xref: /netbsd-src/sys/dev/raidframe/rf_disks.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: rf_disks.c,v 1.49 2003/12/30 21:59:03 oster Exp $	*/
2 /*-
3  * Copyright (c) 1999 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
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) 1995 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Author: Mark Holland
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  */
64 
65 /***************************************************************
66  * rf_disks.c -- code to perform operations on the actual disks
67  ***************************************************************/
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: rf_disks.c,v 1.49 2003/12/30 21:59:03 oster Exp $");
71 
72 #include <dev/raidframe/raidframevar.h>
73 
74 #include "rf_raid.h"
75 #include "rf_alloclist.h"
76 #include "rf_utils.h"
77 #include "rf_general.h"
78 #include "rf_options.h"
79 #include "rf_kintf.h"
80 #include "rf_netbsd.h"
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/proc.h>
85 #include <sys/ioctl.h>
86 #include <sys/fcntl.h>
87 #include <sys/vnode.h>
88 
89 static int rf_AllocDiskStructures(RF_Raid_t *, RF_Config_t *);
90 static void rf_print_label_status( RF_Raid_t *, int, char *,
91 				  RF_ComponentLabel_t *);
92 static int rf_check_label_vitals( RF_Raid_t *, int, int, char *,
93 				  RF_ComponentLabel_t *, int, int );
94 
95 #define DPRINTF6(a,b,c,d,e,f) if (rf_diskDebug) printf(a,b,c,d,e,f)
96 #define DPRINTF7(a,b,c,d,e,f,g) if (rf_diskDebug) printf(a,b,c,d,e,f,g)
97 
98 /**************************************************************************
99  *
100  * initialize the disks comprising the array
101  *
102  * We want the spare disks to have regular row,col numbers so that we can
103  * easily substitue a spare for a failed disk.  But, the driver code assumes
104  * throughout that the array contains numRow by numCol _non-spare_ disks, so
105  * it's not clear how to fit in the spares.  This is an unfortunate holdover
106  * from raidSim.  The quick and dirty fix is to make row zero bigger than the
107  * rest, and put all the spares in it.  This probably needs to get changed
108  * eventually.
109  *
110  **************************************************************************/
111 
112 int
113 rf_ConfigureDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
114 		  RF_Config_t *cfgPtr)
115 {
116 	RF_RaidDisk_t *disks;
117 	RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL;
118 	RF_RowCol_t c;
119 	int bs, ret;
120 	unsigned i, count, foundone = 0, numFailuresThisRow;
121 	int force;
122 
123 	force = cfgPtr->force;
124 
125 	ret = rf_AllocDiskStructures(raidPtr, cfgPtr);
126 	if (ret)
127 		goto fail;
128 
129 	disks = raidPtr->Disks;
130 
131 	numFailuresThisRow = 0;
132 	for (c = 0; c < raidPtr->numCol; c++) {
133 		ret = rf_ConfigureDisk(raidPtr,
134 				       &cfgPtr->devnames[0][c][0],
135 				       &disks[c], c);
136 
137 		if (ret)
138 			goto fail;
139 
140 		if (disks[c].status == rf_ds_optimal) {
141 			raidread_component_label(
142 						 raidPtr->raid_cinfo[c].ci_dev,
143 						 raidPtr->raid_cinfo[c].ci_vp,
144 						 &raidPtr->raid_cinfo[c].ci_label);
145 		}
146 
147 		if (disks[c].status != rf_ds_optimal) {
148 			numFailuresThisRow++;
149 		} else {
150 			if (disks[c].numBlocks < min_numblks)
151 				min_numblks = disks[c].numBlocks;
152 			DPRINTF6("Disk at col %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n",
153 				 c, disks[c].devname,
154 				 (long int) disks[c].numBlocks,
155 				 disks[c].blockSize,
156 				 (long int) disks[c].numBlocks *
157 				 disks[c].blockSize / 1024 / 1024);
158 		}
159 	}
160 	/* XXX fix for n-fault tolerant */
161 	/* XXX this should probably check to see how many failures
162 	   we can handle for this configuration! */
163 	if (numFailuresThisRow > 0)
164 		raidPtr->status = rf_rs_degraded;
165 
166 	/* all disks must be the same size & have the same block size, bs must
167 	 * be a power of 2 */
168 	bs = 0;
169 	foundone = 0;
170 	for (c = 0; c < raidPtr->numCol; c++) {
171 		if (disks[c].status == rf_ds_optimal) {
172 			bs = disks[c].blockSize;
173 			foundone = 1;
174 			break;
175 		}
176 	}
177 	if (!foundone) {
178 		RF_ERRORMSG("RAIDFRAME: Did not find any live disks in the array.\n");
179 		ret = EINVAL;
180 		goto fail;
181 	}
182 	for (count = 0, i = 1; i; i <<= 1)
183 		if (bs & i)
184 			count++;
185 	if (count != 1) {
186 		RF_ERRORMSG1("Error: block size on disks (%d) must be a power of 2\n", bs);
187 		ret = EINVAL;
188 		goto fail;
189 	}
190 
191 	if (rf_CheckLabels( raidPtr, cfgPtr )) {
192 		printf("raid%d: There were fatal errors\n", raidPtr->raidid);
193 		if (force != 0) {
194 			printf("raid%d: Fatal errors being ignored.\n",
195 			       raidPtr->raidid);
196 		} else {
197 			ret = EINVAL;
198 			goto fail;
199 		}
200 	}
201 
202 	for (c = 0; c < raidPtr->numCol; c++) {
203 		if (disks[c].status == rf_ds_optimal) {
204 			if (disks[c].blockSize != bs) {
205 				RF_ERRORMSG1("Error: block size of disk at c %d different from disk at c 0\n", c);
206 				ret = EINVAL;
207 				goto fail;
208 			}
209 			if (disks[c].numBlocks != min_numblks) {
210 				RF_ERRORMSG2("WARNING: truncating disk at c %d to %d blocks\n",
211 					     c, (int) min_numblks);
212 				disks[c].numBlocks = min_numblks;
213 			}
214 		}
215 	}
216 
217 	raidPtr->sectorsPerDisk = min_numblks;
218 	raidPtr->logBytesPerSector = ffs(bs) - 1;
219 	raidPtr->bytesPerSector = bs;
220 	raidPtr->sectorMask = bs - 1;
221 	return (0);
222 
223 fail:
224 
225 	rf_UnconfigureVnodes( raidPtr );
226 
227 	return (ret);
228 }
229 
230 
231 /****************************************************************************
232  * set up the data structures describing the spare disks in the array
233  * recall from the above comment that the spare disk descriptors are stored
234  * in row zero, which is specially expanded to hold them.
235  ****************************************************************************/
236 int
237 rf_ConfigureSpareDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
238 		       RF_Config_t *cfgPtr)
239 {
240 	int     i, ret;
241 	unsigned int bs;
242 	RF_RaidDisk_t *disks;
243 	int     num_spares_done;
244 
245 	num_spares_done = 0;
246 
247 	/* The space for the spares should have already been allocated by
248 	 * ConfigureDisks() */
249 
250 	disks = &raidPtr->Disks[raidPtr->numCol];
251 	for (i = 0; i < raidPtr->numSpare; i++) {
252 		ret = rf_ConfigureDisk(raidPtr, &cfgPtr->spare_names[i][0],
253 				       &disks[i], raidPtr->numCol + i);
254 		if (ret)
255 			goto fail;
256 		if (disks[i].status != rf_ds_optimal) {
257 			RF_ERRORMSG1("Warning: spare disk %s failed TUR\n",
258 				     &cfgPtr->spare_names[i][0]);
259 		} else {
260 			disks[i].status = rf_ds_spare;	/* change status to
261 							 * spare */
262 			DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", i,
263 			    disks[i].devname,
264 			    (long int) disks[i].numBlocks, disks[i].blockSize,
265 			    (long int) disks[i].numBlocks *
266 				 disks[i].blockSize / 1024 / 1024);
267 		}
268 		num_spares_done++;
269 	}
270 
271 	/* check sizes and block sizes on spare disks */
272 	bs = 1 << raidPtr->logBytesPerSector;
273 	for (i = 0; i < raidPtr->numSpare; i++) {
274 		if (disks[i].blockSize != bs) {
275 			RF_ERRORMSG3("Block size of %d on spare disk %s is not the same as on other disks (%d)\n", disks[i].blockSize, disks[i].devname, bs);
276 			ret = EINVAL;
277 			goto fail;
278 		}
279 		if (disks[i].numBlocks < raidPtr->sectorsPerDisk) {
280 			RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
281 				     disks[i].devname, disks[i].blockSize,
282 				     (long int) raidPtr->sectorsPerDisk);
283 			ret = EINVAL;
284 			goto fail;
285 		} else
286 			if (disks[i].numBlocks > raidPtr->sectorsPerDisk) {
287 				RF_ERRORMSG2("Warning: truncating spare disk %s to %ld blocks\n", disks[i].devname, (long int) raidPtr->sectorsPerDisk);
288 
289 				disks[i].numBlocks = raidPtr->sectorsPerDisk;
290 			}
291 	}
292 
293 	return (0);
294 
295 fail:
296 
297 	/* Release the hold on the main components.  We've failed to allocate
298 	 * a spare, and since we're failing, we need to free things..
299 
300 	 XXX failing to allocate a spare is *not* that big of a deal...
301 	 We *can* survive without it, if need be, esp. if we get hot
302 	 adding working.
303 
304 	 If we don't fail out here, then we need a way to remove this spare...
305 	 that should be easier to do here than if we are "live"...
306 
307 	 */
308 
309 	rf_UnconfigureVnodes( raidPtr );
310 
311 	return (ret);
312 }
313 
314 static int
315 rf_AllocDiskStructures(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr)
316 {
317 	int ret;
318 
319 	/* We allocate RF_MAXSPARE on the first row so that we
320 	   have room to do hot-swapping of spares */
321 	RF_MallocAndAdd(raidPtr->Disks, (raidPtr->numCol + RF_MAXSPARE) *
322 			sizeof(RF_RaidDisk_t), (RF_RaidDisk_t *),
323 			raidPtr->cleanupList);
324 	if (raidPtr->Disks == NULL) {
325 		ret = ENOMEM;
326 		goto fail;
327 	}
328 
329 	/* get space for device specific stuff.. */
330 	RF_MallocAndAdd(raidPtr->raid_cinfo,
331 			(raidPtr->numCol + raidPtr->numSpare) *
332 			sizeof(struct raidcinfo), (struct raidcinfo *),
333 			raidPtr->cleanupList);
334 
335 	if (raidPtr->raid_cinfo == NULL) {
336 		ret = ENOMEM;
337 		goto fail;
338 	}
339 
340 	return(0);
341 fail:
342 	rf_UnconfigureVnodes( raidPtr );
343 
344 	return(ret);
345 }
346 
347 
348 /* configure a single disk during auto-configuration at boot */
349 int
350 rf_AutoConfigureDisks(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr,
351 		      RF_AutoConfig_t *auto_config)
352 {
353 	RF_RaidDisk_t *disks;
354 	RF_RaidDisk_t *diskPtr;
355 	RF_RowCol_t c;
356 	RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL;
357 	int bs, ret;
358 	int numFailuresThisRow;
359 	RF_AutoConfig_t *ac;
360 	int parity_good;
361 	int mod_counter;
362 	int mod_counter_found;
363 
364 #if DEBUG
365 	printf("Starting autoconfiguration of RAID set...\n");
366 #endif
367 
368 	ret = rf_AllocDiskStructures(raidPtr, cfgPtr);
369 	if (ret)
370 		goto fail;
371 
372 	disks = raidPtr->Disks;
373 
374 	/* assume the parity will be fine.. */
375 	parity_good = RF_RAID_CLEAN;
376 
377 	/* Check for mod_counters that are too low */
378 	mod_counter_found = 0;
379 	mod_counter = 0;
380 	ac = auto_config;
381 	while(ac!=NULL) {
382 		if (mod_counter_found==0) {
383 			mod_counter = ac->clabel->mod_counter;
384 			mod_counter_found = 1;
385 		} else {
386 			if (ac->clabel->mod_counter > mod_counter) {
387 				mod_counter = ac->clabel->mod_counter;
388 			}
389 		}
390 		ac->flag = 0; /* clear the general purpose flag */
391 		ac = ac->next;
392 	}
393 
394 	bs = 0;
395 
396 	numFailuresThisRow = 0;
397 	for (c = 0; c < raidPtr->numCol; c++) {
398 		diskPtr = &disks[c];
399 
400 		/* find this row/col in the autoconfig */
401 #if DEBUG
402 		printf("Looking for %d in autoconfig\n",c);
403 #endif
404 		ac = auto_config;
405 		while(ac!=NULL) {
406 			if (ac->clabel==NULL) {
407 				/* big-time bad news. */
408 				goto fail;
409 			}
410 			if ((ac->clabel->column == c) &&
411 			    (ac->clabel->mod_counter == mod_counter)) {
412 				/* it's this one... */
413 				/* flag it as 'used', so we don't
414 				   free it later. */
415 				ac->flag = 1;
416 #if DEBUG
417 				printf("Found: %s at %d\n",
418 				       ac->devname,c);
419 #endif
420 
421 				break;
422 			}
423 			ac=ac->next;
424 		}
425 
426 		if (ac==NULL) {
427 			/* we didn't find an exact match with a
428 			   correct mod_counter above... can we find
429 			   one with an incorrect mod_counter to use
430 			   instead?  (this one, if we find it, will be
431 			   marked as failed once the set configures)
432 			*/
433 
434 			ac = auto_config;
435 			while(ac!=NULL) {
436 				if (ac->clabel==NULL) {
437 					/* big-time bad news. */
438 					goto fail;
439 				}
440 				if (ac->clabel->column == c) {
441 					/* it's this one...
442 					   flag it as 'used', so we
443 					   don't free it later. */
444 					ac->flag = 1;
445 #if DEBUG
446 					printf("Found(low mod_counter): %s at %d\n",
447 					       ac->devname,c);
448 #endif
449 
450 					break;
451 				}
452 				ac=ac->next;
453 			}
454 		}
455 
456 
457 
458 		if (ac!=NULL) {
459 			/* Found it.  Configure it.. */
460 			diskPtr->blockSize = ac->clabel->blockSize;
461 			diskPtr->numBlocks = ac->clabel->numBlocks;
462 			/* Note: rf_protectedSectors is already
463 			   factored into numBlocks here */
464 			raidPtr->raid_cinfo[c].ci_vp = ac->vp;
465 			raidPtr->raid_cinfo[c].ci_dev = ac->dev;
466 
467 			memcpy(&raidPtr->raid_cinfo[c].ci_label,
468 			       ac->clabel, sizeof(*ac->clabel));
469 			sprintf(diskPtr->devname, "/dev/%s",
470 				ac->devname);
471 
472 			/* note the fact that this component was
473 			   autoconfigured.  You'll need this info
474 			   later.  Trust me :) */
475 			diskPtr->auto_configured = 1;
476 			diskPtr->dev = ac->dev;
477 
478 			/*
479 			 * we allow the user to specify that
480 			 * only a fraction of the disks should
481 			 * be used this is just for debug: it
482 			 * speeds up the parity scan
483 			 */
484 
485 			diskPtr->numBlocks = diskPtr->numBlocks *
486 				rf_sizePercentage / 100;
487 
488 			/* XXX these will get set multiple times,
489 			   but since we're autoconfiguring, they'd
490 			   better be always the same each time!
491 			   If not, this is the least of your worries */
492 
493 			bs = diskPtr->blockSize;
494 			min_numblks = diskPtr->numBlocks;
495 
496 			/* this gets done multiple times, but that's
497 			   fine -- the serial number will be the same
498 			   for all components, guaranteed */
499 			raidPtr->serial_number = ac->clabel->serial_number;
500 			/* check the last time the label was modified */
501 
502 			if (ac->clabel->mod_counter != mod_counter) {
503 				/* Even though we've filled in all of
504 				   the above, we don't trust this
505 				   component since it's modification
506 				   counter is not in sync with the
507 				   rest, and we really consider it to
508 				   be failed.  */
509 				disks[c].status = rf_ds_failed;
510 				numFailuresThisRow++;
511 			} else {
512 				if (ac->clabel->clean != RF_RAID_CLEAN) {
513 					parity_good = RF_RAID_DIRTY;
514 				}
515 			}
516 		} else {
517 			/* Didn't find it at all!!  Component must
518 			   really be dead */
519 			disks[c].status = rf_ds_failed;
520 			sprintf(disks[c].devname, "component%d", c);
521 			numFailuresThisRow++;
522 		}
523 	}
524 	/* XXX fix for n-fault tolerant */
525 	/* XXX this should probably check to see how many failures
526 	   we can handle for this configuration! */
527 	if (numFailuresThisRow > 0)
528 		raidPtr->status = rf_rs_degraded;
529 
530 	/* close the device for the ones that didn't get used */
531 
532 	ac = auto_config;
533 	while(ac!=NULL) {
534 		if (ac->flag == 0) {
535 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
536 			VOP_CLOSE(ac->vp, FREAD | FWRITE, NOCRED, 0);
537 			vput(ac->vp);
538 			ac->vp = NULL;
539 #if DEBUG
540 			printf("Released %s from auto-config set.\n",
541 			       ac->devname);
542 #endif
543 		}
544 		ac = ac->next;
545 	}
546 
547 	raidPtr->mod_counter = mod_counter;
548 
549 	/* note the state of the parity, if any */
550 	raidPtr->parity_good = parity_good;
551 	raidPtr->sectorsPerDisk = min_numblks;
552 	raidPtr->logBytesPerSector = ffs(bs) - 1;
553 	raidPtr->bytesPerSector = bs;
554 	raidPtr->sectorMask = bs - 1;
555 	return (0);
556 
557 fail:
558 
559 	rf_UnconfigureVnodes( raidPtr );
560 
561 	return (ret);
562 
563 }
564 
565 /* configure a single disk in the array */
566 int
567 rf_ConfigureDisk(RF_Raid_t *raidPtr, char *buf, RF_RaidDisk_t *diskPtr,
568 		 RF_RowCol_t col)
569 {
570 	char   *p;
571 	struct partinfo dpart;
572 	struct vnode *vp;
573 	struct vattr va;
574 	struct proc *proc;
575 	int     error;
576 
577 	p = rf_find_non_white(buf);
578 	if (p[strlen(p) - 1] == '\n') {
579 		/* strip off the newline */
580 		p[strlen(p) - 1] = '\0';
581 	}
582 	(void) strcpy(diskPtr->devname, p);
583 
584 	proc = raidPtr->engine_thread;
585 
586 	/* Let's start by claiming the component is fine and well... */
587 	diskPtr->status = rf_ds_optimal;
588 
589 	raidPtr->raid_cinfo[col].ci_vp = NULL;
590 	raidPtr->raid_cinfo[col].ci_dev = 0;
591 
592 	error = raidlookup(diskPtr->devname, proc, &vp);
593 	if (error) {
594 		printf("raidlookup on device: %s failed!\n", diskPtr->devname);
595 		if (error == ENXIO) {
596 			/* the component isn't there... must be dead :-( */
597 			diskPtr->status = rf_ds_failed;
598 		} else {
599 			return (error);
600 		}
601 	}
602 	if (diskPtr->status == rf_ds_optimal) {
603 
604 		if ((error = VOP_GETATTR(vp, &va, proc->p_ucred, proc)) != 0) {
605 			return (error);
606 		}
607 		error = VOP_IOCTL(vp, DIOCGPART, &dpart,
608 				  FREAD, proc->p_ucred, proc);
609 		if (error) {
610 			return (error);
611 		}
612 
613 		diskPtr->blockSize = dpart.disklab->d_secsize;
614 
615 		diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors;
616 		diskPtr->partitionSize = dpart.part->p_size;
617 
618 		raidPtr->raid_cinfo[col].ci_vp = vp;
619 		raidPtr->raid_cinfo[col].ci_dev = va.va_rdev;
620 
621 		/* This component was not automatically configured */
622 		diskPtr->auto_configured = 0;
623 		diskPtr->dev = va.va_rdev;
624 
625 		/* we allow the user to specify that only a fraction of the
626 		 * disks should be used this is just for debug:  it speeds up
627 		 * the parity scan */
628 		diskPtr->numBlocks = diskPtr->numBlocks *
629 			rf_sizePercentage / 100;
630 	}
631 	return (0);
632 }
633 
634 static void
635 rf_print_label_status(RF_Raid_t *raidPtr, int column, char *dev_name,
636 		      RF_ComponentLabel_t *ci_label)
637 {
638 
639 	printf("raid%d: Component %s being configured at col: %d\n",
640 	       raidPtr->raidid, dev_name, column );
641 	printf("         Column: %d Num Columns: %d\n",
642 	       ci_label->column,
643 	       ci_label->num_columns);
644 	printf("         Version: %d Serial Number: %d Mod Counter: %d\n",
645 	       ci_label->version, ci_label->serial_number,
646 	       ci_label->mod_counter);
647 	printf("         Clean: %s Status: %d\n",
648 	       ci_label->clean ? "Yes" : "No", ci_label->status );
649 }
650 
651 static int rf_check_label_vitals(RF_Raid_t *raidPtr, int row, int column,
652 				 char *dev_name, RF_ComponentLabel_t *ci_label,
653 				 int serial_number, int mod_counter)
654 {
655 	int fatal_error = 0;
656 
657 	if (serial_number != ci_label->serial_number) {
658 		printf("%s has a different serial number: %d %d\n",
659 		       dev_name, serial_number, ci_label->serial_number);
660 		fatal_error = 1;
661 	}
662 	if (mod_counter != ci_label->mod_counter) {
663 		printf("%s has a different modfication count: %d %d\n",
664 		       dev_name, mod_counter, ci_label->mod_counter);
665 	}
666 
667 	if (row != ci_label->row) {
668 		printf("Row out of alignment for: %s\n", dev_name);
669 		fatal_error = 1;
670 	}
671 	if (column != ci_label->column) {
672 		printf("Column out of alignment for: %s\n", dev_name);
673 		fatal_error = 1;
674 	}
675 	if (raidPtr->numCol != ci_label->num_columns) {
676 		printf("Number of columns do not match for: %s\n", dev_name);
677 		fatal_error = 1;
678 	}
679 	if (ci_label->clean == 0) {
680 		/* it's not clean, but that's not fatal */
681 		printf("%s is not clean!\n", dev_name);
682 	}
683 	return(fatal_error);
684 }
685 
686 
687 /*
688 
689    rf_CheckLabels() - check all the component labels for consistency.
690    Return an error if there is anything major amiss.
691 
692  */
693 
694 int
695 rf_CheckLabels(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr)
696 {
697 	int c;
698 	char *dev_name;
699 	RF_ComponentLabel_t *ci_label;
700 	int serial_number = 0;
701 	int mod_number = 0;
702 	int fatal_error = 0;
703 	int mod_values[4];
704 	int mod_count[4];
705 	int ser_values[4];
706 	int ser_count[4];
707 	int num_ser;
708 	int num_mod;
709 	int i;
710 	int found;
711 	int hosed_column;
712 	int too_fatal;
713 	int parity_good;
714 	int force;
715 
716 	hosed_column = -1;
717 	too_fatal = 0;
718 	force = cfgPtr->force;
719 
720 	/*
721 	   We're going to try to be a little intelligent here.  If one
722 	   component's label is bogus, and we can identify that it's the
723 	   *only* one that's gone, we'll mark it as "failed" and allow
724 	   the configuration to proceed.  This will be the *only* case
725 	   that we'll proceed if there would be (otherwise) fatal errors.
726 
727 	   Basically we simply keep a count of how many components had
728 	   what serial number.  If all but one agree, we simply mark
729 	   the disagreeing component as being failed, and allow
730 	   things to come up "normally".
731 
732 	   We do this first for serial numbers, and then for "mod_counter".
733 
734 	 */
735 
736 	num_ser = 0;
737 	num_mod = 0;
738 
739 	for (c = 0; c < raidPtr->numCol; c++) {
740 		ci_label = &raidPtr->raid_cinfo[c].ci_label;
741 		found=0;
742 		for(i=0;i<num_ser;i++) {
743 			if (ser_values[i] == ci_label->serial_number) {
744 				ser_count[i]++;
745 				found=1;
746 				break;
747 			}
748 		}
749 		if (!found) {
750 			ser_values[num_ser] = ci_label->serial_number;
751 			ser_count[num_ser] = 1;
752 			num_ser++;
753 			if (num_ser>2) {
754 				fatal_error = 1;
755 				break;
756 			}
757 		}
758 		found=0;
759 		for(i=0;i<num_mod;i++) {
760 			if (mod_values[i] == ci_label->mod_counter) {
761 				mod_count[i]++;
762 				found=1;
763 				break;
764 			}
765 		}
766 		if (!found) {
767 			mod_values[num_mod] = ci_label->mod_counter;
768 			mod_count[num_mod] = 1;
769 			num_mod++;
770 			if (num_mod>2) {
771 				fatal_error = 1;
772 				break;
773 			}
774 		}
775 	}
776 #if DEBUG
777 	printf("raid%d: Summary of serial numbers:\n", raidPtr->raidid);
778 	for(i=0;i<num_ser;i++) {
779 		printf("%d %d\n", ser_values[i], ser_count[i]);
780 	}
781 	printf("raid%d: Summary of mod counters:\n", raidPtr->raidid);
782 	for(i=0;i<num_mod;i++) {
783 		printf("%d %d\n", mod_values[i], mod_count[i]);
784 	}
785 #endif
786 	serial_number = ser_values[0];
787 	if (num_ser == 2) {
788 		if ((ser_count[0] == 1) || (ser_count[1] == 1)) {
789 			/* Locate the maverick component */
790 			if (ser_count[1] > ser_count[0]) {
791 				serial_number = ser_values[1];
792 			}
793 
794 			for (c = 0; c < raidPtr->numCol; c++) {
795 				ci_label = &raidPtr->raid_cinfo[c].ci_label;
796 				if (serial_number != ci_label->serial_number) {
797 					hosed_column = c;
798 					break;
799 				}
800 			}
801 			printf("Hosed component: %s\n",
802 			       &cfgPtr->devnames[0][hosed_column][0]);
803 			if (!force) {
804 				/* we'll fail this component, as if there are
805 				   other major errors, we arn't forcing things
806 				   and we'll abort the config anyways */
807 				raidPtr->Disks[hosed_column].status
808 					= rf_ds_failed;
809 				raidPtr->numFailures++;
810 				raidPtr->status = rf_rs_degraded;
811 			}
812 		} else {
813 			too_fatal = 1;
814 		}
815 		if (cfgPtr->parityConfig == '0') {
816 			/* We've identified two different serial numbers.
817 			   RAID 0 can't cope with that, so we'll punt */
818 			too_fatal = 1;
819 		}
820 
821 	}
822 
823 	/* record the serial number for later.  If we bail later, setting
824 	   this doesn't matter, otherwise we've got the best guess at the
825 	   correct serial number */
826 	raidPtr->serial_number = serial_number;
827 
828 	mod_number = mod_values[0];
829 	if (num_mod == 2) {
830 		if ((mod_count[0] == 1) || (mod_count[1] == 1)) {
831 			/* Locate the maverick component */
832 			if (mod_count[1] > mod_count[0]) {
833 				mod_number = mod_values[1];
834 			} else if (mod_count[1] < mod_count[0]) {
835 				mod_number = mod_values[0];
836 			} else {
837 				/* counts of different modification values
838 				   are the same.   Assume greater value is
839 				   the correct one, all other things
840 				   considered */
841 				if (mod_values[0] > mod_values[1]) {
842 					mod_number = mod_values[0];
843 				} else {
844 					mod_number = mod_values[1];
845 				}
846 
847 			}
848 
849 			for (c = 0; c < raidPtr->numCol; c++) {
850 				ci_label = &raidPtr->raid_cinfo[c].ci_label;
851 				if (mod_number != ci_label->mod_counter) {
852 					if (hosed_column == c) {
853 						/* same one.  Can
854 						   deal with it.  */
855 					} else {
856 						hosed_column = c;
857 						if (num_ser != 1) {
858 							too_fatal = 1;
859 							break;
860 						}
861 					}
862 				}
863 			}
864 			printf("Hosed component: %s\n",
865 			       &cfgPtr->devnames[0][hosed_column][0]);
866 			if (!force) {
867 				/* we'll fail this component, as if there are
868 				   other major errors, we arn't forcing things
869 				   and we'll abort the config anyways */
870 				if (raidPtr->Disks[hosed_column].status != rf_ds_failed) {
871 					raidPtr->Disks[hosed_column].status
872 						= rf_ds_failed;
873 					raidPtr->numFailures++;
874 					raidPtr->status = rf_rs_degraded;
875 				}
876 			}
877 		} else {
878 			too_fatal = 1;
879 		}
880 		if (cfgPtr->parityConfig == '0') {
881 			/* We've identified two different mod counters.
882 			   RAID 0 can't cope with that, so we'll punt */
883 			too_fatal = 1;
884 		}
885 	}
886 
887 	raidPtr->mod_counter = mod_number;
888 
889 	if (too_fatal) {
890 		/* we've had both a serial number mismatch, and a mod_counter
891 		   mismatch -- and they involved two different components!!
892 		   Bail -- make things fail so that the user must force
893 		   the issue... */
894 		hosed_column = -1;
895 	}
896 
897 	if (num_ser > 2) {
898 		printf("raid%d: Too many different serial numbers!\n",
899 		       raidPtr->raidid);
900 	}
901 
902 	if (num_mod > 2) {
903 		printf("raid%d: Too many different mod counters!\n",
904 		       raidPtr->raidid);
905 	}
906 
907 	/* we start by assuming the parity will be good, and flee from
908 	   that notion at the slightest sign of trouble */
909 
910 	parity_good = RF_RAID_CLEAN;
911 
912 	for (c = 0; c < raidPtr->numCol; c++) {
913 		dev_name = &cfgPtr->devnames[0][c][0];
914 		ci_label = &raidPtr->raid_cinfo[c].ci_label;
915 
916 		if (c == hosed_column) {
917 			printf("raid%d: Ignoring %s\n",
918 			       raidPtr->raidid, dev_name);
919 		} else {
920 			rf_print_label_status( raidPtr, c, dev_name, ci_label);
921 			if (rf_check_label_vitals( raidPtr, 0, c,
922 						   dev_name, ci_label,
923 						   serial_number,
924 						   mod_number )) {
925 				fatal_error = 1;
926 			}
927 			if (ci_label->clean != RF_RAID_CLEAN) {
928 				parity_good = RF_RAID_DIRTY;
929 			}
930 		}
931 	}
932 
933 	if (fatal_error) {
934 		parity_good = RF_RAID_DIRTY;
935 	}
936 
937 	/* we note the state of the parity */
938 	raidPtr->parity_good = parity_good;
939 
940 	return(fatal_error);
941 }
942 
943 int
944 rf_add_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
945 {
946 	RF_RaidDisk_t *disks;
947 	RF_DiskQueue_t *spareQueues;
948 	int ret;
949 	unsigned int bs;
950 	int spare_number;
951 
952 	ret=0;
953 
954 	if (raidPtr->numSpare >= RF_MAXSPARE) {
955 		RF_ERRORMSG1("Too many spares: %d\n", raidPtr->numSpare);
956 		return(EINVAL);
957 	}
958 
959 	RF_LOCK_MUTEX(raidPtr->mutex);
960 	while (raidPtr->adding_hot_spare==1) {
961 		ltsleep(&(raidPtr->adding_hot_spare), PRIBIO, "raidhs", 0,
962 			&(raidPtr->mutex));
963 	}
964 	raidPtr->adding_hot_spare=1;
965 	RF_UNLOCK_MUTEX(raidPtr->mutex);
966 
967 	/* the beginning of the spares... */
968 	disks = &raidPtr->Disks[raidPtr->numCol];
969 
970 	spare_number = raidPtr->numSpare;
971 
972 	ret = rf_ConfigureDisk(raidPtr, sparePtr->component_name,
973 			       &disks[spare_number],
974 			       raidPtr->numCol + spare_number);
975 
976 	if (ret)
977 		goto fail;
978 	if (disks[spare_number].status != rf_ds_optimal) {
979 		RF_ERRORMSG1("Warning: spare disk %s failed TUR\n",
980 			     sparePtr->component_name);
981 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
982 		ret=EINVAL;
983 		goto fail;
984 	} else {
985 		disks[spare_number].status = rf_ds_spare;
986 		DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", spare_number,
987 			 disks[spare_number].devname,
988 			 (long int) disks[spare_number].numBlocks,
989 			 disks[spare_number].blockSize,
990 			 (long int) disks[spare_number].numBlocks *
991 			 disks[spare_number].blockSize / 1024 / 1024);
992 	}
993 
994 
995 	/* check sizes and block sizes on the spare disk */
996 	bs = 1 << raidPtr->logBytesPerSector;
997 	if (disks[spare_number].blockSize != bs) {
998 		RF_ERRORMSG3("Block size of %d on spare disk %s is not the same as on other disks (%d)\n", disks[spare_number].blockSize, disks[spare_number].devname, bs);
999 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1000 		ret = EINVAL;
1001 		goto fail;
1002 	}
1003 	if (disks[spare_number].numBlocks < raidPtr->sectorsPerDisk) {
1004 		RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
1005 			     disks[spare_number].devname,
1006 			     disks[spare_number].blockSize,
1007 			     (long int) raidPtr->sectorsPerDisk);
1008 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1009 		ret = EINVAL;
1010 		goto fail;
1011 	} else {
1012 		if (disks[spare_number].numBlocks >
1013 		    raidPtr->sectorsPerDisk) {
1014 			RF_ERRORMSG2("Warning: truncating spare disk %s to %ld blocks\n", disks[spare_number].devname,
1015 				     (long int) raidPtr->sectorsPerDisk);
1016 
1017 			disks[spare_number].numBlocks = raidPtr->sectorsPerDisk;
1018 		}
1019 	}
1020 
1021 	spareQueues = &raidPtr->Queues[raidPtr->numCol];
1022 	ret = rf_ConfigureDiskQueue( raidPtr, &spareQueues[spare_number],
1023 				 raidPtr->numCol + spare_number,
1024 				 raidPtr->qType,
1025 				 raidPtr->sectorsPerDisk,
1026 				 raidPtr->Disks[raidPtr->numCol +
1027 						  spare_number].dev,
1028 				 raidPtr->maxOutstanding,
1029 				 &raidPtr->shutdownList,
1030 				 raidPtr->cleanupList);
1031 
1032 	RF_LOCK_MUTEX(raidPtr->mutex);
1033 	raidPtr->numSpare++;
1034 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1035 
1036 fail:
1037 	RF_LOCK_MUTEX(raidPtr->mutex);
1038 	raidPtr->adding_hot_spare=0;
1039 	wakeup(&(raidPtr->adding_hot_spare));
1040 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1041 
1042 	return(ret);
1043 }
1044 
1045 int
1046 rf_remove_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
1047 {
1048 	int spare_number;
1049 
1050 
1051 	if (raidPtr->numSpare==0) {
1052 		printf("No spares to remove!\n");
1053 		return(EINVAL);
1054 	}
1055 
1056 	spare_number = sparePtr->column;
1057 
1058 	return(EINVAL); /* XXX not implemented yet */
1059 #if 0
1060 	if (spare_number < 0 || spare_number > raidPtr->numSpare) {
1061 		return(EINVAL);
1062 	}
1063 
1064 	/* verify that this spare isn't in use... */
1065 
1066 
1067 
1068 
1069 	/* it's gone.. */
1070 
1071 	raidPtr->numSpare--;
1072 
1073 	return(0);
1074 #endif
1075 }
1076 
1077 
1078 int
1079 rf_delete_component(RF_Raid_t *raidPtr, RF_SingleComponent_t *component)
1080 {
1081 	RF_RaidDisk_t *disks;
1082 
1083 	if ((component->column < 0) ||
1084 	    (component->column >= raidPtr->numCol)) {
1085 		return(EINVAL);
1086 	}
1087 
1088 	disks = &raidPtr->Disks[component->column];
1089 
1090 	/* 1. This component must be marked as 'failed' */
1091 
1092 	return(EINVAL); /* Not implemented yet. */
1093 }
1094 
1095 int
1096 rf_incorporate_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *component)
1097 {
1098 
1099 	/* Issues here include how to 'move' this in if there is IO
1100 	   taking place (e.g. component queues and such) */
1101 
1102 	return(EINVAL); /* Not implemented yet. */
1103 }
1104