xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.199 2006/01/08 22:26:30 oster 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 1.199 2006/01/08 22:26:30 oster 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 /* XXX Not sure if the following should be replacing the raidPtrs above,
237    or if it should be used in conjunction with that...
238 */
239 
240 struct raid_softc {
241 	int     sc_flags;	/* flags */
242 	int     sc_cflags;	/* configuration flags */
243 	size_t  sc_size;        /* size of the raid device */
244 	char    sc_xname[20];	/* XXX external name */
245 	struct disk sc_dkdev;	/* generic disk device info */
246 	struct bufq_state *buf_queue;	/* used for the device queue */
247 };
248 /* sc_flags */
249 #define RAIDF_INITED	0x01	/* unit has been initialized */
250 #define RAIDF_WLABEL	0x02	/* label area is writable */
251 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
252 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
253 #define RAIDF_LOCKED	0x80	/* unit is locked */
254 
255 #define	raidunit(x)	DISKUNIT(x)
256 int numraid = 0;
257 
258 /*
259  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
260  * Be aware that large numbers can allow the driver to consume a lot of
261  * kernel memory, especially on writes, and in degraded mode reads.
262  *
263  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
264  * a single 64K write will typically require 64K for the old data,
265  * 64K for the old parity, and 64K for the new parity, for a total
266  * of 192K (if the parity buffer is not re-used immediately).
267  * Even it if is used immediately, that's still 128K, which when multiplied
268  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
269  *
270  * Now in degraded mode, for example, a 64K read on the above setup may
271  * require data reconstruction, which will require *all* of the 4 remaining
272  * disks to participate -- 4 * 32K/disk == 128K again.
273  */
274 
275 #ifndef RAIDOUTSTANDING
276 #define RAIDOUTSTANDING   6
277 #endif
278 
279 #define RAIDLABELDEV(dev)	\
280 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
281 
282 /* declared here, and made public, for the benefit of KVM stuff.. */
283 struct raid_softc *raid_softc;
284 
285 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
286 				     struct disklabel *);
287 static void raidgetdisklabel(dev_t);
288 static void raidmakedisklabel(struct raid_softc *);
289 
290 static int raidlock(struct raid_softc *);
291 static void raidunlock(struct raid_softc *);
292 
293 static void rf_markalldirty(RF_Raid_t *);
294 
295 struct device *raidrootdev;
296 
297 void rf_ReconThread(struct rf_recon_req *);
298 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
299 void rf_CopybackThread(RF_Raid_t *raidPtr);
300 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
301 int rf_autoconfig(struct device *self);
302 void rf_buildroothack(RF_ConfigSet_t *);
303 
304 RF_AutoConfig_t *rf_find_raid_components(void);
305 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
306 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
307 static int rf_reasonable_label(RF_ComponentLabel_t *);
308 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
309 int rf_set_autoconfig(RF_Raid_t *, int);
310 int rf_set_rootpartition(RF_Raid_t *, int);
311 void rf_release_all_vps(RF_ConfigSet_t *);
312 void rf_cleanup_config_set(RF_ConfigSet_t *);
313 int rf_have_enough_components(RF_ConfigSet_t *);
314 int rf_auto_config_set(RF_ConfigSet_t *, int *);
315 
316 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
317 				  allow autoconfig to take place.
318 			          Note that this is overridden by having
319 			          RAID_AUTOCONFIG as an option in the
320 			          kernel config file.  */
321 
322 struct RF_Pools_s rf_pools;
323 
324 void
325 raidattach(int num)
326 {
327 	int raidID;
328 	int i, rc;
329 
330 #ifdef DEBUG
331 	printf("raidattach: Asked for %d units\n", num);
332 #endif
333 
334 	if (num <= 0) {
335 #ifdef DIAGNOSTIC
336 		panic("raidattach: count <= 0");
337 #endif
338 		return;
339 	}
340 	/* This is where all the initialization stuff gets done. */
341 
342 	numraid = num;
343 
344 	/* Make some space for requested number of units... */
345 
346 	RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **));
347 	if (raidPtrs == NULL) {
348 		panic("raidPtrs is NULL!!");
349 	}
350 
351 	rf_mutex_init(&rf_sparet_wait_mutex);
352 
353 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
354 
355 	for (i = 0; i < num; i++)
356 		raidPtrs[i] = NULL;
357 	rc = rf_BootRaidframe();
358 	if (rc == 0)
359 		printf("Kernelized RAIDframe activated\n");
360 	else
361 		panic("Serious error booting RAID!!");
362 
363 	/* put together some datastructures like the CCD device does.. This
364 	 * lets us lock the device and what-not when it gets opened. */
365 
366 	raid_softc = (struct raid_softc *)
367 		malloc(num * sizeof(struct raid_softc),
368 		       M_RAIDFRAME, M_NOWAIT);
369 	if (raid_softc == NULL) {
370 		printf("WARNING: no memory for RAIDframe driver\n");
371 		return;
372 	}
373 
374 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
375 
376 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
377 					      M_RAIDFRAME, M_NOWAIT);
378 	if (raidrootdev == NULL) {
379 		panic("No memory for RAIDframe driver!!?!?!");
380 	}
381 
382 	for (raidID = 0; raidID < num; raidID++) {
383 		bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0);
384 		pseudo_disk_init(&raid_softc[raidID].sc_dkdev);
385 
386 		raidrootdev[raidID].dv_class  = DV_DISK;
387 		raidrootdev[raidID].dv_cfdata = NULL;
388 		raidrootdev[raidID].dv_unit   = raidID;
389 		raidrootdev[raidID].dv_parent = NULL;
390 		raidrootdev[raidID].dv_flags  = 0;
391 		snprintf(raidrootdev[raidID].dv_xname,
392 		    sizeof(raidrootdev[raidID].dv_xname), "raid%d", raidID);
393 
394 		RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t),
395 			  (RF_Raid_t *));
396 		if (raidPtrs[raidID] == NULL) {
397 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
398 			numraid = raidID;
399 			return;
400 		}
401 	}
402 
403 #ifdef RAID_AUTOCONFIG
404 	raidautoconfig = 1;
405 #endif
406 
407 	/*
408 	 * Register a finalizer which will be used to auto-config RAID
409 	 * sets once all real hardware devices have been found.
410 	 */
411 	if (config_finalize_register(NULL, rf_autoconfig) != 0)
412 		printf("WARNING: unable to register RAIDframe finalizer\n");
413 }
414 
415 int
416 rf_autoconfig(struct device *self)
417 {
418 	RF_AutoConfig_t *ac_list;
419 	RF_ConfigSet_t *config_sets;
420 
421 	if (raidautoconfig == 0)
422 		return (0);
423 
424 	/* XXX This code can only be run once. */
425 	raidautoconfig = 0;
426 
427 	/* 1. locate all RAID components on the system */
428 #ifdef DEBUG
429 	printf("Searching for RAID components...\n");
430 #endif
431 	ac_list = rf_find_raid_components();
432 
433 	/* 2. Sort them into their respective sets. */
434 	config_sets = rf_create_auto_sets(ac_list);
435 
436 	/*
437 	 * 3. Evaluate each set andconfigure the valid ones.
438 	 * This gets done in rf_buildroothack().
439 	 */
440 	rf_buildroothack(config_sets);
441 
442 	return (1);
443 }
444 
445 void
446 rf_buildroothack(RF_ConfigSet_t *config_sets)
447 {
448 	RF_ConfigSet_t *cset;
449 	RF_ConfigSet_t *next_cset;
450 	int retcode;
451 	int raidID;
452 	int rootID;
453 	int num_root;
454 
455 	rootID = 0;
456 	num_root = 0;
457 	cset = config_sets;
458 	while(cset != NULL ) {
459 		next_cset = cset->next;
460 		if (rf_have_enough_components(cset) &&
461 		    cset->ac->clabel->autoconfigure==1) {
462 			retcode = rf_auto_config_set(cset,&raidID);
463 			if (!retcode) {
464 				if (cset->rootable) {
465 					rootID = raidID;
466 					num_root++;
467 				}
468 			} else {
469 				/* The autoconfig didn't work :( */
470 #if DEBUG
471 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
472 #endif
473 				rf_release_all_vps(cset);
474 			}
475 		} else {
476 			/* we're not autoconfiguring this set...
477 			   release the associated resources */
478 			rf_release_all_vps(cset);
479 		}
480 		/* cleanup */
481 		rf_cleanup_config_set(cset);
482 		cset = next_cset;
483 	}
484 
485 	/* we found something bootable... */
486 
487 	if (num_root == 1) {
488 		booted_device = &raidrootdev[rootID];
489 	} else if (num_root > 1) {
490 		/* we can't guess.. require the user to answer... */
491 		boothowto |= RB_ASKNAME;
492 	}
493 }
494 
495 
496 int
497 raidsize(dev_t dev)
498 {
499 	struct raid_softc *rs;
500 	struct disklabel *lp;
501 	int     part, unit, omask, size;
502 
503 	unit = raidunit(dev);
504 	if (unit >= numraid)
505 		return (-1);
506 	rs = &raid_softc[unit];
507 
508 	if ((rs->sc_flags & RAIDF_INITED) == 0)
509 		return (-1);
510 
511 	part = DISKPART(dev);
512 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
513 	lp = rs->sc_dkdev.dk_label;
514 
515 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp))
516 		return (-1);
517 
518 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
519 		size = -1;
520 	else
521 		size = lp->d_partitions[part].p_size *
522 		    (lp->d_secsize / DEV_BSIZE);
523 
524 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp))
525 		return (-1);
526 
527 	return (size);
528 
529 }
530 
531 int
532 raiddump(dev_t dev, daddr_t blkno, caddr_t va, size_t  size)
533 {
534 	/* Not implemented. */
535 	return ENXIO;
536 }
537 /* ARGSUSED */
538 int
539 raidopen(dev_t dev, int flags, int fmt, struct lwp *l)
540 {
541 	int     unit = raidunit(dev);
542 	struct raid_softc *rs;
543 	struct disklabel *lp;
544 	int     part, pmask;
545 	int     error = 0;
546 
547 	if (unit >= numraid)
548 		return (ENXIO);
549 	rs = &raid_softc[unit];
550 
551 	if ((error = raidlock(rs)) != 0)
552 		return (error);
553 	lp = rs->sc_dkdev.dk_label;
554 
555 	part = DISKPART(dev);
556 	pmask = (1 << part);
557 
558 	if ((rs->sc_flags & RAIDF_INITED) &&
559 	    (rs->sc_dkdev.dk_openmask == 0))
560 		raidgetdisklabel(dev);
561 
562 	/* make sure that this partition exists */
563 
564 	if (part != RAW_PART) {
565 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
566 		    ((part >= lp->d_npartitions) ||
567 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
568 			error = ENXIO;
569 			raidunlock(rs);
570 			return (error);
571 		}
572 	}
573 	/* Prevent this unit from being unconfigured while open. */
574 	switch (fmt) {
575 	case S_IFCHR:
576 		rs->sc_dkdev.dk_copenmask |= pmask;
577 		break;
578 
579 	case S_IFBLK:
580 		rs->sc_dkdev.dk_bopenmask |= pmask;
581 		break;
582 	}
583 
584 	if ((rs->sc_dkdev.dk_openmask == 0) &&
585 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
586 		/* First one... mark things as dirty... Note that we *MUST*
587 		 have done a configure before this.  I DO NOT WANT TO BE
588 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
589 		 THAT THEY BELONG TOGETHER!!!!! */
590 		/* XXX should check to see if we're only open for reading
591 		   here... If so, we needn't do this, but then need some
592 		   other way of keeping track of what's happened.. */
593 
594 		rf_markalldirty( raidPtrs[unit] );
595 	}
596 
597 
598 	rs->sc_dkdev.dk_openmask =
599 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
600 
601 	raidunlock(rs);
602 
603 	return (error);
604 
605 
606 }
607 /* ARGSUSED */
608 int
609 raidclose(dev_t dev, int flags, int fmt, struct lwp *l)
610 {
611 	int     unit = raidunit(dev);
612 	struct raid_softc *rs;
613 	int     error = 0;
614 	int     part;
615 
616 	if (unit >= numraid)
617 		return (ENXIO);
618 	rs = &raid_softc[unit];
619 
620 	if ((error = raidlock(rs)) != 0)
621 		return (error);
622 
623 	part = DISKPART(dev);
624 
625 	/* ...that much closer to allowing unconfiguration... */
626 	switch (fmt) {
627 	case S_IFCHR:
628 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
629 		break;
630 
631 	case S_IFBLK:
632 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
633 		break;
634 	}
635 	rs->sc_dkdev.dk_openmask =
636 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
637 
638 	if ((rs->sc_dkdev.dk_openmask == 0) &&
639 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
640 		/* Last one... device is not unconfigured yet.
641 		   Device shutdown has taken care of setting the
642 		   clean bits if RAIDF_INITED is not set
643 		   mark things as clean... */
644 
645 		rf_update_component_labels(raidPtrs[unit],
646 						 RF_FINAL_COMPONENT_UPDATE);
647 		if (doing_shutdown) {
648 			/* last one, and we're going down, so
649 			   lights out for this RAID set too. */
650 			error = rf_Shutdown(raidPtrs[unit]);
651 
652 			/* It's no longer initialized... */
653 			rs->sc_flags &= ~RAIDF_INITED;
654 
655 			/* Detach the disk. */
656 			pseudo_disk_detach(&rs->sc_dkdev);
657 		}
658 	}
659 
660 	raidunlock(rs);
661 	return (0);
662 
663 }
664 
665 void
666 raidstrategy(struct buf *bp)
667 {
668 	int s;
669 
670 	unsigned int raidID = raidunit(bp->b_dev);
671 	RF_Raid_t *raidPtr;
672 	struct raid_softc *rs = &raid_softc[raidID];
673 	int     wlabel;
674 
675 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
676 		bp->b_error = ENXIO;
677 		bp->b_flags |= B_ERROR;
678 		goto done;
679 	}
680 	if (raidID >= numraid || !raidPtrs[raidID]) {
681 		bp->b_error = ENODEV;
682 		bp->b_flags |= B_ERROR;
683 		goto done;
684 	}
685 	raidPtr = raidPtrs[raidID];
686 	if (!raidPtr->valid) {
687 		bp->b_error = ENODEV;
688 		bp->b_flags |= B_ERROR;
689 		goto done;
690 	}
691 	if (bp->b_bcount == 0) {
692 		db1_printf(("b_bcount is zero..\n"));
693 		goto done;
694 	}
695 
696 	/*
697 	 * Do bounds checking and adjust transfer.  If there's an
698 	 * error, the bounds check will flag that for us.
699 	 */
700 
701 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
702 	if (DISKPART(bp->b_dev) == RAW_PART) {
703 		uint64_t size; /* device size in DEV_BSIZE unit */
704 
705 		if (raidPtr->logBytesPerSector > DEV_BSHIFT) {
706 			size = raidPtr->totalSectors <<
707 			    (raidPtr->logBytesPerSector - DEV_BSHIFT);
708 		} else {
709 			size = raidPtr->totalSectors >>
710 			    (DEV_BSHIFT - raidPtr->logBytesPerSector);
711 		}
712 		if (bounds_check_with_mediasize(bp, DEV_BSIZE, size) <= 0) {
713 			goto done;
714 		}
715 	} else {
716 		if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) {
717 			db1_printf(("Bounds check failed!!:%d %d\n",
718 				(int) bp->b_blkno, (int) wlabel));
719 			goto done;
720 		}
721 	}
722 	s = splbio();
723 
724 	bp->b_resid = 0;
725 
726 	/* stuff it onto our queue */
727 	BUFQ_PUT(rs->buf_queue, bp);
728 
729 	/* scheduled the IO to happen at the next convenient time */
730 	wakeup(&(raidPtrs[raidID]->iodone));
731 
732 	splx(s);
733 	return;
734 
735 done:
736 	bp->b_resid = bp->b_bcount;
737 	biodone(bp);
738 }
739 /* ARGSUSED */
740 int
741 raidread(dev_t dev, struct uio *uio, int flags)
742 {
743 	int     unit = raidunit(dev);
744 	struct raid_softc *rs;
745 
746 	if (unit >= numraid)
747 		return (ENXIO);
748 	rs = &raid_softc[unit];
749 
750 	if ((rs->sc_flags & RAIDF_INITED) == 0)
751 		return (ENXIO);
752 
753 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
754 
755 }
756 /* ARGSUSED */
757 int
758 raidwrite(dev_t dev, struct uio *uio, int flags)
759 {
760 	int     unit = raidunit(dev);
761 	struct raid_softc *rs;
762 
763 	if (unit >= numraid)
764 		return (ENXIO);
765 	rs = &raid_softc[unit];
766 
767 	if ((rs->sc_flags & RAIDF_INITED) == 0)
768 		return (ENXIO);
769 
770 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
771 
772 }
773 
774 int
775 raidioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct lwp *l)
776 {
777 	int     unit = raidunit(dev);
778 	int     error = 0;
779 	int     part, pmask;
780 	struct raid_softc *rs;
781 	RF_Config_t *k_cfg, *u_cfg;
782 	RF_Raid_t *raidPtr;
783 	RF_RaidDisk_t *diskPtr;
784 	RF_AccTotals_t *totals;
785 	RF_DeviceConfig_t *d_cfg, **ucfgp;
786 	u_char *specific_buf;
787 	int retcode = 0;
788 	int column;
789 	int raidid;
790 	struct rf_recon_req *rrcopy, *rr;
791 	RF_ComponentLabel_t *clabel;
792 	RF_ComponentLabel_t ci_label;
793 	RF_ComponentLabel_t **clabel_ptr;
794 	RF_SingleComponent_t *sparePtr,*componentPtr;
795 	RF_SingleComponent_t hot_spare;
796 	RF_SingleComponent_t component;
797 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
798 	int i, j, d;
799 #ifdef __HAVE_OLD_DISKLABEL
800 	struct disklabel newlabel;
801 #endif
802 
803 	if (unit >= numraid)
804 		return (ENXIO);
805 	rs = &raid_softc[unit];
806 	raidPtr = raidPtrs[unit];
807 
808 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
809 		(int) DISKPART(dev), (int) unit, (int) cmd));
810 
811 	/* Must be open for writes for these commands... */
812 	switch (cmd) {
813 	case DIOCSDINFO:
814 	case DIOCWDINFO:
815 #ifdef __HAVE_OLD_DISKLABEL
816 	case ODIOCWDINFO:
817 	case ODIOCSDINFO:
818 #endif
819 	case DIOCWLABEL:
820 		if ((flag & FWRITE) == 0)
821 			return (EBADF);
822 	}
823 
824 	/* Must be initialized for these... */
825 	switch (cmd) {
826 	case DIOCGDINFO:
827 	case DIOCSDINFO:
828 	case DIOCWDINFO:
829 #ifdef __HAVE_OLD_DISKLABEL
830 	case ODIOCGDINFO:
831 	case ODIOCWDINFO:
832 	case ODIOCSDINFO:
833 	case ODIOCGDEFLABEL:
834 #endif
835 	case DIOCGPART:
836 	case DIOCWLABEL:
837 	case DIOCGDEFLABEL:
838 	case RAIDFRAME_SHUTDOWN:
839 	case RAIDFRAME_REWRITEPARITY:
840 	case RAIDFRAME_GET_INFO:
841 	case RAIDFRAME_RESET_ACCTOTALS:
842 	case RAIDFRAME_GET_ACCTOTALS:
843 	case RAIDFRAME_KEEP_ACCTOTALS:
844 	case RAIDFRAME_GET_SIZE:
845 	case RAIDFRAME_FAIL_DISK:
846 	case RAIDFRAME_COPYBACK:
847 	case RAIDFRAME_CHECK_RECON_STATUS:
848 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
849 	case RAIDFRAME_GET_COMPONENT_LABEL:
850 	case RAIDFRAME_SET_COMPONENT_LABEL:
851 	case RAIDFRAME_ADD_HOT_SPARE:
852 	case RAIDFRAME_REMOVE_HOT_SPARE:
853 	case RAIDFRAME_INIT_LABELS:
854 	case RAIDFRAME_REBUILD_IN_PLACE:
855 	case RAIDFRAME_CHECK_PARITY:
856 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
857 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
858 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
859 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
860 	case RAIDFRAME_SET_AUTOCONFIG:
861 	case RAIDFRAME_SET_ROOT:
862 	case RAIDFRAME_DELETE_COMPONENT:
863 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
864 		if ((rs->sc_flags & RAIDF_INITED) == 0)
865 			return (ENXIO);
866 	}
867 
868 	switch (cmd) {
869 
870 		/* configure the system */
871 	case RAIDFRAME_CONFIGURE:
872 
873 		if (raidPtr->valid) {
874 			/* There is a valid RAID set running on this unit! */
875 			printf("raid%d: Device already configured!\n",unit);
876 			return(EINVAL);
877 		}
878 
879 		/* copy-in the configuration information */
880 		/* data points to a pointer to the configuration structure */
881 
882 		u_cfg = *((RF_Config_t **) data);
883 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
884 		if (k_cfg == NULL) {
885 			return (ENOMEM);
886 		}
887 		retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t));
888 		if (retcode) {
889 			RF_Free(k_cfg, sizeof(RF_Config_t));
890 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
891 				retcode));
892 			return (retcode);
893 		}
894 		/* allocate a buffer for the layout-specific data, and copy it
895 		 * in */
896 		if (k_cfg->layoutSpecificSize) {
897 			if (k_cfg->layoutSpecificSize > 10000) {
898 				/* sanity check */
899 				RF_Free(k_cfg, sizeof(RF_Config_t));
900 				return (EINVAL);
901 			}
902 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
903 			    (u_char *));
904 			if (specific_buf == NULL) {
905 				RF_Free(k_cfg, sizeof(RF_Config_t));
906 				return (ENOMEM);
907 			}
908 			retcode = copyin(k_cfg->layoutSpecific, specific_buf,
909 			    k_cfg->layoutSpecificSize);
910 			if (retcode) {
911 				RF_Free(k_cfg, sizeof(RF_Config_t));
912 				RF_Free(specific_buf,
913 					k_cfg->layoutSpecificSize);
914 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
915 					retcode));
916 				return (retcode);
917 			}
918 		} else
919 			specific_buf = NULL;
920 		k_cfg->layoutSpecific = specific_buf;
921 
922 		/* should do some kind of sanity check on the configuration.
923 		 * Store the sum of all the bytes in the last byte? */
924 
925 		/* configure the system */
926 
927 		/*
928 		 * Clear the entire RAID descriptor, just to make sure
929 		 *  there is no stale data left in the case of a
930 		 *  reconfiguration
931 		 */
932 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
933 		raidPtr->raidid = unit;
934 
935 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
936 
937 		if (retcode == 0) {
938 
939 			/* allow this many simultaneous IO's to
940 			   this RAID device */
941 			raidPtr->openings = RAIDOUTSTANDING;
942 
943 			raidinit(raidPtr);
944 			rf_markalldirty(raidPtr);
945 		}
946 		/* free the buffers.  No return code here. */
947 		if (k_cfg->layoutSpecificSize) {
948 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
949 		}
950 		RF_Free(k_cfg, sizeof(RF_Config_t));
951 
952 		return (retcode);
953 
954 		/* shutdown the system */
955 	case RAIDFRAME_SHUTDOWN:
956 
957 		if ((error = raidlock(rs)) != 0)
958 			return (error);
959 
960 		/*
961 		 * If somebody has a partition mounted, we shouldn't
962 		 * shutdown.
963 		 */
964 
965 		part = DISKPART(dev);
966 		pmask = (1 << part);
967 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
968 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
969 			(rs->sc_dkdev.dk_copenmask & pmask))) {
970 			raidunlock(rs);
971 			return (EBUSY);
972 		}
973 
974 		retcode = rf_Shutdown(raidPtr);
975 
976 		/* It's no longer initialized... */
977 		rs->sc_flags &= ~RAIDF_INITED;
978 
979 		/* Detach the disk. */
980 		pseudo_disk_detach(&rs->sc_dkdev);
981 
982 		raidunlock(rs);
983 
984 		return (retcode);
985 	case RAIDFRAME_GET_COMPONENT_LABEL:
986 		clabel_ptr = (RF_ComponentLabel_t **) data;
987 		/* need to read the component label for the disk indicated
988 		   by row,column in clabel */
989 
990 		/* For practice, let's get it directly fromdisk, rather
991 		   than from the in-core copy */
992 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
993 			   (RF_ComponentLabel_t *));
994 		if (clabel == NULL)
995 			return (ENOMEM);
996 
997 		retcode = copyin( *clabel_ptr, clabel,
998 				  sizeof(RF_ComponentLabel_t));
999 
1000 		if (retcode) {
1001 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1002 			return(retcode);
1003 		}
1004 
1005 		clabel->row = 0; /* Don't allow looking at anything else.*/
1006 
1007 		column = clabel->column;
1008 
1009 		if ((column < 0) || (column >= raidPtr->numCol +
1010 				     raidPtr->numSpare)) {
1011 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1012 			return(EINVAL);
1013 		}
1014 
1015 		retcode = raidread_component_label(raidPtr->Disks[column].dev,
1016 				raidPtr->raid_cinfo[column].ci_vp,
1017 				clabel );
1018 
1019 		if (retcode == 0) {
1020 			retcode = copyout(clabel, *clabel_ptr,
1021 					  sizeof(RF_ComponentLabel_t));
1022 		}
1023 		RF_Free(clabel, sizeof(RF_ComponentLabel_t));
1024 		return (retcode);
1025 
1026 	case RAIDFRAME_SET_COMPONENT_LABEL:
1027 		clabel = (RF_ComponentLabel_t *) data;
1028 
1029 		/* XXX check the label for valid stuff... */
1030 		/* Note that some things *should not* get modified --
1031 		   the user should be re-initing the labels instead of
1032 		   trying to patch things.
1033 		   */
1034 
1035 		raidid = raidPtr->raidid;
1036 #if DEBUG
1037 		printf("raid%d: Got component label:\n", raidid);
1038 		printf("raid%d: Version: %d\n", raidid, clabel->version);
1039 		printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number);
1040 		printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter);
1041 		printf("raid%d: Column: %d\n", raidid, clabel->column);
1042 		printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns);
1043 		printf("raid%d: Clean: %d\n", raidid, clabel->clean);
1044 		printf("raid%d: Status: %d\n", raidid, clabel->status);
1045 #endif
1046 		clabel->row = 0;
1047 		column = clabel->column;
1048 
1049 		if ((column < 0) || (column >= raidPtr->numCol)) {
1050 			return(EINVAL);
1051 		}
1052 
1053 		/* XXX this isn't allowed to do anything for now :-) */
1054 
1055 		/* XXX and before it is, we need to fill in the rest
1056 		   of the fields!?!?!?! */
1057 #if 0
1058 		raidwrite_component_label(
1059                             raidPtr->Disks[column].dev,
1060 			    raidPtr->raid_cinfo[column].ci_vp,
1061 			    clabel );
1062 #endif
1063 		return (0);
1064 
1065 	case RAIDFRAME_INIT_LABELS:
1066 		clabel = (RF_ComponentLabel_t *) data;
1067 		/*
1068 		   we only want the serial number from
1069 		   the above.  We get all the rest of the information
1070 		   from the config that was used to create this RAID
1071 		   set.
1072 		   */
1073 
1074 		raidPtr->serial_number = clabel->serial_number;
1075 
1076 		raid_init_component_label(raidPtr, &ci_label);
1077 		ci_label.serial_number = clabel->serial_number;
1078 		ci_label.row = 0; /* we dont' pretend to support more */
1079 
1080 		for(column=0;column<raidPtr->numCol;column++) {
1081 			diskPtr = &raidPtr->Disks[column];
1082 			if (!RF_DEAD_DISK(diskPtr->status)) {
1083 				ci_label.partitionSize = diskPtr->partitionSize;
1084 				ci_label.column = column;
1085 				raidwrite_component_label(
1086 							  raidPtr->Disks[column].dev,
1087 							  raidPtr->raid_cinfo[column].ci_vp,
1088 							  &ci_label );
1089 			}
1090 		}
1091 
1092 		return (retcode);
1093 	case RAIDFRAME_SET_AUTOCONFIG:
1094 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1095 		printf("raid%d: New autoconfig value is: %d\n",
1096 		       raidPtr->raidid, d);
1097 		*(int *) data = d;
1098 		return (retcode);
1099 
1100 	case RAIDFRAME_SET_ROOT:
1101 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1102 		printf("raid%d: New rootpartition value is: %d\n",
1103 		       raidPtr->raidid, d);
1104 		*(int *) data = d;
1105 		return (retcode);
1106 
1107 		/* initialize all parity */
1108 	case RAIDFRAME_REWRITEPARITY:
1109 
1110 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1111 			/* Parity for RAID 0 is trivially correct */
1112 			raidPtr->parity_good = RF_RAID_CLEAN;
1113 			return(0);
1114 		}
1115 
1116 		if (raidPtr->parity_rewrite_in_progress == 1) {
1117 			/* Re-write is already in progress! */
1118 			return(EINVAL);
1119 		}
1120 
1121 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1122 					   rf_RewriteParityThread,
1123 					   raidPtr,"raid_parity");
1124 		return (retcode);
1125 
1126 
1127 	case RAIDFRAME_ADD_HOT_SPARE:
1128 		sparePtr = (RF_SingleComponent_t *) data;
1129 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1130 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1131 		return(retcode);
1132 
1133 	case RAIDFRAME_REMOVE_HOT_SPARE:
1134 		return(retcode);
1135 
1136 	case RAIDFRAME_DELETE_COMPONENT:
1137 		componentPtr = (RF_SingleComponent_t *)data;
1138 		memcpy( &component, componentPtr,
1139 			sizeof(RF_SingleComponent_t));
1140 		retcode = rf_delete_component(raidPtr, &component);
1141 		return(retcode);
1142 
1143 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1144 		componentPtr = (RF_SingleComponent_t *)data;
1145 		memcpy( &component, componentPtr,
1146 			sizeof(RF_SingleComponent_t));
1147 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1148 		return(retcode);
1149 
1150 	case RAIDFRAME_REBUILD_IN_PLACE:
1151 
1152 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1153 			/* Can't do this on a RAID 0!! */
1154 			return(EINVAL);
1155 		}
1156 
1157 		if (raidPtr->recon_in_progress == 1) {
1158 			/* a reconstruct is already in progress! */
1159 			return(EINVAL);
1160 		}
1161 
1162 		componentPtr = (RF_SingleComponent_t *) data;
1163 		memcpy( &component, componentPtr,
1164 			sizeof(RF_SingleComponent_t));
1165 		component.row = 0; /* we don't support any more */
1166 		column = component.column;
1167 
1168 		if ((column < 0) || (column >= raidPtr->numCol)) {
1169 			return(EINVAL);
1170 		}
1171 
1172 		RF_LOCK_MUTEX(raidPtr->mutex);
1173 		if ((raidPtr->Disks[column].status == rf_ds_optimal) &&
1174 		    (raidPtr->numFailures > 0)) {
1175 			/* XXX 0 above shouldn't be constant!!! */
1176 			/* some component other than this has failed.
1177 			   Let's not make things worse than they already
1178 			   are... */
1179 			printf("raid%d: Unable to reconstruct to disk at:\n",
1180 			       raidPtr->raidid);
1181 			printf("raid%d:     Col: %d   Too many failures.\n",
1182 			       raidPtr->raidid, column);
1183 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1184 			return (EINVAL);
1185 		}
1186 		if (raidPtr->Disks[column].status ==
1187 		    rf_ds_reconstructing) {
1188 			printf("raid%d: Unable to reconstruct to disk at:\n",
1189 			       raidPtr->raidid);
1190 			printf("raid%d:    Col: %d   Reconstruction already occuring!\n", raidPtr->raidid, column);
1191 
1192 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1193 			return (EINVAL);
1194 		}
1195 		if (raidPtr->Disks[column].status == rf_ds_spared) {
1196 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1197 			return (EINVAL);
1198 		}
1199 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1200 
1201 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1202 		if (rrcopy == NULL)
1203 			return(ENOMEM);
1204 
1205 		rrcopy->raidPtr = (void *) raidPtr;
1206 		rrcopy->col = column;
1207 
1208 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1209 					   rf_ReconstructInPlaceThread,
1210 					   rrcopy,"raid_reconip");
1211 		return(retcode);
1212 
1213 	case RAIDFRAME_GET_INFO:
1214 		if (!raidPtr->valid)
1215 			return (ENODEV);
1216 		ucfgp = (RF_DeviceConfig_t **) data;
1217 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1218 			  (RF_DeviceConfig_t *));
1219 		if (d_cfg == NULL)
1220 			return (ENOMEM);
1221 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1222 		d_cfg->rows = 1; /* there is only 1 row now */
1223 		d_cfg->cols = raidPtr->numCol;
1224 		d_cfg->ndevs = raidPtr->numCol;
1225 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1226 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1227 			return (ENOMEM);
1228 		}
1229 		d_cfg->nspares = raidPtr->numSpare;
1230 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1231 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1232 			return (ENOMEM);
1233 		}
1234 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1235 		d = 0;
1236 		for (j = 0; j < d_cfg->cols; j++) {
1237 			d_cfg->devs[d] = raidPtr->Disks[j];
1238 			d++;
1239 		}
1240 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1241 			d_cfg->spares[i] = raidPtr->Disks[j];
1242 		}
1243 		retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t));
1244 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1245 
1246 		return (retcode);
1247 
1248 	case RAIDFRAME_CHECK_PARITY:
1249 		*(int *) data = raidPtr->parity_good;
1250 		return (0);
1251 
1252 	case RAIDFRAME_RESET_ACCTOTALS:
1253 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1254 		return (0);
1255 
1256 	case RAIDFRAME_GET_ACCTOTALS:
1257 		totals = (RF_AccTotals_t *) data;
1258 		*totals = raidPtr->acc_totals;
1259 		return (0);
1260 
1261 	case RAIDFRAME_KEEP_ACCTOTALS:
1262 		raidPtr->keep_acc_totals = *(int *)data;
1263 		return (0);
1264 
1265 	case RAIDFRAME_GET_SIZE:
1266 		*(int *) data = raidPtr->totalSectors;
1267 		return (0);
1268 
1269 		/* fail a disk & optionally start reconstruction */
1270 	case RAIDFRAME_FAIL_DISK:
1271 
1272 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1273 			/* Can't do this on a RAID 0!! */
1274 			return(EINVAL);
1275 		}
1276 
1277 		rr = (struct rf_recon_req *) data;
1278 		rr->row = 0;
1279 		if (rr->col < 0 || rr->col >= raidPtr->numCol)
1280 			return (EINVAL);
1281 
1282 
1283 		RF_LOCK_MUTEX(raidPtr->mutex);
1284 		if (raidPtr->status == rf_rs_reconstructing) {
1285 			/* you can't fail a disk while we're reconstructing! */
1286 			/* XXX wrong for RAID6 */
1287 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1288 			return (EINVAL);
1289 		}
1290 		if ((raidPtr->Disks[rr->col].status ==
1291 		     rf_ds_optimal) && (raidPtr->numFailures > 0)) {
1292 			/* some other component has failed.  Let's not make
1293 			   things worse. XXX wrong for RAID6 */
1294 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1295 			return (EINVAL);
1296 		}
1297 		if (raidPtr->Disks[rr->col].status == rf_ds_spared) {
1298 			/* Can't fail a spared disk! */
1299 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1300 			return (EINVAL);
1301 		}
1302 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1303 
1304 		/* make a copy of the recon request so that we don't rely on
1305 		 * the user's buffer */
1306 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1307 		if (rrcopy == NULL)
1308 			return(ENOMEM);
1309 		memcpy(rrcopy, rr, sizeof(*rr));
1310 		rrcopy->raidPtr = (void *) raidPtr;
1311 
1312 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1313 					   rf_ReconThread,
1314 					   rrcopy,"raid_recon");
1315 		return (0);
1316 
1317 		/* invoke a copyback operation after recon on whatever disk
1318 		 * needs it, if any */
1319 	case RAIDFRAME_COPYBACK:
1320 
1321 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1322 			/* This makes no sense on a RAID 0!! */
1323 			return(EINVAL);
1324 		}
1325 
1326 		if (raidPtr->copyback_in_progress == 1) {
1327 			/* Copyback is already in progress! */
1328 			return(EINVAL);
1329 		}
1330 
1331 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1332 					   rf_CopybackThread,
1333 					   raidPtr,"raid_copyback");
1334 		return (retcode);
1335 
1336 		/* return the percentage completion of reconstruction */
1337 	case RAIDFRAME_CHECK_RECON_STATUS:
1338 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1339 			/* This makes no sense on a RAID 0, so tell the
1340 			   user it's done. */
1341 			*(int *) data = 100;
1342 			return(0);
1343 		}
1344 		if (raidPtr->status != rf_rs_reconstructing)
1345 			*(int *) data = 100;
1346 		else {
1347 			if (raidPtr->reconControl->numRUsTotal > 0) {
1348 				*(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal);
1349 			} else {
1350 				*(int *) data = 0;
1351 			}
1352 		}
1353 		return (0);
1354 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1355 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1356 		if (raidPtr->status != rf_rs_reconstructing) {
1357 			progressInfo.remaining = 0;
1358 			progressInfo.completed = 100;
1359 			progressInfo.total = 100;
1360 		} else {
1361 			progressInfo.total =
1362 				raidPtr->reconControl->numRUsTotal;
1363 			progressInfo.completed =
1364 				raidPtr->reconControl->numRUsComplete;
1365 			progressInfo.remaining = progressInfo.total -
1366 				progressInfo.completed;
1367 		}
1368 		retcode = copyout(&progressInfo, *progressInfoPtr,
1369 				  sizeof(RF_ProgressInfo_t));
1370 		return (retcode);
1371 
1372 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1373 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1374 			/* This makes no sense on a RAID 0, so tell the
1375 			   user it's done. */
1376 			*(int *) data = 100;
1377 			return(0);
1378 		}
1379 		if (raidPtr->parity_rewrite_in_progress == 1) {
1380 			*(int *) data = 100 *
1381 				raidPtr->parity_rewrite_stripes_done /
1382 				raidPtr->Layout.numStripe;
1383 		} else {
1384 			*(int *) data = 100;
1385 		}
1386 		return (0);
1387 
1388 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1389 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1390 		if (raidPtr->parity_rewrite_in_progress == 1) {
1391 			progressInfo.total = raidPtr->Layout.numStripe;
1392 			progressInfo.completed =
1393 				raidPtr->parity_rewrite_stripes_done;
1394 			progressInfo.remaining = progressInfo.total -
1395 				progressInfo.completed;
1396 		} else {
1397 			progressInfo.remaining = 0;
1398 			progressInfo.completed = 100;
1399 			progressInfo.total = 100;
1400 		}
1401 		retcode = copyout(&progressInfo, *progressInfoPtr,
1402 				  sizeof(RF_ProgressInfo_t));
1403 		return (retcode);
1404 
1405 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1406 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1407 			/* This makes no sense on a RAID 0 */
1408 			*(int *) data = 100;
1409 			return(0);
1410 		}
1411 		if (raidPtr->copyback_in_progress == 1) {
1412 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1413 				raidPtr->Layout.numStripe;
1414 		} else {
1415 			*(int *) data = 100;
1416 		}
1417 		return (0);
1418 
1419 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1420 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1421 		if (raidPtr->copyback_in_progress == 1) {
1422 			progressInfo.total = raidPtr->Layout.numStripe;
1423 			progressInfo.completed =
1424 				raidPtr->copyback_stripes_done;
1425 			progressInfo.remaining = progressInfo.total -
1426 				progressInfo.completed;
1427 		} else {
1428 			progressInfo.remaining = 0;
1429 			progressInfo.completed = 100;
1430 			progressInfo.total = 100;
1431 		}
1432 		retcode = copyout(&progressInfo, *progressInfoPtr,
1433 				  sizeof(RF_ProgressInfo_t));
1434 		return (retcode);
1435 
1436 		/* the sparetable daemon calls this to wait for the kernel to
1437 		 * need a spare table. this ioctl does not return until a
1438 		 * spare table is needed. XXX -- calling mpsleep here in the
1439 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1440 		 * -- I should either compute the spare table in the kernel,
1441 		 * or have a different -- XXX XXX -- interface (a different
1442 		 * character device) for delivering the table     -- XXX */
1443 #if 0
1444 	case RAIDFRAME_SPARET_WAIT:
1445 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1446 		while (!rf_sparet_wait_queue)
1447 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1448 		waitreq = rf_sparet_wait_queue;
1449 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1450 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1451 
1452 		/* structure assignment */
1453 		*((RF_SparetWait_t *) data) = *waitreq;
1454 
1455 		RF_Free(waitreq, sizeof(*waitreq));
1456 		return (0);
1457 
1458 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1459 		 * code in it that will cause the dameon to exit */
1460 	case RAIDFRAME_ABORT_SPARET_WAIT:
1461 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1462 		waitreq->fcol = -1;
1463 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1464 		waitreq->next = rf_sparet_wait_queue;
1465 		rf_sparet_wait_queue = waitreq;
1466 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1467 		wakeup(&rf_sparet_wait_queue);
1468 		return (0);
1469 
1470 		/* used by the spare table daemon to deliver a spare table
1471 		 * into the kernel */
1472 	case RAIDFRAME_SEND_SPARET:
1473 
1474 		/* install the spare table */
1475 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1476 
1477 		/* respond to the requestor.  the return status of the spare
1478 		 * table installation is passed in the "fcol" field */
1479 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1480 		waitreq->fcol = retcode;
1481 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1482 		waitreq->next = rf_sparet_resp_queue;
1483 		rf_sparet_resp_queue = waitreq;
1484 		wakeup(&rf_sparet_resp_queue);
1485 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1486 
1487 		return (retcode);
1488 #endif
1489 
1490 	default:
1491 		break; /* fall through to the os-specific code below */
1492 
1493 	}
1494 
1495 	if (!raidPtr->valid)
1496 		return (EINVAL);
1497 
1498 	/*
1499 	 * Add support for "regular" device ioctls here.
1500 	 */
1501 
1502 	switch (cmd) {
1503 	case DIOCGDINFO:
1504 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1505 		break;
1506 #ifdef __HAVE_OLD_DISKLABEL
1507 	case ODIOCGDINFO:
1508 		newlabel = *(rs->sc_dkdev.dk_label);
1509 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1510 			return ENOTTY;
1511 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1512 		break;
1513 #endif
1514 
1515 	case DIOCGPART:
1516 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1517 		((struct partinfo *) data)->part =
1518 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1519 		break;
1520 
1521 	case DIOCWDINFO:
1522 	case DIOCSDINFO:
1523 #ifdef __HAVE_OLD_DISKLABEL
1524 	case ODIOCWDINFO:
1525 	case ODIOCSDINFO:
1526 #endif
1527 	{
1528 		struct disklabel *lp;
1529 #ifdef __HAVE_OLD_DISKLABEL
1530 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1531 			memset(&newlabel, 0, sizeof newlabel);
1532 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1533 			lp = &newlabel;
1534 		} else
1535 #endif
1536 		lp = (struct disklabel *)data;
1537 
1538 		if ((error = raidlock(rs)) != 0)
1539 			return (error);
1540 
1541 		rs->sc_flags |= RAIDF_LABELLING;
1542 
1543 		error = setdisklabel(rs->sc_dkdev.dk_label,
1544 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1545 		if (error == 0) {
1546 			if (cmd == DIOCWDINFO
1547 #ifdef __HAVE_OLD_DISKLABEL
1548 			    || cmd == ODIOCWDINFO
1549 #endif
1550 			   )
1551 				error = writedisklabel(RAIDLABELDEV(dev),
1552 				    raidstrategy, rs->sc_dkdev.dk_label,
1553 				    rs->sc_dkdev.dk_cpulabel);
1554 		}
1555 		rs->sc_flags &= ~RAIDF_LABELLING;
1556 
1557 		raidunlock(rs);
1558 
1559 		if (error)
1560 			return (error);
1561 		break;
1562 	}
1563 
1564 	case DIOCWLABEL:
1565 		if (*(int *) data != 0)
1566 			rs->sc_flags |= RAIDF_WLABEL;
1567 		else
1568 			rs->sc_flags &= ~RAIDF_WLABEL;
1569 		break;
1570 
1571 	case DIOCGDEFLABEL:
1572 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1573 		break;
1574 
1575 #ifdef __HAVE_OLD_DISKLABEL
1576 	case ODIOCGDEFLABEL:
1577 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1578 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1579 			return ENOTTY;
1580 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1581 		break;
1582 #endif
1583 
1584 	default:
1585 		retcode = ENOTTY;
1586 	}
1587 	return (retcode);
1588 
1589 }
1590 
1591 
1592 /* raidinit -- complete the rest of the initialization for the
1593    RAIDframe device.  */
1594 
1595 
1596 static void
1597 raidinit(RF_Raid_t *raidPtr)
1598 {
1599 	struct raid_softc *rs;
1600 	int     unit;
1601 
1602 	unit = raidPtr->raidid;
1603 
1604 	rs = &raid_softc[unit];
1605 
1606 	/* XXX should check return code first... */
1607 	rs->sc_flags |= RAIDF_INITED;
1608 
1609 	/* XXX doesn't check bounds. */
1610 	snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit);
1611 
1612 	rs->sc_dkdev.dk_name = rs->sc_xname;
1613 
1614 	/* disk_attach actually creates space for the CPU disklabel, among
1615 	 * other things, so it's critical to call this *BEFORE* we try putzing
1616 	 * with disklabels. */
1617 
1618 	pseudo_disk_attach(&rs->sc_dkdev);
1619 
1620 	/* XXX There may be a weird interaction here between this, and
1621 	 * protectedSectors, as used in RAIDframe.  */
1622 
1623 	rs->sc_size = raidPtr->totalSectors;
1624 }
1625 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0)
1626 /* wake up the daemon & tell it to get us a spare table
1627  * XXX
1628  * the entries in the queues should be tagged with the raidPtr
1629  * so that in the extremely rare case that two recons happen at once,
1630  * we know for which device were requesting a spare table
1631  * XXX
1632  *
1633  * XXX This code is not currently used. GO
1634  */
1635 int
1636 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req)
1637 {
1638 	int     retcode;
1639 
1640 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1641 	req->next = rf_sparet_wait_queue;
1642 	rf_sparet_wait_queue = req;
1643 	wakeup(&rf_sparet_wait_queue);
1644 
1645 	/* mpsleep unlocks the mutex */
1646 	while (!rf_sparet_resp_queue) {
1647 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1648 		    "raidframe getsparetable", 0);
1649 	}
1650 	req = rf_sparet_resp_queue;
1651 	rf_sparet_resp_queue = req->next;
1652 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1653 
1654 	retcode = req->fcol;
1655 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1656 					 * alloc'd */
1657 	return (retcode);
1658 }
1659 #endif
1660 
1661 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1662  * bp & passes it down.
1663  * any calls originating in the kernel must use non-blocking I/O
1664  * do some extra sanity checking to return "appropriate" error values for
1665  * certain conditions (to make some standard utilities work)
1666  *
1667  * Formerly known as: rf_DoAccessKernel
1668  */
1669 void
1670 raidstart(RF_Raid_t *raidPtr)
1671 {
1672 	RF_SectorCount_t num_blocks, pb, sum;
1673 	RF_RaidAddr_t raid_addr;
1674 	struct partition *pp;
1675 	daddr_t blocknum;
1676 	int     unit;
1677 	struct raid_softc *rs;
1678 	int     do_async;
1679 	struct buf *bp;
1680 	int rc;
1681 
1682 	unit = raidPtr->raidid;
1683 	rs = &raid_softc[unit];
1684 
1685 	/* quick check to see if anything has died recently */
1686 	RF_LOCK_MUTEX(raidPtr->mutex);
1687 	if (raidPtr->numNewFailures > 0) {
1688 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1689 		rf_update_component_labels(raidPtr,
1690 					   RF_NORMAL_COMPONENT_UPDATE);
1691 		RF_LOCK_MUTEX(raidPtr->mutex);
1692 		raidPtr->numNewFailures--;
1693 	}
1694 
1695 	/* Check to see if we're at the limit... */
1696 	while (raidPtr->openings > 0) {
1697 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1698 
1699 		/* get the next item, if any, from the queue */
1700 		if ((bp = BUFQ_GET(rs->buf_queue)) == NULL) {
1701 			/* nothing more to do */
1702 			return;
1703 		}
1704 
1705 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1706 		 * partition.. Need to make it absolute to the underlying
1707 		 * device.. */
1708 
1709 		blocknum = bp->b_blkno;
1710 		if (DISKPART(bp->b_dev) != RAW_PART) {
1711 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1712 			blocknum += pp->p_offset;
1713 		}
1714 
1715 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1716 			    (int) blocknum));
1717 
1718 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1719 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1720 
1721 		/* *THIS* is where we adjust what block we're going to...
1722 		 * but DO NOT TOUCH bp->b_blkno!!! */
1723 		raid_addr = blocknum;
1724 
1725 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1726 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1727 		sum = raid_addr + num_blocks + pb;
1728 		if (1 || rf_debugKernelAccess) {
1729 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1730 				    (int) raid_addr, (int) sum, (int) num_blocks,
1731 				    (int) pb, (int) bp->b_resid));
1732 		}
1733 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1734 		    || (sum < num_blocks) || (sum < pb)) {
1735 			bp->b_error = ENOSPC;
1736 			bp->b_flags |= B_ERROR;
1737 			bp->b_resid = bp->b_bcount;
1738 			biodone(bp);
1739 			RF_LOCK_MUTEX(raidPtr->mutex);
1740 			continue;
1741 		}
1742 		/*
1743 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1744 		 */
1745 
1746 		if (bp->b_bcount & raidPtr->sectorMask) {
1747 			bp->b_error = EINVAL;
1748 			bp->b_flags |= B_ERROR;
1749 			bp->b_resid = bp->b_bcount;
1750 			biodone(bp);
1751 			RF_LOCK_MUTEX(raidPtr->mutex);
1752 			continue;
1753 
1754 		}
1755 		db1_printf(("Calling DoAccess..\n"));
1756 
1757 
1758 		RF_LOCK_MUTEX(raidPtr->mutex);
1759 		raidPtr->openings--;
1760 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1761 
1762 		/*
1763 		 * Everything is async.
1764 		 */
1765 		do_async = 1;
1766 
1767 		disk_busy(&rs->sc_dkdev);
1768 
1769 		/* XXX we're still at splbio() here... do we *really*
1770 		   need to be? */
1771 
1772 		/* don't ever condition on bp->b_flags & B_WRITE.
1773 		 * always condition on B_READ instead */
1774 
1775 		rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1776 				 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1777 				 do_async, raid_addr, num_blocks,
1778 				 bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1779 
1780 		if (rc) {
1781 			bp->b_error = rc;
1782 			bp->b_flags |= B_ERROR;
1783 			bp->b_resid = bp->b_bcount;
1784 			biodone(bp);
1785 			/* continue loop */
1786 		}
1787 
1788 		RF_LOCK_MUTEX(raidPtr->mutex);
1789 	}
1790 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1791 }
1792 
1793 
1794 
1795 
1796 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1797 
1798 int
1799 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req)
1800 {
1801 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1802 	struct buf *bp;
1803 
1804 	req->queue = queue;
1805 
1806 #if DIAGNOSTIC
1807 	if (queue->raidPtr->raidid >= numraid) {
1808 		printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid,
1809 		    numraid);
1810 		panic("Invalid Unit number in rf_DispatchKernelIO");
1811 	}
1812 #endif
1813 
1814 	bp = req->bp;
1815 
1816 	switch (req->type) {
1817 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1818 		/* XXX need to do something extra here.. */
1819 		/* I'm leaving this in, as I've never actually seen it used,
1820 		 * and I'd like folks to report it... GO */
1821 		printf(("WAKEUP CALLED\n"));
1822 		queue->numOutstanding++;
1823 
1824 		bp->b_flags = 0;
1825 		bp->b_fspriv.bf_private = req;
1826 
1827 		KernelWakeupFunc(bp);
1828 		break;
1829 
1830 	case RF_IO_TYPE_READ:
1831 	case RF_IO_TYPE_WRITE:
1832 #if RF_ACC_TRACE > 0
1833 		if (req->tracerec) {
1834 			RF_ETIMER_START(req->tracerec->timer);
1835 		}
1836 #endif
1837 		InitBP(bp, queue->rf_cinfo->ci_vp,
1838 		    op, queue->rf_cinfo->ci_dev,
1839 		    req->sectorOffset, req->numSector,
1840 		    req->buf, KernelWakeupFunc, (void *) req,
1841 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1842 
1843 		if (rf_debugKernelAccess) {
1844 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1845 				(long) bp->b_blkno));
1846 		}
1847 		queue->numOutstanding++;
1848 		queue->last_deq_sector = req->sectorOffset;
1849 		/* acc wouldn't have been let in if there were any pending
1850 		 * reqs at any other priority */
1851 		queue->curPriority = req->priority;
1852 
1853 		db1_printf(("Going for %c to unit %d col %d\n",
1854 			    req->type, queue->raidPtr->raidid,
1855 			    queue->col));
1856 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1857 			(int) req->sectorOffset, (int) req->numSector,
1858 			(int) (req->numSector <<
1859 			    queue->raidPtr->logBytesPerSector),
1860 			(int) queue->raidPtr->logBytesPerSector));
1861 		VOP_STRATEGY(bp->b_vp, bp);
1862 
1863 		break;
1864 
1865 	default:
1866 		panic("bad req->type in rf_DispatchKernelIO");
1867 	}
1868 	db1_printf(("Exiting from DispatchKernelIO\n"));
1869 
1870 	return (0);
1871 }
1872 /* this is the callback function associated with a I/O invoked from
1873    kernel code.
1874  */
1875 static void
1876 KernelWakeupFunc(struct buf *bp)
1877 {
1878 	RF_DiskQueueData_t *req = NULL;
1879 	RF_DiskQueue_t *queue;
1880 	int s;
1881 
1882 	s = splbio();
1883 	db1_printf(("recovering the request queue:\n"));
1884 	req = bp->b_fspriv.bf_private;
1885 
1886 	queue = (RF_DiskQueue_t *) req->queue;
1887 
1888 #if RF_ACC_TRACE > 0
1889 	if (req->tracerec) {
1890 		RF_ETIMER_STOP(req->tracerec->timer);
1891 		RF_ETIMER_EVAL(req->tracerec->timer);
1892 		RF_LOCK_MUTEX(rf_tracing_mutex);
1893 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1894 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1895 		req->tracerec->num_phys_ios++;
1896 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1897 	}
1898 #endif
1899 
1900 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1901 	 * ballistic, and mark the component as hosed... */
1902 
1903 	if (bp->b_flags & B_ERROR) {
1904 		/* Mark the disk as dead */
1905 		/* but only mark it once... */
1906 		/* and only if it wouldn't leave this RAID set
1907 		   completely broken */
1908 		if (((queue->raidPtr->Disks[queue->col].status ==
1909 		      rf_ds_optimal) ||
1910 		     (queue->raidPtr->Disks[queue->col].status ==
1911 		      rf_ds_used_spare)) &&
1912 		     (queue->raidPtr->numFailures <
1913 		         queue->raidPtr->Layout.map->faultsTolerated)) {
1914 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1915 			       queue->raidPtr->raidid,
1916 			       queue->raidPtr->Disks[queue->col].devname);
1917 			queue->raidPtr->Disks[queue->col].status =
1918 			    rf_ds_failed;
1919 			queue->raidPtr->status = rf_rs_degraded;
1920 			queue->raidPtr->numFailures++;
1921 			queue->raidPtr->numNewFailures++;
1922 		} else {	/* Disk is already dead... */
1923 			/* printf("Disk already marked as dead!\n"); */
1924 		}
1925 
1926 	}
1927 
1928 	/* Fill in the error value */
1929 
1930 	req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0;
1931 
1932 	simple_lock(&queue->raidPtr->iodone_lock);
1933 
1934 	/* Drop this one on the "finished" queue... */
1935 	TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries);
1936 
1937 	/* Let the raidio thread know there is work to be done. */
1938 	wakeup(&(queue->raidPtr->iodone));
1939 
1940 	simple_unlock(&queue->raidPtr->iodone_lock);
1941 
1942 	splx(s);
1943 }
1944 
1945 
1946 
1947 /*
1948  * initialize a buf structure for doing an I/O in the kernel.
1949  */
1950 static void
1951 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev,
1952        RF_SectorNum_t startSect, RF_SectorCount_t numSect, caddr_t bf,
1953        void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector,
1954        struct proc *b_proc)
1955 {
1956 	/* bp->b_flags       = B_PHYS | rw_flag; */
1957 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
1958 	bp->b_bcount = numSect << logBytesPerSector;
1959 	bp->b_bufsize = bp->b_bcount;
1960 	bp->b_error = 0;
1961 	bp->b_dev = dev;
1962 	bp->b_data = bf;
1963 	bp->b_blkno = startSect;
1964 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
1965 	if (bp->b_bcount == 0) {
1966 		panic("bp->b_bcount is zero in InitBP!!");
1967 	}
1968 	bp->b_proc = b_proc;
1969 	bp->b_iodone = cbFunc;
1970 	bp->b_fspriv.bf_private = cbArg;
1971 	bp->b_vp = b_vp;
1972 	if ((bp->b_flags & B_READ) == 0) {
1973 		bp->b_vp->v_numoutput++;
1974 	}
1975 
1976 }
1977 
1978 static void
1979 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs,
1980 		    struct disklabel *lp)
1981 {
1982 	memset(lp, 0, sizeof(*lp));
1983 
1984 	/* fabricate a label... */
1985 	lp->d_secperunit = raidPtr->totalSectors;
1986 	lp->d_secsize = raidPtr->bytesPerSector;
1987 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
1988 	lp->d_ntracks = 4 * raidPtr->numCol;
1989 	lp->d_ncylinders = raidPtr->totalSectors /
1990 		(lp->d_nsectors * lp->d_ntracks);
1991 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
1992 
1993 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
1994 	lp->d_type = DTYPE_RAID;
1995 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
1996 	lp->d_rpm = 3600;
1997 	lp->d_interleave = 1;
1998 	lp->d_flags = 0;
1999 
2000 	lp->d_partitions[RAW_PART].p_offset = 0;
2001 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2002 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2003 	lp->d_npartitions = RAW_PART + 1;
2004 
2005 	lp->d_magic = DISKMAGIC;
2006 	lp->d_magic2 = DISKMAGIC;
2007 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2008 
2009 }
2010 /*
2011  * Read the disklabel from the raid device.  If one is not present, fake one
2012  * up.
2013  */
2014 static void
2015 raidgetdisklabel(dev_t dev)
2016 {
2017 	int     unit = raidunit(dev);
2018 	struct raid_softc *rs = &raid_softc[unit];
2019 	const char   *errstring;
2020 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2021 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2022 	RF_Raid_t *raidPtr;
2023 
2024 	db1_printf(("Getting the disklabel...\n"));
2025 
2026 	memset(clp, 0, sizeof(*clp));
2027 
2028 	raidPtr = raidPtrs[unit];
2029 
2030 	raidgetdefaultlabel(raidPtr, rs, lp);
2031 
2032 	/*
2033 	 * Call the generic disklabel extraction routine.
2034 	 */
2035 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2036 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2037 	if (errstring)
2038 		raidmakedisklabel(rs);
2039 	else {
2040 		int     i;
2041 		struct partition *pp;
2042 
2043 		/*
2044 		 * Sanity check whether the found disklabel is valid.
2045 		 *
2046 		 * This is necessary since total size of the raid device
2047 		 * may vary when an interleave is changed even though exactly
2048 		 * same componets are used, and old disklabel may used
2049 		 * if that is found.
2050 		 */
2051 		if (lp->d_secperunit != rs->sc_size)
2052 			printf("raid%d: WARNING: %s: "
2053 			    "total sector size in disklabel (%d) != "
2054 			    "the size of raid (%ld)\n", unit, rs->sc_xname,
2055 			    lp->d_secperunit, (long) rs->sc_size);
2056 		for (i = 0; i < lp->d_npartitions; i++) {
2057 			pp = &lp->d_partitions[i];
2058 			if (pp->p_offset + pp->p_size > rs->sc_size)
2059 				printf("raid%d: WARNING: %s: end of partition `%c' "
2060 				       "exceeds the size of raid (%ld)\n",
2061 				       unit, rs->sc_xname, 'a' + i, (long) rs->sc_size);
2062 		}
2063 	}
2064 
2065 }
2066 /*
2067  * Take care of things one might want to take care of in the event
2068  * that a disklabel isn't present.
2069  */
2070 static void
2071 raidmakedisklabel(struct raid_softc *rs)
2072 {
2073 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2074 	db1_printf(("Making a label..\n"));
2075 
2076 	/*
2077 	 * For historical reasons, if there's no disklabel present
2078 	 * the raw partition must be marked FS_BSDFFS.
2079 	 */
2080 
2081 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2082 
2083 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2084 
2085 	lp->d_checksum = dkcksum(lp);
2086 }
2087 /*
2088  * Lookup the provided name in the filesystem.  If the file exists,
2089  * is a valid block device, and isn't being used by anyone else,
2090  * set *vpp to the file's vnode.
2091  * You'll find the original of this in ccd.c
2092  */
2093 int
2094 raidlookup(char *path, struct lwp *l, struct vnode **vpp)
2095 {
2096 	struct nameidata nd;
2097 	struct vnode *vp;
2098 	struct proc *p;
2099 	struct vattr va;
2100 	int     error;
2101 
2102 	p = l ? l->l_proc : NULL;
2103 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, l);
2104 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2105 		return (error);
2106 	}
2107 	vp = nd.ni_vp;
2108 	if (vp->v_usecount > 1) {
2109 		VOP_UNLOCK(vp, 0);
2110 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2111 		return (EBUSY);
2112 	}
2113 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, l)) != 0) {
2114 		VOP_UNLOCK(vp, 0);
2115 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2116 		return (error);
2117 	}
2118 	/* XXX: eventually we should handle VREG, too. */
2119 	if (va.va_type != VBLK) {
2120 		VOP_UNLOCK(vp, 0);
2121 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2122 		return (ENOTBLK);
2123 	}
2124 	VOP_UNLOCK(vp, 0);
2125 	*vpp = vp;
2126 	return (0);
2127 }
2128 /*
2129  * Wait interruptibly for an exclusive lock.
2130  *
2131  * XXX
2132  * Several drivers do this; it should be abstracted and made MP-safe.
2133  * (Hmm... where have we seen this warning before :->  GO )
2134  */
2135 static int
2136 raidlock(struct raid_softc *rs)
2137 {
2138 	int     error;
2139 
2140 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2141 		rs->sc_flags |= RAIDF_WANTED;
2142 		if ((error =
2143 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2144 			return (error);
2145 	}
2146 	rs->sc_flags |= RAIDF_LOCKED;
2147 	return (0);
2148 }
2149 /*
2150  * Unlock and wake up any waiters.
2151  */
2152 static void
2153 raidunlock(struct raid_softc *rs)
2154 {
2155 
2156 	rs->sc_flags &= ~RAIDF_LOCKED;
2157 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2158 		rs->sc_flags &= ~RAIDF_WANTED;
2159 		wakeup(rs);
2160 	}
2161 }
2162 
2163 
2164 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2165 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2166 
2167 int
2168 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2169 {
2170 	RF_ComponentLabel_t clabel;
2171 	raidread_component_label(dev, b_vp, &clabel);
2172 	clabel.mod_counter = mod_counter;
2173 	clabel.clean = RF_RAID_CLEAN;
2174 	raidwrite_component_label(dev, b_vp, &clabel);
2175 	return(0);
2176 }
2177 
2178 
2179 int
2180 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2181 {
2182 	RF_ComponentLabel_t clabel;
2183 	raidread_component_label(dev, b_vp, &clabel);
2184 	clabel.mod_counter = mod_counter;
2185 	clabel.clean = RF_RAID_DIRTY;
2186 	raidwrite_component_label(dev, b_vp, &clabel);
2187 	return(0);
2188 }
2189 
2190 /* ARGSUSED */
2191 int
2192 raidread_component_label(dev_t dev, struct vnode *b_vp,
2193 			 RF_ComponentLabel_t *clabel)
2194 {
2195 	struct buf *bp;
2196 	const struct bdevsw *bdev;
2197 	int error;
2198 
2199 	/* XXX should probably ensure that we don't try to do this if
2200 	   someone has changed rf_protected_sectors. */
2201 
2202 	if (b_vp == NULL) {
2203 		/* For whatever reason, this component is not valid.
2204 		   Don't try to read a component label from it. */
2205 		return(EINVAL);
2206 	}
2207 
2208 	/* get a block of the appropriate size... */
2209 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2210 	bp->b_dev = dev;
2211 
2212 	/* get our ducks in a row for the read */
2213 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2214 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2215 	bp->b_flags |= B_READ;
2216  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2217 
2218 	bdev = bdevsw_lookup(bp->b_dev);
2219 	if (bdev == NULL)
2220 		return (ENXIO);
2221 	(*bdev->d_strategy)(bp);
2222 
2223 	error = biowait(bp);
2224 
2225 	if (!error) {
2226 		memcpy(clabel, bp->b_data,
2227 		       sizeof(RF_ComponentLabel_t));
2228         }
2229 
2230 	brelse(bp);
2231 	return(error);
2232 }
2233 /* ARGSUSED */
2234 int
2235 raidwrite_component_label(dev_t dev, struct vnode *b_vp,
2236 			  RF_ComponentLabel_t *clabel)
2237 {
2238 	struct buf *bp;
2239 	const struct bdevsw *bdev;
2240 	int error;
2241 
2242 	/* get a block of the appropriate size... */
2243 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2244 	bp->b_dev = dev;
2245 
2246 	/* get our ducks in a row for the write */
2247 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2248 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2249 	bp->b_flags |= B_WRITE;
2250  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2251 
2252 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2253 
2254 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2255 
2256 	bdev = bdevsw_lookup(bp->b_dev);
2257 	if (bdev == NULL)
2258 		return (ENXIO);
2259 	(*bdev->d_strategy)(bp);
2260 	error = biowait(bp);
2261 	brelse(bp);
2262 	if (error) {
2263 #if 1
2264 		printf("Failed to write RAID component info!\n");
2265 #endif
2266 	}
2267 
2268 	return(error);
2269 }
2270 
2271 void
2272 rf_markalldirty(RF_Raid_t *raidPtr)
2273 {
2274 	RF_ComponentLabel_t clabel;
2275 	int sparecol;
2276 	int c;
2277 	int j;
2278 	int scol = -1;
2279 
2280 	raidPtr->mod_counter++;
2281 	for (c = 0; c < raidPtr->numCol; c++) {
2282 		/* we don't want to touch (at all) a disk that has
2283 		   failed */
2284 		if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) {
2285 			raidread_component_label(
2286 						 raidPtr->Disks[c].dev,
2287 						 raidPtr->raid_cinfo[c].ci_vp,
2288 						 &clabel);
2289 			if (clabel.status == rf_ds_spared) {
2290 				/* XXX do something special...
2291 				   but whatever you do, don't
2292 				   try to access it!! */
2293 			} else {
2294 				raidmarkdirty(
2295 					      raidPtr->Disks[c].dev,
2296 					      raidPtr->raid_cinfo[c].ci_vp,
2297 					      raidPtr->mod_counter);
2298 			}
2299 		}
2300 	}
2301 
2302 	for( c = 0; c < raidPtr->numSpare ; c++) {
2303 		sparecol = raidPtr->numCol + c;
2304 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2305 			/*
2306 
2307 			   we claim this disk is "optimal" if it's
2308 			   rf_ds_used_spare, as that means it should be
2309 			   directly substitutable for the disk it replaced.
2310 			   We note that too...
2311 
2312 			 */
2313 
2314 			for(j=0;j<raidPtr->numCol;j++) {
2315 				if (raidPtr->Disks[j].spareCol == sparecol) {
2316 					scol = j;
2317 					break;
2318 				}
2319 			}
2320 
2321 			raidread_component_label(
2322 				 raidPtr->Disks[sparecol].dev,
2323 				 raidPtr->raid_cinfo[sparecol].ci_vp,
2324 				 &clabel);
2325 			/* make sure status is noted */
2326 
2327 			raid_init_component_label(raidPtr, &clabel);
2328 
2329 			clabel.row = 0;
2330 			clabel.column = scol;
2331 			/* Note: we *don't* change status from rf_ds_used_spare
2332 			   to rf_ds_optimal */
2333 			/* clabel.status = rf_ds_optimal; */
2334 
2335 			raidmarkdirty(raidPtr->Disks[sparecol].dev,
2336 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2337 				      raidPtr->mod_counter);
2338 		}
2339 	}
2340 }
2341 
2342 
2343 void
2344 rf_update_component_labels(RF_Raid_t *raidPtr, int final)
2345 {
2346 	RF_ComponentLabel_t clabel;
2347 	int sparecol;
2348 	int c;
2349 	int j;
2350 	int scol;
2351 
2352 	scol = -1;
2353 
2354 	/* XXX should do extra checks to make sure things really are clean,
2355 	   rather than blindly setting the clean bit... */
2356 
2357 	raidPtr->mod_counter++;
2358 
2359 	for (c = 0; c < raidPtr->numCol; c++) {
2360 		if (raidPtr->Disks[c].status == rf_ds_optimal) {
2361 			raidread_component_label(
2362 						 raidPtr->Disks[c].dev,
2363 						 raidPtr->raid_cinfo[c].ci_vp,
2364 						 &clabel);
2365 				/* make sure status is noted */
2366 			clabel.status = rf_ds_optimal;
2367 				/* bump the counter */
2368 			clabel.mod_counter = raidPtr->mod_counter;
2369 
2370 			raidwrite_component_label(
2371 						  raidPtr->Disks[c].dev,
2372 						  raidPtr->raid_cinfo[c].ci_vp,
2373 						  &clabel);
2374 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2375 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2376 					raidmarkclean(
2377 						      raidPtr->Disks[c].dev,
2378 						      raidPtr->raid_cinfo[c].ci_vp,
2379 						      raidPtr->mod_counter);
2380 				}
2381 			}
2382 		}
2383 		/* else we don't touch it.. */
2384 	}
2385 
2386 	for( c = 0; c < raidPtr->numSpare ; c++) {
2387 		sparecol = raidPtr->numCol + c;
2388 		/* Need to ensure that the reconstruct actually completed! */
2389 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2390 			/*
2391 
2392 			   we claim this disk is "optimal" if it's
2393 			   rf_ds_used_spare, as that means it should be
2394 			   directly substitutable for the disk it replaced.
2395 			   We note that too...
2396 
2397 			 */
2398 
2399 			for(j=0;j<raidPtr->numCol;j++) {
2400 				if (raidPtr->Disks[j].spareCol == sparecol) {
2401 					scol = j;
2402 					break;
2403 				}
2404 			}
2405 
2406 			/* XXX shouldn't *really* need this... */
2407 			raidread_component_label(
2408 				      raidPtr->Disks[sparecol].dev,
2409 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2410 				      &clabel);
2411 			/* make sure status is noted */
2412 
2413 			raid_init_component_label(raidPtr, &clabel);
2414 
2415 			clabel.mod_counter = raidPtr->mod_counter;
2416 			clabel.column = scol;
2417 			clabel.status = rf_ds_optimal;
2418 
2419 			raidwrite_component_label(
2420 				      raidPtr->Disks[sparecol].dev,
2421 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2422 				      &clabel);
2423 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2424 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2425 					raidmarkclean( raidPtr->Disks[sparecol].dev,
2426 						       raidPtr->raid_cinfo[sparecol].ci_vp,
2427 						       raidPtr->mod_counter);
2428 				}
2429 			}
2430 		}
2431 	}
2432 }
2433 
2434 void
2435 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured)
2436 {
2437 	struct proc *p;
2438 	struct lwp *l;
2439 
2440 	p = raidPtr->engine_thread;
2441 	l = LIST_FIRST(&p->p_lwps);
2442 
2443 	if (vp != NULL) {
2444 		if (auto_configured == 1) {
2445 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2446 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2447 			vput(vp);
2448 
2449 		} else {
2450 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2451 		}
2452 	}
2453 }
2454 
2455 
2456 void
2457 rf_UnconfigureVnodes(RF_Raid_t *raidPtr)
2458 {
2459 	int r,c;
2460 	struct vnode *vp;
2461 	int acd;
2462 
2463 
2464 	/* We take this opportunity to close the vnodes like we should.. */
2465 
2466 	for (c = 0; c < raidPtr->numCol; c++) {
2467 		vp = raidPtr->raid_cinfo[c].ci_vp;
2468 		acd = raidPtr->Disks[c].auto_configured;
2469 		rf_close_component(raidPtr, vp, acd);
2470 		raidPtr->raid_cinfo[c].ci_vp = NULL;
2471 		raidPtr->Disks[c].auto_configured = 0;
2472 	}
2473 
2474 	for (r = 0; r < raidPtr->numSpare; r++) {
2475 		vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp;
2476 		acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured;
2477 		rf_close_component(raidPtr, vp, acd);
2478 		raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL;
2479 		raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0;
2480 	}
2481 }
2482 
2483 
2484 void
2485 rf_ReconThread(struct rf_recon_req *req)
2486 {
2487 	int     s;
2488 	RF_Raid_t *raidPtr;
2489 
2490 	s = splbio();
2491 	raidPtr = (RF_Raid_t *) req->raidPtr;
2492 	raidPtr->recon_in_progress = 1;
2493 
2494 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col,
2495 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2496 
2497 	RF_Free(req, sizeof(*req));
2498 
2499 	raidPtr->recon_in_progress = 0;
2500 	splx(s);
2501 
2502 	/* That's all... */
2503 	kthread_exit(0);        /* does not return */
2504 }
2505 
2506 void
2507 rf_RewriteParityThread(RF_Raid_t *raidPtr)
2508 {
2509 	int retcode;
2510 	int s;
2511 
2512 	raidPtr->parity_rewrite_stripes_done = 0;
2513 	raidPtr->parity_rewrite_in_progress = 1;
2514 	s = splbio();
2515 	retcode = rf_RewriteParity(raidPtr);
2516 	splx(s);
2517 	if (retcode) {
2518 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2519 	} else {
2520 		/* set the clean bit!  If we shutdown correctly,
2521 		   the clean bit on each component label will get
2522 		   set */
2523 		raidPtr->parity_good = RF_RAID_CLEAN;
2524 	}
2525 	raidPtr->parity_rewrite_in_progress = 0;
2526 
2527 	/* Anyone waiting for us to stop?  If so, inform them... */
2528 	if (raidPtr->waitShutdown) {
2529 		wakeup(&raidPtr->parity_rewrite_in_progress);
2530 	}
2531 
2532 	/* That's all... */
2533 	kthread_exit(0);        /* does not return */
2534 }
2535 
2536 
2537 void
2538 rf_CopybackThread(RF_Raid_t *raidPtr)
2539 {
2540 	int s;
2541 
2542 	raidPtr->copyback_in_progress = 1;
2543 	s = splbio();
2544 	rf_CopybackReconstructedData(raidPtr);
2545 	splx(s);
2546 	raidPtr->copyback_in_progress = 0;
2547 
2548 	/* That's all... */
2549 	kthread_exit(0);        /* does not return */
2550 }
2551 
2552 
2553 void
2554 rf_ReconstructInPlaceThread(struct rf_recon_req *req)
2555 {
2556 	int s;
2557 	RF_Raid_t *raidPtr;
2558 
2559 	s = splbio();
2560 	raidPtr = req->raidPtr;
2561 	raidPtr->recon_in_progress = 1;
2562 	rf_ReconstructInPlace(raidPtr, req->col);
2563 	RF_Free(req, sizeof(*req));
2564 	raidPtr->recon_in_progress = 0;
2565 	splx(s);
2566 
2567 	/* That's all... */
2568 	kthread_exit(0);        /* does not return */
2569 }
2570 
2571 RF_AutoConfig_t *
2572 rf_find_raid_components()
2573 {
2574 	struct vnode *vp;
2575 	struct disklabel label;
2576 	struct device *dv;
2577 	dev_t dev;
2578 	int bmajor;
2579 	int error;
2580 	int i;
2581 	int good_one;
2582 	RF_ComponentLabel_t *clabel;
2583 	RF_AutoConfig_t *ac_list;
2584 	RF_AutoConfig_t *ac;
2585 
2586 
2587 	/* initialize the AutoConfig list */
2588 	ac_list = NULL;
2589 
2590 	/* we begin by trolling through *all* the devices on the system */
2591 
2592 	for (dv = alldevs.tqh_first; dv != NULL;
2593 	     dv = dv->dv_list.tqe_next) {
2594 
2595 		/* we are only interested in disks... */
2596 		if (dv->dv_class != DV_DISK)
2597 			continue;
2598 
2599 		/* we don't care about floppies... */
2600 		if (!strcmp(dv->dv_cfdata->cf_name,"fd")) {
2601 			continue;
2602 		}
2603 
2604 		/* we don't care about CD's... */
2605 		if (!strcmp(dv->dv_cfdata->cf_name,"cd")) {
2606 			continue;
2607 		}
2608 
2609 		/* hdfd is the Atari/Hades floppy driver */
2610 		if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) {
2611 			continue;
2612 		}
2613 		/* fdisa is the Atari/Milan floppy driver */
2614 		if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) {
2615 			continue;
2616 		}
2617 
2618 		/* need to find the device_name_to_block_device_major stuff */
2619 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2620 
2621 		/* get a vnode for the raw partition of this disk */
2622 
2623 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2624 		if (bdevvp(dev, &vp))
2625 			panic("RAID can't alloc vnode");
2626 
2627 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2628 
2629 		if (error) {
2630 			/* "Who cares."  Continue looking
2631 			   for something that exists*/
2632 			vput(vp);
2633 			continue;
2634 		}
2635 
2636 		/* Ok, the disk exists.  Go get the disklabel. */
2637 		error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0);
2638 		if (error) {
2639 			/*
2640 			 * XXX can't happen - open() would
2641 			 * have errored out (or faked up one)
2642 			 */
2643 			if (error != ENOTTY)
2644 				printf("RAIDframe: can't get label for dev "
2645 				    "%s (%d)\n", dv->dv_xname, error);
2646 		}
2647 
2648 		/* don't need this any more.  We'll allocate it again
2649 		   a little later if we really do... */
2650 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2651 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2652 		vput(vp);
2653 
2654 		if (error)
2655 			continue;
2656 
2657 		for (i=0; i < label.d_npartitions; i++) {
2658 			/* We only support partitions marked as RAID */
2659 			if (label.d_partitions[i].p_fstype != FS_RAID)
2660 				continue;
2661 
2662 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2663 			if (bdevvp(dev, &vp))
2664 				panic("RAID can't alloc vnode");
2665 
2666 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2667 			if (error) {
2668 				/* Whatever... */
2669 				vput(vp);
2670 				continue;
2671 			}
2672 
2673 			good_one = 0;
2674 
2675 			clabel = (RF_ComponentLabel_t *)
2676 				malloc(sizeof(RF_ComponentLabel_t),
2677 				       M_RAIDFRAME, M_NOWAIT);
2678 			if (clabel == NULL) {
2679 				/* XXX CLEANUP HERE */
2680 				printf("RAID auto config: out of memory!\n");
2681 				return(NULL); /* XXX probably should panic? */
2682 			}
2683 
2684 			if (!raidread_component_label(dev, vp, clabel)) {
2685 				/* Got the label.  Does it look reasonable? */
2686 				if (rf_reasonable_label(clabel) &&
2687 				    (clabel->partitionSize <=
2688 				     label.d_partitions[i].p_size)) {
2689 #if DEBUG
2690 					printf("Component on: %s%c: %d\n",
2691 					       dv->dv_xname, 'a'+i,
2692 					       label.d_partitions[i].p_size);
2693 					rf_print_component_label(clabel);
2694 #endif
2695 					/* if it's reasonable, add it,
2696 					   else ignore it. */
2697 					ac = (RF_AutoConfig_t *)
2698 						malloc(sizeof(RF_AutoConfig_t),
2699 						       M_RAIDFRAME,
2700 						       M_NOWAIT);
2701 					if (ac == NULL) {
2702 						/* XXX should panic?? */
2703 						return(NULL);
2704 					}
2705 
2706 					snprintf(ac->devname,
2707 					    sizeof(ac->devname), "%s%c",
2708 					    dv->dv_xname, 'a'+i);
2709 					ac->dev = dev;
2710 					ac->vp = vp;
2711 					ac->clabel = clabel;
2712 					ac->next = ac_list;
2713 					ac_list = ac;
2714 					good_one = 1;
2715 				}
2716 			}
2717 			if (!good_one) {
2718 				/* cleanup */
2719 				free(clabel, M_RAIDFRAME);
2720 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2721 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2722 				vput(vp);
2723 			}
2724 		}
2725 	}
2726 	return(ac_list);
2727 }
2728 
2729 static int
2730 rf_reasonable_label(RF_ComponentLabel_t *clabel)
2731 {
2732 
2733 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2734 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2735 	    ((clabel->clean == RF_RAID_CLEAN) ||
2736 	     (clabel->clean == RF_RAID_DIRTY)) &&
2737 	    clabel->row >=0 &&
2738 	    clabel->column >= 0 &&
2739 	    clabel->num_rows > 0 &&
2740 	    clabel->num_columns > 0 &&
2741 	    clabel->row < clabel->num_rows &&
2742 	    clabel->column < clabel->num_columns &&
2743 	    clabel->blockSize > 0 &&
2744 	    clabel->numBlocks > 0) {
2745 		/* label looks reasonable enough... */
2746 		return(1);
2747 	}
2748 	return(0);
2749 }
2750 
2751 
2752 #if DEBUG
2753 void
2754 rf_print_component_label(RF_ComponentLabel_t *clabel)
2755 {
2756 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2757 	       clabel->row, clabel->column,
2758 	       clabel->num_rows, clabel->num_columns);
2759 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2760 	       clabel->version, clabel->serial_number,
2761 	       clabel->mod_counter);
2762 	printf("   Clean: %s Status: %d\n",
2763 	       clabel->clean ? "Yes" : "No", clabel->status );
2764 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2765 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2766 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2767 	       (char) clabel->parityConfig, clabel->blockSize,
2768 	       clabel->numBlocks);
2769 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2770 	printf("   Contains root partition: %s\n",
2771 	       clabel->root_partition ? "Yes" : "No" );
2772 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2773 #if 0
2774 	   printf("   Config order: %d\n", clabel->config_order);
2775 #endif
2776 
2777 }
2778 #endif
2779 
2780 RF_ConfigSet_t *
2781 rf_create_auto_sets(RF_AutoConfig_t *ac_list)
2782 {
2783 	RF_AutoConfig_t *ac;
2784 	RF_ConfigSet_t *config_sets;
2785 	RF_ConfigSet_t *cset;
2786 	RF_AutoConfig_t *ac_next;
2787 
2788 
2789 	config_sets = NULL;
2790 
2791 	/* Go through the AutoConfig list, and figure out which components
2792 	   belong to what sets.  */
2793 	ac = ac_list;
2794 	while(ac!=NULL) {
2795 		/* we're going to putz with ac->next, so save it here
2796 		   for use at the end of the loop */
2797 		ac_next = ac->next;
2798 
2799 		if (config_sets == NULL) {
2800 			/* will need at least this one... */
2801 			config_sets = (RF_ConfigSet_t *)
2802 				malloc(sizeof(RF_ConfigSet_t),
2803 				       M_RAIDFRAME, M_NOWAIT);
2804 			if (config_sets == NULL) {
2805 				panic("rf_create_auto_sets: No memory!");
2806 			}
2807 			/* this one is easy :) */
2808 			config_sets->ac = ac;
2809 			config_sets->next = NULL;
2810 			config_sets->rootable = 0;
2811 			ac->next = NULL;
2812 		} else {
2813 			/* which set does this component fit into? */
2814 			cset = config_sets;
2815 			while(cset!=NULL) {
2816 				if (rf_does_it_fit(cset, ac)) {
2817 					/* looks like it matches... */
2818 					ac->next = cset->ac;
2819 					cset->ac = ac;
2820 					break;
2821 				}
2822 				cset = cset->next;
2823 			}
2824 			if (cset==NULL) {
2825 				/* didn't find a match above... new set..*/
2826 				cset = (RF_ConfigSet_t *)
2827 					malloc(sizeof(RF_ConfigSet_t),
2828 					       M_RAIDFRAME, M_NOWAIT);
2829 				if (cset == NULL) {
2830 					panic("rf_create_auto_sets: No memory!");
2831 				}
2832 				cset->ac = ac;
2833 				ac->next = NULL;
2834 				cset->next = config_sets;
2835 				cset->rootable = 0;
2836 				config_sets = cset;
2837 			}
2838 		}
2839 		ac = ac_next;
2840 	}
2841 
2842 
2843 	return(config_sets);
2844 }
2845 
2846 static int
2847 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac)
2848 {
2849 	RF_ComponentLabel_t *clabel1, *clabel2;
2850 
2851 	/* If this one matches the *first* one in the set, that's good
2852 	   enough, since the other members of the set would have been
2853 	   through here too... */
2854 	/* note that we are not checking partitionSize here..
2855 
2856 	   Note that we are also not checking the mod_counters here.
2857 	   If everything else matches execpt the mod_counter, that's
2858 	   good enough for this test.  We will deal with the mod_counters
2859 	   a little later in the autoconfiguration process.
2860 
2861 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2862 
2863 	   The reason we don't check for this is that failed disks
2864 	   will have lower modification counts.  If those disks are
2865 	   not added to the set they used to belong to, then they will
2866 	   form their own set, which may result in 2 different sets,
2867 	   for example, competing to be configured at raid0, and
2868 	   perhaps competing to be the root filesystem set.  If the
2869 	   wrong ones get configured, or both attempt to become /,
2870 	   weird behaviour and or serious lossage will occur.  Thus we
2871 	   need to bring them into the fold here, and kick them out at
2872 	   a later point.
2873 
2874 	*/
2875 
2876 	clabel1 = cset->ac->clabel;
2877 	clabel2 = ac->clabel;
2878 	if ((clabel1->version == clabel2->version) &&
2879 	    (clabel1->serial_number == clabel2->serial_number) &&
2880 	    (clabel1->num_rows == clabel2->num_rows) &&
2881 	    (clabel1->num_columns == clabel2->num_columns) &&
2882 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2883 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2884 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2885 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2886 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2887 	    (clabel1->blockSize == clabel2->blockSize) &&
2888 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2889 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2890 	    (clabel1->root_partition == clabel2->root_partition) &&
2891 	    (clabel1->last_unit == clabel2->last_unit) &&
2892 	    (clabel1->config_order == clabel2->config_order)) {
2893 		/* if it get's here, it almost *has* to be a match */
2894 	} else {
2895 		/* it's not consistent with somebody in the set..
2896 		   punt */
2897 		return(0);
2898 	}
2899 	/* all was fine.. it must fit... */
2900 	return(1);
2901 }
2902 
2903 int
2904 rf_have_enough_components(RF_ConfigSet_t *cset)
2905 {
2906 	RF_AutoConfig_t *ac;
2907 	RF_AutoConfig_t *auto_config;
2908 	RF_ComponentLabel_t *clabel;
2909 	int c;
2910 	int num_cols;
2911 	int num_missing;
2912 	int mod_counter;
2913 	int mod_counter_found;
2914 	int even_pair_failed;
2915 	char parity_type;
2916 
2917 
2918 	/* check to see that we have enough 'live' components
2919 	   of this set.  If so, we can configure it if necessary */
2920 
2921 	num_cols = cset->ac->clabel->num_columns;
2922 	parity_type = cset->ac->clabel->parityConfig;
2923 
2924 	/* XXX Check for duplicate components!?!?!? */
2925 
2926 	/* Determine what the mod_counter is supposed to be for this set. */
2927 
2928 	mod_counter_found = 0;
2929 	mod_counter = 0;
2930 	ac = cset->ac;
2931 	while(ac!=NULL) {
2932 		if (mod_counter_found==0) {
2933 			mod_counter = ac->clabel->mod_counter;
2934 			mod_counter_found = 1;
2935 		} else {
2936 			if (ac->clabel->mod_counter > mod_counter) {
2937 				mod_counter = ac->clabel->mod_counter;
2938 			}
2939 		}
2940 		ac = ac->next;
2941 	}
2942 
2943 	num_missing = 0;
2944 	auto_config = cset->ac;
2945 
2946 	even_pair_failed = 0;
2947 	for(c=0; c<num_cols; c++) {
2948 		ac = auto_config;
2949 		while(ac!=NULL) {
2950 			if ((ac->clabel->column == c) &&
2951 			    (ac->clabel->mod_counter == mod_counter)) {
2952 				/* it's this one... */
2953 #if DEBUG
2954 				printf("Found: %s at %d\n",
2955 				       ac->devname,c);
2956 #endif
2957 				break;
2958 			}
2959 			ac=ac->next;
2960 		}
2961 		if (ac==NULL) {
2962 				/* Didn't find one here! */
2963 				/* special case for RAID 1, especially
2964 				   where there are more than 2
2965 				   components (where RAIDframe treats
2966 				   things a little differently :( ) */
2967 			if (parity_type == '1') {
2968 				if (c%2 == 0) { /* even component */
2969 					even_pair_failed = 1;
2970 				} else { /* odd component.  If
2971 					    we're failed, and
2972 					    so is the even
2973 					    component, it's
2974 					    "Good Night, Charlie" */
2975 					if (even_pair_failed == 1) {
2976 						return(0);
2977 					}
2978 				}
2979 			} else {
2980 				/* normal accounting */
2981 				num_missing++;
2982 			}
2983 		}
2984 		if ((parity_type == '1') && (c%2 == 1)) {
2985 				/* Just did an even component, and we didn't
2986 				   bail.. reset the even_pair_failed flag,
2987 				   and go on to the next component.... */
2988 			even_pair_failed = 0;
2989 		}
2990 	}
2991 
2992 	clabel = cset->ac->clabel;
2993 
2994 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
2995 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
2996 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
2997 		/* XXX this needs to be made *much* more general */
2998 		/* Too many failures */
2999 		return(0);
3000 	}
3001 	/* otherwise, all is well, and we've got enough to take a kick
3002 	   at autoconfiguring this set */
3003 	return(1);
3004 }
3005 
3006 void
3007 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config,
3008 			RF_Raid_t *raidPtr)
3009 {
3010 	RF_ComponentLabel_t *clabel;
3011 	int i;
3012 
3013 	clabel = ac->clabel;
3014 
3015 	/* 1. Fill in the common stuff */
3016 	config->numRow = clabel->num_rows = 1;
3017 	config->numCol = clabel->num_columns;
3018 	config->numSpare = 0; /* XXX should this be set here? */
3019 	config->sectPerSU = clabel->sectPerSU;
3020 	config->SUsPerPU = clabel->SUsPerPU;
3021 	config->SUsPerRU = clabel->SUsPerRU;
3022 	config->parityConfig = clabel->parityConfig;
3023 	/* XXX... */
3024 	strcpy(config->diskQueueType,"fifo");
3025 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3026 	config->layoutSpecificSize = 0; /* XXX ?? */
3027 
3028 	while(ac!=NULL) {
3029 		/* row/col values will be in range due to the checks
3030 		   in reasonable_label() */
3031 		strcpy(config->devnames[0][ac->clabel->column],
3032 		       ac->devname);
3033 		ac = ac->next;
3034 	}
3035 
3036 	for(i=0;i<RF_MAXDBGV;i++) {
3037 		config->debugVars[i][0] = 0;
3038 	}
3039 }
3040 
3041 int
3042 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value)
3043 {
3044 	RF_ComponentLabel_t clabel;
3045 	struct vnode *vp;
3046 	dev_t dev;
3047 	int column;
3048 	int sparecol;
3049 
3050 	raidPtr->autoconfigure = new_value;
3051 
3052 	for(column=0; column<raidPtr->numCol; column++) {
3053 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3054 			dev = raidPtr->Disks[column].dev;
3055 			vp = raidPtr->raid_cinfo[column].ci_vp;
3056 			raidread_component_label(dev, vp, &clabel);
3057 			clabel.autoconfigure = new_value;
3058 			raidwrite_component_label(dev, vp, &clabel);
3059 		}
3060 	}
3061 	for(column = 0; column < raidPtr->numSpare ; column++) {
3062 		sparecol = raidPtr->numCol + column;
3063 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3064 			dev = raidPtr->Disks[sparecol].dev;
3065 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3066 			raidread_component_label(dev, vp, &clabel);
3067 			clabel.autoconfigure = new_value;
3068 			raidwrite_component_label(dev, vp, &clabel);
3069 		}
3070 	}
3071 	return(new_value);
3072 }
3073 
3074 int
3075 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value)
3076 {
3077 	RF_ComponentLabel_t clabel;
3078 	struct vnode *vp;
3079 	dev_t dev;
3080 	int column;
3081 	int sparecol;
3082 
3083 	raidPtr->root_partition = new_value;
3084 	for(column=0; column<raidPtr->numCol; column++) {
3085 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3086 			dev = raidPtr->Disks[column].dev;
3087 			vp = raidPtr->raid_cinfo[column].ci_vp;
3088 			raidread_component_label(dev, vp, &clabel);
3089 			clabel.root_partition = new_value;
3090 			raidwrite_component_label(dev, vp, &clabel);
3091 		}
3092 	}
3093 	for(column = 0; column < raidPtr->numSpare ; column++) {
3094 		sparecol = raidPtr->numCol + column;
3095 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3096 			dev = raidPtr->Disks[sparecol].dev;
3097 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3098 			raidread_component_label(dev, vp, &clabel);
3099 			clabel.root_partition = new_value;
3100 			raidwrite_component_label(dev, vp, &clabel);
3101 		}
3102 	}
3103 	return(new_value);
3104 }
3105 
3106 void
3107 rf_release_all_vps(RF_ConfigSet_t *cset)
3108 {
3109 	RF_AutoConfig_t *ac;
3110 
3111 	ac = cset->ac;
3112 	while(ac!=NULL) {
3113 		/* Close the vp, and give it back */
3114 		if (ac->vp) {
3115 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3116 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3117 			vput(ac->vp);
3118 			ac->vp = NULL;
3119 		}
3120 		ac = ac->next;
3121 	}
3122 }
3123 
3124 
3125 void
3126 rf_cleanup_config_set(RF_ConfigSet_t *cset)
3127 {
3128 	RF_AutoConfig_t *ac;
3129 	RF_AutoConfig_t *next_ac;
3130 
3131 	ac = cset->ac;
3132 	while(ac!=NULL) {
3133 		next_ac = ac->next;
3134 		/* nuke the label */
3135 		free(ac->clabel, M_RAIDFRAME);
3136 		/* cleanup the config structure */
3137 		free(ac, M_RAIDFRAME);
3138 		/* "next.." */
3139 		ac = next_ac;
3140 	}
3141 	/* and, finally, nuke the config set */
3142 	free(cset, M_RAIDFRAME);
3143 }
3144 
3145 
3146 void
3147 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel)
3148 {
3149 	/* current version number */
3150 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3151 	clabel->serial_number = raidPtr->serial_number;
3152 	clabel->mod_counter = raidPtr->mod_counter;
3153 	clabel->num_rows = 1;
3154 	clabel->num_columns = raidPtr->numCol;
3155 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3156 	clabel->status = rf_ds_optimal; /* "It's good!" */
3157 
3158 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3159 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3160 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3161 
3162 	clabel->blockSize = raidPtr->bytesPerSector;
3163 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3164 
3165 	/* XXX not portable */
3166 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3167 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3168 	clabel->autoconfigure = raidPtr->autoconfigure;
3169 	clabel->root_partition = raidPtr->root_partition;
3170 	clabel->last_unit = raidPtr->raidid;
3171 	clabel->config_order = raidPtr->config_order;
3172 }
3173 
3174 int
3175 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit)
3176 {
3177 	RF_Raid_t *raidPtr;
3178 	RF_Config_t *config;
3179 	int raidID;
3180 	int retcode;
3181 
3182 #if DEBUG
3183 	printf("RAID autoconfigure\n");
3184 #endif
3185 
3186 	retcode = 0;
3187 	*unit = -1;
3188 
3189 	/* 1. Create a config structure */
3190 
3191 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3192 				       M_RAIDFRAME,
3193 				       M_NOWAIT);
3194 	if (config==NULL) {
3195 		printf("Out of mem!?!?\n");
3196 				/* XXX do something more intelligent here. */
3197 		return(1);
3198 	}
3199 
3200 	memset(config, 0, sizeof(RF_Config_t));
3201 
3202 	/*
3203 	   2. Figure out what RAID ID this one is supposed to live at
3204 	   See if we can get the same RAID dev that it was configured
3205 	   on last time..
3206 	*/
3207 
3208 	raidID = cset->ac->clabel->last_unit;
3209 	if ((raidID < 0) || (raidID >= numraid)) {
3210 		/* let's not wander off into lala land. */
3211 		raidID = numraid - 1;
3212 	}
3213 	if (raidPtrs[raidID]->valid != 0) {
3214 
3215 		/*
3216 		   Nope... Go looking for an alternative...
3217 		   Start high so we don't immediately use raid0 if that's
3218 		   not taken.
3219 		*/
3220 
3221 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3222 			if (raidPtrs[raidID]->valid == 0) {
3223 				/* can use this one! */
3224 				break;
3225 			}
3226 		}
3227 	}
3228 
3229 	if (raidID < 0) {
3230 		/* punt... */
3231 		printf("Unable to auto configure this set!\n");
3232 		printf("(Out of RAID devs!)\n");
3233 		return(1);
3234 	}
3235 
3236 #if DEBUG
3237 	printf("Configuring raid%d:\n",raidID);
3238 #endif
3239 
3240 	raidPtr = raidPtrs[raidID];
3241 
3242 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3243 	raidPtr->raidid = raidID;
3244 	raidPtr->openings = RAIDOUTSTANDING;
3245 
3246 	/* 3. Build the configuration structure */
3247 	rf_create_configuration(cset->ac, config, raidPtr);
3248 
3249 	/* 4. Do the configuration */
3250 	retcode = rf_Configure(raidPtr, config, cset->ac);
3251 
3252 	if (retcode == 0) {
3253 
3254 		raidinit(raidPtrs[raidID]);
3255 
3256 		rf_markalldirty(raidPtrs[raidID]);
3257 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3258 		if (cset->ac->clabel->root_partition==1) {
3259 			/* everything configured just fine.  Make a note
3260 			   that this set is eligible to be root. */
3261 			cset->rootable = 1;
3262 			/* XXX do this here? */
3263 			raidPtrs[raidID]->root_partition = 1;
3264 		}
3265 	}
3266 
3267 	/* 5. Cleanup */
3268 	free(config, M_RAIDFRAME);
3269 
3270 	*unit = raidID;
3271 	return(retcode);
3272 }
3273 
3274 void
3275 rf_disk_unbusy(RF_RaidAccessDesc_t *desc)
3276 {
3277 	struct buf *bp;
3278 
3279 	bp = (struct buf *)desc->bp;
3280 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3281 	    (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ));
3282 }
3283 
3284 void
3285 rf_pool_init(struct pool *p, size_t size, const char *w_chan,
3286 	     size_t xmin, size_t xmax)
3287 {
3288 	pool_init(p, size, 0, 0, 0, w_chan, NULL);
3289 	pool_sethiwat(p, xmax);
3290 	pool_prime(p, xmin);
3291 	pool_setlowat(p, xmin);
3292 }
3293 
3294 /*
3295  * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see
3296  * if there is IO pending and if that IO could possibly be done for a
3297  * given RAID set.  Returns 0 if IO is waiting and can be done, 1
3298  * otherwise.
3299  *
3300  */
3301 
3302 int
3303 rf_buf_queue_check(int raidid)
3304 {
3305 	if ((BUFQ_PEEK(raid_softc[raidid].buf_queue) != NULL) &&
3306 	    raidPtrs[raidid]->openings > 0) {
3307 		/* there is work to do */
3308 		return 0;
3309 	}
3310 	/* default is nothing to do */
3311 	return 1;
3312 }
3313