xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision aa73cae19608873cc4d1f712c4a0f8f8435f1ffa)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.186 2005/02/27 00:27:45 perry 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 1.186 2005/02/27 00:27:45 perry 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 * bp);
210 static void InitBP(struct buf * bp, struct vnode *, unsigned rw_flag,
211 		   dev_t dev, RF_SectorNum_t startSect,
212 		   RF_SectorCount_t numSect, caddr_t buf,
213 		   void (*cbFunc) (struct buf *), void *cbArg,
214 		   int logBytesPerSector, struct proc * b_proc);
215 static void raidinit(RF_Raid_t *);
216 
217 void raidattach(int);
218 
219 dev_type_open(raidopen);
220 dev_type_close(raidclose);
221 dev_type_read(raidread);
222 dev_type_write(raidwrite);
223 dev_type_ioctl(raidioctl);
224 dev_type_strategy(raidstrategy);
225 dev_type_dump(raiddump);
226 dev_type_size(raidsize);
227 
228 const struct bdevsw raid_bdevsw = {
229 	raidopen, raidclose, raidstrategy, raidioctl,
230 	raiddump, raidsize, D_DISK
231 };
232 
233 const struct cdevsw raid_cdevsw = {
234 	raidopen, raidclose, raidread, raidwrite, raidioctl,
235 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
236 };
237 
238 /*
239  * Pilfered from ccd.c
240  */
241 
242 struct raidbuf {
243 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
244 	struct buf *rf_obp;	/* ptr. to original I/O buf */
245 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
246 };
247 
248 /* XXX Not sure if the following should be replacing the raidPtrs above,
249    or if it should be used in conjunction with that...
250 */
251 
252 struct raid_softc {
253 	int     sc_flags;	/* flags */
254 	int     sc_cflags;	/* configuration flags */
255 	size_t  sc_size;        /* size of the raid device */
256 	char    sc_xname[20];	/* XXX external name */
257 	struct disk sc_dkdev;	/* generic disk device info */
258 	struct bufq_state buf_queue;	/* used for the device queue */
259 };
260 /* sc_flags */
261 #define RAIDF_INITED	0x01	/* unit has been initialized */
262 #define RAIDF_WLABEL	0x02	/* label area is writable */
263 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
264 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
265 #define RAIDF_LOCKED	0x80	/* unit is locked */
266 
267 #define	raidunit(x)	DISKUNIT(x)
268 int numraid = 0;
269 
270 /*
271  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
272  * Be aware that large numbers can allow the driver to consume a lot of
273  * kernel memory, especially on writes, and in degraded mode reads.
274  *
275  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
276  * a single 64K write will typically require 64K for the old data,
277  * 64K for the old parity, and 64K for the new parity, for a total
278  * of 192K (if the parity buffer is not re-used immediately).
279  * Even it if is used immediately, that's still 128K, which when multiplied
280  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
281  *
282  * Now in degraded mode, for example, a 64K read on the above setup may
283  * require data reconstruction, which will require *all* of the 4 remaining
284  * disks to participate -- 4 * 32K/disk == 128K again.
285  */
286 
287 #ifndef RAIDOUTSTANDING
288 #define RAIDOUTSTANDING   6
289 #endif
290 
291 #define RAIDLABELDEV(dev)	\
292 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
293 
294 /* declared here, and made public, for the benefit of KVM stuff.. */
295 struct raid_softc *raid_softc;
296 
297 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
298 				     struct disklabel *);
299 static void raidgetdisklabel(dev_t);
300 static void raidmakedisklabel(struct raid_softc *);
301 
302 static int raidlock(struct raid_softc *);
303 static void raidunlock(struct raid_softc *);
304 
305 static void rf_markalldirty(RF_Raid_t *);
306 
307 struct device *raidrootdev;
308 
309 void rf_ReconThread(struct rf_recon_req *);
310 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
311 void rf_CopybackThread(RF_Raid_t *raidPtr);
312 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
313 int rf_autoconfig(struct device *self);
314 void rf_buildroothack(RF_ConfigSet_t *);
315 
316 RF_AutoConfig_t *rf_find_raid_components(void);
317 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
318 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
319 static int rf_reasonable_label(RF_ComponentLabel_t *);
320 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
321 int rf_set_autoconfig(RF_Raid_t *, int);
322 int rf_set_rootpartition(RF_Raid_t *, int);
323 void rf_release_all_vps(RF_ConfigSet_t *);
324 void rf_cleanup_config_set(RF_ConfigSet_t *);
325 int rf_have_enough_components(RF_ConfigSet_t *);
326 int rf_auto_config_set(RF_ConfigSet_t *, int *);
327 
328 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
329 				  allow autoconfig to take place.
330 			          Note that this is overridden by having
331 			          RAID_AUTOCONFIG as an option in the
332 			          kernel config file.  */
333 
334 struct RF_Pools_s rf_pools;
335 
336 void
337 raidattach(int num)
338 {
339 	int raidID;
340 	int i, rc;
341 
342 #ifdef DEBUG
343 	printf("raidattach: Asked for %d units\n", num);
344 #endif
345 
346 	if (num <= 0) {
347 #ifdef DIAGNOSTIC
348 		panic("raidattach: count <= 0");
349 #endif
350 		return;
351 	}
352 	/* This is where all the initialization stuff gets done. */
353 
354 	numraid = num;
355 
356 	/* Make some space for requested number of units... */
357 
358 	RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **));
359 	if (raidPtrs == NULL) {
360 		panic("raidPtrs is NULL!!");
361 	}
362 
363 	/* Initialize the component buffer pool. */
364 	rf_pool_init(&rf_pools.cbuf, sizeof(struct raidbuf),
365 		     "raidpl", num * RAIDOUTSTANDING,
366 		     2 * num * RAIDOUTSTANDING);
367 
368 	rf_mutex_init(&rf_sparet_wait_mutex);
369 
370 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
371 
372 	for (i = 0; i < num; i++)
373 		raidPtrs[i] = NULL;
374 	rc = rf_BootRaidframe();
375 	if (rc == 0)
376 		printf("Kernelized RAIDframe activated\n");
377 	else
378 		panic("Serious error booting RAID!!");
379 
380 	/* put together some datastructures like the CCD device does.. This
381 	 * lets us lock the device and what-not when it gets opened. */
382 
383 	raid_softc = (struct raid_softc *)
384 		malloc(num * sizeof(struct raid_softc),
385 		       M_RAIDFRAME, M_NOWAIT);
386 	if (raid_softc == NULL) {
387 		printf("WARNING: no memory for RAIDframe driver\n");
388 		return;
389 	}
390 
391 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
392 
393 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
394 					      M_RAIDFRAME, M_NOWAIT);
395 	if (raidrootdev == NULL) {
396 		panic("No memory for RAIDframe driver!!?!?!");
397 	}
398 
399 	for (raidID = 0; raidID < num; raidID++) {
400 		bufq_alloc(&raid_softc[raidID].buf_queue, BUFQ_FCFS);
401 
402 		raidrootdev[raidID].dv_class  = DV_DISK;
403 		raidrootdev[raidID].dv_cfdata = NULL;
404 		raidrootdev[raidID].dv_unit   = raidID;
405 		raidrootdev[raidID].dv_parent = NULL;
406 		raidrootdev[raidID].dv_flags  = 0;
407 		snprintf(raidrootdev[raidID].dv_xname,
408 		    sizeof(raidrootdev[raidID].dv_xname), "raid%d", raidID);
409 
410 		RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t),
411 			  (RF_Raid_t *));
412 		if (raidPtrs[raidID] == NULL) {
413 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
414 			numraid = raidID;
415 			return;
416 		}
417 	}
418 
419 #ifdef RAID_AUTOCONFIG
420 	raidautoconfig = 1;
421 #endif
422 
423 	/*
424 	 * Register a finalizer which will be used to auto-config RAID
425 	 * sets once all real hardware devices have been found.
426 	 */
427 	if (config_finalize_register(NULL, rf_autoconfig) != 0)
428 		printf("WARNING: unable to register RAIDframe finalizer\n");
429 }
430 
431 int
432 rf_autoconfig(struct device *self)
433 {
434 	RF_AutoConfig_t *ac_list;
435 	RF_ConfigSet_t *config_sets;
436 
437 	if (raidautoconfig == 0)
438 		return (0);
439 
440 	/* XXX This code can only be run once. */
441 	raidautoconfig = 0;
442 
443 	/* 1. locate all RAID components on the system */
444 #ifdef DEBUG
445 	printf("Searching for RAID components...\n");
446 #endif
447 	ac_list = rf_find_raid_components();
448 
449 	/* 2. Sort them into their respective sets. */
450 	config_sets = rf_create_auto_sets(ac_list);
451 
452 	/*
453 	 * 3. Evaluate each set andconfigure the valid ones.
454 	 * This gets done in rf_buildroothack().
455 	 */
456 	rf_buildroothack(config_sets);
457 
458 	return (1);
459 }
460 
461 void
462 rf_buildroothack(RF_ConfigSet_t *config_sets)
463 {
464 	RF_ConfigSet_t *cset;
465 	RF_ConfigSet_t *next_cset;
466 	int retcode;
467 	int raidID;
468 	int rootID;
469 	int num_root;
470 
471 	rootID = 0;
472 	num_root = 0;
473 	cset = config_sets;
474 	while(cset != NULL ) {
475 		next_cset = cset->next;
476 		if (rf_have_enough_components(cset) &&
477 		    cset->ac->clabel->autoconfigure==1) {
478 			retcode = rf_auto_config_set(cset,&raidID);
479 			if (!retcode) {
480 				if (cset->rootable) {
481 					rootID = raidID;
482 					num_root++;
483 				}
484 			} else {
485 				/* The autoconfig didn't work :( */
486 #if DEBUG
487 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
488 #endif
489 				rf_release_all_vps(cset);
490 			}
491 		} else {
492 			/* we're not autoconfiguring this set...
493 			   release the associated resources */
494 			rf_release_all_vps(cset);
495 		}
496 		/* cleanup */
497 		rf_cleanup_config_set(cset);
498 		cset = next_cset;
499 	}
500 
501 	/* we found something bootable... */
502 
503 	if (num_root == 1) {
504 		booted_device = &raidrootdev[rootID];
505 	} else if (num_root > 1) {
506 		/* we can't guess.. require the user to answer... */
507 		boothowto |= RB_ASKNAME;
508 	}
509 }
510 
511 
512 int
513 raidsize(dev_t dev)
514 {
515 	struct raid_softc *rs;
516 	struct disklabel *lp;
517 	int     part, unit, omask, size;
518 
519 	unit = raidunit(dev);
520 	if (unit >= numraid)
521 		return (-1);
522 	rs = &raid_softc[unit];
523 
524 	if ((rs->sc_flags & RAIDF_INITED) == 0)
525 		return (-1);
526 
527 	part = DISKPART(dev);
528 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
529 	lp = rs->sc_dkdev.dk_label;
530 
531 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curproc))
532 		return (-1);
533 
534 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
535 		size = -1;
536 	else
537 		size = lp->d_partitions[part].p_size *
538 		    (lp->d_secsize / DEV_BSIZE);
539 
540 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curproc))
541 		return (-1);
542 
543 	return (size);
544 
545 }
546 
547 int
548 raiddump(dev_t dev, daddr_t blkno, caddr_t va, size_t  size)
549 {
550 	/* Not implemented. */
551 	return ENXIO;
552 }
553 /* ARGSUSED */
554 int
555 raidopen(dev_t dev, int flags, int fmt, struct proc *p)
556 {
557 	int     unit = raidunit(dev);
558 	struct raid_softc *rs;
559 	struct disklabel *lp;
560 	int     part, pmask;
561 	int     error = 0;
562 
563 	if (unit >= numraid)
564 		return (ENXIO);
565 	rs = &raid_softc[unit];
566 
567 	if ((error = raidlock(rs)) != 0)
568 		return (error);
569 	lp = rs->sc_dkdev.dk_label;
570 
571 	part = DISKPART(dev);
572 	pmask = (1 << part);
573 
574 	if ((rs->sc_flags & RAIDF_INITED) &&
575 	    (rs->sc_dkdev.dk_openmask == 0))
576 		raidgetdisklabel(dev);
577 
578 	/* make sure that this partition exists */
579 
580 	if (part != RAW_PART) {
581 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
582 		    ((part >= lp->d_npartitions) ||
583 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
584 			error = ENXIO;
585 			raidunlock(rs);
586 			return (error);
587 		}
588 	}
589 	/* Prevent this unit from being unconfigured while open. */
590 	switch (fmt) {
591 	case S_IFCHR:
592 		rs->sc_dkdev.dk_copenmask |= pmask;
593 		break;
594 
595 	case S_IFBLK:
596 		rs->sc_dkdev.dk_bopenmask |= pmask;
597 		break;
598 	}
599 
600 	if ((rs->sc_dkdev.dk_openmask == 0) &&
601 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
602 		/* First one... mark things as dirty... Note that we *MUST*
603 		 have done a configure before this.  I DO NOT WANT TO BE
604 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
605 		 THAT THEY BELONG TOGETHER!!!!! */
606 		/* XXX should check to see if we're only open for reading
607 		   here... If so, we needn't do this, but then need some
608 		   other way of keeping track of what's happened.. */
609 
610 		rf_markalldirty( raidPtrs[unit] );
611 	}
612 
613 
614 	rs->sc_dkdev.dk_openmask =
615 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
616 
617 	raidunlock(rs);
618 
619 	return (error);
620 
621 
622 }
623 /* ARGSUSED */
624 int
625 raidclose(dev_t dev, int flags, int fmt, struct proc *p)
626 {
627 	int     unit = raidunit(dev);
628 	struct raid_softc *rs;
629 	int     error = 0;
630 	int     part;
631 
632 	if (unit >= numraid)
633 		return (ENXIO);
634 	rs = &raid_softc[unit];
635 
636 	if ((error = raidlock(rs)) != 0)
637 		return (error);
638 
639 	part = DISKPART(dev);
640 
641 	/* ...that much closer to allowing unconfiguration... */
642 	switch (fmt) {
643 	case S_IFCHR:
644 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
645 		break;
646 
647 	case S_IFBLK:
648 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
649 		break;
650 	}
651 	rs->sc_dkdev.dk_openmask =
652 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
653 
654 	if ((rs->sc_dkdev.dk_openmask == 0) &&
655 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
656 		/* Last one... device is not unconfigured yet.
657 		   Device shutdown has taken care of setting the
658 		   clean bits if RAIDF_INITED is not set
659 		   mark things as clean... */
660 
661 		rf_update_component_labels(raidPtrs[unit],
662 						 RF_FINAL_COMPONENT_UPDATE);
663 		if (doing_shutdown) {
664 			/* last one, and we're going down, so
665 			   lights out for this RAID set too. */
666 			error = rf_Shutdown(raidPtrs[unit]);
667 
668 			/* It's no longer initialized... */
669 			rs->sc_flags &= ~RAIDF_INITED;
670 
671 			/* Detach the disk. */
672 			disk_detach(&rs->sc_dkdev);
673 		}
674 	}
675 
676 	raidunlock(rs);
677 	return (0);
678 
679 }
680 
681 void
682 raidstrategy(struct buf *bp)
683 {
684 	int s;
685 
686 	unsigned int raidID = raidunit(bp->b_dev);
687 	RF_Raid_t *raidPtr;
688 	struct raid_softc *rs = &raid_softc[raidID];
689 	int     wlabel;
690 
691 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
692 		bp->b_error = ENXIO;
693 		bp->b_flags |= B_ERROR;
694 		bp->b_resid = bp->b_bcount;
695 		biodone(bp);
696 		return;
697 	}
698 	if (raidID >= numraid || !raidPtrs[raidID]) {
699 		bp->b_error = ENODEV;
700 		bp->b_flags |= B_ERROR;
701 		bp->b_resid = bp->b_bcount;
702 		biodone(bp);
703 		return;
704 	}
705 	raidPtr = raidPtrs[raidID];
706 	if (!raidPtr->valid) {
707 		bp->b_error = ENODEV;
708 		bp->b_flags |= B_ERROR;
709 		bp->b_resid = bp->b_bcount;
710 		biodone(bp);
711 		return;
712 	}
713 	if (bp->b_bcount == 0) {
714 		db1_printf(("b_bcount is zero..\n"));
715 		biodone(bp);
716 		return;
717 	}
718 
719 	/*
720 	 * Do bounds checking and adjust transfer.  If there's an
721 	 * error, the bounds check will flag that for us.
722 	 */
723 
724 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
725 	if (DISKPART(bp->b_dev) != RAW_PART)
726 		if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) {
727 			db1_printf(("Bounds check failed!!:%d %d\n",
728 				(int) bp->b_blkno, (int) wlabel));
729 			biodone(bp);
730 			return;
731 		}
732 	s = splbio();
733 
734 	bp->b_resid = 0;
735 
736 	/* stuff it onto our queue */
737 	BUFQ_PUT(&rs->buf_queue, bp);
738 
739 	raidstart(raidPtrs[raidID]);
740 
741 	splx(s);
742 }
743 /* ARGSUSED */
744 int
745 raidread(dev_t dev, struct uio *uio, int flags)
746 {
747 	int     unit = raidunit(dev);
748 	struct raid_softc *rs;
749 
750 	if (unit >= numraid)
751 		return (ENXIO);
752 	rs = &raid_softc[unit];
753 
754 	if ((rs->sc_flags & RAIDF_INITED) == 0)
755 		return (ENXIO);
756 
757 	return (physio(raidstrategy, NULL, dev, B_READ, minphys, uio));
758 
759 }
760 /* ARGSUSED */
761 int
762 raidwrite(dev_t dev, struct uio *uio, int flags)
763 {
764 	int     unit = raidunit(dev);
765 	struct raid_softc *rs;
766 
767 	if (unit >= numraid)
768 		return (ENXIO);
769 	rs = &raid_softc[unit];
770 
771 	if ((rs->sc_flags & RAIDF_INITED) == 0)
772 		return (ENXIO);
773 
774 	return (physio(raidstrategy, NULL, dev, B_WRITE, minphys, uio));
775 
776 }
777 
778 int
779 raidioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
780 {
781 	int     unit = raidunit(dev);
782 	int     error = 0;
783 	int     part, pmask;
784 	struct raid_softc *rs;
785 	RF_Config_t *k_cfg, *u_cfg;
786 	RF_Raid_t *raidPtr;
787 	RF_RaidDisk_t *diskPtr;
788 	RF_AccTotals_t *totals;
789 	RF_DeviceConfig_t *d_cfg, **ucfgp;
790 	u_char *specific_buf;
791 	int retcode = 0;
792 	int column;
793 	int raidid;
794 	struct rf_recon_req *rrcopy, *rr;
795 	RF_ComponentLabel_t *clabel;
796 	RF_ComponentLabel_t ci_label;
797 	RF_ComponentLabel_t **clabel_ptr;
798 	RF_SingleComponent_t *sparePtr,*componentPtr;
799 	RF_SingleComponent_t hot_spare;
800 	RF_SingleComponent_t component;
801 	RF_ProgressInfo_t progressInfo, **progressInfoPtr;
802 	int i, j, d;
803 #ifdef __HAVE_OLD_DISKLABEL
804 	struct disklabel newlabel;
805 #endif
806 
807 	if (unit >= numraid)
808 		return (ENXIO);
809 	rs = &raid_softc[unit];
810 	raidPtr = raidPtrs[unit];
811 
812 	db1_printf(("raidioctl: %d %d %d %d\n", (int) dev,
813 		(int) DISKPART(dev), (int) unit, (int) cmd));
814 
815 	/* Must be open for writes for these commands... */
816 	switch (cmd) {
817 	case DIOCSDINFO:
818 	case DIOCWDINFO:
819 #ifdef __HAVE_OLD_DISKLABEL
820 	case ODIOCWDINFO:
821 	case ODIOCSDINFO:
822 #endif
823 	case DIOCWLABEL:
824 		if ((flag & FWRITE) == 0)
825 			return (EBADF);
826 	}
827 
828 	/* Must be initialized for these... */
829 	switch (cmd) {
830 	case DIOCGDINFO:
831 	case DIOCSDINFO:
832 	case DIOCWDINFO:
833 #ifdef __HAVE_OLD_DISKLABEL
834 	case ODIOCGDINFO:
835 	case ODIOCWDINFO:
836 	case ODIOCSDINFO:
837 	case ODIOCGDEFLABEL:
838 #endif
839 	case DIOCGPART:
840 	case DIOCWLABEL:
841 	case DIOCGDEFLABEL:
842 	case RAIDFRAME_SHUTDOWN:
843 	case RAIDFRAME_REWRITEPARITY:
844 	case RAIDFRAME_GET_INFO:
845 	case RAIDFRAME_RESET_ACCTOTALS:
846 	case RAIDFRAME_GET_ACCTOTALS:
847 	case RAIDFRAME_KEEP_ACCTOTALS:
848 	case RAIDFRAME_GET_SIZE:
849 	case RAIDFRAME_FAIL_DISK:
850 	case RAIDFRAME_COPYBACK:
851 	case RAIDFRAME_CHECK_RECON_STATUS:
852 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
853 	case RAIDFRAME_GET_COMPONENT_LABEL:
854 	case RAIDFRAME_SET_COMPONENT_LABEL:
855 	case RAIDFRAME_ADD_HOT_SPARE:
856 	case RAIDFRAME_REMOVE_HOT_SPARE:
857 	case RAIDFRAME_INIT_LABELS:
858 	case RAIDFRAME_REBUILD_IN_PLACE:
859 	case RAIDFRAME_CHECK_PARITY:
860 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
861 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
862 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
863 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
864 	case RAIDFRAME_SET_AUTOCONFIG:
865 	case RAIDFRAME_SET_ROOT:
866 	case RAIDFRAME_DELETE_COMPONENT:
867 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
868 		if ((rs->sc_flags & RAIDF_INITED) == 0)
869 			return (ENXIO);
870 	}
871 
872 	switch (cmd) {
873 
874 		/* configure the system */
875 	case RAIDFRAME_CONFIGURE:
876 
877 		if (raidPtr->valid) {
878 			/* There is a valid RAID set running on this unit! */
879 			printf("raid%d: Device already configured!\n",unit);
880 			return(EINVAL);
881 		}
882 
883 		/* copy-in the configuration information */
884 		/* data points to a pointer to the configuration structure */
885 
886 		u_cfg = *((RF_Config_t **) data);
887 		RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
888 		if (k_cfg == NULL) {
889 			return (ENOMEM);
890 		}
891 		retcode = copyin(u_cfg, k_cfg, sizeof(RF_Config_t));
892 		if (retcode) {
893 			RF_Free(k_cfg, sizeof(RF_Config_t));
894 			db1_printf(("rf_ioctl: retcode=%d copyin.1\n",
895 				retcode));
896 			return (retcode);
897 		}
898 		/* allocate a buffer for the layout-specific data, and copy it
899 		 * in */
900 		if (k_cfg->layoutSpecificSize) {
901 			if (k_cfg->layoutSpecificSize > 10000) {
902 				/* sanity check */
903 				RF_Free(k_cfg, sizeof(RF_Config_t));
904 				return (EINVAL);
905 			}
906 			RF_Malloc(specific_buf, k_cfg->layoutSpecificSize,
907 			    (u_char *));
908 			if (specific_buf == NULL) {
909 				RF_Free(k_cfg, sizeof(RF_Config_t));
910 				return (ENOMEM);
911 			}
912 			retcode = copyin(k_cfg->layoutSpecific, specific_buf,
913 			    k_cfg->layoutSpecificSize);
914 			if (retcode) {
915 				RF_Free(k_cfg, sizeof(RF_Config_t));
916 				RF_Free(specific_buf,
917 					k_cfg->layoutSpecificSize);
918 				db1_printf(("rf_ioctl: retcode=%d copyin.2\n",
919 					retcode));
920 				return (retcode);
921 			}
922 		} else
923 			specific_buf = NULL;
924 		k_cfg->layoutSpecific = specific_buf;
925 
926 		/* should do some kind of sanity check on the configuration.
927 		 * Store the sum of all the bytes in the last byte? */
928 
929 		/* configure the system */
930 
931 		/*
932 		 * Clear the entire RAID descriptor, just to make sure
933 		 *  there is no stale data left in the case of a
934 		 *  reconfiguration
935 		 */
936 		memset((char *) raidPtr, 0, sizeof(RF_Raid_t));
937 		raidPtr->raidid = unit;
938 
939 		retcode = rf_Configure(raidPtr, k_cfg, NULL);
940 
941 		if (retcode == 0) {
942 
943 			/* allow this many simultaneous IO's to
944 			   this RAID device */
945 			raidPtr->openings = RAIDOUTSTANDING;
946 
947 			raidinit(raidPtr);
948 			rf_markalldirty(raidPtr);
949 		}
950 		/* free the buffers.  No return code here. */
951 		if (k_cfg->layoutSpecificSize) {
952 			RF_Free(specific_buf, k_cfg->layoutSpecificSize);
953 		}
954 		RF_Free(k_cfg, sizeof(RF_Config_t));
955 
956 		return (retcode);
957 
958 		/* shutdown the system */
959 	case RAIDFRAME_SHUTDOWN:
960 
961 		if ((error = raidlock(rs)) != 0)
962 			return (error);
963 
964 		/*
965 		 * If somebody has a partition mounted, we shouldn't
966 		 * shutdown.
967 		 */
968 
969 		part = DISKPART(dev);
970 		pmask = (1 << part);
971 		if ((rs->sc_dkdev.dk_openmask & ~pmask) ||
972 		    ((rs->sc_dkdev.dk_bopenmask & pmask) &&
973 			(rs->sc_dkdev.dk_copenmask & pmask))) {
974 			raidunlock(rs);
975 			return (EBUSY);
976 		}
977 
978 		retcode = rf_Shutdown(raidPtr);
979 
980 		/* It's no longer initialized... */
981 		rs->sc_flags &= ~RAIDF_INITED;
982 
983 		/* Detach the disk. */
984 		disk_detach(&rs->sc_dkdev);
985 
986 		raidunlock(rs);
987 
988 		return (retcode);
989 	case RAIDFRAME_GET_COMPONENT_LABEL:
990 		clabel_ptr = (RF_ComponentLabel_t **) data;
991 		/* need to read the component label for the disk indicated
992 		   by row,column in clabel */
993 
994 		/* For practice, let's get it directly fromdisk, rather
995 		   than from the in-core copy */
996 		RF_Malloc( clabel, sizeof( RF_ComponentLabel_t ),
997 			   (RF_ComponentLabel_t *));
998 		if (clabel == NULL)
999 			return (ENOMEM);
1000 
1001 		memset((char *) clabel, 0, sizeof(RF_ComponentLabel_t));
1002 
1003 		retcode = copyin( *clabel_ptr, clabel,
1004 				  sizeof(RF_ComponentLabel_t));
1005 
1006 		if (retcode) {
1007 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1008 			return(retcode);
1009 		}
1010 
1011 		clabel->row = 0; /* Don't allow looking at anything else.*/
1012 
1013 		column = clabel->column;
1014 
1015 		if ((column < 0) || (column >= raidPtr->numCol +
1016 				     raidPtr->numSpare)) {
1017 			RF_Free( clabel, sizeof(RF_ComponentLabel_t));
1018 			return(EINVAL);
1019 		}
1020 
1021 		raidread_component_label(raidPtr->Disks[column].dev,
1022 				raidPtr->raid_cinfo[column].ci_vp,
1023 				clabel );
1024 
1025 		retcode = copyout(clabel, *clabel_ptr,
1026 				  sizeof(RF_ComponentLabel_t));
1027 		RF_Free(clabel, sizeof(RF_ComponentLabel_t));
1028 		return (retcode);
1029 
1030 	case RAIDFRAME_SET_COMPONENT_LABEL:
1031 		clabel = (RF_ComponentLabel_t *) data;
1032 
1033 		/* XXX check the label for valid stuff... */
1034 		/* Note that some things *should not* get modified --
1035 		   the user should be re-initing the labels instead of
1036 		   trying to patch things.
1037 		   */
1038 
1039 		raidid = raidPtr->raidid;
1040 #if DEBUG
1041 		printf("raid%d: Got component label:\n", raidid);
1042 		printf("raid%d: Version: %d\n", raidid, clabel->version);
1043 		printf("raid%d: Serial Number: %d\n", raidid, clabel->serial_number);
1044 		printf("raid%d: Mod counter: %d\n", raidid, clabel->mod_counter);
1045 		printf("raid%d: Column: %d\n", raidid, clabel->column);
1046 		printf("raid%d: Num Columns: %d\n", raidid, clabel->num_columns);
1047 		printf("raid%d: Clean: %d\n", raidid, clabel->clean);
1048 		printf("raid%d: Status: %d\n", raidid, clabel->status);
1049 #endif
1050 		clabel->row = 0;
1051 		column = clabel->column;
1052 
1053 		if ((column < 0) || (column >= raidPtr->numCol)) {
1054 			return(EINVAL);
1055 		}
1056 
1057 		/* XXX this isn't allowed to do anything for now :-) */
1058 
1059 		/* XXX and before it is, we need to fill in the rest
1060 		   of the fields!?!?!?! */
1061 #if 0
1062 		raidwrite_component_label(
1063                             raidPtr->Disks[column].dev,
1064 			    raidPtr->raid_cinfo[column].ci_vp,
1065 			    clabel );
1066 #endif
1067 		return (0);
1068 
1069 	case RAIDFRAME_INIT_LABELS:
1070 		clabel = (RF_ComponentLabel_t *) data;
1071 		/*
1072 		   we only want the serial number from
1073 		   the above.  We get all the rest of the information
1074 		   from the config that was used to create this RAID
1075 		   set.
1076 		   */
1077 
1078 		raidPtr->serial_number = clabel->serial_number;
1079 
1080 		raid_init_component_label(raidPtr, &ci_label);
1081 		ci_label.serial_number = clabel->serial_number;
1082 		ci_label.row = 0; /* we dont' pretend to support more */
1083 
1084 		for(column=0;column<raidPtr->numCol;column++) {
1085 			diskPtr = &raidPtr->Disks[column];
1086 			if (!RF_DEAD_DISK(diskPtr->status)) {
1087 				ci_label.partitionSize = diskPtr->partitionSize;
1088 				ci_label.column = column;
1089 				raidwrite_component_label(
1090 							  raidPtr->Disks[column].dev,
1091 							  raidPtr->raid_cinfo[column].ci_vp,
1092 							  &ci_label );
1093 			}
1094 		}
1095 
1096 		return (retcode);
1097 	case RAIDFRAME_SET_AUTOCONFIG:
1098 		d = rf_set_autoconfig(raidPtr, *(int *) data);
1099 		printf("raid%d: New autoconfig value is: %d\n",
1100 		       raidPtr->raidid, d);
1101 		*(int *) data = d;
1102 		return (retcode);
1103 
1104 	case RAIDFRAME_SET_ROOT:
1105 		d = rf_set_rootpartition(raidPtr, *(int *) data);
1106 		printf("raid%d: New rootpartition value is: %d\n",
1107 		       raidPtr->raidid, d);
1108 		*(int *) data = d;
1109 		return (retcode);
1110 
1111 		/* initialize all parity */
1112 	case RAIDFRAME_REWRITEPARITY:
1113 
1114 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1115 			/* Parity for RAID 0 is trivially correct */
1116 			raidPtr->parity_good = RF_RAID_CLEAN;
1117 			return(0);
1118 		}
1119 
1120 		if (raidPtr->parity_rewrite_in_progress == 1) {
1121 			/* Re-write is already in progress! */
1122 			return(EINVAL);
1123 		}
1124 
1125 		retcode = RF_CREATE_THREAD(raidPtr->parity_rewrite_thread,
1126 					   rf_RewriteParityThread,
1127 					   raidPtr,"raid_parity");
1128 		return (retcode);
1129 
1130 
1131 	case RAIDFRAME_ADD_HOT_SPARE:
1132 		sparePtr = (RF_SingleComponent_t *) data;
1133 		memcpy( &hot_spare, sparePtr, sizeof(RF_SingleComponent_t));
1134 		retcode = rf_add_hot_spare(raidPtr, &hot_spare);
1135 		return(retcode);
1136 
1137 	case RAIDFRAME_REMOVE_HOT_SPARE:
1138 		return(retcode);
1139 
1140 	case RAIDFRAME_DELETE_COMPONENT:
1141 		componentPtr = (RF_SingleComponent_t *)data;
1142 		memcpy( &component, componentPtr,
1143 			sizeof(RF_SingleComponent_t));
1144 		retcode = rf_delete_component(raidPtr, &component);
1145 		return(retcode);
1146 
1147 	case RAIDFRAME_INCORPORATE_HOT_SPARE:
1148 		componentPtr = (RF_SingleComponent_t *)data;
1149 		memcpy( &component, componentPtr,
1150 			sizeof(RF_SingleComponent_t));
1151 		retcode = rf_incorporate_hot_spare(raidPtr, &component);
1152 		return(retcode);
1153 
1154 	case RAIDFRAME_REBUILD_IN_PLACE:
1155 
1156 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1157 			/* Can't do this on a RAID 0!! */
1158 			return(EINVAL);
1159 		}
1160 
1161 		if (raidPtr->recon_in_progress == 1) {
1162 			/* a reconstruct is already in progress! */
1163 			return(EINVAL);
1164 		}
1165 
1166 		componentPtr = (RF_SingleComponent_t *) data;
1167 		memcpy( &component, componentPtr,
1168 			sizeof(RF_SingleComponent_t));
1169 		component.row = 0; /* we don't support any more */
1170 		column = component.column;
1171 
1172 		if ((column < 0) || (column >= raidPtr->numCol)) {
1173 			return(EINVAL);
1174 		}
1175 
1176 		RF_LOCK_MUTEX(raidPtr->mutex);
1177 		if ((raidPtr->Disks[column].status == rf_ds_optimal) &&
1178 		    (raidPtr->numFailures > 0)) {
1179 			/* XXX 0 above shouldn't be constant!!! */
1180 			/* some component other than this has failed.
1181 			   Let's not make things worse than they already
1182 			   are... */
1183 			printf("raid%d: Unable to reconstruct to disk at:\n",
1184 			       raidPtr->raidid);
1185 			printf("raid%d:     Col: %d   Too many failures.\n",
1186 			       raidPtr->raidid, column);
1187 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1188 			return (EINVAL);
1189 		}
1190 		if (raidPtr->Disks[column].status ==
1191 		    rf_ds_reconstructing) {
1192 			printf("raid%d: Unable to reconstruct to disk at:\n",
1193 			       raidPtr->raidid);
1194 			printf("raid%d:    Col: %d   Reconstruction already occuring!\n", raidPtr->raidid, column);
1195 
1196 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1197 			return (EINVAL);
1198 		}
1199 		if (raidPtr->Disks[column].status == rf_ds_spared) {
1200 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1201 			return (EINVAL);
1202 		}
1203 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1204 
1205 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1206 		if (rrcopy == NULL)
1207 			return(ENOMEM);
1208 
1209 		rrcopy->raidPtr = (void *) raidPtr;
1210 		rrcopy->col = column;
1211 
1212 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1213 					   rf_ReconstructInPlaceThread,
1214 					   rrcopy,"raid_reconip");
1215 		return(retcode);
1216 
1217 	case RAIDFRAME_GET_INFO:
1218 		if (!raidPtr->valid)
1219 			return (ENODEV);
1220 		ucfgp = (RF_DeviceConfig_t **) data;
1221 		RF_Malloc(d_cfg, sizeof(RF_DeviceConfig_t),
1222 			  (RF_DeviceConfig_t *));
1223 		if (d_cfg == NULL)
1224 			return (ENOMEM);
1225 		memset((char *) d_cfg, 0, sizeof(RF_DeviceConfig_t));
1226 		d_cfg->rows = 1; /* there is only 1 row now */
1227 		d_cfg->cols = raidPtr->numCol;
1228 		d_cfg->ndevs = raidPtr->numCol;
1229 		if (d_cfg->ndevs >= RF_MAX_DISKS) {
1230 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1231 			return (ENOMEM);
1232 		}
1233 		d_cfg->nspares = raidPtr->numSpare;
1234 		if (d_cfg->nspares >= RF_MAX_DISKS) {
1235 			RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1236 			return (ENOMEM);
1237 		}
1238 		d_cfg->maxqdepth = raidPtr->maxQueueDepth;
1239 		d = 0;
1240 		for (j = 0; j < d_cfg->cols; j++) {
1241 			d_cfg->devs[d] = raidPtr->Disks[j];
1242 			d++;
1243 		}
1244 		for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++) {
1245 			d_cfg->spares[i] = raidPtr->Disks[j];
1246 		}
1247 		retcode = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig_t));
1248 		RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
1249 
1250 		return (retcode);
1251 
1252 	case RAIDFRAME_CHECK_PARITY:
1253 		*(int *) data = raidPtr->parity_good;
1254 		return (0);
1255 
1256 	case RAIDFRAME_RESET_ACCTOTALS:
1257 		memset(&raidPtr->acc_totals, 0, sizeof(raidPtr->acc_totals));
1258 		return (0);
1259 
1260 	case RAIDFRAME_GET_ACCTOTALS:
1261 		totals = (RF_AccTotals_t *) data;
1262 		*totals = raidPtr->acc_totals;
1263 		return (0);
1264 
1265 	case RAIDFRAME_KEEP_ACCTOTALS:
1266 		raidPtr->keep_acc_totals = *(int *)data;
1267 		return (0);
1268 
1269 	case RAIDFRAME_GET_SIZE:
1270 		*(int *) data = raidPtr->totalSectors;
1271 		return (0);
1272 
1273 		/* fail a disk & optionally start reconstruction */
1274 	case RAIDFRAME_FAIL_DISK:
1275 
1276 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1277 			/* Can't do this on a RAID 0!! */
1278 			return(EINVAL);
1279 		}
1280 
1281 		rr = (struct rf_recon_req *) data;
1282 		rr->row = 0;
1283 		if (rr->col < 0 || rr->col >= raidPtr->numCol)
1284 			return (EINVAL);
1285 
1286 
1287 		RF_LOCK_MUTEX(raidPtr->mutex);
1288 		if (raidPtr->status == rf_rs_reconstructing) {
1289 			/* you can't fail a disk while we're reconstructing! */
1290 			/* XXX wrong for RAID6 */
1291 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1292 			return (EINVAL);
1293 		}
1294 		if ((raidPtr->Disks[rr->col].status ==
1295 		     rf_ds_optimal) && (raidPtr->numFailures > 0)) {
1296 			/* some other component has failed.  Let's not make
1297 			   things worse. XXX wrong for RAID6 */
1298 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1299 			return (EINVAL);
1300 		}
1301 		if (raidPtr->Disks[rr->col].status == rf_ds_spared) {
1302 			/* Can't fail a spared disk! */
1303 			RF_UNLOCK_MUTEX(raidPtr->mutex);
1304 			return (EINVAL);
1305 		}
1306 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1307 
1308 		/* make a copy of the recon request so that we don't rely on
1309 		 * the user's buffer */
1310 		RF_Malloc(rrcopy, sizeof(*rrcopy), (struct rf_recon_req *));
1311 		if (rrcopy == NULL)
1312 			return(ENOMEM);
1313 		memcpy(rrcopy, rr, sizeof(*rr));
1314 		rrcopy->raidPtr = (void *) raidPtr;
1315 
1316 		retcode = RF_CREATE_THREAD(raidPtr->recon_thread,
1317 					   rf_ReconThread,
1318 					   rrcopy,"raid_recon");
1319 		return (0);
1320 
1321 		/* invoke a copyback operation after recon on whatever disk
1322 		 * needs it, if any */
1323 	case RAIDFRAME_COPYBACK:
1324 
1325 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1326 			/* This makes no sense on a RAID 0!! */
1327 			return(EINVAL);
1328 		}
1329 
1330 		if (raidPtr->copyback_in_progress == 1) {
1331 			/* Copyback is already in progress! */
1332 			return(EINVAL);
1333 		}
1334 
1335 		retcode = RF_CREATE_THREAD(raidPtr->copyback_thread,
1336 					   rf_CopybackThread,
1337 					   raidPtr,"raid_copyback");
1338 		return (retcode);
1339 
1340 		/* return the percentage completion of reconstruction */
1341 	case RAIDFRAME_CHECK_RECON_STATUS:
1342 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1343 			/* This makes no sense on a RAID 0, so tell the
1344 			   user it's done. */
1345 			*(int *) data = 100;
1346 			return(0);
1347 		}
1348 		if (raidPtr->status != rf_rs_reconstructing)
1349 			*(int *) data = 100;
1350 		else {
1351 			if (raidPtr->reconControl->numRUsTotal > 0) {
1352 				*(int *) data = (raidPtr->reconControl->numRUsComplete * 100 / raidPtr->reconControl->numRUsTotal);
1353 			} else {
1354 				*(int *) data = 0;
1355 			}
1356 		}
1357 		return (0);
1358 	case RAIDFRAME_CHECK_RECON_STATUS_EXT:
1359 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1360 		if (raidPtr->status != rf_rs_reconstructing) {
1361 			progressInfo.remaining = 0;
1362 			progressInfo.completed = 100;
1363 			progressInfo.total = 100;
1364 		} else {
1365 			progressInfo.total =
1366 				raidPtr->reconControl->numRUsTotal;
1367 			progressInfo.completed =
1368 				raidPtr->reconControl->numRUsComplete;
1369 			progressInfo.remaining = progressInfo.total -
1370 				progressInfo.completed;
1371 		}
1372 		retcode = copyout(&progressInfo, *progressInfoPtr,
1373 				  sizeof(RF_ProgressInfo_t));
1374 		return (retcode);
1375 
1376 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS:
1377 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1378 			/* This makes no sense on a RAID 0, so tell the
1379 			   user it's done. */
1380 			*(int *) data = 100;
1381 			return(0);
1382 		}
1383 		if (raidPtr->parity_rewrite_in_progress == 1) {
1384 			*(int *) data = 100 *
1385 				raidPtr->parity_rewrite_stripes_done /
1386 				raidPtr->Layout.numStripe;
1387 		} else {
1388 			*(int *) data = 100;
1389 		}
1390 		return (0);
1391 
1392 	case RAIDFRAME_CHECK_PARITYREWRITE_STATUS_EXT:
1393 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1394 		if (raidPtr->parity_rewrite_in_progress == 1) {
1395 			progressInfo.total = raidPtr->Layout.numStripe;
1396 			progressInfo.completed =
1397 				raidPtr->parity_rewrite_stripes_done;
1398 			progressInfo.remaining = progressInfo.total -
1399 				progressInfo.completed;
1400 		} else {
1401 			progressInfo.remaining = 0;
1402 			progressInfo.completed = 100;
1403 			progressInfo.total = 100;
1404 		}
1405 		retcode = copyout(&progressInfo, *progressInfoPtr,
1406 				  sizeof(RF_ProgressInfo_t));
1407 		return (retcode);
1408 
1409 	case RAIDFRAME_CHECK_COPYBACK_STATUS:
1410 		if (raidPtr->Layout.map->faultsTolerated == 0) {
1411 			/* This makes no sense on a RAID 0 */
1412 			*(int *) data = 100;
1413 			return(0);
1414 		}
1415 		if (raidPtr->copyback_in_progress == 1) {
1416 			*(int *) data = 100 * raidPtr->copyback_stripes_done /
1417 				raidPtr->Layout.numStripe;
1418 		} else {
1419 			*(int *) data = 100;
1420 		}
1421 		return (0);
1422 
1423 	case RAIDFRAME_CHECK_COPYBACK_STATUS_EXT:
1424 		progressInfoPtr = (RF_ProgressInfo_t **) data;
1425 		if (raidPtr->copyback_in_progress == 1) {
1426 			progressInfo.total = raidPtr->Layout.numStripe;
1427 			progressInfo.completed =
1428 				raidPtr->copyback_stripes_done;
1429 			progressInfo.remaining = progressInfo.total -
1430 				progressInfo.completed;
1431 		} else {
1432 			progressInfo.remaining = 0;
1433 			progressInfo.completed = 100;
1434 			progressInfo.total = 100;
1435 		}
1436 		retcode = copyout(&progressInfo, *progressInfoPtr,
1437 				  sizeof(RF_ProgressInfo_t));
1438 		return (retcode);
1439 
1440 		/* the sparetable daemon calls this to wait for the kernel to
1441 		 * need a spare table. this ioctl does not return until a
1442 		 * spare table is needed. XXX -- calling mpsleep here in the
1443 		 * ioctl code is almost certainly wrong and evil. -- XXX XXX
1444 		 * -- I should either compute the spare table in the kernel,
1445 		 * or have a different -- XXX XXX -- interface (a different
1446 		 * character device) for delivering the table     -- XXX */
1447 #if 0
1448 	case RAIDFRAME_SPARET_WAIT:
1449 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1450 		while (!rf_sparet_wait_queue)
1451 			mpsleep(&rf_sparet_wait_queue, (PZERO + 1) | PCATCH, "sparet wait", 0, (void *) simple_lock_addr(rf_sparet_wait_mutex), MS_LOCK_SIMPLE);
1452 		waitreq = rf_sparet_wait_queue;
1453 		rf_sparet_wait_queue = rf_sparet_wait_queue->next;
1454 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1455 
1456 		/* structure assignment */
1457 		*((RF_SparetWait_t *) data) = *waitreq;
1458 
1459 		RF_Free(waitreq, sizeof(*waitreq));
1460 		return (0);
1461 
1462 		/* wakes up a process waiting on SPARET_WAIT and puts an error
1463 		 * code in it that will cause the dameon to exit */
1464 	case RAIDFRAME_ABORT_SPARET_WAIT:
1465 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1466 		waitreq->fcol = -1;
1467 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1468 		waitreq->next = rf_sparet_wait_queue;
1469 		rf_sparet_wait_queue = waitreq;
1470 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1471 		wakeup(&rf_sparet_wait_queue);
1472 		return (0);
1473 
1474 		/* used by the spare table daemon to deliver a spare table
1475 		 * into the kernel */
1476 	case RAIDFRAME_SEND_SPARET:
1477 
1478 		/* install the spare table */
1479 		retcode = rf_SetSpareTable(raidPtr, *(void **) data);
1480 
1481 		/* respond to the requestor.  the return status of the spare
1482 		 * table installation is passed in the "fcol" field */
1483 		RF_Malloc(waitreq, sizeof(*waitreq), (RF_SparetWait_t *));
1484 		waitreq->fcol = retcode;
1485 		RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1486 		waitreq->next = rf_sparet_resp_queue;
1487 		rf_sparet_resp_queue = waitreq;
1488 		wakeup(&rf_sparet_resp_queue);
1489 		RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1490 
1491 		return (retcode);
1492 #endif
1493 
1494 	default:
1495 		break; /* fall through to the os-specific code below */
1496 
1497 	}
1498 
1499 	if (!raidPtr->valid)
1500 		return (EINVAL);
1501 
1502 	/*
1503 	 * Add support for "regular" device ioctls here.
1504 	 */
1505 
1506 	switch (cmd) {
1507 	case DIOCGDINFO:
1508 		*(struct disklabel *) data = *(rs->sc_dkdev.dk_label);
1509 		break;
1510 #ifdef __HAVE_OLD_DISKLABEL
1511 	case ODIOCGDINFO:
1512 		newlabel = *(rs->sc_dkdev.dk_label);
1513 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1514 			return ENOTTY;
1515 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1516 		break;
1517 #endif
1518 
1519 	case DIOCGPART:
1520 		((struct partinfo *) data)->disklab = rs->sc_dkdev.dk_label;
1521 		((struct partinfo *) data)->part =
1522 		    &rs->sc_dkdev.dk_label->d_partitions[DISKPART(dev)];
1523 		break;
1524 
1525 	case DIOCWDINFO:
1526 	case DIOCSDINFO:
1527 #ifdef __HAVE_OLD_DISKLABEL
1528 	case ODIOCWDINFO:
1529 	case ODIOCSDINFO:
1530 #endif
1531 	{
1532 		struct disklabel *lp;
1533 #ifdef __HAVE_OLD_DISKLABEL
1534 		if (cmd == ODIOCSDINFO || cmd == ODIOCWDINFO) {
1535 			memset(&newlabel, 0, sizeof newlabel);
1536 			memcpy(&newlabel, data, sizeof (struct olddisklabel));
1537 			lp = &newlabel;
1538 		} else
1539 #endif
1540 		lp = (struct disklabel *)data;
1541 
1542 		if ((error = raidlock(rs)) != 0)
1543 			return (error);
1544 
1545 		rs->sc_flags |= RAIDF_LABELLING;
1546 
1547 		error = setdisklabel(rs->sc_dkdev.dk_label,
1548 		    lp, 0, rs->sc_dkdev.dk_cpulabel);
1549 		if (error == 0) {
1550 			if (cmd == DIOCWDINFO
1551 #ifdef __HAVE_OLD_DISKLABEL
1552 			    || cmd == ODIOCWDINFO
1553 #endif
1554 			   )
1555 				error = writedisklabel(RAIDLABELDEV(dev),
1556 				    raidstrategy, rs->sc_dkdev.dk_label,
1557 				    rs->sc_dkdev.dk_cpulabel);
1558 		}
1559 		rs->sc_flags &= ~RAIDF_LABELLING;
1560 
1561 		raidunlock(rs);
1562 
1563 		if (error)
1564 			return (error);
1565 		break;
1566 	}
1567 
1568 	case DIOCWLABEL:
1569 		if (*(int *) data != 0)
1570 			rs->sc_flags |= RAIDF_WLABEL;
1571 		else
1572 			rs->sc_flags &= ~RAIDF_WLABEL;
1573 		break;
1574 
1575 	case DIOCGDEFLABEL:
1576 		raidgetdefaultlabel(raidPtr, rs, (struct disklabel *) data);
1577 		break;
1578 
1579 #ifdef __HAVE_OLD_DISKLABEL
1580 	case ODIOCGDEFLABEL:
1581 		raidgetdefaultlabel(raidPtr, rs, &newlabel);
1582 		if (newlabel.d_npartitions > OLDMAXPARTITIONS)
1583 			return ENOTTY;
1584 		memcpy(data, &newlabel, sizeof (struct olddisklabel));
1585 		break;
1586 #endif
1587 
1588 	default:
1589 		retcode = ENOTTY;
1590 	}
1591 	return (retcode);
1592 
1593 }
1594 
1595 
1596 /* raidinit -- complete the rest of the initialization for the
1597    RAIDframe device.  */
1598 
1599 
1600 static void
1601 raidinit(RF_Raid_t *raidPtr)
1602 {
1603 	struct raid_softc *rs;
1604 	int     unit;
1605 
1606 	unit = raidPtr->raidid;
1607 
1608 	rs = &raid_softc[unit];
1609 
1610 	/* XXX should check return code first... */
1611 	rs->sc_flags |= RAIDF_INITED;
1612 
1613 	/* XXX doesn't check bounds. */
1614 	snprintf(rs->sc_xname, sizeof(rs->sc_xname), "raid%d", unit);
1615 
1616 	rs->sc_dkdev.dk_name = rs->sc_xname;
1617 
1618 	/* disk_attach actually creates space for the CPU disklabel, among
1619 	 * other things, so it's critical to call this *BEFORE* we try putzing
1620 	 * with disklabels. */
1621 
1622 	disk_attach(&rs->sc_dkdev);
1623 
1624 	/* XXX There may be a weird interaction here between this, and
1625 	 * protectedSectors, as used in RAIDframe.  */
1626 
1627 	rs->sc_size = raidPtr->totalSectors;
1628 }
1629 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0)
1630 /* wake up the daemon & tell it to get us a spare table
1631  * XXX
1632  * the entries in the queues should be tagged with the raidPtr
1633  * so that in the extremely rare case that two recons happen at once,
1634  * we know for which device were requesting a spare table
1635  * XXX
1636  *
1637  * XXX This code is not currently used. GO
1638  */
1639 int
1640 rf_GetSpareTableFromDaemon(RF_SparetWait_t *req)
1641 {
1642 	int     retcode;
1643 
1644 	RF_LOCK_MUTEX(rf_sparet_wait_mutex);
1645 	req->next = rf_sparet_wait_queue;
1646 	rf_sparet_wait_queue = req;
1647 	wakeup(&rf_sparet_wait_queue);
1648 
1649 	/* mpsleep unlocks the mutex */
1650 	while (!rf_sparet_resp_queue) {
1651 		tsleep(&rf_sparet_resp_queue, PRIBIO,
1652 		    "raidframe getsparetable", 0);
1653 	}
1654 	req = rf_sparet_resp_queue;
1655 	rf_sparet_resp_queue = req->next;
1656 	RF_UNLOCK_MUTEX(rf_sparet_wait_mutex);
1657 
1658 	retcode = req->fcol;
1659 	RF_Free(req, sizeof(*req));	/* this is not the same req as we
1660 					 * alloc'd */
1661 	return (retcode);
1662 }
1663 #endif
1664 
1665 /* a wrapper around rf_DoAccess that extracts appropriate info from the
1666  * bp & passes it down.
1667  * any calls originating in the kernel must use non-blocking I/O
1668  * do some extra sanity checking to return "appropriate" error values for
1669  * certain conditions (to make some standard utilities work)
1670  *
1671  * Formerly known as: rf_DoAccessKernel
1672  */
1673 void
1674 raidstart(RF_Raid_t *raidPtr)
1675 {
1676 	RF_SectorCount_t num_blocks, pb, sum;
1677 	RF_RaidAddr_t raid_addr;
1678 	struct partition *pp;
1679 	daddr_t blocknum;
1680 	int     unit;
1681 	struct raid_softc *rs;
1682 	int     do_async;
1683 	struct buf *bp;
1684 	int rc;
1685 
1686 	unit = raidPtr->raidid;
1687 	rs = &raid_softc[unit];
1688 
1689 	/* quick check to see if anything has died recently */
1690 	RF_LOCK_MUTEX(raidPtr->mutex);
1691 	if (raidPtr->numNewFailures > 0) {
1692 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1693 		rf_update_component_labels(raidPtr,
1694 					   RF_NORMAL_COMPONENT_UPDATE);
1695 		RF_LOCK_MUTEX(raidPtr->mutex);
1696 		raidPtr->numNewFailures--;
1697 	}
1698 
1699 	/* Check to see if we're at the limit... */
1700 	while (raidPtr->openings > 0) {
1701 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1702 
1703 		/* get the next item, if any, from the queue */
1704 		if ((bp = BUFQ_GET(&rs->buf_queue)) == NULL) {
1705 			/* nothing more to do */
1706 			return;
1707 		}
1708 
1709 		/* Ok, for the bp we have here, bp->b_blkno is relative to the
1710 		 * partition.. Need to make it absolute to the underlying
1711 		 * device.. */
1712 
1713 		blocknum = bp->b_blkno;
1714 		if (DISKPART(bp->b_dev) != RAW_PART) {
1715 			pp = &rs->sc_dkdev.dk_label->d_partitions[DISKPART(bp->b_dev)];
1716 			blocknum += pp->p_offset;
1717 		}
1718 
1719 		db1_printf(("Blocks: %d, %d\n", (int) bp->b_blkno,
1720 			    (int) blocknum));
1721 
1722 		db1_printf(("bp->b_bcount = %d\n", (int) bp->b_bcount));
1723 		db1_printf(("bp->b_resid = %d\n", (int) bp->b_resid));
1724 
1725 		/* *THIS* is where we adjust what block we're going to...
1726 		 * but DO NOT TOUCH bp->b_blkno!!! */
1727 		raid_addr = blocknum;
1728 
1729 		num_blocks = bp->b_bcount >> raidPtr->logBytesPerSector;
1730 		pb = (bp->b_bcount & raidPtr->sectorMask) ? 1 : 0;
1731 		sum = raid_addr + num_blocks + pb;
1732 		if (1 || rf_debugKernelAccess) {
1733 			db1_printf(("raid_addr=%d sum=%d num_blocks=%d(+%d) (%d)\n",
1734 				    (int) raid_addr, (int) sum, (int) num_blocks,
1735 				    (int) pb, (int) bp->b_resid));
1736 		}
1737 		if ((sum > raidPtr->totalSectors) || (sum < raid_addr)
1738 		    || (sum < num_blocks) || (sum < pb)) {
1739 			bp->b_error = ENOSPC;
1740 			bp->b_flags |= B_ERROR;
1741 			bp->b_resid = bp->b_bcount;
1742 			biodone(bp);
1743 			RF_LOCK_MUTEX(raidPtr->mutex);
1744 			continue;
1745 		}
1746 		/*
1747 		 * XXX rf_DoAccess() should do this, not just DoAccessKernel()
1748 		 */
1749 
1750 		if (bp->b_bcount & raidPtr->sectorMask) {
1751 			bp->b_error = EINVAL;
1752 			bp->b_flags |= B_ERROR;
1753 			bp->b_resid = bp->b_bcount;
1754 			biodone(bp);
1755 			RF_LOCK_MUTEX(raidPtr->mutex);
1756 			continue;
1757 
1758 		}
1759 		db1_printf(("Calling DoAccess..\n"));
1760 
1761 
1762 		RF_LOCK_MUTEX(raidPtr->mutex);
1763 		raidPtr->openings--;
1764 		RF_UNLOCK_MUTEX(raidPtr->mutex);
1765 
1766 		/*
1767 		 * Everything is async.
1768 		 */
1769 		do_async = 1;
1770 
1771 		disk_busy(&rs->sc_dkdev);
1772 
1773 		/* XXX we're still at splbio() here... do we *really*
1774 		   need to be? */
1775 
1776 		/* don't ever condition on bp->b_flags & B_WRITE.
1777 		 * always condition on B_READ instead */
1778 
1779 		rc = rf_DoAccess(raidPtr, (bp->b_flags & B_READ) ?
1780 				 RF_IO_TYPE_READ : RF_IO_TYPE_WRITE,
1781 				 do_async, raid_addr, num_blocks,
1782 				 bp->b_data, bp, RF_DAG_NONBLOCKING_IO);
1783 
1784 		if (rc) {
1785 			bp->b_error = rc;
1786 			bp->b_flags |= B_ERROR;
1787 			bp->b_resid = bp->b_bcount;
1788 			biodone(bp);
1789 			/* continue loop */
1790 		}
1791 
1792 		RF_LOCK_MUTEX(raidPtr->mutex);
1793 	}
1794 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1795 }
1796 
1797 
1798 
1799 
1800 /* invoke an I/O from kernel mode.  Disk queue should be locked upon entry */
1801 
1802 int
1803 rf_DispatchKernelIO(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req)
1804 {
1805 	int     op = (req->type == RF_IO_TYPE_READ) ? B_READ : B_WRITE;
1806 	struct buf *bp;
1807 	struct raidbuf *raidbp = NULL;
1808 
1809 	req->queue = queue;
1810 
1811 #if DIAGNOSTIC
1812 	if (queue->raidPtr->raidid >= numraid) {
1813 		printf("Invalid unit number: %d %d\n", queue->raidPtr->raidid,
1814 		    numraid);
1815 		panic("Invalid Unit number in rf_DispatchKernelIO");
1816 	}
1817 #endif
1818 
1819 	bp = req->bp;
1820 #if 1
1821 	/* XXX when there is a physical disk failure, someone is passing us a
1822 	 * buffer that contains old stuff!!  Attempt to deal with this problem
1823 	 * without taking a performance hit... (not sure where the real bug
1824 	 * is.  It's buried in RAIDframe somewhere) :-(  GO ) */
1825 
1826 	if (bp->b_flags & B_ERROR) {
1827 		bp->b_flags &= ~B_ERROR;
1828 	}
1829 	if (bp->b_error != 0) {
1830 		bp->b_error = 0;
1831 	}
1832 #endif
1833 	raidbp = pool_get(&rf_pools.cbuf, PR_NOWAIT);
1834 	if (raidbp == NULL) {
1835 		bp->b_flags |= B_ERROR;
1836 		bp->b_error = ENOMEM;
1837 		return (ENOMEM);
1838 	}
1839 	BUF_INIT(&raidbp->rf_buf);
1840 
1841 	/*
1842 	 * context for raidiodone
1843 	 */
1844 	raidbp->rf_obp = bp;
1845 	raidbp->req = req;
1846 
1847 	BIO_COPYPRIO(&raidbp->rf_buf, bp);
1848 
1849 	switch (req->type) {
1850 	case RF_IO_TYPE_NOP:	/* used primarily to unlock a locked queue */
1851 		/* XXX need to do something extra here.. */
1852 		/* I'm leaving this in, as I've never actually seen it used,
1853 		 * and I'd like folks to report it... GO */
1854 		printf(("WAKEUP CALLED\n"));
1855 		queue->numOutstanding++;
1856 
1857 		/* XXX need to glue the original buffer into this??  */
1858 
1859 		KernelWakeupFunc(&raidbp->rf_buf);
1860 		break;
1861 
1862 	case RF_IO_TYPE_READ:
1863 	case RF_IO_TYPE_WRITE:
1864 #if RF_ACC_TRACE > 0
1865 		if (req->tracerec) {
1866 			RF_ETIMER_START(req->tracerec->timer);
1867 		}
1868 #endif
1869 		InitBP(&raidbp->rf_buf, queue->rf_cinfo->ci_vp,
1870 		    op | bp->b_flags, queue->rf_cinfo->ci_dev,
1871 		    req->sectorOffset, req->numSector,
1872 		    req->buf, KernelWakeupFunc, (void *) req,
1873 		    queue->raidPtr->logBytesPerSector, req->b_proc);
1874 
1875 		if (rf_debugKernelAccess) {
1876 			db1_printf(("dispatch: bp->b_blkno = %ld\n",
1877 				(long) bp->b_blkno));
1878 		}
1879 		queue->numOutstanding++;
1880 		queue->last_deq_sector = req->sectorOffset;
1881 		/* acc wouldn't have been let in if there were any pending
1882 		 * reqs at any other priority */
1883 		queue->curPriority = req->priority;
1884 
1885 		db1_printf(("Going for %c to unit %d col %d\n",
1886 			    req->type, queue->raidPtr->raidid,
1887 			    queue->col));
1888 		db1_printf(("sector %d count %d (%d bytes) %d\n",
1889 			(int) req->sectorOffset, (int) req->numSector,
1890 			(int) (req->numSector <<
1891 			    queue->raidPtr->logBytesPerSector),
1892 			(int) queue->raidPtr->logBytesPerSector));
1893 		if ((raidbp->rf_buf.b_flags & B_READ) == 0) {
1894 			raidbp->rf_buf.b_vp->v_numoutput++;
1895 		}
1896 		VOP_STRATEGY(raidbp->rf_buf.b_vp, &raidbp->rf_buf);
1897 
1898 		break;
1899 
1900 	default:
1901 		panic("bad req->type in rf_DispatchKernelIO");
1902 	}
1903 	db1_printf(("Exiting from DispatchKernelIO\n"));
1904 
1905 	return (0);
1906 }
1907 /* this is the callback function associated with a I/O invoked from
1908    kernel code.
1909  */
1910 static void
1911 KernelWakeupFunc(struct buf *vbp)
1912 {
1913 	RF_DiskQueueData_t *req = NULL;
1914 	RF_DiskQueue_t *queue;
1915 	struct raidbuf *raidbp = (struct raidbuf *) vbp;
1916 	struct buf *bp;
1917 	int s;
1918 
1919 	s = splbio();
1920 	db1_printf(("recovering the request queue:\n"));
1921 	req = raidbp->req;
1922 
1923 	bp = raidbp->rf_obp;
1924 
1925 	queue = (RF_DiskQueue_t *) req->queue;
1926 
1927 	if (raidbp->rf_buf.b_flags & B_ERROR) {
1928 		bp->b_flags |= B_ERROR;
1929 		bp->b_error = raidbp->rf_buf.b_error ?
1930 		    raidbp->rf_buf.b_error : EIO;
1931 	}
1932 
1933 	/* XXX methinks this could be wrong... */
1934 #if 1
1935 	bp->b_resid = raidbp->rf_buf.b_resid;
1936 #endif
1937 #if RF_ACC_TRACE > 0
1938 	if (req->tracerec) {
1939 		RF_ETIMER_STOP(req->tracerec->timer);
1940 		RF_ETIMER_EVAL(req->tracerec->timer);
1941 		RF_LOCK_MUTEX(rf_tracing_mutex);
1942 		req->tracerec->diskwait_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1943 		req->tracerec->phys_io_us += RF_ETIMER_VAL_US(req->tracerec->timer);
1944 		req->tracerec->num_phys_ios++;
1945 		RF_UNLOCK_MUTEX(rf_tracing_mutex);
1946 	}
1947 #endif
1948 	bp->b_bcount = raidbp->rf_buf.b_bcount;	/* XXXX ?? */
1949 
1950 	/* XXX Ok, let's get aggressive... If B_ERROR is set, let's go
1951 	 * ballistic, and mark the component as hosed... */
1952 
1953 	if (bp->b_flags & B_ERROR) {
1954 		/* Mark the disk as dead */
1955 		/* but only mark it once... */
1956 		/* and only if it wouldn't leave this RAID set
1957 		   completely broken */
1958 		if ((queue->raidPtr->Disks[queue->col].status ==
1959 		    rf_ds_optimal) && (queue->raidPtr->numFailures <
1960 				       queue->raidPtr->Layout.map->faultsTolerated)) {
1961 			printf("raid%d: IO Error.  Marking %s as failed.\n",
1962 			       queue->raidPtr->raidid,
1963 			       queue->raidPtr->Disks[queue->col].devname);
1964 			queue->raidPtr->Disks[queue->col].status =
1965 			    rf_ds_failed;
1966 			queue->raidPtr->status = rf_rs_degraded;
1967 			queue->raidPtr->numFailures++;
1968 			queue->raidPtr->numNewFailures++;
1969 		} else {	/* Disk is already dead... */
1970 			/* printf("Disk already marked as dead!\n"); */
1971 		}
1972 
1973 	}
1974 
1975 	pool_put(&rf_pools.cbuf, raidbp);
1976 
1977 	/* Fill in the error value */
1978 
1979 	req->error = (bp->b_flags & B_ERROR) ? bp->b_error : 0;
1980 
1981 	simple_lock(&queue->raidPtr->iodone_lock);
1982 
1983 	/* Drop this one on the "finished" queue... */
1984 	TAILQ_INSERT_TAIL(&(queue->raidPtr->iodone), req, iodone_entries);
1985 
1986 	/* Let the raidio thread know there is work to be done. */
1987 	wakeup(&(queue->raidPtr->iodone));
1988 
1989 	simple_unlock(&queue->raidPtr->iodone_lock);
1990 
1991 	splx(s);
1992 }
1993 
1994 
1995 
1996 /*
1997  * initialize a buf structure for doing an I/O in the kernel.
1998  */
1999 static void
2000 InitBP(struct buf *bp, struct vnode *b_vp, unsigned rw_flag, dev_t dev,
2001        RF_SectorNum_t startSect, RF_SectorCount_t numSect, caddr_t buf,
2002        void (*cbFunc) (struct buf *), void *cbArg, int logBytesPerSector,
2003        struct proc *b_proc)
2004 {
2005 	/* bp->b_flags       = B_PHYS | rw_flag; */
2006 	bp->b_flags = B_CALL | rw_flag;	/* XXX need B_PHYS here too??? */
2007 	bp->b_bcount = numSect << logBytesPerSector;
2008 	bp->b_bufsize = bp->b_bcount;
2009 	bp->b_error = 0;
2010 	bp->b_dev = dev;
2011 	bp->b_data = buf;
2012 	bp->b_blkno = startSect;
2013 	bp->b_resid = bp->b_bcount;	/* XXX is this right!??!?!! */
2014 	if (bp->b_bcount == 0) {
2015 		panic("bp->b_bcount is zero in InitBP!!");
2016 	}
2017 	bp->b_proc = b_proc;
2018 	bp->b_iodone = cbFunc;
2019 	bp->b_vp = b_vp;
2020 
2021 }
2022 
2023 static void
2024 raidgetdefaultlabel(RF_Raid_t *raidPtr, struct raid_softc *rs,
2025 		    struct disklabel *lp)
2026 {
2027 	memset(lp, 0, sizeof(*lp));
2028 
2029 	/* fabricate a label... */
2030 	lp->d_secperunit = raidPtr->totalSectors;
2031 	lp->d_secsize = raidPtr->bytesPerSector;
2032 	lp->d_nsectors = raidPtr->Layout.dataSectorsPerStripe;
2033 	lp->d_ntracks = 4 * raidPtr->numCol;
2034 	lp->d_ncylinders = raidPtr->totalSectors /
2035 		(lp->d_nsectors * lp->d_ntracks);
2036 	lp->d_secpercyl = lp->d_ntracks * lp->d_nsectors;
2037 
2038 	strncpy(lp->d_typename, "raid", sizeof(lp->d_typename));
2039 	lp->d_type = DTYPE_RAID;
2040 	strncpy(lp->d_packname, "fictitious", sizeof(lp->d_packname));
2041 	lp->d_rpm = 3600;
2042 	lp->d_interleave = 1;
2043 	lp->d_flags = 0;
2044 
2045 	lp->d_partitions[RAW_PART].p_offset = 0;
2046 	lp->d_partitions[RAW_PART].p_size = raidPtr->totalSectors;
2047 	lp->d_partitions[RAW_PART].p_fstype = FS_UNUSED;
2048 	lp->d_npartitions = RAW_PART + 1;
2049 
2050 	lp->d_magic = DISKMAGIC;
2051 	lp->d_magic2 = DISKMAGIC;
2052 	lp->d_checksum = dkcksum(rs->sc_dkdev.dk_label);
2053 
2054 }
2055 /*
2056  * Read the disklabel from the raid device.  If one is not present, fake one
2057  * up.
2058  */
2059 static void
2060 raidgetdisklabel(dev_t dev)
2061 {
2062 	int     unit = raidunit(dev);
2063 	struct raid_softc *rs = &raid_softc[unit];
2064 	const char   *errstring;
2065 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2066 	struct cpu_disklabel *clp = rs->sc_dkdev.dk_cpulabel;
2067 	RF_Raid_t *raidPtr;
2068 
2069 	db1_printf(("Getting the disklabel...\n"));
2070 
2071 	memset(clp, 0, sizeof(*clp));
2072 
2073 	raidPtr = raidPtrs[unit];
2074 
2075 	raidgetdefaultlabel(raidPtr, rs, lp);
2076 
2077 	/*
2078 	 * Call the generic disklabel extraction routine.
2079 	 */
2080 	errstring = readdisklabel(RAIDLABELDEV(dev), raidstrategy,
2081 	    rs->sc_dkdev.dk_label, rs->sc_dkdev.dk_cpulabel);
2082 	if (errstring)
2083 		raidmakedisklabel(rs);
2084 	else {
2085 		int     i;
2086 		struct partition *pp;
2087 
2088 		/*
2089 		 * Sanity check whether the found disklabel is valid.
2090 		 *
2091 		 * This is necessary since total size of the raid device
2092 		 * may vary when an interleave is changed even though exactly
2093 		 * same componets are used, and old disklabel may used
2094 		 * if that is found.
2095 		 */
2096 		if (lp->d_secperunit != rs->sc_size)
2097 			printf("raid%d: WARNING: %s: "
2098 			    "total sector size in disklabel (%d) != "
2099 			    "the size of raid (%ld)\n", unit, rs->sc_xname,
2100 			    lp->d_secperunit, (long) rs->sc_size);
2101 		for (i = 0; i < lp->d_npartitions; i++) {
2102 			pp = &lp->d_partitions[i];
2103 			if (pp->p_offset + pp->p_size > rs->sc_size)
2104 				printf("raid%d: WARNING: %s: end of partition `%c' "
2105 				       "exceeds the size of raid (%ld)\n",
2106 				       unit, rs->sc_xname, 'a' + i, (long) rs->sc_size);
2107 		}
2108 	}
2109 
2110 }
2111 /*
2112  * Take care of things one might want to take care of in the event
2113  * that a disklabel isn't present.
2114  */
2115 static void
2116 raidmakedisklabel(struct raid_softc *rs)
2117 {
2118 	struct disklabel *lp = rs->sc_dkdev.dk_label;
2119 	db1_printf(("Making a label..\n"));
2120 
2121 	/*
2122 	 * For historical reasons, if there's no disklabel present
2123 	 * the raw partition must be marked FS_BSDFFS.
2124 	 */
2125 
2126 	lp->d_partitions[RAW_PART].p_fstype = FS_BSDFFS;
2127 
2128 	strncpy(lp->d_packname, "default label", sizeof(lp->d_packname));
2129 
2130 	lp->d_checksum = dkcksum(lp);
2131 }
2132 /*
2133  * Lookup the provided name in the filesystem.  If the file exists,
2134  * is a valid block device, and isn't being used by anyone else,
2135  * set *vpp to the file's vnode.
2136  * You'll find the original of this in ccd.c
2137  */
2138 int
2139 raidlookup(char *path, struct proc *p, struct vnode **vpp)
2140 {
2141 	struct nameidata nd;
2142 	struct vnode *vp;
2143 	struct vattr va;
2144 	int     error;
2145 
2146 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
2147 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2148 		return (error);
2149 	}
2150 	vp = nd.ni_vp;
2151 	if (vp->v_usecount > 1) {
2152 		VOP_UNLOCK(vp, 0);
2153 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2154 		return (EBUSY);
2155 	}
2156 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)) != 0) {
2157 		VOP_UNLOCK(vp, 0);
2158 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2159 		return (error);
2160 	}
2161 	/* XXX: eventually we should handle VREG, too. */
2162 	if (va.va_type != VBLK) {
2163 		VOP_UNLOCK(vp, 0);
2164 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2165 		return (ENOTBLK);
2166 	}
2167 	VOP_UNLOCK(vp, 0);
2168 	*vpp = vp;
2169 	return (0);
2170 }
2171 /*
2172  * Wait interruptibly for an exclusive lock.
2173  *
2174  * XXX
2175  * Several drivers do this; it should be abstracted and made MP-safe.
2176  * (Hmm... where have we seen this warning before :->  GO )
2177  */
2178 static int
2179 raidlock(struct raid_softc *rs)
2180 {
2181 	int     error;
2182 
2183 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2184 		rs->sc_flags |= RAIDF_WANTED;
2185 		if ((error =
2186 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2187 			return (error);
2188 	}
2189 	rs->sc_flags |= RAIDF_LOCKED;
2190 	return (0);
2191 }
2192 /*
2193  * Unlock and wake up any waiters.
2194  */
2195 static void
2196 raidunlock(struct raid_softc *rs)
2197 {
2198 
2199 	rs->sc_flags &= ~RAIDF_LOCKED;
2200 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2201 		rs->sc_flags &= ~RAIDF_WANTED;
2202 		wakeup(rs);
2203 	}
2204 }
2205 
2206 
2207 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2208 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2209 
2210 int
2211 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2212 {
2213 	RF_ComponentLabel_t clabel;
2214 	raidread_component_label(dev, b_vp, &clabel);
2215 	clabel.mod_counter = mod_counter;
2216 	clabel.clean = RF_RAID_CLEAN;
2217 	raidwrite_component_label(dev, b_vp, &clabel);
2218 	return(0);
2219 }
2220 
2221 
2222 int
2223 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2224 {
2225 	RF_ComponentLabel_t clabel;
2226 	raidread_component_label(dev, b_vp, &clabel);
2227 	clabel.mod_counter = mod_counter;
2228 	clabel.clean = RF_RAID_DIRTY;
2229 	raidwrite_component_label(dev, b_vp, &clabel);
2230 	return(0);
2231 }
2232 
2233 /* ARGSUSED */
2234 int
2235 raidread_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 	/* XXX should probably ensure that we don't try to do this if
2243 	   someone has changed rf_protected_sectors. */
2244 
2245 	if (b_vp == NULL) {
2246 		/* For whatever reason, this component is not valid.
2247 		   Don't try to read a component label from it. */
2248 		return(EINVAL);
2249 	}
2250 
2251 	/* get a block of the appropriate size... */
2252 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2253 	bp->b_dev = dev;
2254 
2255 	/* get our ducks in a row for the read */
2256 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2257 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2258 	bp->b_flags |= B_READ;
2259  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2260 
2261 	bdev = bdevsw_lookup(bp->b_dev);
2262 	if (bdev == NULL)
2263 		return (ENXIO);
2264 	(*bdev->d_strategy)(bp);
2265 
2266 	error = biowait(bp);
2267 
2268 	if (!error) {
2269 		memcpy(clabel, bp->b_data,
2270 		       sizeof(RF_ComponentLabel_t));
2271         }
2272 
2273 	brelse(bp);
2274 	return(error);
2275 }
2276 /* ARGSUSED */
2277 int
2278 raidwrite_component_label(dev_t dev, struct vnode *b_vp,
2279 			  RF_ComponentLabel_t *clabel)
2280 {
2281 	struct buf *bp;
2282 	const struct bdevsw *bdev;
2283 	int error;
2284 
2285 	/* get a block of the appropriate size... */
2286 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2287 	bp->b_dev = dev;
2288 
2289 	/* get our ducks in a row for the write */
2290 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2291 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2292 	bp->b_flags |= B_WRITE;
2293  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2294 
2295 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2296 
2297 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2298 
2299 	bdev = bdevsw_lookup(bp->b_dev);
2300 	if (bdev == NULL)
2301 		return (ENXIO);
2302 	(*bdev->d_strategy)(bp);
2303 	error = biowait(bp);
2304 	brelse(bp);
2305 	if (error) {
2306 #if 1
2307 		printf("Failed to write RAID component info!\n");
2308 #endif
2309 	}
2310 
2311 	return(error);
2312 }
2313 
2314 void
2315 rf_markalldirty(RF_Raid_t *raidPtr)
2316 {
2317 	RF_ComponentLabel_t clabel;
2318 	int sparecol;
2319 	int c;
2320 	int j;
2321 	int scol = -1;
2322 
2323 	raidPtr->mod_counter++;
2324 	for (c = 0; c < raidPtr->numCol; c++) {
2325 		/* we don't want to touch (at all) a disk that has
2326 		   failed */
2327 		if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) {
2328 			raidread_component_label(
2329 						 raidPtr->Disks[c].dev,
2330 						 raidPtr->raid_cinfo[c].ci_vp,
2331 						 &clabel);
2332 			if (clabel.status == rf_ds_spared) {
2333 				/* XXX do something special...
2334 				   but whatever you do, don't
2335 				   try to access it!! */
2336 			} else {
2337 				raidmarkdirty(
2338 					      raidPtr->Disks[c].dev,
2339 					      raidPtr->raid_cinfo[c].ci_vp,
2340 					      raidPtr->mod_counter);
2341 			}
2342 		}
2343 	}
2344 
2345 	for( c = 0; c < raidPtr->numSpare ; c++) {
2346 		sparecol = raidPtr->numCol + c;
2347 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2348 			/*
2349 
2350 			   we claim this disk is "optimal" if it's
2351 			   rf_ds_used_spare, as that means it should be
2352 			   directly substitutable for the disk it replaced.
2353 			   We note that too...
2354 
2355 			 */
2356 
2357 			for(j=0;j<raidPtr->numCol;j++) {
2358 				if (raidPtr->Disks[j].spareCol == sparecol) {
2359 					scol = j;
2360 					break;
2361 				}
2362 			}
2363 
2364 			raidread_component_label(
2365 				 raidPtr->Disks[sparecol].dev,
2366 				 raidPtr->raid_cinfo[sparecol].ci_vp,
2367 				 &clabel);
2368 			/* make sure status is noted */
2369 
2370 			raid_init_component_label(raidPtr, &clabel);
2371 
2372 			clabel.row = 0;
2373 			clabel.column = scol;
2374 			/* Note: we *don't* change status from rf_ds_used_spare
2375 			   to rf_ds_optimal */
2376 			/* clabel.status = rf_ds_optimal; */
2377 
2378 			raidmarkdirty(raidPtr->Disks[sparecol].dev,
2379 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2380 				      raidPtr->mod_counter);
2381 		}
2382 	}
2383 }
2384 
2385 
2386 void
2387 rf_update_component_labels(RF_Raid_t *raidPtr, int final)
2388 {
2389 	RF_ComponentLabel_t clabel;
2390 	int sparecol;
2391 	int c;
2392 	int j;
2393 	int scol;
2394 
2395 	scol = -1;
2396 
2397 	/* XXX should do extra checks to make sure things really are clean,
2398 	   rather than blindly setting the clean bit... */
2399 
2400 	raidPtr->mod_counter++;
2401 
2402 	for (c = 0; c < raidPtr->numCol; c++) {
2403 		if (raidPtr->Disks[c].status == rf_ds_optimal) {
2404 			raidread_component_label(
2405 						 raidPtr->Disks[c].dev,
2406 						 raidPtr->raid_cinfo[c].ci_vp,
2407 						 &clabel);
2408 				/* make sure status is noted */
2409 			clabel.status = rf_ds_optimal;
2410 				/* bump the counter */
2411 			clabel.mod_counter = raidPtr->mod_counter;
2412 
2413 			raidwrite_component_label(
2414 						  raidPtr->Disks[c].dev,
2415 						  raidPtr->raid_cinfo[c].ci_vp,
2416 						  &clabel);
2417 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2418 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2419 					raidmarkclean(
2420 						      raidPtr->Disks[c].dev,
2421 						      raidPtr->raid_cinfo[c].ci_vp,
2422 						      raidPtr->mod_counter);
2423 				}
2424 			}
2425 		}
2426 		/* else we don't touch it.. */
2427 	}
2428 
2429 	for( c = 0; c < raidPtr->numSpare ; c++) {
2430 		sparecol = raidPtr->numCol + c;
2431 		/* Need to ensure that the reconstruct actually completed! */
2432 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2433 			/*
2434 
2435 			   we claim this disk is "optimal" if it's
2436 			   rf_ds_used_spare, as that means it should be
2437 			   directly substitutable for the disk it replaced.
2438 			   We note that too...
2439 
2440 			 */
2441 
2442 			for(j=0;j<raidPtr->numCol;j++) {
2443 				if (raidPtr->Disks[j].spareCol == sparecol) {
2444 					scol = j;
2445 					break;
2446 				}
2447 			}
2448 
2449 			/* XXX shouldn't *really* need this... */
2450 			raidread_component_label(
2451 				      raidPtr->Disks[sparecol].dev,
2452 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2453 				      &clabel);
2454 			/* make sure status is noted */
2455 
2456 			raid_init_component_label(raidPtr, &clabel);
2457 
2458 			clabel.mod_counter = raidPtr->mod_counter;
2459 			clabel.column = scol;
2460 			clabel.status = rf_ds_optimal;
2461 
2462 			raidwrite_component_label(
2463 				      raidPtr->Disks[sparecol].dev,
2464 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2465 				      &clabel);
2466 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2467 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2468 					raidmarkclean( raidPtr->Disks[sparecol].dev,
2469 						       raidPtr->raid_cinfo[sparecol].ci_vp,
2470 						       raidPtr->mod_counter);
2471 				}
2472 			}
2473 		}
2474 	}
2475 }
2476 
2477 void
2478 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured)
2479 {
2480 	struct proc *p;
2481 
2482 	p = raidPtr->engine_thread;
2483 
2484 	if (vp != NULL) {
2485 		if (auto_configured == 1) {
2486 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2487 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2488 			vput(vp);
2489 
2490 		} else {
2491 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, p);
2492 		}
2493 	}
2494 }
2495 
2496 
2497 void
2498 rf_UnconfigureVnodes(RF_Raid_t *raidPtr)
2499 {
2500 	int r,c;
2501 	struct vnode *vp;
2502 	int acd;
2503 
2504 
2505 	/* We take this opportunity to close the vnodes like we should.. */
2506 
2507 	for (c = 0; c < raidPtr->numCol; c++) {
2508 		vp = raidPtr->raid_cinfo[c].ci_vp;
2509 		acd = raidPtr->Disks[c].auto_configured;
2510 		rf_close_component(raidPtr, vp, acd);
2511 		raidPtr->raid_cinfo[c].ci_vp = NULL;
2512 		raidPtr->Disks[c].auto_configured = 0;
2513 	}
2514 
2515 	for (r = 0; r < raidPtr->numSpare; r++) {
2516 		vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp;
2517 		acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured;
2518 		rf_close_component(raidPtr, vp, acd);
2519 		raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL;
2520 		raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0;
2521 	}
2522 }
2523 
2524 
2525 void
2526 rf_ReconThread(struct rf_recon_req *req)
2527 {
2528 	int     s;
2529 	RF_Raid_t *raidPtr;
2530 
2531 	s = splbio();
2532 	raidPtr = (RF_Raid_t *) req->raidPtr;
2533 	raidPtr->recon_in_progress = 1;
2534 
2535 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col,
2536 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2537 
2538 	RF_Free(req, sizeof(*req));
2539 
2540 	raidPtr->recon_in_progress = 0;
2541 	splx(s);
2542 
2543 	/* That's all... */
2544 	kthread_exit(0);        /* does not return */
2545 }
2546 
2547 void
2548 rf_RewriteParityThread(RF_Raid_t *raidPtr)
2549 {
2550 	int retcode;
2551 	int s;
2552 
2553 	raidPtr->parity_rewrite_stripes_done = 0;
2554 	raidPtr->parity_rewrite_in_progress = 1;
2555 	s = splbio();
2556 	retcode = rf_RewriteParity(raidPtr);
2557 	splx(s);
2558 	if (retcode) {
2559 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2560 	} else {
2561 		/* set the clean bit!  If we shutdown correctly,
2562 		   the clean bit on each component label will get
2563 		   set */
2564 		raidPtr->parity_good = RF_RAID_CLEAN;
2565 	}
2566 	raidPtr->parity_rewrite_in_progress = 0;
2567 
2568 	/* Anyone waiting for us to stop?  If so, inform them... */
2569 	if (raidPtr->waitShutdown) {
2570 		wakeup(&raidPtr->parity_rewrite_in_progress);
2571 	}
2572 
2573 	/* That's all... */
2574 	kthread_exit(0);        /* does not return */
2575 }
2576 
2577 
2578 void
2579 rf_CopybackThread(RF_Raid_t *raidPtr)
2580 {
2581 	int s;
2582 
2583 	raidPtr->copyback_in_progress = 1;
2584 	s = splbio();
2585 	rf_CopybackReconstructedData(raidPtr);
2586 	splx(s);
2587 	raidPtr->copyback_in_progress = 0;
2588 
2589 	/* That's all... */
2590 	kthread_exit(0);        /* does not return */
2591 }
2592 
2593 
2594 void
2595 rf_ReconstructInPlaceThread(struct rf_recon_req *req)
2596 {
2597 	int s;
2598 	RF_Raid_t *raidPtr;
2599 
2600 	s = splbio();
2601 	raidPtr = req->raidPtr;
2602 	raidPtr->recon_in_progress = 1;
2603 	rf_ReconstructInPlace(raidPtr, req->col);
2604 	RF_Free(req, sizeof(*req));
2605 	raidPtr->recon_in_progress = 0;
2606 	splx(s);
2607 
2608 	/* That's all... */
2609 	kthread_exit(0);        /* does not return */
2610 }
2611 
2612 RF_AutoConfig_t *
2613 rf_find_raid_components()
2614 {
2615 	struct vnode *vp;
2616 	struct disklabel label;
2617 	struct device *dv;
2618 	dev_t dev;
2619 	int bmajor;
2620 	int error;
2621 	int i;
2622 	int good_one;
2623 	RF_ComponentLabel_t *clabel;
2624 	RF_AutoConfig_t *ac_list;
2625 	RF_AutoConfig_t *ac;
2626 
2627 
2628 	/* initialize the AutoConfig list */
2629 	ac_list = NULL;
2630 
2631 	/* we begin by trolling through *all* the devices on the system */
2632 
2633 	for (dv = alldevs.tqh_first; dv != NULL;
2634 	     dv = dv->dv_list.tqe_next) {
2635 
2636 		/* we are only interested in disks... */
2637 		if (dv->dv_class != DV_DISK)
2638 			continue;
2639 
2640 		/* we don't care about floppies... */
2641 		if (!strcmp(dv->dv_cfdata->cf_name,"fd")) {
2642 			continue;
2643 		}
2644 
2645 		/* we don't care about CD's... */
2646 		if (!strcmp(dv->dv_cfdata->cf_name,"cd")) {
2647 			continue;
2648 		}
2649 
2650 		/* hdfd is the Atari/Hades floppy driver */
2651 		if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) {
2652 			continue;
2653 		}
2654 		/* fdisa is the Atari/Milan floppy driver */
2655 		if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) {
2656 			continue;
2657 		}
2658 
2659 		/* need to find the device_name_to_block_device_major stuff */
2660 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2661 
2662 		/* get a vnode for the raw partition of this disk */
2663 
2664 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2665 		if (bdevvp(dev, &vp))
2666 			panic("RAID can't alloc vnode");
2667 
2668 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2669 
2670 		if (error) {
2671 			/* "Who cares."  Continue looking
2672 			   for something that exists*/
2673 			vput(vp);
2674 			continue;
2675 		}
2676 
2677 		/* Ok, the disk exists.  Go get the disklabel. */
2678 		error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0);
2679 		if (error) {
2680 			/*
2681 			 * XXX can't happen - open() would
2682 			 * have errored out (or faked up one)
2683 			 */
2684 			if (error != ENOTTY)
2685 				printf("RAIDframe: can't get label for dev "
2686 				    "%s (%d)\n", dv->dv_xname, error);
2687 		}
2688 
2689 		/* don't need this any more.  We'll allocate it again
2690 		   a little later if we really do... */
2691 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2692 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2693 		vput(vp);
2694 
2695 		if (error)
2696 			continue;
2697 
2698 		for (i=0; i < label.d_npartitions; i++) {
2699 			/* We only support partitions marked as RAID */
2700 			if (label.d_partitions[i].p_fstype != FS_RAID)
2701 				continue;
2702 
2703 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2704 			if (bdevvp(dev, &vp))
2705 				panic("RAID can't alloc vnode");
2706 
2707 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2708 			if (error) {
2709 				/* Whatever... */
2710 				vput(vp);
2711 				continue;
2712 			}
2713 
2714 			good_one = 0;
2715 
2716 			clabel = (RF_ComponentLabel_t *)
2717 				malloc(sizeof(RF_ComponentLabel_t),
2718 				       M_RAIDFRAME, M_NOWAIT);
2719 			if (clabel == NULL) {
2720 				/* XXX CLEANUP HERE */
2721 				printf("RAID auto config: out of memory!\n");
2722 				return(NULL); /* XXX probably should panic? */
2723 			}
2724 
2725 			if (!raidread_component_label(dev, vp, clabel)) {
2726 				/* Got the label.  Does it look reasonable? */
2727 				if (rf_reasonable_label(clabel) &&
2728 				    (clabel->partitionSize <=
2729 				     label.d_partitions[i].p_size)) {
2730 #if DEBUG
2731 					printf("Component on: %s%c: %d\n",
2732 					       dv->dv_xname, 'a'+i,
2733 					       label.d_partitions[i].p_size);
2734 					rf_print_component_label(clabel);
2735 #endif
2736 					/* if it's reasonable, add it,
2737 					   else ignore it. */
2738 					ac = (RF_AutoConfig_t *)
2739 						malloc(sizeof(RF_AutoConfig_t),
2740 						       M_RAIDFRAME,
2741 						       M_NOWAIT);
2742 					if (ac == NULL) {
2743 						/* XXX should panic?? */
2744 						return(NULL);
2745 					}
2746 
2747 					snprintf(ac->devname,
2748 					    sizeof(ac->devname), "%s%c",
2749 					    dv->dv_xname, 'a'+i);
2750 					ac->dev = dev;
2751 					ac->vp = vp;
2752 					ac->clabel = clabel;
2753 					ac->next = ac_list;
2754 					ac_list = ac;
2755 					good_one = 1;
2756 				}
2757 			}
2758 			if (!good_one) {
2759 				/* cleanup */
2760 				free(clabel, M_RAIDFRAME);
2761 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2762 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2763 				vput(vp);
2764 			}
2765 		}
2766 	}
2767 	return(ac_list);
2768 }
2769 
2770 static int
2771 rf_reasonable_label(RF_ComponentLabel_t *clabel)
2772 {
2773 
2774 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2775 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2776 	    ((clabel->clean == RF_RAID_CLEAN) ||
2777 	     (clabel->clean == RF_RAID_DIRTY)) &&
2778 	    clabel->row >=0 &&
2779 	    clabel->column >= 0 &&
2780 	    clabel->num_rows > 0 &&
2781 	    clabel->num_columns > 0 &&
2782 	    clabel->row < clabel->num_rows &&
2783 	    clabel->column < clabel->num_columns &&
2784 	    clabel->blockSize > 0 &&
2785 	    clabel->numBlocks > 0) {
2786 		/* label looks reasonable enough... */
2787 		return(1);
2788 	}
2789 	return(0);
2790 }
2791 
2792 
2793 #if DEBUG
2794 void
2795 rf_print_component_label(RF_ComponentLabel_t *clabel)
2796 {
2797 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2798 	       clabel->row, clabel->column,
2799 	       clabel->num_rows, clabel->num_columns);
2800 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2801 	       clabel->version, clabel->serial_number,
2802 	       clabel->mod_counter);
2803 	printf("   Clean: %s Status: %d\n",
2804 	       clabel->clean ? "Yes" : "No", clabel->status );
2805 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2806 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2807 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2808 	       (char) clabel->parityConfig, clabel->blockSize,
2809 	       clabel->numBlocks);
2810 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2811 	printf("   Contains root partition: %s\n",
2812 	       clabel->root_partition ? "Yes" : "No" );
2813 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2814 #if 0
2815 	   printf("   Config order: %d\n", clabel->config_order);
2816 #endif
2817 
2818 }
2819 #endif
2820 
2821 RF_ConfigSet_t *
2822 rf_create_auto_sets(RF_AutoConfig_t *ac_list)
2823 {
2824 	RF_AutoConfig_t *ac;
2825 	RF_ConfigSet_t *config_sets;
2826 	RF_ConfigSet_t *cset;
2827 	RF_AutoConfig_t *ac_next;
2828 
2829 
2830 	config_sets = NULL;
2831 
2832 	/* Go through the AutoConfig list, and figure out which components
2833 	   belong to what sets.  */
2834 	ac = ac_list;
2835 	while(ac!=NULL) {
2836 		/* we're going to putz with ac->next, so save it here
2837 		   for use at the end of the loop */
2838 		ac_next = ac->next;
2839 
2840 		if (config_sets == NULL) {
2841 			/* will need at least this one... */
2842 			config_sets = (RF_ConfigSet_t *)
2843 				malloc(sizeof(RF_ConfigSet_t),
2844 				       M_RAIDFRAME, M_NOWAIT);
2845 			if (config_sets == NULL) {
2846 				panic("rf_create_auto_sets: No memory!");
2847 			}
2848 			/* this one is easy :) */
2849 			config_sets->ac = ac;
2850 			config_sets->next = NULL;
2851 			config_sets->rootable = 0;
2852 			ac->next = NULL;
2853 		} else {
2854 			/* which set does this component fit into? */
2855 			cset = config_sets;
2856 			while(cset!=NULL) {
2857 				if (rf_does_it_fit(cset, ac)) {
2858 					/* looks like it matches... */
2859 					ac->next = cset->ac;
2860 					cset->ac = ac;
2861 					break;
2862 				}
2863 				cset = cset->next;
2864 			}
2865 			if (cset==NULL) {
2866 				/* didn't find a match above... new set..*/
2867 				cset = (RF_ConfigSet_t *)
2868 					malloc(sizeof(RF_ConfigSet_t),
2869 					       M_RAIDFRAME, M_NOWAIT);
2870 				if (cset == NULL) {
2871 					panic("rf_create_auto_sets: No memory!");
2872 				}
2873 				cset->ac = ac;
2874 				ac->next = NULL;
2875 				cset->next = config_sets;
2876 				cset->rootable = 0;
2877 				config_sets = cset;
2878 			}
2879 		}
2880 		ac = ac_next;
2881 	}
2882 
2883 
2884 	return(config_sets);
2885 }
2886 
2887 static int
2888 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac)
2889 {
2890 	RF_ComponentLabel_t *clabel1, *clabel2;
2891 
2892 	/* If this one matches the *first* one in the set, that's good
2893 	   enough, since the other members of the set would have been
2894 	   through here too... */
2895 	/* note that we are not checking partitionSize here..
2896 
2897 	   Note that we are also not checking the mod_counters here.
2898 	   If everything else matches execpt the mod_counter, that's
2899 	   good enough for this test.  We will deal with the mod_counters
2900 	   a little later in the autoconfiguration process.
2901 
2902 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2903 
2904 	   The reason we don't check for this is that failed disks
2905 	   will have lower modification counts.  If those disks are
2906 	   not added to the set they used to belong to, then they will
2907 	   form their own set, which may result in 2 different sets,
2908 	   for example, competing to be configured at raid0, and
2909 	   perhaps competing to be the root filesystem set.  If the
2910 	   wrong ones get configured, or both attempt to become /,
2911 	   weird behaviour and or serious lossage will occur.  Thus we
2912 	   need to bring them into the fold here, and kick them out at
2913 	   a later point.
2914 
2915 	*/
2916 
2917 	clabel1 = cset->ac->clabel;
2918 	clabel2 = ac->clabel;
2919 	if ((clabel1->version == clabel2->version) &&
2920 	    (clabel1->serial_number == clabel2->serial_number) &&
2921 	    (clabel1->num_rows == clabel2->num_rows) &&
2922 	    (clabel1->num_columns == clabel2->num_columns) &&
2923 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2924 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2925 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2926 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2927 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2928 	    (clabel1->blockSize == clabel2->blockSize) &&
2929 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2930 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2931 	    (clabel1->root_partition == clabel2->root_partition) &&
2932 	    (clabel1->last_unit == clabel2->last_unit) &&
2933 	    (clabel1->config_order == clabel2->config_order)) {
2934 		/* if it get's here, it almost *has* to be a match */
2935 	} else {
2936 		/* it's not consistent with somebody in the set..
2937 		   punt */
2938 		return(0);
2939 	}
2940 	/* all was fine.. it must fit... */
2941 	return(1);
2942 }
2943 
2944 int
2945 rf_have_enough_components(RF_ConfigSet_t *cset)
2946 {
2947 	RF_AutoConfig_t *ac;
2948 	RF_AutoConfig_t *auto_config;
2949 	RF_ComponentLabel_t *clabel;
2950 	int c;
2951 	int num_cols;
2952 	int num_missing;
2953 	int mod_counter;
2954 	int mod_counter_found;
2955 	int even_pair_failed;
2956 	char parity_type;
2957 
2958 
2959 	/* check to see that we have enough 'live' components
2960 	   of this set.  If so, we can configure it if necessary */
2961 
2962 	num_cols = cset->ac->clabel->num_columns;
2963 	parity_type = cset->ac->clabel->parityConfig;
2964 
2965 	/* XXX Check for duplicate components!?!?!? */
2966 
2967 	/* Determine what the mod_counter is supposed to be for this set. */
2968 
2969 	mod_counter_found = 0;
2970 	mod_counter = 0;
2971 	ac = cset->ac;
2972 	while(ac!=NULL) {
2973 		if (mod_counter_found==0) {
2974 			mod_counter = ac->clabel->mod_counter;
2975 			mod_counter_found = 1;
2976 		} else {
2977 			if (ac->clabel->mod_counter > mod_counter) {
2978 				mod_counter = ac->clabel->mod_counter;
2979 			}
2980 		}
2981 		ac = ac->next;
2982 	}
2983 
2984 	num_missing = 0;
2985 	auto_config = cset->ac;
2986 
2987 	even_pair_failed = 0;
2988 	for(c=0; c<num_cols; c++) {
2989 		ac = auto_config;
2990 		while(ac!=NULL) {
2991 			if ((ac->clabel->column == c) &&
2992 			    (ac->clabel->mod_counter == mod_counter)) {
2993 				/* it's this one... */
2994 #if DEBUG
2995 				printf("Found: %s at %d\n",
2996 				       ac->devname,c);
2997 #endif
2998 				break;
2999 			}
3000 			ac=ac->next;
3001 		}
3002 		if (ac==NULL) {
3003 				/* Didn't find one here! */
3004 				/* special case for RAID 1, especially
3005 				   where there are more than 2
3006 				   components (where RAIDframe treats
3007 				   things a little differently :( ) */
3008 			if (parity_type == '1') {
3009 				if (c%2 == 0) { /* even component */
3010 					even_pair_failed = 1;
3011 				} else { /* odd component.  If
3012 					    we're failed, and
3013 					    so is the even
3014 					    component, it's
3015 					    "Good Night, Charlie" */
3016 					if (even_pair_failed == 1) {
3017 						return(0);
3018 					}
3019 				}
3020 			} else {
3021 				/* normal accounting */
3022 				num_missing++;
3023 			}
3024 		}
3025 		if ((parity_type == '1') && (c%2 == 1)) {
3026 				/* Just did an even component, and we didn't
3027 				   bail.. reset the even_pair_failed flag,
3028 				   and go on to the next component.... */
3029 			even_pair_failed = 0;
3030 		}
3031 	}
3032 
3033 	clabel = cset->ac->clabel;
3034 
3035 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3036 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3037 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3038 		/* XXX this needs to be made *much* more general */
3039 		/* Too many failures */
3040 		return(0);
3041 	}
3042 	/* otherwise, all is well, and we've got enough to take a kick
3043 	   at autoconfiguring this set */
3044 	return(1);
3045 }
3046 
3047 void
3048 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config,
3049 			RF_Raid_t *raidPtr)
3050 {
3051 	RF_ComponentLabel_t *clabel;
3052 	int i;
3053 
3054 	clabel = ac->clabel;
3055 
3056 	/* 1. Fill in the common stuff */
3057 	config->numRow = clabel->num_rows = 1;
3058 	config->numCol = clabel->num_columns;
3059 	config->numSpare = 0; /* XXX should this be set here? */
3060 	config->sectPerSU = clabel->sectPerSU;
3061 	config->SUsPerPU = clabel->SUsPerPU;
3062 	config->SUsPerRU = clabel->SUsPerRU;
3063 	config->parityConfig = clabel->parityConfig;
3064 	/* XXX... */
3065 	strcpy(config->diskQueueType,"fifo");
3066 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3067 	config->layoutSpecificSize = 0; /* XXX ?? */
3068 
3069 	while(ac!=NULL) {
3070 		/* row/col values will be in range due to the checks
3071 		   in reasonable_label() */
3072 		strcpy(config->devnames[0][ac->clabel->column],
3073 		       ac->devname);
3074 		ac = ac->next;
3075 	}
3076 
3077 	for(i=0;i<RF_MAXDBGV;i++) {
3078 		config->debugVars[i][0] = 0;
3079 	}
3080 }
3081 
3082 int
3083 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value)
3084 {
3085 	RF_ComponentLabel_t clabel;
3086 	struct vnode *vp;
3087 	dev_t dev;
3088 	int column;
3089 	int sparecol;
3090 
3091 	raidPtr->autoconfigure = new_value;
3092 
3093 	for(column=0; column<raidPtr->numCol; column++) {
3094 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3095 			dev = raidPtr->Disks[column].dev;
3096 			vp = raidPtr->raid_cinfo[column].ci_vp;
3097 			raidread_component_label(dev, vp, &clabel);
3098 			clabel.autoconfigure = new_value;
3099 			raidwrite_component_label(dev, vp, &clabel);
3100 		}
3101 	}
3102 	for(column = 0; column < raidPtr->numSpare ; column++) {
3103 		sparecol = raidPtr->numCol + column;
3104 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3105 			dev = raidPtr->Disks[sparecol].dev;
3106 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3107 			raidread_component_label(dev, vp, &clabel);
3108 			clabel.autoconfigure = new_value;
3109 			raidwrite_component_label(dev, vp, &clabel);
3110 		}
3111 	}
3112 	return(new_value);
3113 }
3114 
3115 int
3116 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value)
3117 {
3118 	RF_ComponentLabel_t clabel;
3119 	struct vnode *vp;
3120 	dev_t dev;
3121 	int column;
3122 	int sparecol;
3123 
3124 	raidPtr->root_partition = new_value;
3125 	for(column=0; column<raidPtr->numCol; column++) {
3126 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3127 			dev = raidPtr->Disks[column].dev;
3128 			vp = raidPtr->raid_cinfo[column].ci_vp;
3129 			raidread_component_label(dev, vp, &clabel);
3130 			clabel.root_partition = new_value;
3131 			raidwrite_component_label(dev, vp, &clabel);
3132 		}
3133 	}
3134 	for(column = 0; column < raidPtr->numSpare ; column++) {
3135 		sparecol = raidPtr->numCol + column;
3136 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3137 			dev = raidPtr->Disks[sparecol].dev;
3138 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3139 			raidread_component_label(dev, vp, &clabel);
3140 			clabel.root_partition = new_value;
3141 			raidwrite_component_label(dev, vp, &clabel);
3142 		}
3143 	}
3144 	return(new_value);
3145 }
3146 
3147 void
3148 rf_release_all_vps(RF_ConfigSet_t *cset)
3149 {
3150 	RF_AutoConfig_t *ac;
3151 
3152 	ac = cset->ac;
3153 	while(ac!=NULL) {
3154 		/* Close the vp, and give it back */
3155 		if (ac->vp) {
3156 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3157 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3158 			vput(ac->vp);
3159 			ac->vp = NULL;
3160 		}
3161 		ac = ac->next;
3162 	}
3163 }
3164 
3165 
3166 void
3167 rf_cleanup_config_set(RF_ConfigSet_t *cset)
3168 {
3169 	RF_AutoConfig_t *ac;
3170 	RF_AutoConfig_t *next_ac;
3171 
3172 	ac = cset->ac;
3173 	while(ac!=NULL) {
3174 		next_ac = ac->next;
3175 		/* nuke the label */
3176 		free(ac->clabel, M_RAIDFRAME);
3177 		/* cleanup the config structure */
3178 		free(ac, M_RAIDFRAME);
3179 		/* "next.." */
3180 		ac = next_ac;
3181 	}
3182 	/* and, finally, nuke the config set */
3183 	free(cset, M_RAIDFRAME);
3184 }
3185 
3186 
3187 void
3188 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel)
3189 {
3190 	/* current version number */
3191 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3192 	clabel->serial_number = raidPtr->serial_number;
3193 	clabel->mod_counter = raidPtr->mod_counter;
3194 	clabel->num_rows = 1;
3195 	clabel->num_columns = raidPtr->numCol;
3196 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3197 	clabel->status = rf_ds_optimal; /* "It's good!" */
3198 
3199 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3200 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3201 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3202 
3203 	clabel->blockSize = raidPtr->bytesPerSector;
3204 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3205 
3206 	/* XXX not portable */
3207 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3208 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3209 	clabel->autoconfigure = raidPtr->autoconfigure;
3210 	clabel->root_partition = raidPtr->root_partition;
3211 	clabel->last_unit = raidPtr->raidid;
3212 	clabel->config_order = raidPtr->config_order;
3213 }
3214 
3215 int
3216 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit)
3217 {
3218 	RF_Raid_t *raidPtr;
3219 	RF_Config_t *config;
3220 	int raidID;
3221 	int retcode;
3222 
3223 #if DEBUG
3224 	printf("RAID autoconfigure\n");
3225 #endif
3226 
3227 	retcode = 0;
3228 	*unit = -1;
3229 
3230 	/* 1. Create a config structure */
3231 
3232 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3233 				       M_RAIDFRAME,
3234 				       M_NOWAIT);
3235 	if (config==NULL) {
3236 		printf("Out of mem!?!?\n");
3237 				/* XXX do something more intelligent here. */
3238 		return(1);
3239 	}
3240 
3241 	memset(config, 0, sizeof(RF_Config_t));
3242 
3243 	/*
3244 	   2. Figure out what RAID ID this one is supposed to live at
3245 	   See if we can get the same RAID dev that it was configured
3246 	   on last time..
3247 	*/
3248 
3249 	raidID = cset->ac->clabel->last_unit;
3250 	if ((raidID < 0) || (raidID >= numraid)) {
3251 		/* let's not wander off into lala land. */
3252 		raidID = numraid - 1;
3253 	}
3254 	if (raidPtrs[raidID]->valid != 0) {
3255 
3256 		/*
3257 		   Nope... Go looking for an alternative...
3258 		   Start high so we don't immediately use raid0 if that's
3259 		   not taken.
3260 		*/
3261 
3262 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3263 			if (raidPtrs[raidID]->valid == 0) {
3264 				/* can use this one! */
3265 				break;
3266 			}
3267 		}
3268 	}
3269 
3270 	if (raidID < 0) {
3271 		/* punt... */
3272 		printf("Unable to auto configure this set!\n");
3273 		printf("(Out of RAID devs!)\n");
3274 		return(1);
3275 	}
3276 
3277 #if DEBUG
3278 	printf("Configuring raid%d:\n",raidID);
3279 #endif
3280 
3281 	raidPtr = raidPtrs[raidID];
3282 
3283 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3284 	raidPtr->raidid = raidID;
3285 	raidPtr->openings = RAIDOUTSTANDING;
3286 
3287 	/* 3. Build the configuration structure */
3288 	rf_create_configuration(cset->ac, config, raidPtr);
3289 
3290 	/* 4. Do the configuration */
3291 	retcode = rf_Configure(raidPtr, config, cset->ac);
3292 
3293 	if (retcode == 0) {
3294 
3295 		raidinit(raidPtrs[raidID]);
3296 
3297 		rf_markalldirty(raidPtrs[raidID]);
3298 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3299 		if (cset->ac->clabel->root_partition==1) {
3300 			/* everything configured just fine.  Make a note
3301 			   that this set is eligible to be root. */
3302 			cset->rootable = 1;
3303 			/* XXX do this here? */
3304 			raidPtrs[raidID]->root_partition = 1;
3305 		}
3306 	}
3307 
3308 	/* 5. Cleanup */
3309 	free(config, M_RAIDFRAME);
3310 
3311 	*unit = raidID;
3312 	return(retcode);
3313 }
3314 
3315 void
3316 rf_disk_unbusy(RF_RaidAccessDesc_t *desc)
3317 {
3318 	struct buf *bp;
3319 
3320 	bp = (struct buf *)desc->bp;
3321 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3322 	    (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ));
3323 }
3324 
3325 void
3326 rf_pool_init(struct pool *p, size_t size, char *w_chan,
3327 	     size_t min, size_t max)
3328 {
3329 	pool_init(p, size, 0, 0, 0, w_chan, NULL);
3330 	pool_sethiwat(p, max);
3331 	pool_prime(p, min);
3332 	pool_setlowat(p, min);
3333 }
3334