xref: /netbsd-src/sys/dev/raidframe/rf_netbsdkintf.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: rf_netbsdkintf.c,v 1.192 2005/12/11 12:23:37 christos Exp $	*/
2 /*-
3  * Copyright (c) 1996, 1997, 1998 The NetBSD Foundation, Inc.
4  * All rights reserved.
5  *
6  * This code is derived from software contributed to The NetBSD Foundation
7  * by Greg Oster; Jason R. Thorpe.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *        This product includes software developed by the NetBSD
20  *        Foundation, Inc. and its contributors.
21  * 4. Neither the name of The NetBSD Foundation nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 1990, 1993
40  *      The Regents of the University of California.  All rights reserved.
41  *
42  * This code is derived from software contributed to Berkeley by
43  * the Systems Programming Group of the University of Utah Computer
44  * Science Department.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. Neither the name of the University nor the names of its contributors
55  *    may be used to endorse or promote products derived from this software
56  *    without specific prior written permission.
57  *
58  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
68  * SUCH DAMAGE.
69  *
70  * from: Utah $Hdr: cd.c 1.6 90/11/28$
71  *
72  *      @(#)cd.c        8.2 (Berkeley) 11/16/93
73  */
74 
75 /*
76  * Copyright (c) 1988 University of Utah.
77  *
78  * This code is derived from software contributed to Berkeley by
79  * the Systems Programming Group of the University of Utah Computer
80  * Science Department.
81  *
82  * Redistribution and use in source and binary forms, with or without
83  * modification, are permitted provided that the following conditions
84  * are met:
85  * 1. Redistributions of source code must retain the above copyright
86  *    notice, this list of conditions and the following disclaimer.
87  * 2. Redistributions in binary form must reproduce the above copyright
88  *    notice, this list of conditions and the following disclaimer in the
89  *    documentation and/or other materials provided with the distribution.
90  * 3. All advertising materials mentioning features or use of this software
91  *    must display the following acknowledgement:
92  *      This product includes software developed by the University of
93  *      California, Berkeley and its contributors.
94  * 4. Neither the name of the University nor the names of its contributors
95  *    may be used to endorse or promote products derived from this software
96  *    without specific prior written permission.
97  *
98  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
99  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
100  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
101  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
102  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
103  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
104  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
105  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
106  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
107  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
108  * SUCH DAMAGE.
109  *
110  * from: Utah $Hdr: cd.c 1.6 90/11/28$
111  *
112  *      @(#)cd.c        8.2 (Berkeley) 11/16/93
113  */
114 
115 /*
116  * Copyright (c) 1995 Carnegie-Mellon University.
117  * All rights reserved.
118  *
119  * Authors: Mark Holland, Jim Zelenka
120  *
121  * Permission to use, copy, modify and distribute this software and
122  * its documentation is hereby granted, provided that both the copyright
123  * notice and this permission notice appear in all copies of the
124  * software, derivative works or modified versions, and any portions
125  * thereof, and that both notices appear in supporting documentation.
126  *
127  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
128  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
129  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
130  *
131  * Carnegie Mellon requests users of this software to return to
132  *
133  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
134  *  School of Computer Science
135  *  Carnegie Mellon University
136  *  Pittsburgh PA 15213-3890
137  *
138  * any improvements or extensions that they make and grant Carnegie the
139  * rights to redistribute these changes.
140  */
141 
142 /***********************************************************
143  *
144  * rf_kintf.c -- the kernel interface routines for RAIDframe
145  *
146  ***********************************************************/
147 
148 #include <sys/cdefs.h>
149 __KERNEL_RCSID(0, "$NetBSD: rf_netbsdkintf.c,v 1.192 2005/12/11 12:23:37 christos Exp $");
150 
151 #include <sys/param.h>
152 #include <sys/errno.h>
153 #include <sys/pool.h>
154 #include <sys/proc.h>
155 #include <sys/queue.h>
156 #include <sys/disk.h>
157 #include <sys/device.h>
158 #include <sys/stat.h>
159 #include <sys/ioctl.h>
160 #include <sys/fcntl.h>
161 #include <sys/systm.h>
162 #include <sys/namei.h>
163 #include <sys/vnode.h>
164 #include <sys/disklabel.h>
165 #include <sys/conf.h>
166 #include <sys/lock.h>
167 #include <sys/buf.h>
168 #include <sys/bufq.h>
169 #include <sys/user.h>
170 #include <sys/reboot.h>
171 
172 #include <dev/raidframe/raidframevar.h>
173 #include <dev/raidframe/raidframeio.h>
174 #include "raid.h"
175 #include "opt_raid_autoconfig.h"
176 #include "rf_raid.h"
177 #include "rf_copyback.h"
178 #include "rf_dag.h"
179 #include "rf_dagflags.h"
180 #include "rf_desc.h"
181 #include "rf_diskqueue.h"
182 #include "rf_etimer.h"
183 #include "rf_general.h"
184 #include "rf_kintf.h"
185 #include "rf_options.h"
186 #include "rf_driver.h"
187 #include "rf_parityscan.h"
188 #include "rf_threadstuff.h"
189 
190 #ifdef DEBUG
191 int     rf_kdebug_level = 0;
192 #define db1_printf(a) if (rf_kdebug_level > 0) printf a
193 #else				/* DEBUG */
194 #define db1_printf(a) { }
195 #endif				/* DEBUG */
196 
197 static RF_Raid_t **raidPtrs;	/* global raid device descriptors */
198 
199 RF_DECLARE_STATIC_MUTEX(rf_sparet_wait_mutex)
200 
201 static RF_SparetWait_t *rf_sparet_wait_queue;	/* requests to install a
202 						 * spare table */
203 static RF_SparetWait_t *rf_sparet_resp_queue;	/* responses from
204 						 * installation process */
205 
206 MALLOC_DEFINE(M_RAIDFRAME, "RAIDframe", "RAIDframe structures");
207 
208 /* prototypes */
209 static void KernelWakeupFunc(struct buf *);
210 static void InitBP(struct buf *, struct vnode *, unsigned,
211     dev_t, RF_SectorNum_t, RF_SectorCount_t, caddr_t, void (*) (struct buf *),
212     void *, int, struct proc *);
213 static void raidinit(RF_Raid_t *);
214 
215 void raidattach(int);
216 
217 dev_type_open(raidopen);
218 dev_type_close(raidclose);
219 dev_type_read(raidread);
220 dev_type_write(raidwrite);
221 dev_type_ioctl(raidioctl);
222 dev_type_strategy(raidstrategy);
223 dev_type_dump(raiddump);
224 dev_type_size(raidsize);
225 
226 const struct bdevsw raid_bdevsw = {
227 	raidopen, raidclose, raidstrategy, raidioctl,
228 	raiddump, raidsize, D_DISK
229 };
230 
231 const struct cdevsw raid_cdevsw = {
232 	raidopen, raidclose, raidread, raidwrite, raidioctl,
233 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
234 };
235 
236 /*
237  * Pilfered from ccd.c
238  */
239 
240 struct raidbuf {
241 	struct buf rf_buf;	/* new I/O buf.  MUST BE FIRST!!! */
242 	struct buf *rf_obp;	/* ptr. to original I/O buf */
243 	RF_DiskQueueData_t *req;/* the request that this was part of.. */
244 };
245 
246 /* XXX Not sure if the following should be replacing the raidPtrs above,
247    or if it should be used in conjunction with that...
248 */
249 
250 struct raid_softc {
251 	int     sc_flags;	/* flags */
252 	int     sc_cflags;	/* configuration flags */
253 	size_t  sc_size;        /* size of the raid device */
254 	char    sc_xname[20];	/* XXX external name */
255 	struct disk sc_dkdev;	/* generic disk device info */
256 	struct bufq_state *buf_queue;	/* used for the device queue */
257 };
258 /* sc_flags */
259 #define RAIDF_INITED	0x01	/* unit has been initialized */
260 #define RAIDF_WLABEL	0x02	/* label area is writable */
261 #define RAIDF_LABELLING	0x04	/* unit is currently being labelled */
262 #define RAIDF_WANTED	0x40	/* someone is waiting to obtain a lock */
263 #define RAIDF_LOCKED	0x80	/* unit is locked */
264 
265 #define	raidunit(x)	DISKUNIT(x)
266 int numraid = 0;
267 
268 /*
269  * Allow RAIDOUTSTANDING number of simultaneous IO's to this RAID device.
270  * Be aware that large numbers can allow the driver to consume a lot of
271  * kernel memory, especially on writes, and in degraded mode reads.
272  *
273  * For example: with a stripe width of 64 blocks (32k) and 5 disks,
274  * a single 64K write will typically require 64K for the old data,
275  * 64K for the old parity, and 64K for the new parity, for a total
276  * of 192K (if the parity buffer is not re-used immediately).
277  * Even it if is used immediately, that's still 128K, which when multiplied
278  * by say 10 requests, is 1280K, *on top* of the 640K of incoming data.
279  *
280  * Now in degraded mode, for example, a 64K read on the above setup may
281  * require data reconstruction, which will require *all* of the 4 remaining
282  * disks to participate -- 4 * 32K/disk == 128K again.
283  */
284 
285 #ifndef RAIDOUTSTANDING
286 #define RAIDOUTSTANDING   6
287 #endif
288 
289 #define RAIDLABELDEV(dev)	\
290 	(MAKEDISKDEV(major((dev)), raidunit((dev)), RAW_PART))
291 
292 /* declared here, and made public, for the benefit of KVM stuff.. */
293 struct raid_softc *raid_softc;
294 
295 static void raidgetdefaultlabel(RF_Raid_t *, struct raid_softc *,
296 				     struct disklabel *);
297 static void raidgetdisklabel(dev_t);
298 static void raidmakedisklabel(struct raid_softc *);
299 
300 static int raidlock(struct raid_softc *);
301 static void raidunlock(struct raid_softc *);
302 
303 static void rf_markalldirty(RF_Raid_t *);
304 
305 struct device *raidrootdev;
306 
307 void rf_ReconThread(struct rf_recon_req *);
308 void rf_RewriteParityThread(RF_Raid_t *raidPtr);
309 void rf_CopybackThread(RF_Raid_t *raidPtr);
310 void rf_ReconstructInPlaceThread(struct rf_recon_req *);
311 int rf_autoconfig(struct device *self);
312 void rf_buildroothack(RF_ConfigSet_t *);
313 
314 RF_AutoConfig_t *rf_find_raid_components(void);
315 RF_ConfigSet_t *rf_create_auto_sets(RF_AutoConfig_t *);
316 static int rf_does_it_fit(RF_ConfigSet_t *,RF_AutoConfig_t *);
317 static int rf_reasonable_label(RF_ComponentLabel_t *);
318 void rf_create_configuration(RF_AutoConfig_t *,RF_Config_t *, RF_Raid_t *);
319 int rf_set_autoconfig(RF_Raid_t *, int);
320 int rf_set_rootpartition(RF_Raid_t *, int);
321 void rf_release_all_vps(RF_ConfigSet_t *);
322 void rf_cleanup_config_set(RF_ConfigSet_t *);
323 int rf_have_enough_components(RF_ConfigSet_t *);
324 int rf_auto_config_set(RF_ConfigSet_t *, int *);
325 
326 static int raidautoconfig = 0; /* Debugging, mostly.  Set to 0 to not
327 				  allow autoconfig to take place.
328 			          Note that this is overridden by having
329 			          RAID_AUTOCONFIG as an option in the
330 			          kernel config file.  */
331 
332 struct RF_Pools_s rf_pools;
333 
334 void
335 raidattach(int num)
336 {
337 	int raidID;
338 	int i, rc;
339 
340 #ifdef DEBUG
341 	printf("raidattach: Asked for %d units\n", num);
342 #endif
343 
344 	if (num <= 0) {
345 #ifdef DIAGNOSTIC
346 		panic("raidattach: count <= 0");
347 #endif
348 		return;
349 	}
350 	/* This is where all the initialization stuff gets done. */
351 
352 	numraid = num;
353 
354 	/* Make some space for requested number of units... */
355 
356 	RF_Malloc(raidPtrs, num * sizeof(RF_Raid_t *), (RF_Raid_t **));
357 	if (raidPtrs == NULL) {
358 		panic("raidPtrs is NULL!!");
359 	}
360 
361 	/* Initialize the component buffer pool. */
362 	rf_pool_init(&rf_pools.cbuf, sizeof(struct raidbuf),
363 		     "raidpl", num * RAIDOUTSTANDING,
364 		     2 * num * RAIDOUTSTANDING);
365 
366 	rf_mutex_init(&rf_sparet_wait_mutex);
367 
368 	rf_sparet_wait_queue = rf_sparet_resp_queue = NULL;
369 
370 	for (i = 0; i < num; i++)
371 		raidPtrs[i] = NULL;
372 	rc = rf_BootRaidframe();
373 	if (rc == 0)
374 		printf("Kernelized RAIDframe activated\n");
375 	else
376 		panic("Serious error booting RAID!!");
377 
378 	/* put together some datastructures like the CCD device does.. This
379 	 * lets us lock the device and what-not when it gets opened. */
380 
381 	raid_softc = (struct raid_softc *)
382 		malloc(num * sizeof(struct raid_softc),
383 		       M_RAIDFRAME, M_NOWAIT);
384 	if (raid_softc == NULL) {
385 		printf("WARNING: no memory for RAIDframe driver\n");
386 		return;
387 	}
388 
389 	memset(raid_softc, 0, num * sizeof(struct raid_softc));
390 
391 	raidrootdev = (struct device *)malloc(num * sizeof(struct device),
392 					      M_RAIDFRAME, M_NOWAIT);
393 	if (raidrootdev == NULL) {
394 		panic("No memory for RAIDframe driver!!?!?!");
395 	}
396 
397 	for (raidID = 0; raidID < num; raidID++) {
398 		bufq_alloc(&raid_softc[raidID].buf_queue, "fcfs", 0);
399 		pseudo_disk_init(&raid_softc[raidID].sc_dkdev);
400 
401 		raidrootdev[raidID].dv_class  = DV_DISK;
402 		raidrootdev[raidID].dv_cfdata = NULL;
403 		raidrootdev[raidID].dv_unit   = raidID;
404 		raidrootdev[raidID].dv_parent = NULL;
405 		raidrootdev[raidID].dv_flags  = 0;
406 		snprintf(raidrootdev[raidID].dv_xname,
407 		    sizeof(raidrootdev[raidID].dv_xname), "raid%d", raidID);
408 
409 		RF_Malloc(raidPtrs[raidID], sizeof(RF_Raid_t),
410 			  (RF_Raid_t *));
411 		if (raidPtrs[raidID] == NULL) {
412 			printf("WARNING: raidPtrs[%d] is NULL\n", raidID);
413 			numraid = raidID;
414 			return;
415 		}
416 	}
417 
418 #ifdef RAID_AUTOCONFIG
419 	raidautoconfig = 1;
420 #endif
421 
422 	/*
423 	 * Register a finalizer which will be used to auto-config RAID
424 	 * sets once all real hardware devices have been found.
425 	 */
426 	if (config_finalize_register(NULL, rf_autoconfig) != 0)
427 		printf("WARNING: unable to register RAIDframe finalizer\n");
428 }
429 
430 int
431 rf_autoconfig(struct device *self)
432 {
433 	RF_AutoConfig_t *ac_list;
434 	RF_ConfigSet_t *config_sets;
435 
436 	if (raidautoconfig == 0)
437 		return (0);
438 
439 	/* XXX This code can only be run once. */
440 	raidautoconfig = 0;
441 
442 	/* 1. locate all RAID components on the system */
443 #ifdef DEBUG
444 	printf("Searching for RAID components...\n");
445 #endif
446 	ac_list = rf_find_raid_components();
447 
448 	/* 2. Sort them into their respective sets. */
449 	config_sets = rf_create_auto_sets(ac_list);
450 
451 	/*
452 	 * 3. Evaluate each set andconfigure the valid ones.
453 	 * This gets done in rf_buildroothack().
454 	 */
455 	rf_buildroothack(config_sets);
456 
457 	return (1);
458 }
459 
460 void
461 rf_buildroothack(RF_ConfigSet_t *config_sets)
462 {
463 	RF_ConfigSet_t *cset;
464 	RF_ConfigSet_t *next_cset;
465 	int retcode;
466 	int raidID;
467 	int rootID;
468 	int num_root;
469 
470 	rootID = 0;
471 	num_root = 0;
472 	cset = config_sets;
473 	while(cset != NULL ) {
474 		next_cset = cset->next;
475 		if (rf_have_enough_components(cset) &&
476 		    cset->ac->clabel->autoconfigure==1) {
477 			retcode = rf_auto_config_set(cset,&raidID);
478 			if (!retcode) {
479 				if (cset->rootable) {
480 					rootID = raidID;
481 					num_root++;
482 				}
483 			} else {
484 				/* The autoconfig didn't work :( */
485 #if DEBUG
486 				printf("Autoconfig failed with code %d for raid%d\n", retcode, raidID);
487 #endif
488 				rf_release_all_vps(cset);
489 			}
490 		} else {
491 			/* we're not autoconfiguring this set...
492 			   release the associated resources */
493 			rf_release_all_vps(cset);
494 		}
495 		/* cleanup */
496 		rf_cleanup_config_set(cset);
497 		cset = next_cset;
498 	}
499 
500 	/* we found something bootable... */
501 
502 	if (num_root == 1) {
503 		booted_device = &raidrootdev[rootID];
504 	} else if (num_root > 1) {
505 		/* we can't guess.. require the user to answer... */
506 		boothowto |= RB_ASKNAME;
507 	}
508 }
509 
510 
511 int
512 raidsize(dev_t dev)
513 {
514 	struct raid_softc *rs;
515 	struct disklabel *lp;
516 	int     part, unit, omask, size;
517 
518 	unit = raidunit(dev);
519 	if (unit >= numraid)
520 		return (-1);
521 	rs = &raid_softc[unit];
522 
523 	if ((rs->sc_flags & RAIDF_INITED) == 0)
524 		return (-1);
525 
526 	part = DISKPART(dev);
527 	omask = rs->sc_dkdev.dk_openmask & (1 << part);
528 	lp = rs->sc_dkdev.dk_label;
529 
530 	if (omask == 0 && raidopen(dev, 0, S_IFBLK, curlwp))
531 		return (-1);
532 
533 	if (lp->d_partitions[part].p_fstype != FS_SWAP)
534 		size = -1;
535 	else
536 		size = lp->d_partitions[part].p_size *
537 		    (lp->d_secsize / DEV_BSIZE);
538 
539 	if (omask == 0 && raidclose(dev, 0, S_IFBLK, curlwp))
540 		return (-1);
541 
542 	return (size);
543 
544 }
545 
546 int
547 raiddump(dev_t dev, daddr_t blkno, caddr_t va, size_t  size)
548 {
549 	/* Not implemented. */
550 	return ENXIO;
551 }
552 /* ARGSUSED */
553 int
554 raidopen(dev_t dev, int flags, int fmt, struct lwp *l)
555 {
556 	int     unit = raidunit(dev);
557 	struct raid_softc *rs;
558 	struct disklabel *lp;
559 	int     part, pmask;
560 	int     error = 0;
561 
562 	if (unit >= numraid)
563 		return (ENXIO);
564 	rs = &raid_softc[unit];
565 
566 	if ((error = raidlock(rs)) != 0)
567 		return (error);
568 	lp = rs->sc_dkdev.dk_label;
569 
570 	part = DISKPART(dev);
571 	pmask = (1 << part);
572 
573 	if ((rs->sc_flags & RAIDF_INITED) &&
574 	    (rs->sc_dkdev.dk_openmask == 0))
575 		raidgetdisklabel(dev);
576 
577 	/* make sure that this partition exists */
578 
579 	if (part != RAW_PART) {
580 		if (((rs->sc_flags & RAIDF_INITED) == 0) ||
581 		    ((part >= lp->d_npartitions) ||
582 			(lp->d_partitions[part].p_fstype == FS_UNUSED))) {
583 			error = ENXIO;
584 			raidunlock(rs);
585 			return (error);
586 		}
587 	}
588 	/* Prevent this unit from being unconfigured while open. */
589 	switch (fmt) {
590 	case S_IFCHR:
591 		rs->sc_dkdev.dk_copenmask |= pmask;
592 		break;
593 
594 	case S_IFBLK:
595 		rs->sc_dkdev.dk_bopenmask |= pmask;
596 		break;
597 	}
598 
599 	if ((rs->sc_dkdev.dk_openmask == 0) &&
600 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
601 		/* First one... mark things as dirty... Note that we *MUST*
602 		 have done a configure before this.  I DO NOT WANT TO BE
603 		 SCRIBBLING TO RANDOM COMPONENTS UNTIL IT'S BEEN DETERMINED
604 		 THAT THEY BELONG TOGETHER!!!!! */
605 		/* XXX should check to see if we're only open for reading
606 		   here... If so, we needn't do this, but then need some
607 		   other way of keeping track of what's happened.. */
608 
609 		rf_markalldirty( raidPtrs[unit] );
610 	}
611 
612 
613 	rs->sc_dkdev.dk_openmask =
614 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
615 
616 	raidunlock(rs);
617 
618 	return (error);
619 
620 
621 }
622 /* ARGSUSED */
623 int
624 raidclose(dev_t dev, int flags, int fmt, struct lwp *l)
625 {
626 	int     unit = raidunit(dev);
627 	struct raid_softc *rs;
628 	int     error = 0;
629 	int     part;
630 
631 	if (unit >= numraid)
632 		return (ENXIO);
633 	rs = &raid_softc[unit];
634 
635 	if ((error = raidlock(rs)) != 0)
636 		return (error);
637 
638 	part = DISKPART(dev);
639 
640 	/* ...that much closer to allowing unconfiguration... */
641 	switch (fmt) {
642 	case S_IFCHR:
643 		rs->sc_dkdev.dk_copenmask &= ~(1 << part);
644 		break;
645 
646 	case S_IFBLK:
647 		rs->sc_dkdev.dk_bopenmask &= ~(1 << part);
648 		break;
649 	}
650 	rs->sc_dkdev.dk_openmask =
651 	    rs->sc_dkdev.dk_copenmask | rs->sc_dkdev.dk_bopenmask;
652 
653 	if ((rs->sc_dkdev.dk_openmask == 0) &&
654 	    ((rs->sc_flags & RAIDF_INITED) != 0)) {
655 		/* Last one... device is not unconfigured yet.
656 		   Device shutdown has taken care of setting the
657 		   clean bits if RAIDF_INITED is not set
658 		   mark things as clean... */
659 
660 		rf_update_component_labels(raidPtrs[unit],
661 						 RF_FINAL_COMPONENT_UPDATE);
662 		if (doing_shutdown) {
663 			/* last one, and we're going down, so
664 			   lights out for this RAID set too. */
665 			error = rf_Shutdown(raidPtrs[unit]);
666 
667 			/* It's no longer initialized... */
668 			rs->sc_flags &= ~RAIDF_INITED;
669 
670 			/* Detach the disk. */
671 			pseudo_disk_detach(&rs->sc_dkdev);
672 		}
673 	}
674 
675 	raidunlock(rs);
676 	return (0);
677 
678 }
679 
680 void
681 raidstrategy(struct buf *bp)
682 {
683 	int s;
684 
685 	unsigned int raidID = raidunit(bp->b_dev);
686 	RF_Raid_t *raidPtr;
687 	struct raid_softc *rs = &raid_softc[raidID];
688 	int     wlabel;
689 
690 	if ((rs->sc_flags & RAIDF_INITED) ==0) {
691 		bp->b_error = ENXIO;
692 		bp->b_flags |= B_ERROR;
693 		bp->b_resid = bp->b_bcount;
694 		biodone(bp);
695 		return;
696 	}
697 	if (raidID >= numraid || !raidPtrs[raidID]) {
698 		bp->b_error = ENODEV;
699 		bp->b_flags |= B_ERROR;
700 		bp->b_resid = bp->b_bcount;
701 		biodone(bp);
702 		return;
703 	}
704 	raidPtr = raidPtrs[raidID];
705 	if (!raidPtr->valid) {
706 		bp->b_error = ENODEV;
707 		bp->b_flags |= B_ERROR;
708 		bp->b_resid = bp->b_bcount;
709 		biodone(bp);
710 		return;
711 	}
712 	if (bp->b_bcount == 0) {
713 		db1_printf(("b_bcount is zero..\n"));
714 		biodone(bp);
715 		return;
716 	}
717 
718 	/*
719 	 * Do bounds checking and adjust transfer.  If there's an
720 	 * error, the bounds check will flag that for us.
721 	 */
722 
723 	wlabel = rs->sc_flags & (RAIDF_WLABEL | RAIDF_LABELLING);
724 	if (DISKPART(bp->b_dev) != RAW_PART)
725 		if (bounds_check_with_label(&rs->sc_dkdev, bp, wlabel) <= 0) {
726 			db1_printf(("Bounds check failed!!:%d %d\n",
727 				(int) bp->b_blkno, (int) wlabel));
728 			biodone(bp);
729 			return;
730 		}
731 	s = splbio();
732 
733 	bp->b_resid = 0;
734 
735 	/* stuff it onto our queue */
736 	BUFQ_PUT(rs->buf_queue, bp);
737 
738 	/* scheduled the IO to happen at the next convenient time */
739 	wakeup(&(raidPtrs[raidID]->iodone));
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 lwp *l)
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 		pseudo_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 	pseudo_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 bf,
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 = bf;
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 lwp *l, struct vnode **vpp)
2140 {
2141 	struct nameidata nd;
2142 	struct vnode *vp;
2143 	struct proc *p;
2144 	struct vattr va;
2145 	int     error;
2146 
2147 	p = l ? l->l_proc : NULL;
2148 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, l);
2149 	if ((error = vn_open(&nd, FREAD | FWRITE, 0)) != 0) {
2150 		return (error);
2151 	}
2152 	vp = nd.ni_vp;
2153 	if (vp->v_usecount > 1) {
2154 		VOP_UNLOCK(vp, 0);
2155 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2156 		return (EBUSY);
2157 	}
2158 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, l)) != 0) {
2159 		VOP_UNLOCK(vp, 0);
2160 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2161 		return (error);
2162 	}
2163 	/* XXX: eventually we should handle VREG, too. */
2164 	if (va.va_type != VBLK) {
2165 		VOP_UNLOCK(vp, 0);
2166 		(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2167 		return (ENOTBLK);
2168 	}
2169 	VOP_UNLOCK(vp, 0);
2170 	*vpp = vp;
2171 	return (0);
2172 }
2173 /*
2174  * Wait interruptibly for an exclusive lock.
2175  *
2176  * XXX
2177  * Several drivers do this; it should be abstracted and made MP-safe.
2178  * (Hmm... where have we seen this warning before :->  GO )
2179  */
2180 static int
2181 raidlock(struct raid_softc *rs)
2182 {
2183 	int     error;
2184 
2185 	while ((rs->sc_flags & RAIDF_LOCKED) != 0) {
2186 		rs->sc_flags |= RAIDF_WANTED;
2187 		if ((error =
2188 			tsleep(rs, PRIBIO | PCATCH, "raidlck", 0)) != 0)
2189 			return (error);
2190 	}
2191 	rs->sc_flags |= RAIDF_LOCKED;
2192 	return (0);
2193 }
2194 /*
2195  * Unlock and wake up any waiters.
2196  */
2197 static void
2198 raidunlock(struct raid_softc *rs)
2199 {
2200 
2201 	rs->sc_flags &= ~RAIDF_LOCKED;
2202 	if ((rs->sc_flags & RAIDF_WANTED) != 0) {
2203 		rs->sc_flags &= ~RAIDF_WANTED;
2204 		wakeup(rs);
2205 	}
2206 }
2207 
2208 
2209 #define RF_COMPONENT_INFO_OFFSET  16384 /* bytes */
2210 #define RF_COMPONENT_INFO_SIZE     1024 /* bytes */
2211 
2212 int
2213 raidmarkclean(dev_t dev, struct vnode *b_vp, int mod_counter)
2214 {
2215 	RF_ComponentLabel_t clabel;
2216 	raidread_component_label(dev, b_vp, &clabel);
2217 	clabel.mod_counter = mod_counter;
2218 	clabel.clean = RF_RAID_CLEAN;
2219 	raidwrite_component_label(dev, b_vp, &clabel);
2220 	return(0);
2221 }
2222 
2223 
2224 int
2225 raidmarkdirty(dev_t dev, struct vnode *b_vp, int mod_counter)
2226 {
2227 	RF_ComponentLabel_t clabel;
2228 	raidread_component_label(dev, b_vp, &clabel);
2229 	clabel.mod_counter = mod_counter;
2230 	clabel.clean = RF_RAID_DIRTY;
2231 	raidwrite_component_label(dev, b_vp, &clabel);
2232 	return(0);
2233 }
2234 
2235 /* ARGSUSED */
2236 int
2237 raidread_component_label(dev_t dev, struct vnode *b_vp,
2238 			 RF_ComponentLabel_t *clabel)
2239 {
2240 	struct buf *bp;
2241 	const struct bdevsw *bdev;
2242 	int error;
2243 
2244 	/* XXX should probably ensure that we don't try to do this if
2245 	   someone has changed rf_protected_sectors. */
2246 
2247 	if (b_vp == NULL) {
2248 		/* For whatever reason, this component is not valid.
2249 		   Don't try to read a component label from it. */
2250 		return(EINVAL);
2251 	}
2252 
2253 	/* get a block of the appropriate size... */
2254 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2255 	bp->b_dev = dev;
2256 
2257 	/* get our ducks in a row for the read */
2258 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2259 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2260 	bp->b_flags |= B_READ;
2261  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2262 
2263 	bdev = bdevsw_lookup(bp->b_dev);
2264 	if (bdev == NULL)
2265 		return (ENXIO);
2266 	(*bdev->d_strategy)(bp);
2267 
2268 	error = biowait(bp);
2269 
2270 	if (!error) {
2271 		memcpy(clabel, bp->b_data,
2272 		       sizeof(RF_ComponentLabel_t));
2273         }
2274 
2275 	brelse(bp);
2276 	return(error);
2277 }
2278 /* ARGSUSED */
2279 int
2280 raidwrite_component_label(dev_t dev, struct vnode *b_vp,
2281 			  RF_ComponentLabel_t *clabel)
2282 {
2283 	struct buf *bp;
2284 	const struct bdevsw *bdev;
2285 	int error;
2286 
2287 	/* get a block of the appropriate size... */
2288 	bp = geteblk((int)RF_COMPONENT_INFO_SIZE);
2289 	bp->b_dev = dev;
2290 
2291 	/* get our ducks in a row for the write */
2292 	bp->b_blkno = RF_COMPONENT_INFO_OFFSET / DEV_BSIZE;
2293 	bp->b_bcount = RF_COMPONENT_INFO_SIZE;
2294 	bp->b_flags |= B_WRITE;
2295  	bp->b_resid = RF_COMPONENT_INFO_SIZE / DEV_BSIZE;
2296 
2297 	memset(bp->b_data, 0, RF_COMPONENT_INFO_SIZE );
2298 
2299 	memcpy(bp->b_data, clabel, sizeof(RF_ComponentLabel_t));
2300 
2301 	bdev = bdevsw_lookup(bp->b_dev);
2302 	if (bdev == NULL)
2303 		return (ENXIO);
2304 	(*bdev->d_strategy)(bp);
2305 	error = biowait(bp);
2306 	brelse(bp);
2307 	if (error) {
2308 #if 1
2309 		printf("Failed to write RAID component info!\n");
2310 #endif
2311 	}
2312 
2313 	return(error);
2314 }
2315 
2316 void
2317 rf_markalldirty(RF_Raid_t *raidPtr)
2318 {
2319 	RF_ComponentLabel_t clabel;
2320 	int sparecol;
2321 	int c;
2322 	int j;
2323 	int scol = -1;
2324 
2325 	raidPtr->mod_counter++;
2326 	for (c = 0; c < raidPtr->numCol; c++) {
2327 		/* we don't want to touch (at all) a disk that has
2328 		   failed */
2329 		if (!RF_DEAD_DISK(raidPtr->Disks[c].status)) {
2330 			raidread_component_label(
2331 						 raidPtr->Disks[c].dev,
2332 						 raidPtr->raid_cinfo[c].ci_vp,
2333 						 &clabel);
2334 			if (clabel.status == rf_ds_spared) {
2335 				/* XXX do something special...
2336 				   but whatever you do, don't
2337 				   try to access it!! */
2338 			} else {
2339 				raidmarkdirty(
2340 					      raidPtr->Disks[c].dev,
2341 					      raidPtr->raid_cinfo[c].ci_vp,
2342 					      raidPtr->mod_counter);
2343 			}
2344 		}
2345 	}
2346 
2347 	for( c = 0; c < raidPtr->numSpare ; c++) {
2348 		sparecol = raidPtr->numCol + c;
2349 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2350 			/*
2351 
2352 			   we claim this disk is "optimal" if it's
2353 			   rf_ds_used_spare, as that means it should be
2354 			   directly substitutable for the disk it replaced.
2355 			   We note that too...
2356 
2357 			 */
2358 
2359 			for(j=0;j<raidPtr->numCol;j++) {
2360 				if (raidPtr->Disks[j].spareCol == sparecol) {
2361 					scol = j;
2362 					break;
2363 				}
2364 			}
2365 
2366 			raidread_component_label(
2367 				 raidPtr->Disks[sparecol].dev,
2368 				 raidPtr->raid_cinfo[sparecol].ci_vp,
2369 				 &clabel);
2370 			/* make sure status is noted */
2371 
2372 			raid_init_component_label(raidPtr, &clabel);
2373 
2374 			clabel.row = 0;
2375 			clabel.column = scol;
2376 			/* Note: we *don't* change status from rf_ds_used_spare
2377 			   to rf_ds_optimal */
2378 			/* clabel.status = rf_ds_optimal; */
2379 
2380 			raidmarkdirty(raidPtr->Disks[sparecol].dev,
2381 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2382 				      raidPtr->mod_counter);
2383 		}
2384 	}
2385 }
2386 
2387 
2388 void
2389 rf_update_component_labels(RF_Raid_t *raidPtr, int final)
2390 {
2391 	RF_ComponentLabel_t clabel;
2392 	int sparecol;
2393 	int c;
2394 	int j;
2395 	int scol;
2396 
2397 	scol = -1;
2398 
2399 	/* XXX should do extra checks to make sure things really are clean,
2400 	   rather than blindly setting the clean bit... */
2401 
2402 	raidPtr->mod_counter++;
2403 
2404 	for (c = 0; c < raidPtr->numCol; c++) {
2405 		if (raidPtr->Disks[c].status == rf_ds_optimal) {
2406 			raidread_component_label(
2407 						 raidPtr->Disks[c].dev,
2408 						 raidPtr->raid_cinfo[c].ci_vp,
2409 						 &clabel);
2410 				/* make sure status is noted */
2411 			clabel.status = rf_ds_optimal;
2412 				/* bump the counter */
2413 			clabel.mod_counter = raidPtr->mod_counter;
2414 
2415 			raidwrite_component_label(
2416 						  raidPtr->Disks[c].dev,
2417 						  raidPtr->raid_cinfo[c].ci_vp,
2418 						  &clabel);
2419 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2420 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2421 					raidmarkclean(
2422 						      raidPtr->Disks[c].dev,
2423 						      raidPtr->raid_cinfo[c].ci_vp,
2424 						      raidPtr->mod_counter);
2425 				}
2426 			}
2427 		}
2428 		/* else we don't touch it.. */
2429 	}
2430 
2431 	for( c = 0; c < raidPtr->numSpare ; c++) {
2432 		sparecol = raidPtr->numCol + c;
2433 		/* Need to ensure that the reconstruct actually completed! */
2434 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
2435 			/*
2436 
2437 			   we claim this disk is "optimal" if it's
2438 			   rf_ds_used_spare, as that means it should be
2439 			   directly substitutable for the disk it replaced.
2440 			   We note that too...
2441 
2442 			 */
2443 
2444 			for(j=0;j<raidPtr->numCol;j++) {
2445 				if (raidPtr->Disks[j].spareCol == sparecol) {
2446 					scol = j;
2447 					break;
2448 				}
2449 			}
2450 
2451 			/* XXX shouldn't *really* need this... */
2452 			raidread_component_label(
2453 				      raidPtr->Disks[sparecol].dev,
2454 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2455 				      &clabel);
2456 			/* make sure status is noted */
2457 
2458 			raid_init_component_label(raidPtr, &clabel);
2459 
2460 			clabel.mod_counter = raidPtr->mod_counter;
2461 			clabel.column = scol;
2462 			clabel.status = rf_ds_optimal;
2463 
2464 			raidwrite_component_label(
2465 				      raidPtr->Disks[sparecol].dev,
2466 				      raidPtr->raid_cinfo[sparecol].ci_vp,
2467 				      &clabel);
2468 			if (final == RF_FINAL_COMPONENT_UPDATE) {
2469 				if (raidPtr->parity_good == RF_RAID_CLEAN) {
2470 					raidmarkclean( raidPtr->Disks[sparecol].dev,
2471 						       raidPtr->raid_cinfo[sparecol].ci_vp,
2472 						       raidPtr->mod_counter);
2473 				}
2474 			}
2475 		}
2476 	}
2477 }
2478 
2479 void
2480 rf_close_component(RF_Raid_t *raidPtr, struct vnode *vp, int auto_configured)
2481 {
2482 	struct proc *p;
2483 	struct lwp *l;
2484 
2485 	p = raidPtr->engine_thread;
2486 	l = LIST_FIRST(&p->p_lwps);
2487 
2488 	if (vp != NULL) {
2489 		if (auto_configured == 1) {
2490 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2491 			VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2492 			vput(vp);
2493 
2494 		} else {
2495 			(void) vn_close(vp, FREAD | FWRITE, p->p_ucred, l);
2496 		}
2497 	}
2498 }
2499 
2500 
2501 void
2502 rf_UnconfigureVnodes(RF_Raid_t *raidPtr)
2503 {
2504 	int r,c;
2505 	struct vnode *vp;
2506 	int acd;
2507 
2508 
2509 	/* We take this opportunity to close the vnodes like we should.. */
2510 
2511 	for (c = 0; c < raidPtr->numCol; c++) {
2512 		vp = raidPtr->raid_cinfo[c].ci_vp;
2513 		acd = raidPtr->Disks[c].auto_configured;
2514 		rf_close_component(raidPtr, vp, acd);
2515 		raidPtr->raid_cinfo[c].ci_vp = NULL;
2516 		raidPtr->Disks[c].auto_configured = 0;
2517 	}
2518 
2519 	for (r = 0; r < raidPtr->numSpare; r++) {
2520 		vp = raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp;
2521 		acd = raidPtr->Disks[raidPtr->numCol + r].auto_configured;
2522 		rf_close_component(raidPtr, vp, acd);
2523 		raidPtr->raid_cinfo[raidPtr->numCol + r].ci_vp = NULL;
2524 		raidPtr->Disks[raidPtr->numCol + r].auto_configured = 0;
2525 	}
2526 }
2527 
2528 
2529 void
2530 rf_ReconThread(struct rf_recon_req *req)
2531 {
2532 	int     s;
2533 	RF_Raid_t *raidPtr;
2534 
2535 	s = splbio();
2536 	raidPtr = (RF_Raid_t *) req->raidPtr;
2537 	raidPtr->recon_in_progress = 1;
2538 
2539 	rf_FailDisk((RF_Raid_t *) req->raidPtr, req->col,
2540 		    ((req->flags & RF_FDFLAGS_RECON) ? 1 : 0));
2541 
2542 	RF_Free(req, sizeof(*req));
2543 
2544 	raidPtr->recon_in_progress = 0;
2545 	splx(s);
2546 
2547 	/* That's all... */
2548 	kthread_exit(0);        /* does not return */
2549 }
2550 
2551 void
2552 rf_RewriteParityThread(RF_Raid_t *raidPtr)
2553 {
2554 	int retcode;
2555 	int s;
2556 
2557 	raidPtr->parity_rewrite_stripes_done = 0;
2558 	raidPtr->parity_rewrite_in_progress = 1;
2559 	s = splbio();
2560 	retcode = rf_RewriteParity(raidPtr);
2561 	splx(s);
2562 	if (retcode) {
2563 		printf("raid%d: Error re-writing parity!\n",raidPtr->raidid);
2564 	} else {
2565 		/* set the clean bit!  If we shutdown correctly,
2566 		   the clean bit on each component label will get
2567 		   set */
2568 		raidPtr->parity_good = RF_RAID_CLEAN;
2569 	}
2570 	raidPtr->parity_rewrite_in_progress = 0;
2571 
2572 	/* Anyone waiting for us to stop?  If so, inform them... */
2573 	if (raidPtr->waitShutdown) {
2574 		wakeup(&raidPtr->parity_rewrite_in_progress);
2575 	}
2576 
2577 	/* That's all... */
2578 	kthread_exit(0);        /* does not return */
2579 }
2580 
2581 
2582 void
2583 rf_CopybackThread(RF_Raid_t *raidPtr)
2584 {
2585 	int s;
2586 
2587 	raidPtr->copyback_in_progress = 1;
2588 	s = splbio();
2589 	rf_CopybackReconstructedData(raidPtr);
2590 	splx(s);
2591 	raidPtr->copyback_in_progress = 0;
2592 
2593 	/* That's all... */
2594 	kthread_exit(0);        /* does not return */
2595 }
2596 
2597 
2598 void
2599 rf_ReconstructInPlaceThread(struct rf_recon_req *req)
2600 {
2601 	int s;
2602 	RF_Raid_t *raidPtr;
2603 
2604 	s = splbio();
2605 	raidPtr = req->raidPtr;
2606 	raidPtr->recon_in_progress = 1;
2607 	rf_ReconstructInPlace(raidPtr, req->col);
2608 	RF_Free(req, sizeof(*req));
2609 	raidPtr->recon_in_progress = 0;
2610 	splx(s);
2611 
2612 	/* That's all... */
2613 	kthread_exit(0);        /* does not return */
2614 }
2615 
2616 RF_AutoConfig_t *
2617 rf_find_raid_components()
2618 {
2619 	struct vnode *vp;
2620 	struct disklabel label;
2621 	struct device *dv;
2622 	dev_t dev;
2623 	int bmajor;
2624 	int error;
2625 	int i;
2626 	int good_one;
2627 	RF_ComponentLabel_t *clabel;
2628 	RF_AutoConfig_t *ac_list;
2629 	RF_AutoConfig_t *ac;
2630 
2631 
2632 	/* initialize the AutoConfig list */
2633 	ac_list = NULL;
2634 
2635 	/* we begin by trolling through *all* the devices on the system */
2636 
2637 	for (dv = alldevs.tqh_first; dv != NULL;
2638 	     dv = dv->dv_list.tqe_next) {
2639 
2640 		/* we are only interested in disks... */
2641 		if (dv->dv_class != DV_DISK)
2642 			continue;
2643 
2644 		/* we don't care about floppies... */
2645 		if (!strcmp(dv->dv_cfdata->cf_name,"fd")) {
2646 			continue;
2647 		}
2648 
2649 		/* we don't care about CD's... */
2650 		if (!strcmp(dv->dv_cfdata->cf_name,"cd")) {
2651 			continue;
2652 		}
2653 
2654 		/* hdfd is the Atari/Hades floppy driver */
2655 		if (!strcmp(dv->dv_cfdata->cf_name,"hdfd")) {
2656 			continue;
2657 		}
2658 		/* fdisa is the Atari/Milan floppy driver */
2659 		if (!strcmp(dv->dv_cfdata->cf_name,"fdisa")) {
2660 			continue;
2661 		}
2662 
2663 		/* need to find the device_name_to_block_device_major stuff */
2664 		bmajor = devsw_name2blk(dv->dv_xname, NULL, 0);
2665 
2666 		/* get a vnode for the raw partition of this disk */
2667 
2668 		dev = MAKEDISKDEV(bmajor, dv->dv_unit, RAW_PART);
2669 		if (bdevvp(dev, &vp))
2670 			panic("RAID can't alloc vnode");
2671 
2672 		error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2673 
2674 		if (error) {
2675 			/* "Who cares."  Continue looking
2676 			   for something that exists*/
2677 			vput(vp);
2678 			continue;
2679 		}
2680 
2681 		/* Ok, the disk exists.  Go get the disklabel. */
2682 		error = VOP_IOCTL(vp, DIOCGDINFO, &label, FREAD, NOCRED, 0);
2683 		if (error) {
2684 			/*
2685 			 * XXX can't happen - open() would
2686 			 * have errored out (or faked up one)
2687 			 */
2688 			if (error != ENOTTY)
2689 				printf("RAIDframe: can't get label for dev "
2690 				    "%s (%d)\n", dv->dv_xname, error);
2691 		}
2692 
2693 		/* don't need this any more.  We'll allocate it again
2694 		   a little later if we really do... */
2695 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2696 		VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2697 		vput(vp);
2698 
2699 		if (error)
2700 			continue;
2701 
2702 		for (i=0; i < label.d_npartitions; i++) {
2703 			/* We only support partitions marked as RAID */
2704 			if (label.d_partitions[i].p_fstype != FS_RAID)
2705 				continue;
2706 
2707 			dev = MAKEDISKDEV(bmajor, dv->dv_unit, i);
2708 			if (bdevvp(dev, &vp))
2709 				panic("RAID can't alloc vnode");
2710 
2711 			error = VOP_OPEN(vp, FREAD, NOCRED, 0);
2712 			if (error) {
2713 				/* Whatever... */
2714 				vput(vp);
2715 				continue;
2716 			}
2717 
2718 			good_one = 0;
2719 
2720 			clabel = (RF_ComponentLabel_t *)
2721 				malloc(sizeof(RF_ComponentLabel_t),
2722 				       M_RAIDFRAME, M_NOWAIT);
2723 			if (clabel == NULL) {
2724 				/* XXX CLEANUP HERE */
2725 				printf("RAID auto config: out of memory!\n");
2726 				return(NULL); /* XXX probably should panic? */
2727 			}
2728 
2729 			if (!raidread_component_label(dev, vp, clabel)) {
2730 				/* Got the label.  Does it look reasonable? */
2731 				if (rf_reasonable_label(clabel) &&
2732 				    (clabel->partitionSize <=
2733 				     label.d_partitions[i].p_size)) {
2734 #if DEBUG
2735 					printf("Component on: %s%c: %d\n",
2736 					       dv->dv_xname, 'a'+i,
2737 					       label.d_partitions[i].p_size);
2738 					rf_print_component_label(clabel);
2739 #endif
2740 					/* if it's reasonable, add it,
2741 					   else ignore it. */
2742 					ac = (RF_AutoConfig_t *)
2743 						malloc(sizeof(RF_AutoConfig_t),
2744 						       M_RAIDFRAME,
2745 						       M_NOWAIT);
2746 					if (ac == NULL) {
2747 						/* XXX should panic?? */
2748 						return(NULL);
2749 					}
2750 
2751 					snprintf(ac->devname,
2752 					    sizeof(ac->devname), "%s%c",
2753 					    dv->dv_xname, 'a'+i);
2754 					ac->dev = dev;
2755 					ac->vp = vp;
2756 					ac->clabel = clabel;
2757 					ac->next = ac_list;
2758 					ac_list = ac;
2759 					good_one = 1;
2760 				}
2761 			}
2762 			if (!good_one) {
2763 				/* cleanup */
2764 				free(clabel, M_RAIDFRAME);
2765 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2766 				VOP_CLOSE(vp, FREAD | FWRITE, NOCRED, 0);
2767 				vput(vp);
2768 			}
2769 		}
2770 	}
2771 	return(ac_list);
2772 }
2773 
2774 static int
2775 rf_reasonable_label(RF_ComponentLabel_t *clabel)
2776 {
2777 
2778 	if (((clabel->version==RF_COMPONENT_LABEL_VERSION_1) ||
2779 	     (clabel->version==RF_COMPONENT_LABEL_VERSION)) &&
2780 	    ((clabel->clean == RF_RAID_CLEAN) ||
2781 	     (clabel->clean == RF_RAID_DIRTY)) &&
2782 	    clabel->row >=0 &&
2783 	    clabel->column >= 0 &&
2784 	    clabel->num_rows > 0 &&
2785 	    clabel->num_columns > 0 &&
2786 	    clabel->row < clabel->num_rows &&
2787 	    clabel->column < clabel->num_columns &&
2788 	    clabel->blockSize > 0 &&
2789 	    clabel->numBlocks > 0) {
2790 		/* label looks reasonable enough... */
2791 		return(1);
2792 	}
2793 	return(0);
2794 }
2795 
2796 
2797 #if DEBUG
2798 void
2799 rf_print_component_label(RF_ComponentLabel_t *clabel)
2800 {
2801 	printf("   Row: %d Column: %d Num Rows: %d Num Columns: %d\n",
2802 	       clabel->row, clabel->column,
2803 	       clabel->num_rows, clabel->num_columns);
2804 	printf("   Version: %d Serial Number: %d Mod Counter: %d\n",
2805 	       clabel->version, clabel->serial_number,
2806 	       clabel->mod_counter);
2807 	printf("   Clean: %s Status: %d\n",
2808 	       clabel->clean ? "Yes" : "No", clabel->status );
2809 	printf("   sectPerSU: %d SUsPerPU: %d SUsPerRU: %d\n",
2810 	       clabel->sectPerSU, clabel->SUsPerPU, clabel->SUsPerRU);
2811 	printf("   RAID Level: %c  blocksize: %d numBlocks: %d\n",
2812 	       (char) clabel->parityConfig, clabel->blockSize,
2813 	       clabel->numBlocks);
2814 	printf("   Autoconfig: %s\n", clabel->autoconfigure ? "Yes" : "No" );
2815 	printf("   Contains root partition: %s\n",
2816 	       clabel->root_partition ? "Yes" : "No" );
2817 	printf("   Last configured as: raid%d\n", clabel->last_unit );
2818 #if 0
2819 	   printf("   Config order: %d\n", clabel->config_order);
2820 #endif
2821 
2822 }
2823 #endif
2824 
2825 RF_ConfigSet_t *
2826 rf_create_auto_sets(RF_AutoConfig_t *ac_list)
2827 {
2828 	RF_AutoConfig_t *ac;
2829 	RF_ConfigSet_t *config_sets;
2830 	RF_ConfigSet_t *cset;
2831 	RF_AutoConfig_t *ac_next;
2832 
2833 
2834 	config_sets = NULL;
2835 
2836 	/* Go through the AutoConfig list, and figure out which components
2837 	   belong to what sets.  */
2838 	ac = ac_list;
2839 	while(ac!=NULL) {
2840 		/* we're going to putz with ac->next, so save it here
2841 		   for use at the end of the loop */
2842 		ac_next = ac->next;
2843 
2844 		if (config_sets == NULL) {
2845 			/* will need at least this one... */
2846 			config_sets = (RF_ConfigSet_t *)
2847 				malloc(sizeof(RF_ConfigSet_t),
2848 				       M_RAIDFRAME, M_NOWAIT);
2849 			if (config_sets == NULL) {
2850 				panic("rf_create_auto_sets: No memory!");
2851 			}
2852 			/* this one is easy :) */
2853 			config_sets->ac = ac;
2854 			config_sets->next = NULL;
2855 			config_sets->rootable = 0;
2856 			ac->next = NULL;
2857 		} else {
2858 			/* which set does this component fit into? */
2859 			cset = config_sets;
2860 			while(cset!=NULL) {
2861 				if (rf_does_it_fit(cset, ac)) {
2862 					/* looks like it matches... */
2863 					ac->next = cset->ac;
2864 					cset->ac = ac;
2865 					break;
2866 				}
2867 				cset = cset->next;
2868 			}
2869 			if (cset==NULL) {
2870 				/* didn't find a match above... new set..*/
2871 				cset = (RF_ConfigSet_t *)
2872 					malloc(sizeof(RF_ConfigSet_t),
2873 					       M_RAIDFRAME, M_NOWAIT);
2874 				if (cset == NULL) {
2875 					panic("rf_create_auto_sets: No memory!");
2876 				}
2877 				cset->ac = ac;
2878 				ac->next = NULL;
2879 				cset->next = config_sets;
2880 				cset->rootable = 0;
2881 				config_sets = cset;
2882 			}
2883 		}
2884 		ac = ac_next;
2885 	}
2886 
2887 
2888 	return(config_sets);
2889 }
2890 
2891 static int
2892 rf_does_it_fit(RF_ConfigSet_t *cset, RF_AutoConfig_t *ac)
2893 {
2894 	RF_ComponentLabel_t *clabel1, *clabel2;
2895 
2896 	/* If this one matches the *first* one in the set, that's good
2897 	   enough, since the other members of the set would have been
2898 	   through here too... */
2899 	/* note that we are not checking partitionSize here..
2900 
2901 	   Note that we are also not checking the mod_counters here.
2902 	   If everything else matches execpt the mod_counter, that's
2903 	   good enough for this test.  We will deal with the mod_counters
2904 	   a little later in the autoconfiguration process.
2905 
2906 	    (clabel1->mod_counter == clabel2->mod_counter) &&
2907 
2908 	   The reason we don't check for this is that failed disks
2909 	   will have lower modification counts.  If those disks are
2910 	   not added to the set they used to belong to, then they will
2911 	   form their own set, which may result in 2 different sets,
2912 	   for example, competing to be configured at raid0, and
2913 	   perhaps competing to be the root filesystem set.  If the
2914 	   wrong ones get configured, or both attempt to become /,
2915 	   weird behaviour and or serious lossage will occur.  Thus we
2916 	   need to bring them into the fold here, and kick them out at
2917 	   a later point.
2918 
2919 	*/
2920 
2921 	clabel1 = cset->ac->clabel;
2922 	clabel2 = ac->clabel;
2923 	if ((clabel1->version == clabel2->version) &&
2924 	    (clabel1->serial_number == clabel2->serial_number) &&
2925 	    (clabel1->num_rows == clabel2->num_rows) &&
2926 	    (clabel1->num_columns == clabel2->num_columns) &&
2927 	    (clabel1->sectPerSU == clabel2->sectPerSU) &&
2928 	    (clabel1->SUsPerPU == clabel2->SUsPerPU) &&
2929 	    (clabel1->SUsPerRU == clabel2->SUsPerRU) &&
2930 	    (clabel1->parityConfig == clabel2->parityConfig) &&
2931 	    (clabel1->maxOutstanding == clabel2->maxOutstanding) &&
2932 	    (clabel1->blockSize == clabel2->blockSize) &&
2933 	    (clabel1->numBlocks == clabel2->numBlocks) &&
2934 	    (clabel1->autoconfigure == clabel2->autoconfigure) &&
2935 	    (clabel1->root_partition == clabel2->root_partition) &&
2936 	    (clabel1->last_unit == clabel2->last_unit) &&
2937 	    (clabel1->config_order == clabel2->config_order)) {
2938 		/* if it get's here, it almost *has* to be a match */
2939 	} else {
2940 		/* it's not consistent with somebody in the set..
2941 		   punt */
2942 		return(0);
2943 	}
2944 	/* all was fine.. it must fit... */
2945 	return(1);
2946 }
2947 
2948 int
2949 rf_have_enough_components(RF_ConfigSet_t *cset)
2950 {
2951 	RF_AutoConfig_t *ac;
2952 	RF_AutoConfig_t *auto_config;
2953 	RF_ComponentLabel_t *clabel;
2954 	int c;
2955 	int num_cols;
2956 	int num_missing;
2957 	int mod_counter;
2958 	int mod_counter_found;
2959 	int even_pair_failed;
2960 	char parity_type;
2961 
2962 
2963 	/* check to see that we have enough 'live' components
2964 	   of this set.  If so, we can configure it if necessary */
2965 
2966 	num_cols = cset->ac->clabel->num_columns;
2967 	parity_type = cset->ac->clabel->parityConfig;
2968 
2969 	/* XXX Check for duplicate components!?!?!? */
2970 
2971 	/* Determine what the mod_counter is supposed to be for this set. */
2972 
2973 	mod_counter_found = 0;
2974 	mod_counter = 0;
2975 	ac = cset->ac;
2976 	while(ac!=NULL) {
2977 		if (mod_counter_found==0) {
2978 			mod_counter = ac->clabel->mod_counter;
2979 			mod_counter_found = 1;
2980 		} else {
2981 			if (ac->clabel->mod_counter > mod_counter) {
2982 				mod_counter = ac->clabel->mod_counter;
2983 			}
2984 		}
2985 		ac = ac->next;
2986 	}
2987 
2988 	num_missing = 0;
2989 	auto_config = cset->ac;
2990 
2991 	even_pair_failed = 0;
2992 	for(c=0; c<num_cols; c++) {
2993 		ac = auto_config;
2994 		while(ac!=NULL) {
2995 			if ((ac->clabel->column == c) &&
2996 			    (ac->clabel->mod_counter == mod_counter)) {
2997 				/* it's this one... */
2998 #if DEBUG
2999 				printf("Found: %s at %d\n",
3000 				       ac->devname,c);
3001 #endif
3002 				break;
3003 			}
3004 			ac=ac->next;
3005 		}
3006 		if (ac==NULL) {
3007 				/* Didn't find one here! */
3008 				/* special case for RAID 1, especially
3009 				   where there are more than 2
3010 				   components (where RAIDframe treats
3011 				   things a little differently :( ) */
3012 			if (parity_type == '1') {
3013 				if (c%2 == 0) { /* even component */
3014 					even_pair_failed = 1;
3015 				} else { /* odd component.  If
3016 					    we're failed, and
3017 					    so is the even
3018 					    component, it's
3019 					    "Good Night, Charlie" */
3020 					if (even_pair_failed == 1) {
3021 						return(0);
3022 					}
3023 				}
3024 			} else {
3025 				/* normal accounting */
3026 				num_missing++;
3027 			}
3028 		}
3029 		if ((parity_type == '1') && (c%2 == 1)) {
3030 				/* Just did an even component, and we didn't
3031 				   bail.. reset the even_pair_failed flag,
3032 				   and go on to the next component.... */
3033 			even_pair_failed = 0;
3034 		}
3035 	}
3036 
3037 	clabel = cset->ac->clabel;
3038 
3039 	if (((clabel->parityConfig == '0') && (num_missing > 0)) ||
3040 	    ((clabel->parityConfig == '4') && (num_missing > 1)) ||
3041 	    ((clabel->parityConfig == '5') && (num_missing > 1))) {
3042 		/* XXX this needs to be made *much* more general */
3043 		/* Too many failures */
3044 		return(0);
3045 	}
3046 	/* otherwise, all is well, and we've got enough to take a kick
3047 	   at autoconfiguring this set */
3048 	return(1);
3049 }
3050 
3051 void
3052 rf_create_configuration(RF_AutoConfig_t *ac, RF_Config_t *config,
3053 			RF_Raid_t *raidPtr)
3054 {
3055 	RF_ComponentLabel_t *clabel;
3056 	int i;
3057 
3058 	clabel = ac->clabel;
3059 
3060 	/* 1. Fill in the common stuff */
3061 	config->numRow = clabel->num_rows = 1;
3062 	config->numCol = clabel->num_columns;
3063 	config->numSpare = 0; /* XXX should this be set here? */
3064 	config->sectPerSU = clabel->sectPerSU;
3065 	config->SUsPerPU = clabel->SUsPerPU;
3066 	config->SUsPerRU = clabel->SUsPerRU;
3067 	config->parityConfig = clabel->parityConfig;
3068 	/* XXX... */
3069 	strcpy(config->diskQueueType,"fifo");
3070 	config->maxOutstandingDiskReqs = clabel->maxOutstanding;
3071 	config->layoutSpecificSize = 0; /* XXX ?? */
3072 
3073 	while(ac!=NULL) {
3074 		/* row/col values will be in range due to the checks
3075 		   in reasonable_label() */
3076 		strcpy(config->devnames[0][ac->clabel->column],
3077 		       ac->devname);
3078 		ac = ac->next;
3079 	}
3080 
3081 	for(i=0;i<RF_MAXDBGV;i++) {
3082 		config->debugVars[i][0] = 0;
3083 	}
3084 }
3085 
3086 int
3087 rf_set_autoconfig(RF_Raid_t *raidPtr, int new_value)
3088 {
3089 	RF_ComponentLabel_t clabel;
3090 	struct vnode *vp;
3091 	dev_t dev;
3092 	int column;
3093 	int sparecol;
3094 
3095 	raidPtr->autoconfigure = new_value;
3096 
3097 	for(column=0; column<raidPtr->numCol; column++) {
3098 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3099 			dev = raidPtr->Disks[column].dev;
3100 			vp = raidPtr->raid_cinfo[column].ci_vp;
3101 			raidread_component_label(dev, vp, &clabel);
3102 			clabel.autoconfigure = new_value;
3103 			raidwrite_component_label(dev, vp, &clabel);
3104 		}
3105 	}
3106 	for(column = 0; column < raidPtr->numSpare ; column++) {
3107 		sparecol = raidPtr->numCol + column;
3108 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3109 			dev = raidPtr->Disks[sparecol].dev;
3110 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3111 			raidread_component_label(dev, vp, &clabel);
3112 			clabel.autoconfigure = new_value;
3113 			raidwrite_component_label(dev, vp, &clabel);
3114 		}
3115 	}
3116 	return(new_value);
3117 }
3118 
3119 int
3120 rf_set_rootpartition(RF_Raid_t *raidPtr, int new_value)
3121 {
3122 	RF_ComponentLabel_t clabel;
3123 	struct vnode *vp;
3124 	dev_t dev;
3125 	int column;
3126 	int sparecol;
3127 
3128 	raidPtr->root_partition = new_value;
3129 	for(column=0; column<raidPtr->numCol; column++) {
3130 		if (raidPtr->Disks[column].status == rf_ds_optimal) {
3131 			dev = raidPtr->Disks[column].dev;
3132 			vp = raidPtr->raid_cinfo[column].ci_vp;
3133 			raidread_component_label(dev, vp, &clabel);
3134 			clabel.root_partition = new_value;
3135 			raidwrite_component_label(dev, vp, &clabel);
3136 		}
3137 	}
3138 	for(column = 0; column < raidPtr->numSpare ; column++) {
3139 		sparecol = raidPtr->numCol + column;
3140 		if (raidPtr->Disks[sparecol].status == rf_ds_used_spare) {
3141 			dev = raidPtr->Disks[sparecol].dev;
3142 			vp = raidPtr->raid_cinfo[sparecol].ci_vp;
3143 			raidread_component_label(dev, vp, &clabel);
3144 			clabel.root_partition = new_value;
3145 			raidwrite_component_label(dev, vp, &clabel);
3146 		}
3147 	}
3148 	return(new_value);
3149 }
3150 
3151 void
3152 rf_release_all_vps(RF_ConfigSet_t *cset)
3153 {
3154 	RF_AutoConfig_t *ac;
3155 
3156 	ac = cset->ac;
3157 	while(ac!=NULL) {
3158 		/* Close the vp, and give it back */
3159 		if (ac->vp) {
3160 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
3161 			VOP_CLOSE(ac->vp, FREAD, NOCRED, 0);
3162 			vput(ac->vp);
3163 			ac->vp = NULL;
3164 		}
3165 		ac = ac->next;
3166 	}
3167 }
3168 
3169 
3170 void
3171 rf_cleanup_config_set(RF_ConfigSet_t *cset)
3172 {
3173 	RF_AutoConfig_t *ac;
3174 	RF_AutoConfig_t *next_ac;
3175 
3176 	ac = cset->ac;
3177 	while(ac!=NULL) {
3178 		next_ac = ac->next;
3179 		/* nuke the label */
3180 		free(ac->clabel, M_RAIDFRAME);
3181 		/* cleanup the config structure */
3182 		free(ac, M_RAIDFRAME);
3183 		/* "next.." */
3184 		ac = next_ac;
3185 	}
3186 	/* and, finally, nuke the config set */
3187 	free(cset, M_RAIDFRAME);
3188 }
3189 
3190 
3191 void
3192 raid_init_component_label(RF_Raid_t *raidPtr, RF_ComponentLabel_t *clabel)
3193 {
3194 	/* current version number */
3195 	clabel->version = RF_COMPONENT_LABEL_VERSION;
3196 	clabel->serial_number = raidPtr->serial_number;
3197 	clabel->mod_counter = raidPtr->mod_counter;
3198 	clabel->num_rows = 1;
3199 	clabel->num_columns = raidPtr->numCol;
3200 	clabel->clean = RF_RAID_DIRTY; /* not clean */
3201 	clabel->status = rf_ds_optimal; /* "It's good!" */
3202 
3203 	clabel->sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
3204 	clabel->SUsPerPU = raidPtr->Layout.SUsPerPU;
3205 	clabel->SUsPerRU = raidPtr->Layout.SUsPerRU;
3206 
3207 	clabel->blockSize = raidPtr->bytesPerSector;
3208 	clabel->numBlocks = raidPtr->sectorsPerDisk;
3209 
3210 	/* XXX not portable */
3211 	clabel->parityConfig = raidPtr->Layout.map->parityConfig;
3212 	clabel->maxOutstanding = raidPtr->maxOutstanding;
3213 	clabel->autoconfigure = raidPtr->autoconfigure;
3214 	clabel->root_partition = raidPtr->root_partition;
3215 	clabel->last_unit = raidPtr->raidid;
3216 	clabel->config_order = raidPtr->config_order;
3217 }
3218 
3219 int
3220 rf_auto_config_set(RF_ConfigSet_t *cset, int *unit)
3221 {
3222 	RF_Raid_t *raidPtr;
3223 	RF_Config_t *config;
3224 	int raidID;
3225 	int retcode;
3226 
3227 #if DEBUG
3228 	printf("RAID autoconfigure\n");
3229 #endif
3230 
3231 	retcode = 0;
3232 	*unit = -1;
3233 
3234 	/* 1. Create a config structure */
3235 
3236 	config = (RF_Config_t *)malloc(sizeof(RF_Config_t),
3237 				       M_RAIDFRAME,
3238 				       M_NOWAIT);
3239 	if (config==NULL) {
3240 		printf("Out of mem!?!?\n");
3241 				/* XXX do something more intelligent here. */
3242 		return(1);
3243 	}
3244 
3245 	memset(config, 0, sizeof(RF_Config_t));
3246 
3247 	/*
3248 	   2. Figure out what RAID ID this one is supposed to live at
3249 	   See if we can get the same RAID dev that it was configured
3250 	   on last time..
3251 	*/
3252 
3253 	raidID = cset->ac->clabel->last_unit;
3254 	if ((raidID < 0) || (raidID >= numraid)) {
3255 		/* let's not wander off into lala land. */
3256 		raidID = numraid - 1;
3257 	}
3258 	if (raidPtrs[raidID]->valid != 0) {
3259 
3260 		/*
3261 		   Nope... Go looking for an alternative...
3262 		   Start high so we don't immediately use raid0 if that's
3263 		   not taken.
3264 		*/
3265 
3266 		for(raidID = numraid - 1; raidID >= 0; raidID--) {
3267 			if (raidPtrs[raidID]->valid == 0) {
3268 				/* can use this one! */
3269 				break;
3270 			}
3271 		}
3272 	}
3273 
3274 	if (raidID < 0) {
3275 		/* punt... */
3276 		printf("Unable to auto configure this set!\n");
3277 		printf("(Out of RAID devs!)\n");
3278 		return(1);
3279 	}
3280 
3281 #if DEBUG
3282 	printf("Configuring raid%d:\n",raidID);
3283 #endif
3284 
3285 	raidPtr = raidPtrs[raidID];
3286 
3287 	/* XXX all this stuff should be done SOMEWHERE ELSE! */
3288 	raidPtr->raidid = raidID;
3289 	raidPtr->openings = RAIDOUTSTANDING;
3290 
3291 	/* 3. Build the configuration structure */
3292 	rf_create_configuration(cset->ac, config, raidPtr);
3293 
3294 	/* 4. Do the configuration */
3295 	retcode = rf_Configure(raidPtr, config, cset->ac);
3296 
3297 	if (retcode == 0) {
3298 
3299 		raidinit(raidPtrs[raidID]);
3300 
3301 		rf_markalldirty(raidPtrs[raidID]);
3302 		raidPtrs[raidID]->autoconfigure = 1; /* XXX do this here? */
3303 		if (cset->ac->clabel->root_partition==1) {
3304 			/* everything configured just fine.  Make a note
3305 			   that this set is eligible to be root. */
3306 			cset->rootable = 1;
3307 			/* XXX do this here? */
3308 			raidPtrs[raidID]->root_partition = 1;
3309 		}
3310 	}
3311 
3312 	/* 5. Cleanup */
3313 	free(config, M_RAIDFRAME);
3314 
3315 	*unit = raidID;
3316 	return(retcode);
3317 }
3318 
3319 void
3320 rf_disk_unbusy(RF_RaidAccessDesc_t *desc)
3321 {
3322 	struct buf *bp;
3323 
3324 	bp = (struct buf *)desc->bp;
3325 	disk_unbusy(&raid_softc[desc->raidPtr->raidid].sc_dkdev,
3326 	    (bp->b_bcount - bp->b_resid), (bp->b_flags & B_READ));
3327 }
3328 
3329 void
3330 rf_pool_init(struct pool *p, size_t size, const char *w_chan,
3331 	     size_t xmin, size_t xmax)
3332 {
3333 	pool_init(p, size, 0, 0, 0, w_chan, NULL);
3334 	pool_sethiwat(p, xmax);
3335 	pool_prime(p, xmin);
3336 	pool_setlowat(p, xmin);
3337 }
3338 
3339 /*
3340  * rf_buf_queue_check(int raidid) -- looks into the buf_queue to see
3341  * if there is IO pending and if that IO could possibly be done for a
3342  * given RAID set.  Returns 0 if IO is waiting and can be done, 1
3343  * otherwise.
3344  *
3345  */
3346 
3347 int
3348 rf_buf_queue_check(int raidid)
3349 {
3350 	if ((BUFQ_PEEK(raid_softc[raidid].buf_queue) != NULL) &&
3351 	    raidPtrs[raidid]->openings > 0) {
3352 		/* there is work to do */
3353 		return 0;
3354 	}
3355 	/* default is nothing to do */
3356 	return 1;
3357 }
3358