 				      raidPtr->Disks[0][sparecol].dev,
2468 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2469 				      &clabel);
2470 			/* make sure status is noted */
2471 
2472 			raid_init_component_label(raidPtr, &clabel);
2473 
2474 			clabel.mod_counter = raidPtr->mod_counter;
2475 			clabel.row = srow;
2476 			clabel.column = scol;
2477 			clabel.status = rf_ds_optimal;
2478 
2479 			raidwrite_component_label(
2480 				      raidPtr->Disks[0][sparecol].dev,
2481 				      raidPtr->raid_cinfo[0][sparecol].ci_vp,
2482 				      &clabel);
2483 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2484 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2485 					raidmarkclean( raidPtr->Disks[0][sparecol].dev,
2486 						       raidPtr->raid_cinfo[0][sparecol].ci_vp,
2487 						       raidPtr->mod_counter);
2488 				}
2489 			}
2490 		}
2491 	}
2492 	/* 	printf("Component labels updated\n"); */
2493 }
2494 
2495 void
2496 rf_close_component(raidPtr, vp, auto_configured)
2497 	RF_Raid_t *raidPtr;
2498 	struct vnode *vp;
2499 	int auto_configured;
2500 {
2501 	struct proc *p;
2502 
2503 	p = raidPtr->engine_thread;
2504 
2505 	if (vp != NULL) {
2506 		if (auto_configured == 1) {
2507 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2508 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2509 			vput(vp);
2510 
2511 		} else {
2512 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2513 		}
2514 	} else {
2515 		printf("vnode was NULL\n");
2516 	}
2517 }
2518 
2519 
2520 void
2521 rf_UnconfigureVnodes(raidPtr)
2522 	RF_Raid_t *raidPtr;
2523 {
2524 	int r,c;
2525 	struct proc *p;
2526 	struct vnode *vp;
2527 	int acd;
2528 
2529 
2530 	/* We take this opportunity to close the vnodes like we should.. */
2531 
2532 	p = raidPtr->engine_thread;
2533 
2534 	for (r = 0; r < raidPtr->numRow; r++) {
2535 		for (c = 0; c < raidPtr->numCol; c++) {
2536 			printf("Closing vnode for row: %d col: %d\n", r, c);
2537 			vp = raidPtr->raid_cinfo[r][c].ci_vp;
2538 			acd = raidPtr->Disks[r][c].auto_configured;
2539 			rf_close_component(raidPtr, vp, acd);
2540 			raidPtr->raid_cinfo[r][c].ci_vp = NULL;
2541 			raidPtr->Disks[r][c].auto_configured = 0;
2542 		}
2543 	}
2544 	for (r = 0; r < raidPtr->numSpare; r++) {
2545 		printf("Closing vnode for spare: %d\n", r);
2546 		vp = raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp;
2547 		acd = raidPtr->Disks[0][raidPtr->numCol + r].auto_configured;
2548 		rf_close_component(raidPtr, vp, acd);
2549 		raidPtr->raid_cinfo[0][raidPtr->numCol + r].ci_vp = NULL;
2550 		raidPtr->Disks[0][raidPtr->numCol + r].auto_configured = 0;
2551 	}
2552 }
2553 
2554 
2555 void
2556 rf_ReconThread(req)
2557 	struct rf_recon_req *req;
2558 {
2559 	int     s;
2560 	RF_Raid_t *raidPtr;
2561 
2562 	s = splbio();
2563 	raidPtr = (RF_Raid_t *) req->raidPtr;
2564 	raidPtr->recon_in_progress = 1;
2565 
2566 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->row, req->col,
2567 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2568 
2569 	/* XXX get rid of this! we don't need it at all.. */
2570 	RF_Free(req, sizeof(*req));
2571 
2572 	raidPtr->recon_in_progress = 0;
2573 	splx(s);
2574 
2575 	/* That's all... */
2576 	kthread_exit(0);        /* does not return */
2577 }
2578 
2579 void
2580 rf_RewriteParityThread(raidPtr)
2581 	RF_Raid_t *raidPtr;
2582 {
2583 	int retcode;
2584 	int s;
2585 
2586 	raidPtr->parity_rewrite_in_progress = 1;
2587 	s = splbio();
2588 	retcode = rf_RewriteParity(raidPtr);
2589 	splx(s);
2590 	if (retcode) {
2591 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2592 	} else {
2593 		/* set the clean bit!  If we shutdown correctly,
2594 		   the clean bit on each component label will get
2595 		   set */
2596 		raidPtr->parity_good = RF_RAID_CLEAN;
2597 	}
2598 	raidPtr->parity_rewrite_in_progress = 0;
2599 
2600 	/* Anyone waiting for us to stop?  If so, inform them... */
2601 	if (raidPtr->waitShutdown) {
2602 		wakeup(&raidPtr->parity_rewrite_in_progress);
2603 	}
2604 
2605 	/* That's all... */
2606 	kthread_exit(0);        /* does not return */
2607 }
2608 
2609 
2610 void
2611 rf_CopybackThread(raidPtr)
2612 	RF_Raid_t *raidPtr;
2613 {
2614 	int s;
2615 
2616 	raidPtr->copyback_in_progress = 1;
2617 	s = splbio();
2618 	rf_CopybackReconstructedData(raidPtr);
2619 	splx(s);
2620 	raidPtr->copyback_in_progress = 0;
2621 
2622 	/* That's all... */
2623 	kthread_exit(0);        /* does not return */
2624 }
2625 
2626 
2627 void
2628 rf_ReconstructInPlaceThread(req)
2629 	struct rf_recon_req *req;
2630 {
2631 	int retcode;
2632 	int s;
2633 	RF_Raid_t *raidPtr;
2634 
2635 	s = splbio();
2636 	raidPtr = req->raidPtr;
2637 	raidPtr->recon_in_progress = 1;
2638 	retcode = rf_ReconstructInPlace(raidPtr, req->row, req->col);
2639 	RF_Free(req, sizeof(*req));
2640 	raidPtr->recon_in_progress = 0;
2641 	splx(s);
2642 
2643 	/* That's all... */
2644 	kthread_exit(0);        /* does not return */
2645 }
2646 
2647 void
2648 rf_mountroot_hook(dev)
2649 	struct device *dev;
2650 {
2651 
2652 }
2653 
2654 
2655 RF_AutoConfig_t *
2656 rf_find_raid_components()
2657 {
2658 	struct devnametobdevmaj *dtobdm;
2659 	struct vnode *vp;
2660 	struct disklabel label;
2661 	struct device *dv;
2662 	char *cd_name;
2663 	dev_t dev;
2664 	int error;
2665 	int i;
2666 	int good_one;
2667 	RF_ComponentLabel_t *clabel;
2668 	RF_AutoConfig_t *ac_list;
2669 	RF_AutoConfig_t *ac;
2670 
2671 
2672 	/* initialize the AutoConfig list */
2673 	ac_list = NULL;
2674 
2675 	/* we begin by trolling through *all* the devices on the system */
2676 
2677 	for (dv = alldevs.tqh_first; dv != NULL;
2678 	     dv = dv->dv_list.tqe_next) {
2679 
2680 		/* we are only interested in disks... */
2681 		if (dv->dv_class != DV_DISK)
2682 			continue;
2683 
2684 		/* we don't care about floppies... */
2685 		if (!strcmp(dv->dv_cfdata->cf_driver->cd_name,"fd")) {
2686 			continue;
2687 		}
2688 
2689 		/* need to find the device_name_to_block_device_major stuff */
2690 		cd_name = dv->dv_cfdata->cf_driver->cd_name;
2691 		dtobdm = dev_name2blk;
2692 		while (dtobdm->d_name && strcmp(dtobdm->d_name, cd_name)) {
2693 			dtobdm++;
2694 		}
2695 
2696 		/* get a vnode for the raw partition of this disk */
2697 
2698 		dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, RAW_PART);
2699 		if (bdevvp(dev, &vp))
2700 			panic("RAID can't alloc vnode");
2701 
2702 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2703 
2704 		if (error) {
2705 			/* "Who cares."  Continue looking
2706 			   for something that exists*/
2707 			vput(vp);
2708 			continue;
2709 		}
2710 
2711 		/* Ok, the disk exists.  Go get the disklabel. */
2712 		error = VOP_IOCTL(vp, DIOCGDINFO, (caddr_t)&label,
2713 				  FREAD, NOCRED, 0);
2714 		if (error) {
2715 			/*
2716 			 * XXX can't happen - open() would
2717 			 * have errored out (or faked up one)
2718 			 */
2719 			printf("can't get label for dev %s%c (%d)!?!?\n",
2720 			       dv->dv_xname, 'a' + RAW_PART, error);
2721 		}
2722 
2723 		/* don't need this any more.  We'll allocate it again
2724 		   a little later if we really do... */
2725 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2726 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2727 		vput(vp);
2728 
2729 		for (i=0; i < label.d_npartitions; i++) {
2730 			/* We only support partitions marked as RAID */
2731 			if (label.d_partitions[i].p_fstype != FS_RAID)
2732 				continue;
2733 
2734 			dev = MAKEDISKDEV(dtobdm->d_maj, dv->dv_unit, i);
2735 			if (bdevvp(dev, &vp))
2736 				panic("RAID can't alloc vnode");
2737 
2738 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2739 			if (error) {
2740 				/* Whatever... */
2741 				vput(vp);
2742 				continue;
2743 			}
2744 
2745 			good_one = 0;
2746 
2747 			clabel = (RF_ComponentLabel_t *)
2748 				malloc(sizeof(RF_ComponentLabel_t),
2749 				       M_RAIDFRAME, M_NOWAIT);
2750 			if (clabel == NULL) {
2751 				/* XXX CLEANUP HERE */
2752 				printf("RAID auto config: out of memory!\n");
2753 				return(NULL); /* XXX probably should panic? */
2754 			}
2755 
2756 			if (!raidread_component_label(dev, vp, clabel)) {
2757 				/* Got the label.  Does it look reasonable? */
2758 				if (rf_reasonable_label(clabel) &&
2759 				    (clabel->partitionSize <=
2760 				     label.d_partitions[i].p_size)) {
2761 #if DEBUG
2762 					printf("Component on: %s%c: %d\n",
2763 					       dv->dv_xname, 'a'+i,
2764 					       label.d_partitions[i].p_size);
2765 					rf_print_component_label(clabel);
2766 #endif
2767 					/* if it's reasonable, add it,
2768 					   else ignore it. */
2769 					ac = (RF_AutoConfig_t *)
2770 						malloc(sizeof(RF_AutoConfig_t),
2771 						       M_RAIDFRAME,
2772 						       M_NOWAIT);
2773 					if (ac == NULL) {
2774 						/* XXX should panic?? */
2775 						return(NULL);
2776 					}
2777 
2778 					sprintf(ac->devname, "%s%c",
2779 						dv->dv_xname, 'a'+i);
2780 					ac->dev = dev;
2781 					ac->vp = vp;
2782 					ac->clabel = clabel;
2783 					ac->next = ac_list;
2784 					ac_list = ac;
2785 					good_one = 1;
2786 				}
2787 			}
2788 			if (!good_one) {
2789 				/* cleanup */
2790 				free(clabel, M_RAIDFRAME);
2791 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2792 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2793 				vput(vp);
2794 			}
2795 		}
2796 	}
2797 	return(ac_list);
2798 }
2799 
2800 static int
2801 rf_reasonable_label(clabel)
2802 	RF_ComponentLabel_t *clabel;
2803 {
2804 
2805 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2806 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2807 	    ((clabel->clean == RF_RAID_CLEAN) ||
2808 	     (clabel->clean == RF_RAID_DIRTY)) &&
2809 	    clabel->row >=0 &&
2810 	    clabel->column >= 0 &&
2811 	    clabel->num_rows > 0 &&
2812 	    clabel->num_columns > 0 &&
2813 	    clabel->row < clabel->num_rows &&
2814 	    clabel->column < clabel->num_columns &&
2815 	    clabel->blockSize > 0 &&
2816 	    clabel->numBlocks > 0) {
2817 		/* label looks reasonable enough... */
2818 		return(1);
2819 	}
2820 	return(0);
2821 }
2822 
2823 
2824 void
2825 rf_print_component_label(clabel)
2826 	RF_ComponentLabel_t *clabel;
2827 {
2828 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2829 	       clabel->row, clabel->column,
2830 	       clabel->num_rows, clabel->num_columns);
2831 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2832 	       clabel->version, clabel->serial_number,
2833 	       clabel->mod_counter);
2834 	printf("   Clean: %s Status: %d\n",
2835 	       clabel->clean ? "Yes" : "No", clabel->status );
2836 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2837 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2838 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2839 	       (char) clabel->parityConfig, clabel->blockSize,
2840 	       clabel->numBlocks);
2841 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2842 	printf("   Contains root partition: %s\n",
2843 	       clabel->root_partition ? "Yes" : "No" );
2844 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2845 #if 0
2846 	   printf("   Config order: %d\n", clabel->config_order);
2847 #endif
2848 
2849 }
2850 
2851 RF_ConfigSet_t *
2852 rf_create_auto_sets(ac_list)
2853 	RF_AutoConfig_t *ac_list;
2854 {
2855 	RF_AutoConfig_t *ac;
2856 	RF_ConfigSet_t *config_sets;
2857 	RF_ConfigSet_t *cset;
2858 	RF_AutoConfig_t *ac_next;
2859 
2860 
2861 	config_sets = NULL;
2862 
2863 	/* Go through the AutoConfig list, and figure out which components
2864 	   belong to what sets.  */
2865 	ac = ac_list;
2866 	while(ac!=NULL) {
2867 		/* we're going to putz with ac->next, so save it here
2868 		   for use at the end of the loop */
2869 		ac_next = ac->next;
2870 
2871 		if (config_sets == NULL) {
2872 			/* will need at least this one... */
2873 			config_sets = (RF_ConfigSet_t *)
2874 				malloc(sizeof(RF_ConfigSet_t),
2875 				       M_RAIDFRAME, M_NOWAIT);
2876 			if (config_sets == NULL) {
2877 				panic("rf_create_auto_sets: No memory!\n");
2878 			}
2879 			/* this one is easy :) */
2880 			config_sets->ac = ac;
2881 			config_sets->next = NULL;
2882 			config_sets->rootable = 0;
2883 			ac->next = NULL;
2884 		} else {
2885 			/* which set does this component fit into? */
2886 			cset = config_sets;
2887 			while(cset!=NULL) {
2888 				if (rf_does_it_fit(cset, ac)) {
2889 					/* looks like it matches... */
2890 					ac->next = cset->ac;
2891 					cset->ac = ac;
2892 					break;
2893 				}
2894 				cset = cset->next;
2895 			}
2896 			if (cset==NULL) {
2897 				/* didn't find a match above... new set..*/
2898 				cset = (RF_ConfigSet_t *)
2899 					malloc(sizeof(RF_ConfigSet_t),
2900 					       M_RAIDFRAME, M_NOWAIT);
2901 				if (cset == NULL) {
2902 					panic("rf_create_auto_sets: No memory!\n");
2903 				}
2904 				cset->ac = ac;
2905 				ac->next = NULL;
2906 				cset->next = config_sets;
2907 				cset->rootable = 0;
2908 				config_sets = cset;
2909 			}
2910 		}
2911 		ac = ac_next;
2912 	}
2913 
2914 
2915 	return(config_sets);
2916 }
2917 
2918 static int
2919 rf_does_it_fit(cset, ac)
2920 	RF_ConfigSet_t *cset;
2921 	RF_AutoConfig_t *ac;
2922 {
2923 	RF_ComponentLabel_t *clabel1, *clabel2;
2924 
2925 	/* If this one matches the *first* one in the set, that's good
2926 	   enough, since the other members of the set would have been
2927 	   through here too... */
2928 	/* note that we are not checking partitionSize here..
2929 
2930 	   Note that we are also not checking the mod_counters here.
2931 	   If everything else matches execpt the mod_counter, that's
2932 	   good enough for this test.  We will deal with the mod_counters
2933 	   a little later in the autoconfiguration process.
2934 
2935 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2936 
2937 	   The reason we don't check for this is that failed disks
2938 	   will have lower modification counts.  If those disks are
2939 	   not added to the set they used to belong to, then they will
2940 	   form their own set, which may result in 2 different sets,
2941 	   for example, competing to be configured at raid0, and
2942 	   perhaps competing to be the root filesystem set.  If the
2943 	   wrong ones get configured, or both attempt to become /,
2944 	   weird behaviour and or serious lossage will occur.  Thus we
2945 	   need to bring them into the fold here, and kick them out at
2946 	   a later point.
2947 
2948 	*/
2949 
2950 	clabel1 = cset->ac->clabel;
2951 	clabel2 = ac->clabel;
2952 	if ((clabel1->version == clabel2->version) &&
2953 	    (clabel1->serial_number == clabel2->serial_number) &&
2954 	    (clabel1->num_rows == clabel2->num_rows) &&
2955 	    (clabel1->num_columns == clabel2->num_columns) &&
2956 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2957 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2958 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2959 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2960 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2961 	    (clabel1->blockSize == clabel2->blockSize) &&
2962 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2963 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2964 	    (clabel1->root_partition == clabel2->root_partition) &&
2965 	    (clabel1->last_unit == clabel2->last_unit) &&
2966 	    (clabel1->config_order == clabel2->config_order)) {
2967 		/* if it get's here, it almost *has* to be a match */
2968 	} else {
2969 		/* it's not consistent with somebody in the set..
2970 		   punt */
2971 		return(0);
2972 	}
2973 	/* all was fine.. it must fit... */
2974 	return(1);
2975 }
2976 
2977 int
2978 rf_have_enough_components(cset)
2979 	RF_ConfigSet_t *cset;
2980 {
2981 	RF_AutoConfig_t *ac;
2982 	RF_AutoConfig_t *auto_config;
2983 	RF_ComponentLabel_t *clabel;
2984 	int r,c;
2985 	int num_rows;
2986 	int num_cols;
2987 	int num_missing;
2988 	int mod_counter;
2989 	int mod_counter_found;
2990 	int even_pair_failed;
2991 	char parity_type;
2992 
2993 
2994 	/* check to see that we have enough 'live' components
2995 	   of this set.  If so, we can configure it if necessary */
2996 
2997 	num_rows = cset->ac->clabel->num_rows;
2998 	num_cols = cset->ac->clabel->num_columns;
2999 	parity_type = cset->ac->clabel->parityConfig;
3000 
3001 	/* XXX Check for duplicate components!?!?!? */
3002 
3003 	/* Determine what the mod_counter is supposed to be for this set. */
3004 
3005 	mod_counter_found = 0;
3006 	mod_counter = 0;
3007 	ac = cset->ac;
3008 	while(ac!=NULL) {
3009 		if (mod_counter_found==0) {
3010 			mod_counter = ac->clabel->mod_counter;
3011 			mod_counter_found = 1;
3012 		} else {
3013 			if (ac->clabel->mod_counter > mod_counter) {
3014 				mod_counter = ac->clabel->mod_counter;
3015 			}
3016 		}
3017 		ac = ac->next;
3018 	}
3019 
3020 	num_missing = 0;
3021 	auto_config = cset->ac;
3022 
3023 	for(r=0; r<num_rows; r++) {
3024 		even_pair_failed = 0;
3025 		for(c=0; c<num_cols; c++) {
3026 			ac = auto_config;
3027 			while(ac!=NULL) {
3028 				if ((ac->clabel->row == r) &&
3029 				    (ac->clabel->column == c) &&
3030 				    (ac->clabel->mod_counter == mod_counter)) {
3031 					/* it's this one... */
3032 #if DEBUG
3033 					printf("Found: %s at %d,%d\n",
3034 					       ac->devname,r,c);
3035 #endif
3036 					break;
3037 				}
3038 				ac=ac->next;
3039 			}
3040 			if (ac==NULL) {
3041 				/* Didn't find one here! */
3042 				/* special case for RAID 1, especially
3043 				   where there are more than 2
3044 				   components (where RAIDframe treats
3045 				   things a little differently :( ) */
3046 				if (parity_type == '1') {
3047 					if (c%2 == 0) { /* even component */
3048 						even_pair_failed = 1;
3049 					} else { /* odd component.  If
3050                                                     we're failed, and
3051                                                     so is the even
3052                                                     component, it's
3053                                                     "Good Night, Charlie" */
3054 						if (even_pair_failed == 1) {
3055 							return(0);
3056 						}
3057 					}
3058 				} else {
3059 					/* normal accounting */
3060 					num_missing++;
3061 				}
3062 			}
3063 			if ((parity_type == '1') && (c%2 == 1)) {
3064 				/* Just did an even component, and we didn't
3065 				   bail.. reset the even_pair_failed flag,
3066 				   and go on to the next component.... */
3067 				even_pair_failed = 0;
3068 			}
3069 		}
3070 	}
3071 
3072 	clabel = cset->ac->clabel;
3073 
3074 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3075 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3076 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3077 		/* XXX this needs to be made *much* more general */
3078 		/* Too many failures */
3079 		return(0);
3080 	}
3081 	/* otherwise, all is well, and we've got enough to take a kick
3082 	   at autoconfiguring this set */
3083 	return(1);
3084 }
3085 
3086 void
3087 rf_create_configuration(ac,config,raidPtr)
3088 	RF_AutoConfig_t *ac;
3089 	RF_Config_t *config;
3090 	RF_Raid_t *raidPtr;
3091 {
3092 	RF_ComponentLabel_t *clabel;
3093 	int i;
3094 
3095 	clabel = ac->clabel;
3096 
3097 	/* 1. Fill in the common stuff */
3098 	config->numRow = clabel->num_rows;
3099 	config->numCol = clabel->num_columns;
3100 	config->numSpare = 0; /* XXX should this be set here? */
3101 	config->sectPerSU = clabel->sectPerSU;
3102 	config->SUsPerPU = clabel->SUsPerPU;
3103 	config->SUsPerRU = clabel->SUsPerRU;
3104 	config->parityConfig = clabel->parityConfig;
3105 	/* XXX... */
3106 	strcpy(config->diskQueueType,"fifo");
3107 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3108 	config->layoutSpecificSize = 0; /* XXX ?? */
3109 
3110 	while(ac!=NULL) {
3111 		/* row/col values will be in range due to the checks
3112 		   in reasonable_label() */
3113 		strcpy(config->devnames[ac->clabel->row][ac->clabel->column],
3114 		       ac->devname);
3115 		ac = ac->next;
3116 	}
3117 
3118 	for(i=0;i<RF_MAXDBGV;i++) {
3119 		config->debugVars[i][0] = NULL;
3120 	}
3121 }
3122 
3123 int
3124 rf_set_autoconfig(raidPtr, new_value)
3125 	RF_Raid_t *raidPtr;
3126 	int new_value;
3127 {
3128 	RF_ComponentLabel_t clabel;
3129 	struct vnode *vp;
3130 	dev_t dev;
3131 	int row, column;
3132 
3133 	raidPtr->autoconfigure = new_value;
3134 	for(row=0; row<raidPtr->numRow; row++) {
3135 		for(column=0; column<raidPtr->numCol; column++) {
3136 			if (raidPtr->Disks[row][column].status ==
3137 			    rf_ds_optimal) {
3138 				dev = raidPtr->Disks[row][column].dev;
3139 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3140 				raidread_component_label(dev, vp, &clabel);
3141 				clabel.autoconfigure = new_value;
3142 				raidwrite_component_label(dev, vp, &clabel);
3143 			}
3144 		}
3145 	}
3146 	return(new_value);
3147 }
3148 
3149 int
3150 rf_set_rootpartition(raidPtr, new_value)
3151 	RF_Raid_t *raidPtr;
3152 	int new_value;
3153 {
3154 	RF_ComponentLabel_t clabel;
3155 	struct vnode *vp;
3156 	dev_t dev;
3157 	int row, column;
3158 
3159 	raidPtr->root_partition = new_value;
3160 	for(row=0; row<raidPtr->numRow; row++) {
3161 		for(column=0; column<raidPtr->numCol; column++) {
3162 			if (raidPtr->Disks[row][column].status ==
3163 			    rf_ds_optimal) {
3164 				dev = raidPtr->Disks[row][column].dev;
3165 				vp = raidPtr->raid_cinfo[row][column].ci_vp;
3166 				raidread_component_label(dev, vp, &clabel);
3167 				clabel.root_partition = new_value;
3168 				raidwrite_component_label(dev, vp, &clabel);
3169 			}
3170 		}
3171 	}
3172 	return(new_value);
3173 }
3174 
3175 void
3176 rf_release_all_vps(cset)
3177 	RF_ConfigSet_t *cset;
3178 {
3179 	RF_AutoConfig_t *ac;
3180 
3181 	ac = cset->ac;
3182 	while(ac!=NULL) {
3183 		/* Close the vp, and give it back */
3184 		if (ac->vp) {
3185 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3186 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3187 			vput(ac->vp);
3188 			ac->vp = NULL;
3189 		}
3190 		ac = ac->next;
3191 	}
3192 }
3193 
3194 
3195 void
3196 rf_cleanup_config_set(cset)
3197 	RF_ConfigSet_t *cset;
3198 {
3199 	RF_AutoConfig_t *ac;
3200 	RF_AutoConfig_t *next_ac;
3201 
3202 	ac = cset->ac;
3203 	while(ac!=NULL) {
3204 		next_ac = ac->next;
3205 		/* nuke the label */
3206 		free(ac->clabel, M_RAIDFRAME);
3207 		/* cleanup the config structure */
3208 		free(ac, M_RAIDFRAME);
3209 		/* "next.." */
3210 		ac = next_ac;
3211 	}
3212 	/* and, finally, nuke the config set */
3213 	free(cset, M_RAIDFRAME);
3214 }
3215 
3216 
3217 void
3218 raid_init_component_label(raidPtr, clabel)
3219 	RF_Raid_t *raidPtr;
3220 	RF_ComponentLabel_t *clabel;
3221 {
3222 	/* current version number */
3223 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3224 	clabel->serial_number = raidPtr->serial_number;
3225 	clabel->mod_counter = raidPtr->mod_counter;
3226 	clabel->num_rows = raidPtr->numRow;
3227 	clabel->num_columns = raidPtr->numCol;
3228 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3229 	clabel->status = rf_ds_optimal; /* "It's good!" */
3230 
3231 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3232 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3233 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3234 
3235 	clabel->blockSize = raidPtr->bytesPerSector;
3236 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3237 
3238 	/* XXX not portable */
3239 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3240 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3241 	clabel->autoconfigure = raidPtr->autoconfigure;
3242 	clabel->root_partition = raidPtr->root_partition;
3243 	clabel->last_unit = raidPtr->raidid;
3244 	clabel->config_order = raidPtr->config_order;
3245 }
3246 
3247 int
3248 rf_auto_config_set(cset,unit)
3249 	RF_ConfigSet_t *cset;
3250 	int *unit;
3251 {
3252 	RF_Raid_t *raidPtr;
3253 	RF_Config_t *config;
3254 	int raidID;
3255 	int retcode;
3256 
3257 	printf("RAID autoconfigure\n");
3258 
3259 	retcode = 0;
3260 	*unit = -1;
3261 
3262 	/* 1. Create a config structure */
3263 
3264 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3265 				       M_RAIDFRAME,
3266 				       M_NOWAIT);
3267 	if (config==NULL) {
3268 		printf("Out of mem!?!?\n");
3269 				/* XXX do something more intelligent here. */
3270 		return(1);
3271 	}
3272 
3273 	memset(config, 0, sizeof(RF_Config_t));
3274 
3275 	/* XXX raidID needs to be set correctly.. */
3276 
3277 	/*
3278 	   2. Figure out what RAID ID this one is supposed to live at
3279 	   See if we can get the same RAID dev that it was configured
3280 	   on last time..
3281 	*/
3282 
3283 	raidID = cset->ac->clabel->last_unit;
3284 	if ((raidID < 0) || (raidID >= numraid)) {
3285 		/* let's not wander off into lala land. */
3286 		raidID = numraid - 1;
3287 	}
3288 	if (raidPtrs[raidID]->valid != 0) {
3289 
3290 		/*
3291 		   Nope... Go looking for an alternative...
3292 		   Start high so we don't immediately use raid0 if that's
3293 		   not taken.
3294 		*/
3295 
3296 		for(raidID = numraid; raidID >= 0; raidID--) {
3297 			if (raidPtrs[raidID]->valid == 0) {
3298 				/* can use this one! */
3299 				break;
3300 			}
3301 		}
3302 	}
3303 
3304 	if (raidID < 0) {
3305 		/* punt... */
3306 		printf("Unable to auto configure this set!\n");
3307 		printf("(Out of RAID devs!)\n");
3308 		return(1);
3309 	}
3310 	printf("Configuring raid%d:\n",raidID);
3311 	raidPtr = raidPtrs[raidID];
3312 
3313 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3314 	raidPtr->raidid = raidID;
3315 	raidPtr->openings = RAIDOUTSTANDING;
3316 
3317 	/* 3. Build the configuration structure */
3318 	rf_create_configuration(cset->ac, config, raidPtr);
3319 
3320 	/* 4. Do the configuration */
3321 	retcode = rf_Configure(raidPtr, config, cset->ac);
3322 
3323 	if (retcode == 0) {
3324 
3325 		raidinit(raidPtrs[raidID]);
3326 
3327 		rf_markalldirty(raidPtrs[raidID]);
3328 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3329 		if (cset->ac->clabel->root_partition==1) {
3330 			/* everything configured just fine.  Make a note
3331 			   that this set is eligible to be root. */
3332 			cset->rootable = 1;
3333 			/* XXX do this here? */
3334 			raidPtrs[raidID]->root_partition = 1;
3335 		}
3336 	}
3337 
3338 	/* 5. Cleanup */
3339 	free(config, M_RAIDFRAME);
3340 
3341 	*unit = raidID;
3342 	return(retcode);
3343 }
3344 
3345 void
3346 rf_disk_unbusy(desc)
3347 	RF_RaidAccessDesc_t *desc;
3348 {
3349 	struct buf *bp;
3350 
3351 	bp = (struct buf *)desc->bp;
3352 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3353 			    (bp->b_bcount - bp->b_resid));
3354 }
3355