xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision bf1e9b32e27832f0c493206710fb8b58a980838a)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.187 2005/05/29 22:03:09 christos 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) 1990, 1993
40  *      The Regents of the University of California.  All rights reserved.
41  *
42  * This code is derived from software contributed to Berkeley by
43  * the Systems Programming Group of the University of Utah Computer
44  * Science Department.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  * from: Utah $Hdr: cd.c 1.6 90/11/28$
71  *
72  *      @(#)cd.c        8.2 (Berkeley) 11/16/93
73  */
74 
75 /*
76  * Copyright (c) 1988 University of Utah.
77  *
78  * This code is derived from software contributed to Berkeley by
79  * the Systems Programming Group of the University of Utah Computer
80  * Science Department.
81  *
82  * Redistribution and use in source and binary forms, with or without
83  * modification, are permitted provided that the following conditions
84  * are met:
85  * 1. Redistributions of source code must retain the above copyright
86  *    notice, this list of conditions and the following disclaimer.
87  * 2. Redistributions in binary form must reproduce the above copyright
88  *    notice, this list of conditions and the following disclaimer in the
89  *    documentation and/or other materials provided with the distribution.
90  * 3. All advertising materials mentioning features or use of this software
91  *    must display the following acknowledgement:
92  *      This product includes software developed by the University of
93  *      California, Berkeley and its contributors.
94  * 4. Neither the name of the University nor the names of its contributors
95  *    may be used to endorse or promote products derived from this software
96  *    without specific prior written permission.
97  *
98  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
99  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
100  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
101  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
102  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
103  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
104  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
105  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
106  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
107  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
108  * SUCH DAMAGE.
109  *
110  * from: Utah $Hdr: cd.c 1.6 90/11/28$
111  *
112  *      @(#)cd.c        8.2 (Berkeley) 11/16/93
113  */
114 
115 /*
116  * Copyright (c) 1995 Carnegie-Mellon University.
117  * All rights reserved.
118  *
119  * Authors: Mark Holland, Jim Zelenka
120  *
121  * Permission to use, copy, modify and distribute this software and
122  * its documentation is hereby granted, provided that both the copyright
123  * notice and this permission notice appear in all copies of the
124  * software, derivative works or modified versions, and any portions
125  * thereof, and that both notices appear in supporting documentation.
126  *
127  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
128  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
129  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
130  *
131  * Carnegie Mellon requests users of this software to return to
132  *
133  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
134  *  School of Computer Science
135  *  Carnegie Mellon University
136  *  Pittsburgh PA 15213-3890
137  *
138  * any improvements or extensions that they make and grant Carnegie the
139  * rights to redistribute these changes.
140  */
141 
142 /***********************************************************
143  *
144  * rf_kintf.c -- the kernel interface routines for RAIDframe
145  *
146  ***********************************************************/
147 
148 #include <sys/cdefs.h>
149 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.187 2005/05/29 22:03:09 christos Exp $");
150 
151 #include <sys/param.h>
152 #include <sys/errno.h>
153 #include <sys/pool.h>
154 #include <sys/proc.h>
155 #include <sys/queue.h>
156 #include <sys/disk.h>
157 #include <sys/device.h>
158 #include <sys/stat.h>
159 #include <sys/ioctl.h>
160 #include <sys/fcntl.h>
161 #include <sys/systm.h>
162 #include <sys/namei.h>
163 #include <sys/vnode.h>
164 #include <sys/disklabel.h>
165 #include <sys/conf.h>
166 #include <sys/lock.h>
167 #include <sys/buf.h>
168 #include <sys/bufq.h>
169 #include <sys/user.h>
170 #include <sys/reboot.h>
171 
172 #include <dev/raidframe/raidframevar.h>
173 #include <dev/raidframe/raidframeio.h>
174 #include "raid.h"
175 #include "opt_raid_autoconfig.h"
176 #include "rf_raid.h"
177 #include "rf_copyback.h"
178 #include "rf_dag.h"
179 #include "rf_dagflags.h"
180 #include "rf_desc.h"
181 #include "rf_diskqueue.h"
182 #include "rf_etimer.h"
183 #include "rf_general.h"
184 #include "rf_kintf.h"
185 #include "rf_options.h"
186 #include "rf_driver.h"
187 #include "rf_parityscan.h"
188 #include "rf_threadstuff.h"
189 
190 #ifdef DEBUG
191 int     rf_kdebug_level = 0;
192 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
193 #else				/* DEBUG */
194 #define db1_printf(a) { }
195 #endif				/* DEBUG */
196 
197 static RF_Raid_t **raidPtrs;	/* global raid device descriptors */
198 
199 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
200 
201 static RF_SparetWait_t *rf_sparet_wait_queue;	/* requests to install a
202 						 * spare table */
203 static RF_SparetWait_t *rf_sparet_resp_queue;	/* responses from
204 						 * installation process */
205 
206 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures");
207 
208 /* prototypes */
209 static void KernelWakeupFunc(struct buf *);
210 static void InitBP(struct buf *, struct vnode *, unsigned,
211     dev_t, RF_SectorNum_t, RF_SectorCount_t, caddr_t, void (*) (struct buf *),
212     void *, int, struct proc *);
213 static void raidinit(RF_Raid_t *);
214 
215 void raidattach(int);
216 
217 dev_type_open(raidopen);
218 dev_type_close(raidclose);
219 dev_type_read(raidread);
220 dev_type_write(raidwrite);
221 dev_type_ioctl(raidioctl);
222 dev_type_strategy(raidstrategy);
223 dev_type_dump(raiddump);
224 dev_type_size(raidsize);
225 
226 const struct bdevsw raid_bdevsw = {
227 	raidopen, raidclose, raidstrategy, raidioctl,
228 	raiddump, raidsize, D_DISK
229 };
230 
231 const struct cdevsw raid_cdevsw = {
232 	raidopen, raidclose, raidread, raidwrite, raidioctl,
233 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
234 };
235 
236 /*
237  * Pilfered from ccd.c
238  */
239 
240 struct raidbuf {
241 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
242 	struct buf *rf_obp;	/* ptr. to original I/O buf */
243 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
244 };
245 
246 /* XXX Not sure if the following should be replacing the raidPtrs above,
247    or if it should be used in conjunction with that...
248 */
249 
250 struct raid_softc {
251 	int     sc_flags;	/* flags */
252 	int     sc_cflags;	/* configuration flags */
253 	size_t  sc_size;        /* size of the raid device */
254 	char    sc_xname[20];	/* XXX external name */
255 	struct disk sc_dkdev;	/* generic disk device info */
256 	struct bufq_state buf_queue;	/* used for the device queue */
257 };
258 /* sc_flags */
259 #define RAIDF_INITED	0x01	/* unit has been initialized */
260 #define RAIDF_WLABEL	0x02	/* label area is writable */
261 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
262 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
263 #define RAIDF_LOCKED	0x80	/* unit is locked */
264 
265 #define	raidunit(x)	DISKUNIT(x)
266 int numraid = 0;
267 
268 /*
269  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
270  * Be aware that large numbers can allow the driver to consume a lot of
271  * kernel memory, especially on writes, and in degraded mode reads.
272  *
273  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
274  * a single 64K write will typically require 64K for the old data,
275  * 64K for the old parity, and 64K for the new parity, for a total
276  * of 192K (if the parity buffer is not re-used immediately).
277  * Even it if is used immediately, that's still 128K, which when multiplied
278  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
279  *
280  * Now in degraded mode, for example, a 64K read on the above setup may
281  * require data reconstruction, which will require *all* of the 4 remaining
282  * disks to participate -- 4 * 32K/disk == 128K again.
283  */
284 
285 #ifndef RAIDOUTSTANDING
286 #define RAIDOUTSTANDING   6
287 #endif
288 
289 #define RAIDLABELDEV(dev)	\
290 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
291 
292 /* declared here, and made public, for the benefit of KVM stuff.. */
293 struct raid_softc *raid_softc;
294 
295 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
296 				     struct disklabel *);
297 static void raidgetdisklabel(dev_t);
298 static void raidmakedisklabel(struct raid_softc *);
299 
300 static int raidlock(struct raid_softc *);
301 static void raidunlock(struct raid_softc *);
302 
303 static void rf_markalldirty(RF_Raid_t *);
304 
305 struct device *raidrootdev;
306 
307 void rf_ReconThread(struct rf_recon_req *);
308 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
309 void rf_CopybackThread(RF_Raid_t *raidPtr);
310 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
311 int rf_autoconfig(struct device *self);
312 void rf_buildroothack(RF_ConfigSet_t *);
313 
314 RF_AutoConfig_t *rf_find_raid_components(void);
315 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
316 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
317 static int rf_reasonable_label(RF_ComponentLabel_t *);
318 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
319 int rf_set_autoconfig(RF_Raid_t *, int);
320 int rf_set_rootpartition(RF_Raid_t *, int);
321 void rf_release_all_vps(RF_ConfigSet_t *);
322 void rf_cleanup_config_set(RF_ConfigSet_t *);
323 int rf_have_enough_components(RF_ConfigSet_t *);
324 int rf_auto_config_set(RF_ConfigSet_t *, int *);
325 
326 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
327 				  allow autoconfig to take place.
328 			          Note that this is overridden by having
329 			          RAID_AUTOCONFIG as an option in the
330 			          kernel config file.  */
331 
332 struct RF_Pools_s rf_pools;
333 
334 void
335 raidattach(int num)
336 {
337 	int raidID;
338 	int i, rc;
339 
340 #ifdef DEBUG
341 	printf("raidattach: Asked for %d units\n", num);
342 #endif
343 
344 	if (num <= 0) {
345 #ifdef DIAGNOSTIC
346 		panic("raidattach: count <= 0");
347 #endif
348 		return;
349 	}
350 	/* This is where all the initialization stuff gets done. */
351 
352 	numraid = num;
353 
354 	/* Make some space for requested number of units... */
355 
356 	RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **));
357 	if (raidPtrs == NULL) {
358 		panic("raidPtrs is NULL!!");
359 	}
360 
361 	/* Initialize the component buffer pool. */
362 	rf_pool_init(&rf_pools.cbuf, sizeof(struct raidbuf),
363 		     "raidpl", num * RAIDOUTSTANDING,
364 		     2 * num * RAIDOUTSTANDING);
365 
366 	rf_mutex_init(&rf_sparet_wait_mutex);
367 
368 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
369 
370 	for (i = 0; i < num; i++)
371 		raidPtrs[i] = NULL;
372 	rc = rf_BootRaidframe();
373 	if (rc == 0)
374 		printf("Kernelized RAIDframe activated\n");
375 	else
376 		panic("Serious error booting RAID!!");
377 
378 	/* put together some datastructures like the CCD device does.. This
379 	 * lets us lock the device and what-not when it gets opened. */
380 
381 	raid_softc = (struct raid_softc *)
382 		malloc(num * sizeof(struct raid_softc),
383 		       M_RAIDFRAME, M_NOWAIT);
384 	if (raid_softc == NULL) {
385 		printf("WARNING: no memory for RAIDframe driver\n");
386 		return;
387 	}
388 
389 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
390 
391 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
392 					      M_RAIDFRAME, M_NOWAIT);
393 	if (raidrootdev == NULL) {
394 		panic("No memory for RAIDframe driver!!?!?!");
395 	}
396 
397 	for (raidID = 0; raidID < num; raidID++) {
398 		bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_FCFS);
399 
400 		raidrootdev[raidID].dv_class  = DV_DISK;
401 		raidrootdev[raidID].dv_cfdata = NULL;
402 		raidrootdev[raidID].dv_unit   = raidID;
403 		raidrootdev[raidID].dv_parent = NULL;
404 		raidrootdev[raidID].dv_flags  = 0;
405 		snprintf(raidrootdev[raidID].dv_xname,
406 		    sizeof(raidrootdev[raidID].dv_xname), "raid%d", raidID);
407 
408 		RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t),
409 			  (RF_Raid_t *));
410 		if (raidPtrs[raidID] == NULL) {
411 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
412 			numraid = raidID;
413 			return;
414 		}
415 	}
416 
417 #ifdef RAID_AUTOCONFIG
418 	raidautoconfig = 1;
419 #endif
420 
421 	/*
422 	 * Register a finalizer which will be used to auto-config RAID
423 	 * sets once all real hardware devices have been found.
424 	 */
425 	if (config_finalize_register(NULL, rf_autoconfig) != 0)
426 		printf("WARNING: unable to register RAIDframe finalizer\n");
427 }
428 
429 int
430 rf_autoconfig(struct device *self)
431 {
432 	RF_AutoConfig_t *ac_list;
433 	RF_ConfigSet_t *config_sets;
434 
435 	if (raidautoconfig == 0)
436 		return (0);
437 
438 	/* XXX This code can only be run once. */
439 	raidautoconfig = 0;
440 
441 	/* 1. locate all RAID components on the system */
442 #ifdef DEBUG
443 	printf("Searching for RAID components...\n");
444 #endif
445 	ac_list = rf_find_raid_components();
446 
447 	/* 2. Sort them into their respective sets. */
448 	config_sets = rf_create_auto_sets(ac_list);
449 
450 	/*
451 	 * 3. Evaluate each set andconfigure the valid ones.
452 	 * This gets done in rf_buildroothack().
453 	 */
454 	rf_buildroothack(config_sets);
455 
456 	return (1);
457 }
458 
459 void
460 rf_buildroothack(RF_ConfigSet_t *config_sets)
461 {
462 	RF_ConfigSet_t *cset;
463 	RF_ConfigSet_t *next_cset;
464 	int retcode;
465 	int raidID;
466 	int rootID;
467 	int num_root;
468 
469 	rootID = 0;
470 	num_root = 0;
471 	cset = config_sets;
472 	while(cset != NULL ) {
473 		next_cset = cset->next;
474 		if (rf_have_enough_components(cset) &&
475 		    cset->ac->clabel->autoconfigure==1) {
476 			retcode = rf_auto_config_set(cset,&raidID);
477 			if (!retcode) {
478 				if (cset->rootable) {
479 					rootID = raidID;
480 					num_root++;
481 				}
482 			} else {
483 				/* The autoconfig didn't work :( */
484 #if DEBUG
485 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
486 #endif
487 				rf_release_all_vps(cset);
488 			}
489 		} else {
490 			/* we're not autoconfiguring this set...
491 			   release the associated resources */
492 			rf_release_all_vps(cset);
493 		}
494 		/* cleanup */
495 		rf_cleanup_config_set(cset);
496 		cset = next_cset;
497 	}
498 
499 	/* we found something bootable... */
500 
501 	if (num_root == 1) {
502 		booted_device = &raidrootdev[rootID];
503 	} else if (num_root > 1) {
504 		/* we can't guess.. require the user to answer... */
505 		boothowto |= RB_ASKNAME;
506 	}
507 }
508 
509 
510 int
511 raidsize(dev_t dev)
512 {
513 	struct raid_softc *rs;
514 	struct disklabel *lp;
515 	int     part, unit, omask, size;
516 
517 	unit = raidunit(dev);
518 	if (unit >= numraid)
519 		return (-1);
520 	rs = &raid_softc[unit];
521 
522 	if ((rs->sc_flags & RAIDF_INITED) == 0)
523 		return (-1);
524 
525 	part = DISKPART(dev);
526 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
527 	lp = rs->sc_dkdev.dk_label;
528 
529 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
530 		return (-1);
531 
532 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
533 		size = -1;
534 	else
535 		size = lp->d_partitions[part].p_size *
536 		    (lp->d_secsize / DEV_BSIZE);
537 
538 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
539 		return (-1);
540 
541 	return (size);
542 
543 }
544 
545 int
546 raiddump(dev_t dev, daddr_t blkno, caddr_t va, size_t  size)
547 {
548 	/* Not implemented. */
549 	return ENXIO;
550 }
551 /* ARGSUSED */
552 int
553 raidopen(dev_t dev, int flags, int fmt, struct proc *p)
554 {
555 	int     unit = raidunit(dev);
556 	struct raid_softc *rs;
557 	struct disklabel *lp;
558 	int     part, pmask;
559 	int     error = 0;
560 
561 	if (unit >= numraid)
562 		return (ENXIO);
563 	rs = &raid_softc[unit];
564 
565 	if ((error = raidlock(rs)) != 0)
566 		return (error);
567 	lp = rs->sc_dkdev.dk_label;
568 
569 	part = DISKPART(dev);
570 	pmask = (1 << part);
571 
572 	if ((rs->sc_flags & RAIDF_INITED) &&
573 	    (rs->sc_dkdev.dk_openmask == 0))
574 		raidgetdisklabel(dev);
575 
576 	/* make sure that this partition exists */
577 
578 	if (part != RAW_PART) {
579 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
580 		    ((part >= lp->d_npartitions) ||
581 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
582 			error = ENXIO;
583 			raidunlock(rs);
584 			return (error);
585 		}
586 	}
587 	/* Prevent this unit from being unconfigured while open. */
588 	switch (fmt) {
589 	case S_IFCHR:
590 		rs->sc_dkdev.dk_copenmask |= pmask;
591 		break;
592 
593 	case S_IFBLK:
594 		rs->sc_dkdev.dk_bopenmask |= pmask;
595 		break;
596 	}
597 
598 	if ((rs->sc_dkdev.dk_openmask == 0) &&
599 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
600 		/* First one... mark things as dirty... Note that we *MUST*
601 		 have done a configure before this.  I DO NOT WANT TO BE
602 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
603 		 THAT THEY BELONG TOGETHER!!!!! */
604 		/* XXX should check to see if we're only open for reading
605 		   here... If so, we needn't do this, but then need some
606 		   other way of keeping track of what's happened.. */
607 
608 		rf_markalldirty( raidPtrs[unit] );
609 	}
610 
611 
612 	rs->sc_dkdev.dk_openmask =
613 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
614 
615 	raidunlock(rs);
616 
617 	return (error);
618 
619 
620 }
621 /* ARGSUSED */
622 int
623 raidclose(dev_t dev, int flags, int fmt, struct proc *p)
624 {
625 	int     unit = raidunit(dev);
626 	struct raid_softc *rs;
627 	int     error = 0;
628 	int     part;
629 
630 	if (unit >= numraid)
631 		return (ENXIO);
632 	rs = &raid_softc[unit];
633 
634 	if ((error = raidlock(rs)) != 0)
635 		return (error);
636 
637 	part = DISKPART(dev);
638 
639 	/* ...that much closer to allowing unconfiguration... */
640 	switch (fmt) {
641 	case S_IFCHR:
642 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
643 		break;
644 
645 	case S_IFBLK:
646 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
647 		break;
648 	}
649 	rs->sc_dkdev.dk_openmask =
650 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
651 
652 	if ((rs->sc_dkdev.dk_openmask == 0) &&
653 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
654 		/* Last one... device is not unconfigured yet.
655 		   Device shutdown has taken care of setting the
656 		   clean bits if RAIDF_INITED is not set
657 		   mark things as clean... */
658 
659 		rf_update_component_labels(raidPtrs[unit],
660 						 RF_FINAL_COMPONENT_UPDATE);
661 		if (doing_shutdown) {
662 			/* last one, and we're going down, so
663 			   lights out for this RAID set too. */
664 			error = rf_Shutdown(raidPtrs[unit]);
665 
666 			/* It's no longer initialized... */
667 			rs->sc_flags &= ~RAIDF_INITED;
668 
669 			/* Detach the disk. */
670 			disk_detach(&rs->sc_dkdev);
671 		}
672 	}
673 
674 	raidunlock(rs);
675 	return (0);
676 
677 }
678 
679 void
680 raidstrategy(struct buf *bp)
681 {
682 	int s;
683 
684 	unsigned int raidID = raidunit(bp->b_dev);
685 	RF_Raid_t *raidPtr;
686 	struct raid_softc *rs = &raid_softc[raidID];
687 	int     wlabel;
688 
689 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
690 		bp->b_error = ENXIO;
691 		bp->b_flags |= B_ERROR;
692 		bp->b_resid = bp->b_bcount;
693 		biodone(bp);
694 		return;
695 	}
696 	if (raidID >= numraid || !raidPtrs[raidID]) {
697 		bp->b_error = ENODEV;
698 		bp->b_flags |= B_ERROR;
699 		bp->b_resid = bp->b_bcount;
700 		biodone(bp);
701 		return;
702 	}
703 	raidPtr = raidPtrs[raidID];
704 	if (!raidPtr->valid) {
705 		bp->b_error = ENODEV;
706 		bp->b_flags |= B_ERROR;
707 		bp->b_resid = bp->b_bcount;
708 		biodone(bp);
709 		return;
710 	}
711 	if (bp->b_bcount == 0) {
712 		db1_printf(("b_bcount is zero..\n"));
713 		biodone(bp);
714 		return;
715 	}
716 
717 	/*
718 	 * Do bounds checking and adjust transfer.  If there's an
719 	 * error, the bounds check will flag that for us.
720 	 */
721 
722 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
723 	if (DISKPART(bp->b_dev) != RAW_PART)
724 		if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) {
725 			db1_printf(("Bounds check failed!!:%d %d\n",
726 				(int) bp->b_blkno, (int) wlabel));
727 			biodone(bp);
728 			return;
729 		}
730 	s = splbio();
731 
732 	bp->b_resid = 0;
733 
734 	/* stuff it onto our queue */
735 	BUFQ_PUT(&rs->buf_queue, bp);
736 
737 	raidstart(raidPtrs[raidID]);
738 
739 	splx(s);
740 }
741 /* ARGSUSED */
742 int
743 raidread(dev_t dev, struct uio *uio, int flags)
744 {
745 	int     unit = raidunit(dev);
746 	struct raid_softc *rs;
747 
748 	if (unit >= numraid)
749 		return (ENXIO);
750 	rs = &raid_softc[unit];
751 
752 	if ((rs->sc_flags & RAIDF_INITED) == 0)
753 		return (ENXIO);
754 
755 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
756 
757 }
758 /* ARGSUSED */
759 int
760 raidwrite(dev_t dev, struct uio *uio, int flags)
761 {
762 	int     unit = raidunit(dev);
763 	struct raid_softc *rs;
764 
765 	if (unit >= numraid)
766 		return (ENXIO);
767 	rs = &raid_softc[unit];
768 
769 	if ((rs->sc_flags & RAIDF_INITED) == 0)
770 		return (ENXIO);
771 
772 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
773 
774 }
775 
776 int
777 raidioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
778 {
779 	int     unit = raidunit(dev);
780 	int     error = 0;
781 	int     part, pmask;
782 	struct raid_softc *rs;
783 	RF_Config_t *k_cfg, *u_cfg;
784 	RF_Raid_t *raidPtr;
785 	RF_RaidDisk_t *diskPtr;
786 	RF_AccTotals_t *totals;
787 	RF_DeviceConfig_t *d_cfg, **ucfgp;
788 	u_char *specific_buf;
789 	int retcode = 0;
790 	int column;
791 	int raidid;
792 	struct rf_recon_req *rrcopy, *rr;
793 	RF_ComponentLabel_t *clabel;
794 	RF_ComponentLabel_t ci_label;
795 	RF_ComponentLabel_t **clabel_ptr;
796 	RF_SingleComponent_t *sparePtr,*componentPtr;
797 	RF_SingleComponent_t hot_spare;
798 	RF_SingleComponent_t component;
799 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
800 	int i, j, d;
801 #ifdef __HAVE_OLD_DISKLABEL
802 	struct disklabel newlabel;
803 #endif
804 
805 	if (unit >= numraid)
806 		return (ENXIO);
807 	rs = &raid_softc[unit];
808 	raidPtr = raidPtrs[unit];
809 
810 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
811 		(int) DISKPART(dev), (int) unit, (int) cmd));
812 
813 	/* Must be open for writes for these commands... */
814 	switch (cmd) {
815 	case DIOCSDINFO:
816 	case DIOCWDINFO:
817 #ifdef __HAVE_OLD_DISKLABEL
818 	case ODIOCWDINFO:
819 	case ODIOCSDINFO:
820 #endif
821 	case DIOCWLABEL:
822 		if ((flag & FWRITE) == 0)
823 			return (EBADF);
824 	}
825 
826 	/* Must be initialized for these... */
827 	switch (cmd) {
828 	case DIOCGDINFO:
829 	case DIOCSDINFO:
830 	case DIOCWDINFO:
831 #ifdef __HAVE_OLD_DISKLABEL
832 	case ODIOCGDINFO:
833 	case ODIOCWDINFO:
834 	case ODIOCSDINFO:
835 	case ODIOCGDEFLABEL:
836 #endif
837 	case DIOCGPART:
838 	case DIOCWLABEL:
839 	case DIOCGDEFLABEL:
840 	case RAIDFRAME_SHUTDOWN:
841 	case RAIDFRAME_REWRITEPARITY:
842 	case RAIDFRAME_GET_INFO:
843 	case RAIDFRAME_RESET_ACCTOTALS:
844 	case RAIDFRAME_GET_ACCTOTALS:
845 	case RAIDFRAME_KEEP_ACCTOTALS:
846 	case RAIDFRAME_GET_SIZE:
847 	case RAIDFRAME_FAIL_DISK:
848 	case RAIDFRAME_COPYBACK:
849 	case RAIDFRAME_CHECK_RECON_STATUS:
850 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
851 	case RAIDFRAME_GET_COMPONENT_LABEL:
852 	case RAIDFRAME_SET_COMPONENT_LABEL:
853 	case RAIDFRAME_ADD_HOT_SPARE:
854 	case RAIDFRAME_REMOVE_HOT_SPARE:
855 	case RAIDFRAME_INIT_LABELS:
856 	case RAIDFRAME_REBUILD_IN_PLACE:
857 	case RAIDFRAME_CHECK_PARITY:
858 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
859 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
860 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
861 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
862 	case RAIDFRAME_SET_AUTOCONFIG:
863 	case RAIDFRAME_SET_ROOT:
864 	case RAIDFRAME_DELETE_COMPONENT:
865 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
866 		if ((rs->sc_flags & RAIDF_INITED) == 0)
867 			return (ENXIO);
868 	}
869 
870 	switch (cmd) {
871 
872 		/* configure the system */
873 	case RAIDFRAME_CONFIGURE:
874 
875 		if (raidPtr->valid) {
876 			/* There is a valid RAID set running on this unit! */
877 			printf("raid%d: Device already configured!\n",unit);
878 			return(EINVAL);
879 		}
880 
881 		/* copy-in the configuration information */
882 		/* data points to a pointer to the configuration structure */
883 
884 		u_cfg = *((RF_Config_t **) data);
885 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
886 		if (k_cfg == NULL) {
887 			return (ENOMEM);
888 		}
889 		retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t));
890 		if (retcode) {
891 			RF_Free(k_cfg, sizeof(RF_Config_t));
892 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
893 				retcode));
894 			return (retcode);
895 		}
896 		/* allocate a buffer for the layout-specific data, and copy it
897 		 * in */
898 		if (k_cfg->layoutSpecificSize) {
899 			if (k_cfg->layoutSpecificSize > 10000) {
900 				/* sanity check */
901 				RF_Free(k_cfg, sizeof(RF_Config_t));
902 				return (EINVAL);
903 			}
904 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
905 			    (u_char *));
906 			if (specific_buf == NULL) {
907 				RF_Free(k_cfg, sizeof(RF_Config_t));
908 				return (ENOMEM);
909 			}
910 			retcode = copyin(k_cfg->layoutSpecific, specific_buf,
911 			    k_cfg->layoutSpecificSize);
912 			if (retcode) {
913 				RF_Free(k_cfg, sizeof(RF_Config_t));
914 				RF_Free(specific_buf,
915 					k_cfg->layoutSpecificSize);
916 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
917 					retcode));
918 				return (retcode);
919 			}
920 		} else
921 			specific_buf = NULL;
922 		k_cfg->layoutSpecific = specific_buf;
923 
924 		/* should do some kind of sanity check on the configuration.
925 		 * Store the sum of all the bytes in the last byte? */
926 
927 		/* configure the system */
928 
929 		/*
930 		 * Clear the entire RAID descriptor, just to make sure
931 		 *  there is no stale data left in the case of a
932 		 *  reconfiguration
933 		 */
934 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
935 		raidPtr->raidid = unit;
936 
937 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
938 
939 		if (retcode == 0) {
940 
941 			/* allow this many simultaneous IO's to
942 			   this RAID device */
943 			raidPtr->openings = RAIDOUTSTANDING;
944 
945 			raidinit(raidPtr);
946 			rf_markalldirty(raidPtr);
947 		}
948 		/* free the buffers.  No return code here. */
949 		if (k_cfg->layoutSpecificSize) {
950 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
951 		}
952 		RF_Free(k_cfg, sizeof(RF_Config_t));
953 
954 		return (retcode);
955 
956 		/* shutdown the system */
957 	case RAIDFRAME_SHUTDOWN:
958 
959 		if ((error = raidlock(rs)) != 0)
960 			return (error);
961 
962 		/*
963 		 * If somebody has a partition mounted, we shouldn't
964 		 * shutdown.
965 		 */
966 
967 		part = DISKPART(dev);
968 		pmask = (1 << part);
969 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
970 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
971 			(rs->sc_dkdev.dk_copenmask & pmask))) {
972 			raidunlock(rs);
973 			return (EBUSY);
974 		}
975 
976 		retcode = rf_Shutdown(raidPtr);
977 
978 		/* It's no longer initialized... */
979 		rs->sc_flags &= ~RAIDF_INITED;
980 
981 		/* Detach the disk. */
982 		disk_detach(&rs->sc_dkdev);
983 
984 		raidunlock(rs);
985 
986 		return (retcode);
987 	case RAIDFRAME_GET_COMPONENT_LABEL:
988 		clabel_ptr = (RF_ComponentLabel_t **) data;
989 		/* need to read the component label for the disk indicated
990 		   by row,column in clabel */
991 
992 		/* For practice, let's get it directly fromdisk, rather
993 		   than from the in-core copy */
994 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
995 			   (RF_ComponentLabel_t *));
996 		if (clabel == NULL)
997 			return (ENOMEM);
998 
999 		memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t));
1000 
1001 		retcode = copyin( *clabel_ptr, clabel,
1002 				  sizeof(RF_ComponentLabel_t));
1003 
1004 		if (retcode) {
1005 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1006 			return(retcode);
1007 		}
1008 
1009 		clabel->row = 0; /* Don't allow looking at anything else.*/
1010 
1011 		column = clabel->column;
1012 
1013 		if ((column < 0) || (column >= raidPtr->numCol +
1014 				     raidPtr->numSpare)) {
1015 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1016 			return(EINVAL);
1017 		}
1018 
1019 		raidread_component_label(raidPtr->Disks[column].dev,
1020 				raidPtr->raid_cinfo[column].ci_vp,
1021 				clabel );
1022 
1023 		retcode = copyout(clabel, *clabel_ptr,
1024 				  sizeof(RF_ComponentLabel_t));
1025 		RF_Free(clabel, sizeof(RF_ComponentLabel_t));
1026 		return (retcode);
1027 
1028 	case RAIDFRAME_SET_COMPONENT_LABEL:
1029 		clabel = (RF_ComponentLabel_t *) data;
1030 
1031 		/* XXX check the label for valid stuff... */
1032 		/* Note that some things *should not* get modified --
1033 		   the user should be re-initing the labels instead of
1034 		   trying to patch things.
1035 		   */
1036 
1037 		raidid = raidPtr->raidid;
1038 #if DEBUG
1039 		printf("raid%d: Got component label:\n", raidid);
1040 		printf("raid%d: Version: %d\n", raidid, clabel->version);
1041 		printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number);
1042 		printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter);
1043 		printf("raid%d: Column: %d\n", raidid, clabel->column);
1044 		printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns);
1045 		printf("raid%d: Clean: %d\n", raidid, clabel->clean);
1046 		printf("raid%d: Status: %d\n", raidid, clabel->status);
1047 #endif
1048 		clabel->row = 0;
1049 		column = clabel->column;
1050 
1051 		if ((column < 0) || (column >= raidPtr->numCol)) {
1052 			return(EINVAL);
1053 		}
1054 
1055 		/* XXX this isn't allowed to do anything for now :-) */
1056 
1057 		/* XXX and before it is, we need to fill in the rest
1058 		   of the fields!?!?!?! */
1059 #if 0
1060 		raidwrite_component_label(
1061                             raidPtr->Disks[column].dev,
1062 			    raidPtr->raid_cinfo[column].ci_vp,
1063 			    clabel );
1064 #endif
1065 		return (0);
1066 
1067 	case RAIDFRAME_INIT_LABELS:
1068 		clabel = (RF_ComponentLabel_t *) data;
1069 		/*
1070 		   we only want the serial number from
1071 		   the above.  We get all the rest of the information
1072 		   from the config that was used to create this RAID
1073 		   set.
1074 		   */
1075 
1076 		raidPtr->serial_number = clabel->serial_number;
1077 
1078 		raid_init_component_label(raidPtr, &ci_label);
1079 		ci_label.serial_number = clabel->serial_number;
1080 		ci_label.row = 0; /* we dont' pretend to support more */
1081 
1082 		for(column=0;column<raidPtr->numCol;column++) {
1083 			diskPtr = &raidPtr->Disks[column];
1084 			if (!RF_DEAD_DISK(diskPtr->status)) {
1085 				ci_label.partitionSize = diskPtr->partitionSize;
1086 				ci_label.column = column;
1087 				raidwrite_component_label(
1088 							  raidPtr->Disks[column].dev,
1089 							  raidPtr->raid_cinfo[column].ci_vp,
1090 							  &ci_label );
1091 			}
1092 		}
1093 
1094 		return (retcode);
1095 	case RAIDFRAME_SET_AUTOCONFIG:
1096 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1097 		printf("raid%d: New autoconfig value is: %d\n",
1098 		       raidPtr->raidid, d);
1099 		*(int *) data = d;
1100 		return (retcode);
1101 
1102 	case RAIDFRAME_SET_ROOT:
1103 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1104 		printf("raid%d: New rootpartition value is: %d\n",
1105 		       raidPtr->raidid, d);
1106 		*(int *) data = d;
1107 		return (retcode);
1108 
1109 		/* initialize all parity */
1110 	case RAIDFRAME_REWRITEPARITY:
1111 
1112 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1113 			/* Parity for RAID 0 is trivially correct */
1114 			raidPtr->parity_good = RF_RAID_CLEAN;
1115 			return(0);
1116 		}
1117 
1118 		if (raidPtr->parity_rewrite_in_progress == 1) {
1119 			/* Re-write is already in progress! */
1120 			return(EINVAL);
1121 		}
1122 
1123 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1124 					   rf_RewriteParityThread,
1125 					   raidPtr,"raid_parity");
1126 		return (retcode);
1127 
1128 
1129 	case RAIDFRAME_ADD_HOT_SPARE:
1130 		sparePtr = (RF_SingleComponent_t *) data;
1131 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1132 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1133 		return(retcode);
1134 
1135 	case RAIDFRAME_REMOVE_HOT_SPARE:
1136 		return(retcode);
1137 
1138 	case RAIDFRAME_DELETE_COMPONENT:
1139 		componentPtr = (RF_SingleComponent_t *)data;
1140 		memcpy( &component, componentPtr,
1141 			sizeof(RF_SingleComponent_t));
1142 		retcode = rf_delete_component(raidPtr, &component);
1143 		return(retcode);
1144 
1145 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1146 		componentPtr = (RF_SingleComponent_t *)data;
1147 		memcpy( &component, componentPtr,
1148 			sizeof(RF_SingleComponent_t));
1149 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1150 		return(retcode);
1151 
1152 	case RAIDFRAME_REBUILD_IN_PLACE:
1153 
1154 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1155 			/* Can't do this on a RAID 0!! */
1156 			return(EINVAL);
1157 		}
1158 
1159 		if (raidPtr->recon_in_progress == 1) {
1160 			/* a reconstruct is already in progress! */
1161 			return(EINVAL);
1162 		}
1163 
1164 		componentPtr = (RF_SingleComponent_t *) data;
1165 		memcpy( &component, componentPtr,
1166 			sizeof(RF_SingleComponent_t));
1167 		component.row = 0; /* we don't support any more */
1168 		column = component.column;
1169 
1170 		if ((column < 0) || (column >= raidPtr->numCol)) {
1171 			return(EINVAL);
1172 		}
1173 
1174 		RF_LOCK_MUTEX(raidPtr->mutex);
1175 		if ((raidPtr->Disks[column].status == rf_ds_optimal) &&
1176 		    (raidPtr->numFailures > 0)) {
1177 			/* XXX 0 above shouldn't be constant!!! */
1178 			/* some component other than this has failed.
1179 			   Let's not make things worse than they already
1180 			   are... */
1181 			printf("raid%d: Unable to reconstruct to disk at:\n",
1182 			       raidPtr->raidid);
1183 			printf("raid%d:     Col: %d   Too many failures.\n",
1184 			       raidPtr->raidid, column);
1185 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1186 			return (EINVAL);
1187 		}
1188 		if (raidPtr->Disks[column].status ==
1189 		    rf_ds_reconstructing) {
1190 			printf("raid%d: Unable to reconstruct to disk at:\n",
1191 			       raidPtr->raidid);
1192 			printf("raid%d:    Col: %d   Reconstruction already occuring!\n", raidPtr->raidid, column);
1193 
1194 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1195 			return (EINVAL);
1196 		}
1197 		if (raidPtr->Disks[column].status == rf_ds_spared) {
1198 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1199 			return (EINVAL);
1200 		}
1201 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1202 
1203 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1204 		if (rrcopy == NULL)
1205 			return(ENOMEM);
1206 
1207 		rrcopy->raidPtr = (void *) raidPtr;
1208 		rrcopy->col = column;
1209 
1210 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1211 					   rf_ReconstructInPlaceThread,
1212 					   rrcopy,"raid_reconip");
1213 		return(retcode);
1214 
1215 	case RAIDFRAME_GET_INFO:
1216 		if (!raidPtr->valid)
1217 			return (ENODEV);
1218 		ucfgp = (RF_DeviceConfig_t **) data;
1219 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1220 			  (RF_DeviceConfig_t *));
1221 		if (d_cfg == NULL)
1222 			return (ENOMEM);
1223 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1224 		d_cfg->rows = 1; /* there is only 1 row now */
1225 		d_cfg->cols = raidPtr->numCol;
1226 		d_cfg->ndevs = raidPtr->numCol;
1227 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1228 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1229 			return (ENOMEM);
1230 		}
1231 		d_cfg->nspares = raidPtr->numSpare;
1232 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1233 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1234 			return (ENOMEM);
1235 		}
1236 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1237 		d = 0;
1238 		for (j = 0; j < d_cfg->cols; j++) {
1239 			d_cfg->devs[d] = raidPtr->Disks[j];
1240 			d++;
1241 		}
1242 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1243 			d_cfg->spares[i] = raidPtr->Disks[j];
1244 		}
1245 		retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t));
1246 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1247 
1248 		return (retcode);
1249 
1250 	case RAIDFRAME_CHECK_PARITY:
1251 		*(int *) data = raidPtr->parity_good;
1252 		return (0);
1253 
1254 	case RAIDFRAME_RESET_ACCTOTALS:
1255 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1256 		return (0);
1257 
1258 	case RAIDFRAME_GET_ACCTOTALS:
1259 		totals = (RF_AccTotals_t *) data;
1260 		*totals = raidPtr->acc_totals;
1261 		return (0);
1262 
1263 	case RAIDFRAME_KEEP_ACCTOTALS:
1264 		raidPtr->keep_acc_totals = *(int *)data;
1265 		return (0);
1266 
1267 	case RAIDFRAME_GET_SIZE:
1268 		*(int *) data = raidPtr->totalSectors;
1269 		return (0);
1270 
1271 		/* fail a disk & optionally start reconstruction */
1272 	case RAIDFRAME_FAIL_DISK:
1273 
1274 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1275 			/* Can't do this on a RAID 0!! */
1276 			return(EINVAL);
1277 		}
1278 
1279 		rr = (struct rf_recon_req *) data;
1280 		rr->row = 0;
1281 		if (rr->col < 0 || rr->col >= raidPtr->numCol)
1282 			return (EINVAL);
1283 
1284 
1285 		RF_LOCK_MUTEX(raidPtr->mutex);
1286 		if (raidPtr->status == rf_rs_reconstructing) {
1287 			/* you can't fail a disk while we're reconstructing! */
1288 			/* XXX wrong for RAID6 */
1289 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1290 			return (EINVAL);
1291 		}
1292 		if ((raidPtr->Disks[rr->col].status ==
1293 		     rf_ds_optimal) && (raidPtr->numFailures > 0)) {
1294 			/* some other component has failed.  Let's not make
1295 			   things worse. XXX wrong for RAID6 */
1296 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1297 			return (EINVAL);
1298 		}
1299 		if (raidPtr->Disks[rr->col].status == rf_ds_spared) {
1300 			/* Can't fail a spared disk! */
1301 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1302 			return (EINVAL);
1303 		}
1304 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1305 
1306 		/* make a copy of the recon request so that we don't rely on
1307 		 * the user's buffer */
1308 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1309 		if (rrcopy == NULL)
1310 			return(ENOMEM);
1311 		memcpy(rrcopy, rr, sizeof(*rr));
1312 		rrcopy->raidPtr = (void *) raidPtr;
1313 
1314 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1315 					   rf_ReconThread,
1316 					   rrcopy,"raid_recon");
1317 		return (0);
1318 
1319 		/* invoke a copyback operation after recon on whatever disk
1320 		 * needs it, if any */
1321 	case RAIDFRAME_COPYBACK:
1322 
1323 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1324 			/* This makes no sense on a RAID 0!! */
1325 			return(EINVAL);
1326 		}
1327 
1328 		if (raidPtr->copyback_in_progress == 1) {
1329 			/* Copyback is already in progress! */
1330 			return(EINVAL);
1331 		}
1332 
1333 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1334 					   rf_CopybackThread,
1335 					   raidPtr,"raid_copyback");
1336 		return (retcode);
1337 
1338 		/* return the percentage completion of reconstruction */
1339 	case RAIDFRAME_CHECK_RECON_STATUS:
1340 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1341 			/* This makes no sense on a RAID 0, so tell the
1342 			   user it's done. */
1343 			*(int *) data = 100;
1344 			return(0);
1345 		}
1346 		if (raidPtr->status != rf_rs_reconstructing)
1347 			*(int *) data = 100;
1348 		else {
1349 			if (raidPtr->reconControl->numRUsTotal > 0) {
1350 				*(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal);
1351 			} else {
1352 				*(int *) data = 0;
1353 			}
1354 		}
1355 		return (0);
1356 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1357 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1358 		if (raidPtr->status != rf_rs_reconstructing) {
1359 			progressInfo.remaining = 0;
1360 			progressInfo.completed = 100;
1361 			progressInfo.total = 100;
1362 		} else {
1363 			progressInfo.total =
1364 				raidPtr->reconControl->numRUsTotal;
1365 			progressInfo.completed =
1366 				raidPtr->reconControl->numRUsComplete;
1367 			progressInfo.remaining = progressInfo.total -
1368 				progressInfo.completed;
1369 		}
1370 		retcode = copyout(&progressInfo, *progressInfoPtr,
1371 				  sizeof(RF_ProgressInfo_t));
1372 		return (retcode);
1373 
1374 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1375 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1376 			/* This makes no sense on a RAID 0, so tell the
1377 			   user it's done. */
1378 			*(int *) data = 100;
1379 			return(0);
1380 		}
1381 		if (raidPtr->parity_rewrite_in_progress == 1) {
1382 			*(int *) data = 100 *
1383 				raidPtr->parity_rewrite_stripes_done /
1384 				raidPtr->Layout.numStripe;
1385 		} else {
1386 			*(int *) data = 100;
1387 		}
1388 		return (0);
1389 
1390 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1391 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1392 		if (raidPtr->parity_rewrite_in_progress == 1) {
1393 			progressInfo.total = raidPtr->Layout.numStripe;
1394 			progressInfo.completed =
1395 				raidPtr->parity_rewrite_stripes_done;
1396 			progressInfo.remaining = progressInfo.total -
1397 				progressInfo.completed;
1398 		} else {
1399 			progressInfo.remaining = 0;
1400 			progressInfo.completed = 100;
1401 			progressInfo.total = 100;
1402 		}
1403 		retcode = copyout(&progressInfo, *progressInfoPtr,
1404 				  sizeof(RF_ProgressInfo_t));
1405 		return (retcode);
1406 
1407 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1408 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1409 			/* This makes no sense on a RAID 0 */
1410 			*(int *) data = 100;
1411 			return(0);
1412 		}
1413 		if (raidPtr->copyback_in_progress == 1) {
1414 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1415 				raidPtr->Layout.numStripe;
1416 		} else {
1417 			*(int *) data = 100;
1418 		}
1419 		return (0);
1420 
1421 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1422 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1423 		if (raidPtr->copyback_in_progress == 1) {
1424 			progressInfo.total = raidPtr->Layout.numStripe;
1425 			progressInfo.completed =
1426 				raidPtr->copyback_stripes_done;
1427 			progressInfo.remaining = progressInfo.total -
1428 				progressInfo.completed;
1429 		} else {
1430 			progressInfo.remaining = 0;
1431 			progressInfo.completed = 100;
1432 			progressInfo.total = 100;
1433 		}
1434 		retcode = copyout(&progressInfo, *progressInfoPtr,
1435 				  sizeof(RF_ProgressInfo_t));
1436 		return (retcode);
1437 
1438 		/* the sparetable daemon calls this to wait for the kernel to
1439 		 * need a spare table. this ioctl does not return until a
1440 		 * spare table is needed. XXX -- calling mpsleep here in the
1441 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1442 		 * -- I should either compute the spare table in the kernel,
1443 		 * or have a different -- XXX XXX -- interface (a different
1444 		 * character device) for delivering the table     -- XXX */
1445 #if 0
1446 	case RAIDFRAME_SPARET_WAIT:
1447 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1448 		while (!rf_sparet_wait_queue)
1449 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1450 		waitreq = rf_sparet_wait_queue;
1451 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1452 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1453 
1454 		/* structure assignment */
1455 		*((RF_SparetWait_t *) data) = *waitreq;
1456 
1457 		RF_Free(waitreq, sizeof(*waitreq));
1458 		return (0);
1459 
1460 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1461 		 * code in it that will cause the dameon to exit */
1462 	case RAIDFRAME_ABORT_SPARET_WAIT:
1463 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1464 		waitreq->fcol = -1;
1465 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1466 		waitreq->next = rf_sparet_wait_queue;
1467 		rf_sparet_wait_queue = waitreq;
1468 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1469 		wakeup(&rf_sparet_wait_queue);
1470 		return (0);
1471 
1472 		/* used by the spare table daemon to deliver a spare table
1473 		 * into the kernel */
1474 	case RAIDFRAME_SEND_SPARET:
1475 
1476 		/* install the spare table */
1477 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1478 
1479 		/* respond to the requestor.  the return status of the spare
1480 		 * table installation is passed in the "fcol" field */
1481 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1482 		waitreq->fcol = retcode;
1483 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1484 		waitreq->next = rf_sparet_resp_queue;
1485 		rf_sparet_resp_queue = waitreq;
1486 		wakeup(&rf_sparet_resp_queue);
1487 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1488 
1489 		return (retcode);
1490 #endif
1491 
1492 	default:
1493 		break; /* fall through to the os-specific code below */
1494 
1495 	}
1496 
1497 	if (!raidPtr->valid)
1498 		return (EINVAL);
1499 
1500 	/*
1501 	 * Add support for "regular" device ioctls here.
1502 	 */
1503 
1504 	switch (cmd) {
1505 	case DIOCGDINFO:
1506 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1507 		break;
1508 #ifdef __HAVE_OLD_DISKLABEL
1509 	case ODIOCGDINFO:
1510 		newlabel = *(rs->sc_dkdev.dk_label);
1511 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1512 			return ENOTTY;
1513 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1514 		break;
1515 #endif
1516 
1517 	case DIOCGPART:
1518 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1519 		((struct partinfo *) data)->part =
1520 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1521 		break;
1522 
1523 	case DIOCWDINFO:
1524 	case DIOCSDINFO:
1525 #ifdef __HAVE_OLD_DISKLABEL
1526 	case ODIOCWDINFO:
1527 	case ODIOCSDINFO:
1528 #endif
1529 	{
1530 		struct disklabel *lp;
1531 #ifdef __HAVE_OLD_DISKLABEL
1532 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1533 			memset(&newlabel, 0, sizeof newlabel);
1534 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1535 			lp = &newlabel;
1536 		} else
1537 #endif
1538 		lp = (struct disklabel *)data;
1539 
1540 		if ((error = raidlock(rs)) != 0)
1541 			return (error);
1542 
1543 		rs->sc_flags |= RAIDF_LABELLING;
1544 
1545 		error = setdisklabel(rs->sc_dkdev.dk_label,
1546 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1547 		if (error == 0) {
1548 			if (cmd == DIOCWDINFO
1549 #ifdef __HAVE_OLD_DISKLABEL
1550 			    || cmd == ODIOCWDINFO
1551 #endif
1552 			   )
1553 				error = writedisklabel(RAIDLABELDEV(dev),
1554 				    raidstrategy, rs->sc_dkdev.dk_label,
1555 				    rs->sc_dkdev.dk_cpulabel);
1556 		}
1557 		rs->sc_flags &= ~RAIDF_LABELLING;
1558 
1559 		raidunlock(rs);
1560 
1561 		if (error)
1562 			return (error);
1563 		break;
1564 	}
1565 
1566 	case DIOCWLABEL:
1567 		if (*(int *) data != 0)
1568 			rs->sc_flags |= RAIDF_WLABEL;
1569 		else
1570 			rs->sc_flags &= ~RAIDF_WLABEL;
1571 		break;
1572 
1573 	case DIOCGDEFLABEL:
1574 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1575 		break;
1576 
1577 #ifdef __HAVE_OLD_DISKLABEL
1578 	case ODIOCGDEFLABEL:
1579 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1580 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1581 			return ENOTTY;
1582 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1583 		break;
1584 #endif
1585 
1586 	default:
1587 		retcode = ENOTTY;
1588 	}
1589 	return (retcode);
1590 
1591 }
1592 
1593 
1594 /* raidinit -- complete the rest of the initialization for the
1595    RAIDframe device.  */
1596 
1597 
1598 static void
1599 raidinit(RF_Raid_t *raidPtr)
1600 {
1601 	struct raid_softc *rs;
1602 	int     unit;
1603 
1604 	unit = raidPtr->raidid;
1605 
1606 	rs = &raid_softc[unit];
1607 
1608 	/* XXX should check return code first... */
1609 	rs->sc_flags |= RAIDF_INITED;
1610 
1611 	/* XXX doesn't check bounds. */
1612 	snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit);
1613 
1614 	rs->sc_dkdev.dk_name = rs->sc_xname;
1615 
1616 	/* disk_attach actually creates space for the CPU disklabel, among
1617 	 * other things, so it's critical to call this *BEFORE* we try putzing
1618 	 * with disklabels. */
1619 
1620 	disk_attach(&rs->sc_dkdev);
1621 
1622 	/* XXX There may be a weird interaction here between this, and
1623 	 * protectedSectors, as used in RAIDframe.  */
1624 
1625 	rs->sc_size = raidPtr->totalSectors;
1626 }
1627 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0)
1628 /* wake up the daemon & tell it to get us a spare table
1629  * XXX
1630  * the entries in the queues should be tagged with the raidPtr
1631  * so that in the extremely rare case that two recons happen at once,
1632  * we know for which device were requesting a spare table
1633  * XXX
1634  *
1635  * XXX This code is not currently used. GO
1636  */
1637 int
1638 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req)
1639 {
1640 	int     retcode;
1641 
1642 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1643 	req->next = rf_sparet_wait_queue;
1644 	rf_sparet_wait_queue = req;
1645 	wakeup(&rf_sparet_wait_queue);
1646 
1647 	/* mpsleep unlocks the mutex */
1648 	while (!rf_sparet_resp_queue) {
1649 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1650 		    "raidframe getsparetable", 0);
1651 	}
1652 	req = rf_sparet_resp_queue;
1653 	rf_sparet_resp_queue = req->next;
1654 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1655 
1656 	retcode = req->fcol;
1657 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1658 					 * alloc'd */
1659 	return (retcode);
1660 }
1661 #endif
1662 
1663 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1664  * bp & passes it down.
1665  * any calls originating in the kernel must use non-blocking I/O
1666  * do some extra sanity checking to return "appropriate" error values for
1667  * certain conditions (to make some standard utilities work)
1668  *
1669  * Formerly known as: rf_DoAccessKernel
1670  */
1671 void
1672 raidstart(RF_Raid_t *raidPtr)
1673 {
1674 	RF_SectorCount_t num_blocks, pb, sum;
1675 	RF_RaidAddr_t raid_addr;
1676 	struct partition *pp;
1677 	daddr_t blocknum;
1678 	int     unit;
1679 	struct raid_softc *rs;
1680 	int     do_async;
1681 	struct buf *bp;
1682 	int rc;
1683 
1684 	unit = raidPtr->raidid;
1685 	rs = &raid_softc[unit];
1686 
1687 	/* quick check to see if anything has died recently */
1688 	RF_LOCK_MUTEX(raidPtr->mutex);
1689 	if (raidPtr->numNewFailures > 0) {
1690 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1691 		rf_update_component_labels(raidPtr,
1692 					   RF_NORMAL_COMPONENT_UPDATE);
1693 		RF_LOCK_MUTEX(raidPtr->mutex);
1694 		raidPtr->numNewFailures--;
1695 	}
1696 
1697 	/* Check to see if we're at the limit... */
1698 	while (raidPtr->openings > 0) {
1699 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1700 
1701 		/* get the next item, if any, from the queue */
1702 		if ((bp = BUFQ_GET(&rs->buf_queue)) == NULL) {
1703 			/* nothing more to do */
1704 			return;
1705 		}
1706 
1707 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1708 		 * partition.. Need to make it absolute to the underlying
1709 		 * device.. */
1710 
1711 		blocknum = bp->b_blkno;
1712 		if (DISKPART(bp->b_dev) != RAW_PART) {
1713 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1714 			blocknum += pp->p_offset;
1715 		}
1716 
1717 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1718 			    (int) blocknum));
1719 
1720 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1721 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1722 
1723 		/* *THIS* is where we adjust what block we're going to...
1724 		 * but DO NOT TOUCH bp->b_blkno!!! */
1725 		raid_addr = blocknum;
1726 
1727 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1728 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1729 		sum = raid_addr + num_blocks + pb;
1730 		if (1 || rf_debugKernelAccess) {
1731 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1732 				    (int) raid_addr, (int) sum, (int) num_blocks,
1733 				    (int) pb, (int) bp->b_resid));
1734 		}
1735 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1736 		    || (sum < num_blocks) || (sum < pb)) {
1737 			bp->b_error = ENOSPC;
1738 			bp->b_flags |= B_ERROR;
1739 			bp->b_resid = bp->b_bcount;
1740 			biodone(bp);
1741 			RF_LOCK_MUTEX(raidPtr->mutex);
1742 			continue;
1743 		}
1744 		/*
1745 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1746 		 */
1747 
1748 		if (bp->b_bcount & raidPtr->sectorMask) {
1749 			bp->b_error = EINVAL;
1750 			bp->b_flags |= B_ERROR;
1751 			bp->b_resid = bp->b_bcount;
1752 			biodone(bp);
1753 			RF_LOCK_MUTEX(raidPtr->mutex);
1754 			continue;
1755 
1756 		}
1757 		db1_printf(("Calling DoAccess..\n"));
1758 
1759 
1760 		RF_LOCK_MUTEX(raidPtr->mutex);
1761 		raidPtr->openings--;
1762 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1763 
1764 		/*
1765 		 * Everything is async.
1766 		 */
1767 		do_async = 1;
1768 
1769 		disk_busy(&rs->sc_dkdev);
1770 
1771 		/* XXX we're still at splbio() here... do we *really*
1772 		   need to be? */
1773 
1774 		/* don't ever condition on bp->b_flags & B_WRITE.
1775 		 * always condition on B_READ instead */
1776 
1777 		rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1778 				 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1779 				 do_async, raid_addr, num_blocks,
1780 				 bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1781 
1782 		if (rc) {
1783 			bp->b_error = rc;
1784 			bp->b_flags |= B_ERROR;
1785 			bp->b_resid = bp->b_bcount;
1786 			biodone(bp);
1787 			/* continue loop */
1788 		}
1789 
1790 		RF_LOCK_MUTEX(raidPtr->mutex);
1791 	}
1792 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1793 }
1794 
1795 
1796 
1797 
1798 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1799 
1800 int
1801 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req)
1802 {
1803 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1804 	struct buf *bp;
1805 	struct raidbuf *raidbp = NULL;
1806 
1807 	req->queue = queue;
1808 
1809 #if DIAGNOSTIC
1810 	if (queue->raidPtr->raidid >= numraid) {
1811 		printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid,
1812 		    numraid);
1813 		panic("Invalid Unit number in rf_DispatchKernelIO");
1814 	}
1815 #endif
1816 
1817 	bp = req->bp;
1818 #if 1
1819 	/* XXX when there is a physical disk failure, someone is passing us a
1820 	 * buffer that contains old stuff!!  Attempt to deal with this problem
1821 	 * without taking a performance hit... (not sure where the real bug
1822 	 * is.  It's buried in RAIDframe somewhere) :-(  GO ) */
1823 
1824 	if (bp->b_flags & B_ERROR) {
1825 		bp->b_flags &= ~B_ERROR;
1826 	}
1827 	if (bp->b_error != 0) {
1828 		bp->b_error = 0;
1829 	}
1830 #endif
1831 	raidbp = pool_get(&rf_pools.cbuf, PR_NOWAIT);
1832 	if (raidbp == NULL) {
1833 		bp->b_flags |= B_ERROR;
1834 		bp->b_error = ENOMEM;
1835 		return (ENOMEM);
1836 	}
1837 	BUF_INIT(&raidbp->rf_buf);
1838 
1839 	/*
1840 	 * context for raidiodone
1841 	 */
1842 	raidbp->rf_obp = bp;
1843 	raidbp->req = req;
1844 
1845 	BIO_COPYPRIO(&raidbp->rf_buf, bp);
1846 
1847 	switch (req->type) {
1848 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1849 		/* XXX need to do something extra here.. */
1850 		/* I'm leaving this in, as I've never actually seen it used,
1851 		 * and I'd like folks to report it... GO */
1852 		printf(("WAKEUP CALLED\n"));
1853 		queue->numOutstanding++;
1854 
1855 		/* XXX need to glue the original buffer into this??  */
1856 
1857 		KernelWakeupFunc(&raidbp->rf_buf);
1858 		break;
1859 
1860 	case RF_IO_TYPE_READ:
1861 	case RF_IO_TYPE_WRITE:
1862 #if RF_ACC_TRACE > 0
1863 		if (req->tracerec) {
1864 			RF_ETIMER_START(req->tracerec->timer);
1865 		}
1866 #endif
1867 		InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1868 		    op | bp->b_flags, queue->rf_cinfo->ci_dev,
1869 		    req->sectorOffset, req->numSector,
1870 		    req->buf, KernelWakeupFunc, (void *) req,
1871 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1872 
1873 		if (rf_debugKernelAccess) {
1874 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1875 				(long) bp->b_blkno));
1876 		}
1877 		queue->numOutstanding++;
1878 		queue->last_deq_sector = req->sectorOffset;
1879 		/* acc wouldn't have been let in if there were any pending
1880 		 * reqs at any other priority */
1881 		queue->curPriority = req->priority;
1882 
1883 		db1_printf(("Going for %c to unit %d col %d\n",
1884 			    req->type, queue->raidPtr->raidid,
1885 			    queue->col));
1886 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1887 			(int) req->sectorOffset, (int) req->numSector,
1888 			(int) (req->numSector <<
1889 			    queue->raidPtr->logBytesPerSector),
1890 			(int) queue->raidPtr->logBytesPerSector));
1891 		if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1892 			raidbp->rf_buf.b_vp->v_numoutput++;
1893 		}
1894 		VOP_STRATEGY(raidbp->rf_buf.b_vp, &raidbp->rf_buf);
1895 
1896 		break;
1897 
1898 	default:
1899 		panic("bad req->type in rf_DispatchKernelIO");
1900 	}
1901 	db1_printf(("Exiting from DispatchKernelIO\n"));
1902 
1903 	return (0);
1904 }
1905 /* this is the callback function associated with a I/O invoked from
1906    kernel code.
1907  */
1908 static void
1909 KernelWakeupFunc(struct buf *vbp)
1910 {
1911 	RF_DiskQueueData_t *req = NULL;
1912 	RF_DiskQueue_t *queue;
1913 	struct raidbuf *raidbp = (struct raidbuf *) vbp;
1914 	struct buf *bp;
1915 	int s;
1916 
1917 	s = splbio();
1918 	db1_printf(("recovering the request queue:\n"));
1919 	req = raidbp->req;
1920 
1921 	bp = raidbp->rf_obp;
1922 
1923 	queue = (RF_DiskQueue_t *) req->queue;
1924 
1925 	if (raidbp->rf_buf.b_flags & B_ERROR) {
1926 		bp->b_flags |= B_ERROR;
1927 		bp->b_error = raidbp->rf_buf.b_error ?
1928 		    raidbp->rf_buf.b_error : EIO;
1929 	}
1930 
1931 	/* XXX methinks this could be wrong... */
1932 #if 1
1933 	bp->b_resid = raidbp->rf_buf.b_resid;
1934 #endif
1935 #if RF_ACC_TRACE > 0
1936 	if (req->tracerec) {
1937 		RF_ETIMER_STOP(req->tracerec->timer);
1938 		RF_ETIMER_EVAL(req->tracerec->timer);
1939 		RF_LOCK_MUTEX(rf_tracing_mutex);
1940 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1941 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1942 		req->tracerec->num_phys_ios++;
1943 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1944 	}
1945 #endif
1946 	bp->b_bcount = raidbp->rf_buf.b_bcount;	/* XXXX ?? */
1947 
1948 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1949 	 * ballistic, and mark the component as hosed... */
1950 
1951 	if (bp->b_flags & B_ERROR) {
1952 		/* Mark the disk as dead */
1953 		/* but only mark it once... */
1954 		/* and only if it wouldn't leave this RAID set
1955 		   completely broken */
1956 		if ((queue->raidPtr->Disks[queue->col].status ==
1957 		    rf_ds_optimal) && (queue->raidPtr->numFailures <
1958 				       queue->raidPtr->Layout.map->faultsTolerated)) {
1959 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1960 			       queue->raidPtr->raidid,
1961 			       queue->raidPtr->Disks[queue->col].devname);
1962 			queue->raidPtr->Disks[queue->col].status =
1963 			    rf_ds_failed;
1964 			queue->raidPtr->status = rf_rs_degraded;
1965 			queue->raidPtr->numFailures++;
1966 			queue->raidPtr->numNewFailures++;
1967 		} else {	/* Disk is already dead... */
1968 			/* printf("Disk already marked as dead!\n"); */
1969 		}
1970 
1971 	}
1972 
1973 	pool_put(&rf_pools.cbuf, raidbp);
1974 
1975 	/* Fill in the error value */
1976 
1977 	req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0;
1978 
1979 	simple_lock(&queue->raidPtr->iodone_lock);
1980 
1981 	/* Drop this one on the "finished" queue... */
1982 	TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries);
1983 
1984 	/* Let the raidio thread know there is work to be done. */
1985 	wakeup(&(queue->raidPtr->iodone));
1986 
1987 	simple_unlock(&queue->raidPtr->iodone_lock);
1988 
1989 	splx(s);
1990 }
1991 
1992 
1993 
1994 /*
1995  * initialize a buf structure for doing an I/O in the kernel.
1996  */
1997 static void
1998 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev,
1999        RF_SectorNum_t startSect, RF_SectorCount_t numSect, caddr_t bf,
2000        void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector,
2001        struct proc *b_proc)
2002 {
2003 	/* bp->b_flags       = B_PHYS | rw_flag; */
2004 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
2005 	bp->b_bcount = numSect << logBytesPerSector;
2006 	bp->b_bufsize = bp->b_bcount;
2007 	bp->b_error = 0;
2008 	bp->b_dev = dev;
2009 	bp->b_data = bf;
2010 	bp->b_blkno = startSect;
2011 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
2012 	if (bp->b_bcount == 0) {
2013 		panic("bp->b_bcount is zero in InitBP!!");
2014 	}
2015 	bp->b_proc = b_proc;
2016 	bp->b_iodone = cbFunc;
2017 	bp->b_vp = b_vp;
2018 
2019 }
2020 
2021 static void
2022 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs,
2023 		    struct disklabel *lp)
2024 {
2025 	memset(lp, 0, sizeof(*lp));
2026 
2027 	/* fabricate a label... */
2028 	lp->d_secperunit = raidPtr->totalSectors;
2029 	lp->d_secsize = raidPtr->bytesPerSector;
2030 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
2031 	lp->d_ntracks = 4 * raidPtr->numCol;
2032 	lp->d_ncylinders = raidPtr->totalSectors /
2033 		(lp->d_nsectors * lp->d_ntracks);
2034 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
2035 
2036 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
2037 	lp->d_type = DTYPE_RAID;
2038 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
2039 	lp->d_rpm = 3600;
2040 	lp->d_interleave = 1;
2041 	lp->d_flags = 0;
2042 
2043 	lp->d_partitions[RAW_PART].p_offset = 0;
2044 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2045 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2046 	lp->d_npartitions = RAW_PART + 1;
2047 
2048 	lp->d_magic = DISKMAGIC;
2049 	lp->d_magic2 = DISKMAGIC;
2050 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2051 
2052 }
2053 /*
2054  * Read the disklabel from the raid device.  If one is not present, fake one
2055  * up.
2056  */
2057 static void
2058 raidgetdisklabel(dev_t dev)
2059 {
2060 	int     unit = raidunit(dev);
2061 	struct raid_softc *rs = &raid_softc[unit];
2062 	const char   *errstring;
2063 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2064 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2065 	RF_Raid_t *raidPtr;
2066 
2067 	db1_printf(("Getting the disklabel...\n"));
2068 
2069 	memset(clp, 0, sizeof(*clp));
2070 
2071 	raidPtr = raidPtrs[unit];
2072 
2073 	raidgetdefaultlabel(raidPtr, rs, lp);
2074 
2075 	/*
2076 	 * Call the generic disklabel extraction routine.
2077 	 */
2078 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2079 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2080 	if (errstring)
2081 		raidmakedisklabel(rs);
2082 	else {
2083 		int     i;
2084 		struct partition *pp;
2085 
2086 		/*
2087 		 * Sanity check whether the found disklabel is valid.
2088 		 *
2089 		 * This is necessary since total size of the raid device
2090 		 * may vary when an interleave is changed even though exactly
2091 		 * same componets are used, and old disklabel may used
2092 		 * if that is found.
2093 		 */
2094 		if (lp->d_secperunit != rs->sc_size)
2095 			printf("raid%d: WARNING: %s: "
2096 			    "total sector size in disklabel (%d) != "
2097 			    "the size of raid (%ld)\n", unit, rs->sc_xname,
2098 			    lp->d_secperunit, (long) rs->sc_size);
2099 		for (i = 0; i < lp->d_npartitions; i++) {
2100 			pp = &lp->d_partitions[i];
2101 			if (pp->p_offset + pp->p_size > rs->sc_size)
2102 				printf("raid%d: WARNING: %s: end of partition `%c' "
2103 				       "exceeds the size of raid (%ld)\n",
2104 				       unit, rs->sc_xname, 'a' + i, (long) rs->sc_size);
2105 		}
2106 	}
2107 
2108 }
2109 /*
2110  * Take care of things one might want to take care of in the event
2111  * that a disklabel isn't present.
2112  */
2113 static void
2114 raidmakedisklabel(struct raid_softc *rs)
2115 {
2116 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2117 	db1_printf(("Making a label..\n"));
2118 
2119 	/*
2120 	 * For historical reasons, if there's no disklabel present
2121 	 * the raw partition must be marked FS_BSDFFS.
2122 	 */
2123 
2124 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2125 
2126 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2127 
2128 	lp->d_checksum = dkcksum(lp);
2129 }
2130 /*
2131  * Lookup the provided name in the filesystem.  If the file exists,
2132  * is a valid block device, and isn't being used by anyone else,
2133  * set *vpp to the file's vnode.
2134  * You'll find the original of this in ccd.c
2135  */
2136 int
2137 raidlookup(char *path, struct proc *p, struct vnode **vpp)
2138 {
2139 	struct nameidata nd;
2140 	struct vnode *vp;
2141 	struct vattr va;
2142 	int     error;
2143 
2144 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2145 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2146 		return (error);
2147 	}
2148 	vp = nd.ni_vp;
2149 	if (vp->v_usecount > 1) {
2150 		VOP_UNLOCK(vp, 0);
2151 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2152 		return (EBUSY);
2153 	}
2154 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2155 		VOP_UNLOCK(vp, 0);
2156 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2157 		return (error);
2158 	}
2159 	/* XXX: eventually we should handle VREG, too. */
2160 	if (va.va_type != VBLK) {
2161 		VOP_UNLOCK(vp, 0);
2162 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2163 		return (ENOTBLK);
2164 	}
2165 	VOP_UNLOCK(vp, 0);
2166 	*vpp = vp;
2167 	return (0);
2168 }
2169 /*
2170  * Wait interruptibly for an exclusive lock.
2171  *
2172  * XXX
2173  * Several drivers do this; it should be abstracted and made MP-safe.
2174  * (Hmm... where have we seen this warning before :->  GO )
2175  */
2176 static int
2177 raidlock(struct raid_softc *rs)
2178 {
2179 	int     error;
2180 
2181 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2182 		rs->sc_flags |= RAIDF_WANTED;
2183 		if ((error =
2184 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2185 			return (error);
2186 	}
2187 	rs->sc_flags |= RAIDF_LOCKED;
2188 	return (0);
2189 }
2190 /*
2191  * Unlock and wake up any waiters.
2192  */
2193 static void
2194 raidunlock(struct raid_softc *rs)
2195 {
2196 
2197 	rs->sc_flags &= ~RAIDF_LOCKED;
2198 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2199 		rs->sc_flags &= ~RAIDF_WANTED;
2200 		wakeup(rs);
2201 	}
2202 }
2203 
2204 
2205 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2206 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2207 
2208 int
2209 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2210 {
2211 	RF_ComponentLabel_t clabel;
2212 	raidread_component_label(dev, b_vp, &clabel);
2213 	clabel.mod_counter = mod_counter;
2214 	clabel.clean = RF_RAID_CLEAN;
2215 	raidwrite_component_label(dev, b_vp, &clabel);
2216 	return(0);
2217 }
2218 
2219 
2220 int
2221 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2222 {
2223 	RF_ComponentLabel_t clabel;
2224 	raidread_component_label(dev, b_vp, &clabel);
2225 	clabel.mod_counter = mod_counter;
2226 	clabel.clean = RF_RAID_DIRTY;
2227 	raidwrite_component_label(dev, b_vp, &clabel);
2228 	return(0);
2229 }
2230 
2231 /* ARGSUSED */
2232 int
2233 raidread_component_label(dev_t dev, struct vnode *b_vp,
2234 			 RF_ComponentLabel_t *clabel)
2235 {
2236 	struct buf *bp;
2237 	const struct bdevsw *bdev;
2238 	int error;
2239 
2240 	/* XXX should probably ensure that we don't try to do this if
2241 	   someone has changed rf_protected_sectors. */
2242 
2243 	if (b_vp == NULL) {
2244 		/* For whatever reason, this component is not valid.
2245 		   Don't try to read a component label from it. */
2246 		return(EINVAL);
2247 	}
2248 
2249 	/* get a block of the appropriate size... */
2250 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2251 	bp->b_dev = dev;
2252 
2253 	/* get our ducks in a row for the read */
2254 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2255 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2256 	bp->b_flags |= B_READ;
2257  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2258 
2259 	bdev = bdevsw_lookup(bp->b_dev);
2260 	if (bdev == NULL)
2261 		return (ENXIO);
2262 	(*bdev->d_strategy)(bp);
2263 
2264 	error = biowait(bp);
2265 
2266 	if (!error) {
2267 		memcpy(clabel, bp->b_data,
2268 		       sizeof(RF_ComponentLabel_t));
2269         }
2270 
2271 	brelse(bp);
2272 	return(error);
2273 }
2274 /* ARGSUSED */
2275 int
2276 raidwrite_component_label(dev_t dev, struct vnode *b_vp,
2277 			  RF_ComponentLabel_t *clabel)
2278 {
2279 	struct buf *bp;
2280 	const struct bdevsw *bdev;
2281 	int error;
2282 
2283 	/* get a block of the appropriate size... */
2284 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2285 	bp->b_dev = dev;
2286 
2287 	/* get our ducks in a row for the write */
2288 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2289 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2290 	bp->b_flags |= B_WRITE;
2291  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2292 
2293 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2294 
2295 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2296 
2297 	bdev = bdevsw_lookup(bp->b_dev);
2298 	if (bdev == NULL)
2299 		return (ENXIO);
2300 	(*bdev->d_strategy)(bp);
2301 	error = biowait(bp);
2302 	brelse(bp);
2303 	if (error) {
2304 #if 1
2305 		printf("Failed to write RAID component info!\n");
2306 #endif
2307 	}
2308 
2309 	return(error);
2310 }
2311 
2312 void
2313 rf_markalldirty(RF_Raid_t *raidPtr)
2314 {
2315 	RF_ComponentLabel_t clabel;
2316 	int sparecol;
2317 	int c;
2318 	int j;
2319 	int scol = -1;
2320 
2321 	raidPtr->mod_counter++;
2322 	for (c = 0; c < raidPtr->numCol; c++) {
2323 		/* we don't want to touch (at all) a disk that has
2324 		   failed */
2325 		if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) {
2326 			raidread_component_label(
2327 						 raidPtr->Disks[c].dev,
2328 						 raidPtr->raid_cinfo[c].ci_vp,
2329 						 &clabel);
2330 			if (clabel.status == rf_ds_spared) {
2331 				/* XXX do something special...
2332 				   but whatever you do, don't
2333 				   try to access it!! */
2334 			} else {
2335 				raidmarkdirty(
2336 					      raidPtr->Disks[c].dev,
2337 					      raidPtr->raid_cinfo[c].ci_vp,
2338 					      raidPtr->mod_counter);
2339 			}
2340 		}
2341 	}
2342 
2343 	for( c = 0; c < raidPtr->numSpare ; c++) {
2344 		sparecol = raidPtr->numCol + c;
2345 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2346 			/*
2347 
2348 			   we claim this disk is "optimal" if it's
2349 			   rf_ds_used_spare, as that means it should be
2350 			   directly substitutable for the disk it replaced.
2351 			   We note that too...
2352 
2353 			 */
2354 
2355 			for(j=0;j<raidPtr->numCol;j++) {
2356 				if (raidPtr->Disks[j].spareCol == sparecol) {
2357 					scol = j;
2358 					break;
2359 				}
2360 			}
2361 
2362 			raidread_component_label(
2363 				 raidPtr->Disks[sparecol].dev,
2364 				 raidPtr->raid_cinfo[sparecol].ci_vp,
2365 				 &clabel);
2366 			/* make sure status is noted */
2367 
2368 			raid_init_component_label(raidPtr, &clabel);
2369 
2370 			clabel.row = 0;
2371 			clabel.column = scol;
2372 			/* Note: we *don't* change status from rf_ds_used_spare
2373 			   to rf_ds_optimal */
2374 			/* clabel.status = rf_ds_optimal; */
2375 
2376 			raidmarkdirty(raidPtr->Disks[sparecol].dev,
2377 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2378 				      raidPtr->mod_counter);
2379 		}
2380 	}
2381 }
2382 
2383 
2384 void
2385 rf_update_component_labels(RF_Raid_t *raidPtr, int final)
2386 {
2387 	RF_ComponentLabel_t clabel;
2388 	int sparecol;
2389 	int c;
2390 	int j;
2391 	int scol;
2392 
2393 	scol = -1;
2394 
2395 	/* XXX should do extra checks to make sure things really are clean,
2396 	   rather than blindly setting the clean bit... */
2397 
2398 	raidPtr->mod_counter++;
2399 
2400 	for (c = 0; c < raidPtr->numCol; c++) {
2401 		if (raidPtr->Disks[c].status == rf_ds_optimal) {
2402 			raidread_component_label(
2403 						 raidPtr->Disks[c].dev,
2404 						 raidPtr->raid_cinfo[c].ci_vp,
2405 						 &clabel);
2406 				/* make sure status is noted */
2407 			clabel.status = rf_ds_optimal;
2408 				/* bump the counter */
2409 			clabel.mod_counter = raidPtr->mod_counter;
2410 
2411 			raidwrite_component_label(
2412 						  raidPtr->Disks[c].dev,
2413 						  raidPtr->raid_cinfo[c].ci_vp,
2414 						  &clabel);
2415 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2416 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2417 					raidmarkclean(
2418 						      raidPtr->Disks[c].dev,
2419 						      raidPtr->raid_cinfo[c].ci_vp,
2420 						      raidPtr->mod_counter);
2421 				}
2422 			}
2423 		}
2424 		/* else we don't touch it.. */
2425 	}
2426 
2427 	for( c = 0; c < raidPtr->numSpare ; c++) {
2428 		sparecol = raidPtr->numCol + c;
2429 		/* Need to ensure that the reconstruct actually completed! */
2430 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2431 			/*
2432 
2433 			   we claim this disk is "optimal" if it's
2434 			   rf_ds_used_spare, as that means it should be
2435 			   directly substitutable for the disk it replaced.
2436 			   We note that too...
2437 
2438 			 */
2439 
2440 			for(j=0;j<raidPtr->numCol;j++) {
2441 				if (raidPtr->Disks[j].spareCol == sparecol) {
2442 					scol = j;
2443 					break;
2444 				}
2445 			}
2446 
2447 			/* XXX shouldn't *really* need this... */
2448 			raidread_component_label(
2449 				      raidPtr->Disks[sparecol].dev,
2450 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2451 				      &clabel);
2452 			/* make sure status is noted */
2453 
2454 			raid_init_component_label(raidPtr, &clabel);
2455 
2456 			clabel.mod_counter = raidPtr->mod_counter;
2457 			clabel.column = scol;
2458 			clabel.status = rf_ds_optimal;
2459 
2460 			raidwrite_component_label(
2461 				      raidPtr->Disks[sparecol].dev,
2462 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2463 				      &clabel);
2464 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2465 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2466 					raidmarkclean( raidPtr->Disks[sparecol].dev,
2467 						       raidPtr->raid_cinfo[sparecol].ci_vp,
2468 						       raidPtr->mod_counter);
2469 				}
2470 			}
2471 		}
2472 	}
2473 }
2474 
2475 void
2476 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured)
2477 {
2478 	struct proc *p;
2479 
2480 	p = raidPtr->engine_thread;
2481 
2482 	if (vp != NULL) {
2483 		if (auto_configured == 1) {
2484 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2485 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2486 			vput(vp);
2487 
2488 		} else {
2489 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2490 		}
2491 	}
2492 }
2493 
2494 
2495 void
2496 rf_UnconfigureVnodes(RF_Raid_t *raidPtr)
2497 {
2498 	int r,c;
2499 	struct vnode *vp;
2500 	int acd;
2501 
2502 
2503 	/* We take this opportunity to close the vnodes like we should.. */
2504 
2505 	for (c = 0; c < raidPtr->numCol; c++) {
2506 		vp = raidPtr->raid_cinfo[c].ci_vp;
2507 		acd = raidPtr->Disks[c].auto_configured;
2508 		rf_close_component(raidPtr, vp, acd);
2509 		raidPtr->raid_cinfo[c].ci_vp = NULL;
2510 		raidPtr->Disks[c].auto_configured = 0;
2511 	}
2512 
2513 	for (r = 0; r < raidPtr->numSpare; r++) {
2514 		vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp;
2515 		acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured;
2516 		rf_close_component(raidPtr, vp, acd);
2517 		raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL;
2518 		raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0;
2519 	}
2520 }
2521 
2522 
2523 void
2524 rf_ReconThread(struct rf_recon_req *req)
2525 {
2526 	int     s;
2527 	RF_Raid_t *raidPtr;
2528 
2529 	s = splbio();
2530 	raidPtr = (RF_Raid_t *) req->raidPtr;
2531 	raidPtr->recon_in_progress = 1;
2532 
2533 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col,
2534 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2535 
2536 	RF_Free(req, sizeof(*req));
2537 
2538 	raidPtr->recon_in_progress = 0;
2539 	splx(s);
2540 
2541 	/* That's all... */
2542 	kthread_exit(0);        /* does not return */
2543 }
2544 
2545 void
2546 rf_RewriteParityThread(RF_Raid_t *raidPtr)
2547 {
2548 	int retcode;
2549 	int s;
2550 
2551 	raidPtr->parity_rewrite_stripes_done = 0;
2552 	raidPtr->parity_rewrite_in_progress = 1;
2553 	s = splbio();
2554 	retcode = rf_RewriteParity(raidPtr);
2555 	splx(s);
2556 	if (retcode) {
2557 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2558 	} else {
2559 		/* set the clean bit!  If we shutdown correctly,
2560 		   the clean bit on each component label will get
2561 		   set */
2562 		raidPtr->parity_good = RF_RAID_CLEAN;
2563 	}
2564 	raidPtr->parity_rewrite_in_progress = 0;
2565 
2566 	/* Anyone waiting for us to stop?  If so, inform them... */
2567 	if (raidPtr->waitShutdown) {
2568 		wakeup(&raidPtr->parity_rewrite_in_progress);
2569 	}
2570 
2571 	/* That's all... */
2572 	kthread_exit(0);        /* does not return */
2573 }
2574 
2575 
2576 void
2577 rf_CopybackThread(RF_Raid_t *raidPtr)
2578 {
2579 	int s;
2580 
2581 	raidPtr->copyback_in_progress = 1;
2582 	s = splbio();
2583 	rf_CopybackReconstructedData(raidPtr);
2584 	splx(s);
2585 	raidPtr->copyback_in_progress = 0;
2586 
2587 	/* That's all... */
2588 	kthread_exit(0);        /* does not return */
2589 }
2590 
2591 
2592 void
2593 rf_ReconstructInPlaceThread(struct rf_recon_req *req)
2594 {
2595 	int s;
2596 	RF_Raid_t *raidPtr;
2597 
2598 	s = splbio();
2599 	raidPtr = req->raidPtr;
2600 	raidPtr->recon_in_progress = 1;
2601 	rf_ReconstructInPlace(raidPtr, req->col);
2602 	RF_Free(req, sizeof(*req));
2603 	raidPtr->recon_in_progress = 0;
2604 	splx(s);
2605 
2606 	/* That's all... */
2607 	kthread_exit(0);        /* does not return */
2608 }
2609 
2610 RF_AutoConfig_t *
2611 rf_find_raid_components()
2612 {
2613 	struct vnode *vp;
2614 	struct disklabel label;
2615 	struct device *dv;
2616 	dev_t dev;
2617 	int bmajor;
2618 	int error;
2619 	int i;
2620 	int good_one;
2621 	RF_ComponentLabel_t *clabel;
2622 	RF_AutoConfig_t *ac_list;
2623 	RF_AutoConfig_t *ac;
2624 
2625 
2626 	/* initialize the AutoConfig list */
2627 	ac_list = NULL;
2628 
2629 	/* we begin by trolling through *all* the devices on the system */
2630 
2631 	for (dv = alldevs.tqh_first; dv != NULL;
2632 	     dv = dv->dv_list.tqe_next) {
2633 
2634 		/* we are only interested in disks... */
2635 		if (dv->dv_class != DV_DISK)
2636 			continue;
2637 
2638 		/* we don't care about floppies... */
2639 		if (!strcmp(dv->dv_cfdata->cf_name,"fd")) {
2640 			continue;
2641 		}
2642 
2643 		/* we don't care about CD's... */
2644 		if (!strcmp(dv->dv_cfdata->cf_name,"cd")) {
2645 			continue;
2646 		}
2647 
2648 		/* hdfd is the Atari/Hades floppy driver */
2649 		if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) {
2650 			continue;
2651 		}
2652 		/* fdisa is the Atari/Milan floppy driver */
2653 		if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) {
2654 			continue;
2655 		}
2656 
2657 		/* need to find the device_name_to_block_device_major stuff */
2658 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2659 
2660 		/* get a vnode for the raw partition of this disk */
2661 
2662 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2663 		if (bdevvp(dev, &vp))
2664 			panic("RAID can't alloc vnode");
2665 
2666 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2667 
2668 		if (error) {
2669 			/* "Who cares."  Continue looking
2670 			   for something that exists*/
2671 			vput(vp);
2672 			continue;
2673 		}
2674 
2675 		/* Ok, the disk exists.  Go get the disklabel. */
2676 		error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0);
2677 		if (error) {
2678 			/*
2679 			 * XXX can't happen - open() would
2680 			 * have errored out (or faked up one)
2681 			 */
2682 			if (error != ENOTTY)
2683 				printf("RAIDframe: can't get label for dev "
2684 				    "%s (%d)\n", dv->dv_xname, error);
2685 		}
2686 
2687 		/* don't need this any more.  We'll allocate it again
2688 		   a little later if we really do... */
2689 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2690 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2691 		vput(vp);
2692 
2693 		if (error)
2694 			continue;
2695 
2696 		for (i=0; i < label.d_npartitions; i++) {
2697 			/* We only support partitions marked as RAID */
2698 			if (label.d_partitions[i].p_fstype != FS_RAID)
2699 				continue;
2700 
2701 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2702 			if (bdevvp(dev, &vp))
2703 				panic("RAID can't alloc vnode");
2704 
2705 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2706 			if (error) {
2707 				/* Whatever... */
2708 				vput(vp);
2709 				continue;
2710 			}
2711 
2712 			good_one = 0;
2713 
2714 			clabel = (RF_ComponentLabel_t *)
2715 				malloc(sizeof(RF_ComponentLabel_t),
2716 				       M_RAIDFRAME, M_NOWAIT);
2717 			if (clabel == NULL) {
2718 				/* XXX CLEANUP HERE */
2719 				printf("RAID auto config: out of memory!\n");
2720 				return(NULL); /* XXX probably should panic? */
2721 			}
2722 
2723 			if (!raidread_component_label(dev, vp, clabel)) {
2724 				/* Got the label.  Does it look reasonable? */
2725 				if (rf_reasonable_label(clabel) &&
2726 				    (clabel->partitionSize <=
2727 				     label.d_partitions[i].p_size)) {
2728 #if DEBUG
2729 					printf("Component on: %s%c: %d\n",
2730 					       dv->dv_xname, 'a'+i,
2731 					       label.d_partitions[i].p_size);
2732 					rf_print_component_label(clabel);
2733 #endif
2734 					/* if it's reasonable, add it,
2735 					   else ignore it. */
2736 					ac = (RF_AutoConfig_t *)
2737 						malloc(sizeof(RF_AutoConfig_t),
2738 						       M_RAIDFRAME,
2739 						       M_NOWAIT);
2740 					if (ac == NULL) {
2741 						/* XXX should panic?? */
2742 						return(NULL);
2743 					}
2744 
2745 					snprintf(ac->devname,
2746 					    sizeof(ac->devname), "%s%c",
2747 					    dv->dv_xname, 'a'+i);
2748 					ac->dev = dev;
2749 					ac->vp = vp;
2750 					ac->clabel = clabel;
2751 					ac->next = ac_list;
2752 					ac_list = ac;
2753 					good_one = 1;
2754 				}
2755 			}
2756 			if (!good_one) {
2757 				/* cleanup */
2758 				free(clabel, M_RAIDFRAME);
2759 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2760 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2761 				vput(vp);
2762 			}
2763 		}
2764 	}
2765 	return(ac_list);
2766 }
2767 
2768 static int
2769 rf_reasonable_label(RF_ComponentLabel_t *clabel)
2770 {
2771 
2772 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2773 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2774 	    ((clabel->clean == RF_RAID_CLEAN) ||
2775 	     (clabel->clean == RF_RAID_DIRTY)) &&
2776 	    clabel->row >=0 &&
2777 	    clabel->column >= 0 &&
2778 	    clabel->num_rows > 0 &&
2779 	    clabel->num_columns > 0 &&
2780 	    clabel->row < clabel->num_rows &&
2781 	    clabel->column < clabel->num_columns &&
2782 	    clabel->blockSize > 0 &&
2783 	    clabel->numBlocks > 0) {
2784 		/* label looks reasonable enough... */
2785 		return(1);
2786 	}
2787 	return(0);
2788 }
2789 
2790 
2791 #if DEBUG
2792 void
2793 rf_print_component_label(RF_ComponentLabel_t *clabel)
2794 {
2795 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2796 	       clabel->row, clabel->column,
2797 	       clabel->num_rows, clabel->num_columns);
2798 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2799 	       clabel->version, clabel->serial_number,
2800 	       clabel->mod_counter);
2801 	printf("   Clean: %s Status: %d\n",
2802 	       clabel->clean ? "Yes" : "No", clabel->status );
2803 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2804 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2805 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2806 	       (char) clabel->parityConfig, clabel->blockSize,
2807 	       clabel->numBlocks);
2808 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2809 	printf("   Contains root partition: %s\n",
2810 	       clabel->root_partition ? "Yes" : "No" );
2811 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2812 #if 0
2813 	   printf("   Config order: %d\n", clabel->config_order);
2814 #endif
2815 
2816 }
2817 #endif
2818 
2819 RF_ConfigSet_t *
2820 rf_create_auto_sets(RF_AutoConfig_t *ac_list)
2821 {
2822 	RF_AutoConfig_t *ac;
2823 	RF_ConfigSet_t *config_sets;
2824 	RF_ConfigSet_t *cset;
2825 	RF_AutoConfig_t *ac_next;
2826 
2827 
2828 	config_sets = NULL;
2829 
2830 	/* Go through the AutoConfig list, and figure out which components
2831 	   belong to what sets.  */
2832 	ac = ac_list;
2833 	while(ac!=NULL) {
2834 		/* we're going to putz with ac->next, so save it here
2835 		   for use at the end of the loop */
2836 		ac_next = ac->next;
2837 
2838 		if (config_sets == NULL) {
2839 			/* will need at least this one... */
2840 			config_sets = (RF_ConfigSet_t *)
2841 				malloc(sizeof(RF_ConfigSet_t),
2842 				       M_RAIDFRAME, M_NOWAIT);
2843 			if (config_sets == NULL) {
2844 				panic("rf_create_auto_sets: No memory!");
2845 			}
2846 			/* this one is easy :) */
2847 			config_sets->ac = ac;
2848 			config_sets->next = NULL;
2849 			config_sets->rootable = 0;
2850 			ac->next = NULL;
2851 		} else {
2852 			/* which set does this component fit into? */
2853 			cset = config_sets;
2854 			while(cset!=NULL) {
2855 				if (rf_does_it_fit(cset, ac)) {
2856 					/* looks like it matches... */
2857 					ac->next = cset->ac;
2858 					cset->ac = ac;
2859 					break;
2860 				}
2861 				cset = cset->next;
2862 			}
2863 			if (cset==NULL) {
2864 				/* didn't find a match above... new set..*/
2865 				cset = (RF_ConfigSet_t *)
2866 					malloc(sizeof(RF_ConfigSet_t),
2867 					       M_RAIDFRAME, M_NOWAIT);
2868 				if (cset == NULL) {
2869 					panic("rf_create_auto_sets: No memory!");
2870 				}
2871 				cset->ac = ac;
2872 				ac->next = NULL;
2873 				cset->next = config_sets;
2874 				cset->rootable = 0;
2875 				config_sets = cset;
2876 			}
2877 		}
2878 		ac = ac_next;
2879 	}
2880 
2881 
2882 	return(config_sets);
2883 }
2884 
2885 static int
2886 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac)
2887 {
2888 	RF_ComponentLabel_t *clabel1, *clabel2;
2889 
2890 	/* If this one matches the *first* one in the set, that's good
2891 	   enough, since the other members of the set would have been
2892 	   through here too... */
2893 	/* note that we are not checking partitionSize here..
2894 
2895 	   Note that we are also not checking the mod_counters here.
2896 	   If everything else matches execpt the mod_counter, that's
2897 	   good enough for this test.  We will deal with the mod_counters
2898 	   a little later in the autoconfiguration process.
2899 
2900 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2901 
2902 	   The reason we don't check for this is that failed disks
2903 	   will have lower modification counts.  If those disks are
2904 	   not added to the set they used to belong to, then they will
2905 	   form their own set, which may result in 2 different sets,
2906 	   for example, competing to be configured at raid0, and
2907 	   perhaps competing to be the root filesystem set.  If the
2908 	   wrong ones get configured, or both attempt to become /,
2909 	   weird behaviour and or serious lossage will occur.  Thus we
2910 	   need to bring them into the fold here, and kick them out at
2911 	   a later point.
2912 
2913 	*/
2914 
2915 	clabel1 = cset->ac->clabel;
2916 	clabel2 = ac->clabel;
2917 	if ((clabel1->version == clabel2->version) &&
2918 	    (clabel1->serial_number == clabel2->serial_number) &&
2919 	    (clabel1->num_rows == clabel2->num_rows) &&
2920 	    (clabel1->num_columns == clabel2->num_columns) &&
2921 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2922 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2923 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2924 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2925 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2926 	    (clabel1->blockSize == clabel2->blockSize) &&
2927 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2928 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2929 	    (clabel1->root_partition == clabel2->root_partition) &&
2930 	    (clabel1->last_unit == clabel2->last_unit) &&
2931 	    (clabel1->config_order == clabel2->config_order)) {
2932 		/* if it get's here, it almost *has* to be a match */
2933 	} else {
2934 		/* it's not consistent with somebody in the set..
2935 		   punt */
2936 		return(0);
2937 	}
2938 	/* all was fine.. it must fit... */
2939 	return(1);
2940 }
2941 
2942 int
2943 rf_have_enough_components(RF_ConfigSet_t *cset)
2944 {
2945 	RF_AutoConfig_t *ac;
2946 	RF_AutoConfig_t *auto_config;
2947 	RF_ComponentLabel_t *clabel;
2948 	int c;
2949 	int num_cols;
2950 	int num_missing;
2951 	int mod_counter;
2952 	int mod_counter_found;
2953 	int even_pair_failed;
2954 	char parity_type;
2955 
2956 
2957 	/* check to see that we have enough 'live' components
2958 	   of this set.  If so, we can configure it if necessary */
2959 
2960 	num_cols = cset->ac->clabel->num_columns;
2961 	parity_type = cset->ac->clabel->parityConfig;
2962 
2963 	/* XXX Check for duplicate components!?!?!? */
2964 
2965 	/* Determine what the mod_counter is supposed to be for this set. */
2966 
2967 	mod_counter_found = 0;
2968 	mod_counter = 0;
2969 	ac = cset->ac;
2970 	while(ac!=NULL) {
2971 		if (mod_counter_found==0) {
2972 			mod_counter = ac->clabel->mod_counter;
2973 			mod_counter_found = 1;
2974 		} else {
2975 			if (ac->clabel->mod_counter > mod_counter) {
2976 				mod_counter = ac->clabel->mod_counter;
2977 			}
2978 		}
2979 		ac = ac->next;
2980 	}
2981 
2982 	num_missing = 0;
2983 	auto_config = cset->ac;
2984 
2985 	even_pair_failed = 0;
2986 	for(c=0; c<num_cols; c++) {
2987 		ac = auto_config;
2988 		while(ac!=NULL) {
2989 			if ((ac->clabel->column == c) &&
2990 			    (ac->clabel->mod_counter == mod_counter)) {
2991 				/* it's this one... */
2992 #if DEBUG
2993 				printf("Found: %s at %d\n",
2994 				       ac->devname,c);
2995 #endif
2996 				break;
2997 			}
2998 			ac=ac->next;
2999 		}
3000 		if (ac==NULL) {
3001 				/* Didn't find one here! */
3002 				/* special case for RAID 1, especially
3003 				   where there are more than 2
3004 				   components (where RAIDframe treats
3005 				   things a little differently :( ) */
3006 			if (parity_type == '1') {
3007 				if (c%2 == 0) { /* even component */
3008 					even_pair_failed = 1;
3009 				} else { /* odd component.  If
3010 					    we're failed, and
3011 					    so is the even
3012 					    component, it's
3013 					    "Good Night, Charlie" */
3014 					if (even_pair_failed == 1) {
3015 						return(0);
3016 					}
3017 				}
3018 			} else {
3019 				/* normal accounting */
3020 				num_missing++;
3021 			}
3022 		}
3023 		if ((parity_type == '1') && (c%2 == 1)) {
3024 				/* Just did an even component, and we didn't
3025 				   bail.. reset the even_pair_failed flag,
3026 				   and go on to the next component.... */
3027 			even_pair_failed = 0;
3028 		}
3029 	}
3030 
3031 	clabel = cset->ac->clabel;
3032 
3033 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3034 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3035 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3036 		/* XXX this needs to be made *much* more general */
3037 		/* Too many failures */
3038 		return(0);
3039 	}
3040 	/* otherwise, all is well, and we've got enough to take a kick
3041 	   at autoconfiguring this set */
3042 	return(1);
3043 }
3044 
3045 void
3046 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config,
3047 			RF_Raid_t *raidPtr)
3048 {
3049 	RF_ComponentLabel_t *clabel;
3050 	int i;
3051 
3052 	clabel = ac->clabel;
3053 
3054 	/* 1. Fill in the common stuff */
3055 	config->numRow = clabel->num_rows = 1;
3056 	config->numCol = clabel->num_columns;
3057 	config->numSpare = 0; /* XXX should this be set here? */
3058 	config->sectPerSU = clabel->sectPerSU;
3059 	config->SUsPerPU = clabel->SUsPerPU;
3060 	config->SUsPerRU = clabel->SUsPerRU;
3061 	config->parityConfig = clabel->parityConfig;
3062 	/* XXX... */
3063 	strcpy(config->diskQueueType,"fifo");
3064 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3065 	config->layoutSpecificSize = 0; /* XXX ?? */
3066 
3067 	while(ac!=NULL) {
3068 		/* row/col values will be in range due to the checks
3069 		   in reasonable_label() */
3070 		strcpy(config->devnames[0][ac->clabel->column],
3071 		       ac->devname);
3072 		ac = ac->next;
3073 	}
3074 
3075 	for(i=0;i<RF_MAXDBGV;i++) {
3076 		config->debugVars[i][0] = 0;
3077 	}
3078 }
3079 
3080 int
3081 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value)
3082 {
3083 	RF_ComponentLabel_t clabel;
3084 	struct vnode *vp;
3085 	dev_t dev;
3086 	int column;
3087 	int sparecol;
3088 
3089 	raidPtr->autoconfigure = new_value;
3090 
3091 	for(column=0; column<raidPtr->numCol; column++) {
3092 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3093 			dev = raidPtr->Disks[column].dev;
3094 			vp = raidPtr->raid_cinfo[column].ci_vp;
3095 			raidread_component_label(dev, vp, &clabel);
3096 			clabel.autoconfigure = new_value;
3097 			raidwrite_component_label(dev, vp, &clabel);
3098 		}
3099 	}
3100 	for(column = 0; column < raidPtr->numSpare ; column++) {
3101 		sparecol = raidPtr->numCol + column;
3102 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3103 			dev = raidPtr->Disks[sparecol].dev;
3104 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3105 			raidread_component_label(dev, vp, &clabel);
3106 			clabel.autoconfigure = new_value;
3107 			raidwrite_component_label(dev, vp, &clabel);
3108 		}
3109 	}
3110 	return(new_value);
3111 }
3112 
3113 int
3114 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value)
3115 {
3116 	RF_ComponentLabel_t clabel;
3117 	struct vnode *vp;
3118 	dev_t dev;
3119 	int column;
3120 	int sparecol;
3121 
3122 	raidPtr->root_partition = new_value;
3123 	for(column=0; column<raidPtr->numCol; column++) {
3124 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3125 			dev = raidPtr->Disks[column].dev;
3126 			vp = raidPtr->raid_cinfo[column].ci_vp;
3127 			raidread_component_label(dev, vp, &clabel);
3128 			clabel.root_partition = new_value;
3129 			raidwrite_component_label(dev, vp, &clabel);
3130 		}
3131 	}
3132 	for(column = 0; column < raidPtr->numSpare ; column++) {
3133 		sparecol = raidPtr->numCol + column;
3134 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3135 			dev = raidPtr->Disks[sparecol].dev;
3136 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3137 			raidread_component_label(dev, vp, &clabel);
3138 			clabel.root_partition = new_value;
3139 			raidwrite_component_label(dev, vp, &clabel);
3140 		}
3141 	}
3142 	return(new_value);
3143 }
3144 
3145 void
3146 rf_release_all_vps(RF_ConfigSet_t *cset)
3147 {
3148 	RF_AutoConfig_t *ac;
3149 
3150 	ac = cset->ac;
3151 	while(ac!=NULL) {
3152 		/* Close the vp, and give it back */
3153 		if (ac->vp) {
3154 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3155 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3156 			vput(ac->vp);
3157 			ac->vp = NULL;
3158 		}
3159 		ac = ac->next;
3160 	}
3161 }
3162 
3163 
3164 void
3165 rf_cleanup_config_set(RF_ConfigSet_t *cset)
3166 {
3167 	RF_AutoConfig_t *ac;
3168 	RF_AutoConfig_t *next_ac;
3169 
3170 	ac = cset->ac;
3171 	while(ac!=NULL) {
3172 		next_ac = ac->next;
3173 		/* nuke the label */
3174 		free(ac->clabel, M_RAIDFRAME);
3175 		/* cleanup the config structure */
3176 		free(ac, M_RAIDFRAME);
3177 		/* "next.." */
3178 		ac = next_ac;
3179 	}
3180 	/* and, finally, nuke the config set */
3181 	free(cset, M_RAIDFRAME);
3182 }
3183 
3184 
3185 void
3186 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel)
3187 {
3188 	/* current version number */
3189 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3190 	clabel->serial_number = raidPtr->serial_number;
3191 	clabel->mod_counter = raidPtr->mod_counter;
3192 	clabel->num_rows = 1;
3193 	clabel->num_columns = raidPtr->numCol;
3194 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3195 	clabel->status = rf_ds_optimal; /* "It's good!" */
3196 
3197 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3198 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3199 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3200 
3201 	clabel->blockSize = raidPtr->bytesPerSector;
3202 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3203 
3204 	/* XXX not portable */
3205 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3206 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3207 	clabel->autoconfigure = raidPtr->autoconfigure;
3208 	clabel->root_partition = raidPtr->root_partition;
3209 	clabel->last_unit = raidPtr->raidid;
3210 	clabel->config_order = raidPtr->config_order;
3211 }
3212 
3213 int
3214 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit)
3215 {
3216 	RF_Raid_t *raidPtr;
3217 	RF_Config_t *config;
3218 	int raidID;
3219 	int retcode;
3220 
3221 #if DEBUG
3222 	printf("RAID autoconfigure\n");
3223 #endif
3224 
3225 	retcode = 0;
3226 	*unit = -1;
3227 
3228 	/* 1. Create a config structure */
3229 
3230 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3231 				       M_RAIDFRAME,
3232 				       M_NOWAIT);
3233 	if (config==NULL) {
3234 		printf("Out of mem!?!?\n");
3235 				/* XXX do something more intelligent here. */
3236 		return(1);
3237 	}
3238 
3239 	memset(config, 0, sizeof(RF_Config_t));
3240 
3241 	/*
3242 	   2. Figure out what RAID ID this one is supposed to live at
3243 	   See if we can get the same RAID dev that it was configured
3244 	   on last time..
3245 	*/
3246 
3247 	raidID = cset->ac->clabel->last_unit;
3248 	if ((raidID < 0) || (raidID >= numraid)) {
3249 		/* let's not wander off into lala land. */
3250 		raidID = numraid - 1;
3251 	}
3252 	if (raidPtrs[raidID]->valid != 0) {
3253 
3254 		/*
3255 		   Nope... Go looking for an alternative...
3256 		   Start high so we don't immediately use raid0 if that's
3257 		   not taken.
3258 		*/
3259 
3260 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3261 			if (raidPtrs[raidID]->valid == 0) {
3262 				/* can use this one! */
3263 				break;
3264 			}
3265 		}
3266 	}
3267 
3268 	if (raidID < 0) {
3269 		/* punt... */
3270 		printf("Unable to auto configure this set!\n");
3271 		printf("(Out of RAID devs!)\n");
3272 		return(1);
3273 	}
3274 
3275 #if DEBUG
3276 	printf("Configuring raid%d:\n",raidID);
3277 #endif
3278 
3279 	raidPtr = raidPtrs[raidID];
3280 
3281 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3282 	raidPtr->raidid = raidID;
3283 	raidPtr->openings = RAIDOUTSTANDING;
3284 
3285 	/* 3. Build the configuration structure */
3286 	rf_create_configuration(cset->ac, config, raidPtr);
3287 
3288 	/* 4. Do the configuration */
3289 	retcode = rf_Configure(raidPtr, config, cset->ac);
3290 
3291 	if (retcode == 0) {
3292 
3293 		raidinit(raidPtrs[raidID]);
3294 
3295 		rf_markalldirty(raidPtrs[raidID]);
3296 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3297 		if (cset->ac->clabel->root_partition==1) {
3298 			/* everything configured just fine.  Make a note
3299 			   that this set is eligible to be root. */
3300 			cset->rootable = 1;
3301 			/* XXX do this here? */
3302 			raidPtrs[raidID]->root_partition = 1;
3303 		}
3304 	}
3305 
3306 	/* 5. Cleanup */
3307 	free(config, M_RAIDFRAME);
3308 
3309 	*unit = raidID;
3310 	return(retcode);
3311 }
3312 
3313 void
3314 rf_disk_unbusy(RF_RaidAccessDesc_t *desc)
3315 {
3316 	struct buf *bp;
3317 
3318 	bp = (struct buf *)desc->bp;
3319 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3320 	    (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ));
3321 }
3322 
3323 void
3324 rf_pool_init(struct pool *p, size_t size, const char *w_chan,
3325 	     size_t xmin, size_t xmax)
3326 {
3327 	pool_init(p, size, 0, 0, 0, w_chan, NULL);
3328 	pool_sethiwat(p, xmax);
3329 	pool_prime(p, xmin);
3330 	pool_setlowat(p, xmin);
3331 }
3332