xref: /netbsd-src/sys/dev/raidframe/rf_disks.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: rf_disks.c,v 1.69 2008/01/26 20:44:47 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 1.69 2008/01/26 20:44:47 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 #include <sys/kauth.h>
89 
90 static int rf_AllocDiskStructures(RF_Raid_t *, RF_Config_t *);
91 static void rf_print_label_status( RF_Raid_t *, int, char *,
92 				  RF_ComponentLabel_t *);
93 static int rf_check_label_vitals( RF_Raid_t *, int, int, char *,
94 				  RF_ComponentLabel_t *, int, int );
95 
96 #define DPRINTF6(a,b,c,d,e,f) if (rf_diskDebug) printf(a,b,c,d,e,f)
97 #define DPRINTF7(a,b,c,d,e,f,g) if (rf_diskDebug) printf(a,b,c,d,e,f,g)
98 
99 /**************************************************************************
100  *
101  * initialize the disks comprising the array
102  *
103  * We want the spare disks to have regular row,col numbers so that we can
104  * easily substitue a spare for a failed disk.  But, the driver code assumes
105  * throughout that the array contains numRow by numCol _non-spare_ disks, so
106  * it's not clear how to fit in the spares.  This is an unfortunate holdover
107  * from raidSim.  The quick and dirty fix is to make row zero bigger than the
108  * rest, and put all the spares in it.  This probably needs to get changed
109  * eventually.
110  *
111  **************************************************************************/
112 
113 int
114 rf_ConfigureDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
115 		  RF_Config_t *cfgPtr)
116 {
117 	RF_RaidDisk_t *disks;
118 	RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL;
119 	RF_RowCol_t c;
120 	int bs, ret;
121 	unsigned i, count, foundone = 0, numFailuresThisRow;
122 	int force;
123 
124 	force = cfgPtr->force;
125 
126 	ret = rf_AllocDiskStructures(raidPtr, cfgPtr);
127 	if (ret)
128 		goto fail;
129 
130 	disks = raidPtr->Disks;
131 
132 	numFailuresThisRow = 0;
133 	for (c = 0; c < raidPtr->numCol; c++) {
134 		ret = rf_ConfigureDisk(raidPtr,
135 				       &cfgPtr->devnames[0][c][0],
136 				       &disks[c], c);
137 
138 		if (ret)
139 			goto fail;
140 
141 		if (disks[c].status == rf_ds_optimal) {
142 			raidread_component_label(
143 						 raidPtr->raid_cinfo[c].ci_dev,
144 						 raidPtr->raid_cinfo[c].ci_vp,
145 						 &raidPtr->raid_cinfo[c].ci_label);
146 		}
147 
148 		if (disks[c].status != rf_ds_optimal) {
149 			numFailuresThisRow++;
150 		} else {
151 			if (disks[c].numBlocks < min_numblks)
152 				min_numblks = disks[c].numBlocks;
153 			DPRINTF6("Disk at col %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n",
154 				 c, disks[c].devname,
155 				 (long int) disks[c].numBlocks,
156 				 disks[c].blockSize,
157 				 (long int) disks[c].numBlocks *
158 				 disks[c].blockSize / 1024 / 1024);
159 		}
160 	}
161 	/* XXX fix for n-fault tolerant */
162 	/* XXX this should probably check to see how many failures
163 	   we can handle for this configuration! */
164 	if (numFailuresThisRow > 0)
165 		raidPtr->status = rf_rs_degraded;
166 
167 	/* all disks must be the same size & have the same block size, bs must
168 	 * be a power of 2 */
169 	bs = 0;
170 	foundone = 0;
171 	for (c = 0; c < raidPtr->numCol; c++) {
172 		if (disks[c].status == rf_ds_optimal) {
173 			bs = disks[c].blockSize;
174 			foundone = 1;
175 			break;
176 		}
177 	}
178 	if (!foundone) {
179 		RF_ERRORMSG("RAIDFRAME: Did not find any live disks in the array.\n");
180 		ret = EINVAL;
181 		goto fail;
182 	}
183 	for (count = 0, i = 1; i; i <<= 1)
184 		if (bs & i)
185 			count++;
186 	if (count != 1) {
187 		RF_ERRORMSG1("Error: block size on disks (%d) must be a power of 2\n", bs);
188 		ret = EINVAL;
189 		goto fail;
190 	}
191 
192 	if (rf_CheckLabels( raidPtr, cfgPtr )) {
193 		printf("raid%d: There were fatal errors\n", raidPtr->raidid);
194 		if (force != 0) {
195 			printf("raid%d: Fatal errors being ignored.\n",
196 			       raidPtr->raidid);
197 		} else {
198 			ret = EINVAL;
199 			goto fail;
200 		}
201 	}
202 
203 	for (c = 0; c < raidPtr->numCol; c++) {
204 		if (disks[c].status == rf_ds_optimal) {
205 			if (disks[c].blockSize != bs) {
206 				RF_ERRORMSG1("Error: block size of disk at c %d different from disk at c 0\n", c);
207 				ret = EINVAL;
208 				goto fail;
209 			}
210 			if (disks[c].numBlocks != min_numblks) {
211 				RF_ERRORMSG2("WARNING: truncating disk at c %d to %d blocks\n",
212 					     c, (int) min_numblks);
213 				disks[c].numBlocks = min_numblks;
214 			}
215 		}
216 	}
217 
218 	raidPtr->sectorsPerDisk = min_numblks;
219 	raidPtr->logBytesPerSector = ffs(bs) - 1;
220 	raidPtr->bytesPerSector = bs;
221 	raidPtr->sectorMask = bs - 1;
222 	return (0);
223 
224 fail:
225 
226 	rf_UnconfigureVnodes( raidPtr );
227 
228 	return (ret);
229 }
230 
231 
232 /****************************************************************************
233  * set up the data structures describing the spare disks in the array
234  * recall from the above comment that the spare disk descriptors are stored
235  * in row zero, which is specially expanded to hold them.
236  ****************************************************************************/
237 int
238 rf_ConfigureSpareDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
239 		       RF_Config_t *cfgPtr)
240 {
241 	int     i, ret;
242 	unsigned int bs;
243 	RF_RaidDisk_t *disks;
244 	int     num_spares_done;
245 
246 	num_spares_done = 0;
247 
248 	/* The space for the spares should have already been allocated by
249 	 * ConfigureDisks() */
250 
251 	disks = &raidPtr->Disks[raidPtr->numCol];
252 	for (i = 0; i < raidPtr->numSpare; i++) {
253 		ret = rf_ConfigureDisk(raidPtr, &cfgPtr->spare_names[i][0],
254 				       &disks[i], raidPtr->numCol + i);
255 		if (ret)
256 			goto fail;
257 		if (disks[i].status != rf_ds_optimal) {
258 			RF_ERRORMSG1("Warning: spare disk %s failed TUR\n",
259 				     &cfgPtr->spare_names[i][0]);
260 		} else {
261 			disks[i].status = rf_ds_spare;	/* change status to
262 							 * spare */
263 			DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", i,
264 			    disks[i].devname,
265 			    (long int) disks[i].numBlocks, disks[i].blockSize,
266 			    (long int) disks[i].numBlocks *
267 				 disks[i].blockSize / 1024 / 1024);
268 		}
269 		num_spares_done++;
270 	}
271 
272 	/* check sizes and block sizes on spare disks */
273 	bs = 1 << raidPtr->logBytesPerSector;
274 	for (i = 0; i < raidPtr->numSpare; i++) {
275 		if (disks[i].blockSize != bs) {
276 			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);
277 			ret = EINVAL;
278 			goto fail;
279 		}
280 		if (disks[i].numBlocks < raidPtr->sectorsPerDisk) {
281 			RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
282 				     disks[i].devname, disks[i].blockSize,
283 				     (long int) raidPtr->sectorsPerDisk);
284 			ret = EINVAL;
285 			goto fail;
286 		} else
287 			if (disks[i].numBlocks > raidPtr->sectorsPerDisk) {
288 				RF_ERRORMSG3("Warning: truncating spare disk %s to %ld blocks (from %ld)\n",
289 				    disks[i].devname,
290 				    (long int) raidPtr->sectorsPerDisk,
291 				    (long int) disks[i].numBlocks);
292 
293 				disks[i].numBlocks = raidPtr->sectorsPerDisk;
294 			}
295 	}
296 
297 	return (0);
298 
299 fail:
300 
301 	/* Release the hold on the main components.  We've failed to allocate
302 	 * a spare, and since we're failing, we need to free things..
303 
304 	 XXX failing to allocate a spare is *not* that big of a deal...
305 	 We *can* survive without it, if need be, esp. if we get hot
306 	 adding working.
307 
308 	 If we don't fail out here, then we need a way to remove this spare...
309 	 that should be easier to do here than if we are "live"...
310 
311 	 */
312 
313 	rf_UnconfigureVnodes( raidPtr );
314 
315 	return (ret);
316 }
317 
318 static int
319 rf_AllocDiskStructures(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr)
320 {
321 	int ret;
322 
323 	/* We allocate RF_MAXSPARE on the first row so that we
324 	   have room to do hot-swapping of spares */
325 	RF_MallocAndAdd(raidPtr->Disks, (raidPtr->numCol + RF_MAXSPARE) *
326 			sizeof(RF_RaidDisk_t), (RF_RaidDisk_t *),
327 			raidPtr->cleanupList);
328 	if (raidPtr->Disks == NULL) {
329 		ret = ENOMEM;
330 		goto fail;
331 	}
332 
333 	/* get space for device specific stuff.. */
334 	RF_MallocAndAdd(raidPtr->raid_cinfo,
335 			(raidPtr->numCol + RF_MAXSPARE) *
336 			sizeof(struct raidcinfo), (struct raidcinfo *),
337 			raidPtr->cleanupList);
338 
339 	if (raidPtr->raid_cinfo == NULL) {
340 		ret = ENOMEM;
341 		goto fail;
342 	}
343 
344 	return(0);
345 fail:
346 	rf_UnconfigureVnodes( raidPtr );
347 
348 	return(ret);
349 }
350 
351 
352 /* configure a single disk during auto-configuration at boot */
353 int
354 rf_AutoConfigureDisks(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr,
355 		      RF_AutoConfig_t *auto_config)
356 {
357 	RF_RaidDisk_t *disks;
358 	RF_RaidDisk_t *diskPtr;
359 	RF_RowCol_t c;
360 	RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL;
361 	int bs, ret;
362 	int numFailuresThisRow;
363 	RF_AutoConfig_t *ac;
364 	int parity_good;
365 	int mod_counter;
366 	int mod_counter_found;
367 
368 #if DEBUG
369 	printf("Starting autoconfiguration of RAID set...\n");
370 #endif
371 
372 	ret = rf_AllocDiskStructures(raidPtr, cfgPtr);
373 	if (ret)
374 		goto fail;
375 
376 	disks = raidPtr->Disks;
377 
378 	/* assume the parity will be fine.. */
379 	parity_good = RF_RAID_CLEAN;
380 
381 	/* Check for mod_counters that are too low */
382 	mod_counter_found = 0;
383 	mod_counter = 0;
384 	ac = auto_config;
385 	while(ac!=NULL) {
386 		if (mod_counter_found==0) {
387 			mod_counter = ac->clabel->mod_counter;
388 			mod_counter_found = 1;
389 		} else {
390 			if (ac->clabel->mod_counter > mod_counter) {
391 				mod_counter = ac->clabel->mod_counter;
392 			}
393 		}
394 		ac->flag = 0; /* clear the general purpose flag */
395 		ac = ac->next;
396 	}
397 
398 	bs = 0;
399 
400 	numFailuresThisRow = 0;
401 	for (c = 0; c < raidPtr->numCol; c++) {
402 		diskPtr = &disks[c];
403 
404 		/* find this row/col in the autoconfig */
405 #if DEBUG
406 		printf("Looking for %d in autoconfig\n",c);
407 #endif
408 		ac = auto_config;
409 		while(ac!=NULL) {
410 			if (ac->clabel==NULL) {
411 				/* big-time bad news. */
412 				goto fail;
413 			}
414 			if ((ac->clabel->column == c) &&
415 			    (ac->clabel->mod_counter == mod_counter)) {
416 				/* it's this one... */
417 				/* flag it as 'used', so we don't
418 				   free it later. */
419 				ac->flag = 1;
420 #if DEBUG
421 				printf("Found: %s at %d\n",
422 				       ac->devname,c);
423 #endif
424 
425 				break;
426 			}
427 			ac=ac->next;
428 		}
429 
430 		if (ac==NULL) {
431 			/* we didn't find an exact match with a
432 			   correct mod_counter above... can we find
433 			   one with an incorrect mod_counter to use
434 			   instead?  (this one, if we find it, will be
435 			   marked as failed once the set configures)
436 			*/
437 
438 			ac = auto_config;
439 			while(ac!=NULL) {
440 				if (ac->clabel==NULL) {
441 					/* big-time bad news. */
442 					goto fail;
443 				}
444 				if (ac->clabel->column == c) {
445 					/* it's this one...
446 					   flag it as 'used', so we
447 					   don't free it later. */
448 					ac->flag = 1;
449 #if DEBUG
450 					printf("Found(low mod_counter): %s at %d\n",
451 					       ac->devname,c);
452 #endif
453 
454 					break;
455 				}
456 				ac=ac->next;
457 			}
458 		}
459 
460 
461 
462 		if (ac!=NULL) {
463 			/* Found it.  Configure it.. */
464 			diskPtr->blockSize = ac->clabel->blockSize;
465 			diskPtr->numBlocks = ac->clabel->numBlocks;
466 			/* Note: rf_protectedSectors is already
467 			   factored into numBlocks here */
468 			raidPtr->raid_cinfo[c].ci_vp = ac->vp;
469 			raidPtr->raid_cinfo[c].ci_dev = ac->dev;
470 
471 			memcpy(&raidPtr->raid_cinfo[c].ci_label,
472 			    ac->clabel, sizeof(*ac->clabel));
473 			snprintf(diskPtr->devname, sizeof(diskPtr->devname),
474 			    "/dev/%s", ac->devname);
475 
476 			/* note the fact that this component was
477 			   autoconfigured.  You'll need this info
478 			   later.  Trust me :) */
479 			diskPtr->auto_configured = 1;
480 			diskPtr->dev = ac->dev;
481 
482 			/*
483 			 * we allow the user to specify that
484 			 * only a fraction of the disks should
485 			 * be used this is just for debug: it
486 			 * speeds up the parity scan
487 			 */
488 
489 			diskPtr->numBlocks = diskPtr->numBlocks *
490 				rf_sizePercentage / 100;
491 
492 			/* XXX these will get set multiple times,
493 			   but since we're autoconfiguring, they'd
494 			   better be always the same each time!
495 			   If not, this is the least of your worries */
496 
497 			bs = diskPtr->blockSize;
498 			min_numblks = diskPtr->numBlocks;
499 
500 			/* this gets done multiple times, but that's
501 			   fine -- the serial number will be the same
502 			   for all components, guaranteed */
503 			raidPtr->serial_number = ac->clabel->serial_number;
504 			/* check the last time the label was modified */
505 
506 			if (ac->clabel->mod_counter != mod_counter) {
507 				/* Even though we've filled in all of
508 				   the above, we don't trust this
509 				   component since it's modification
510 				   counter is not in sync with the
511 				   rest, and we really consider it to
512 				   be failed.  */
513 				disks[c].status = rf_ds_failed;
514 				numFailuresThisRow++;
515 			} else {
516 				if (ac->clabel->clean != RF_RAID_CLEAN) {
517 					parity_good = RF_RAID_DIRTY;
518 				}
519 			}
520 		} else {
521 			/* Didn't find it at all!!  Component must
522 			   really be dead */
523 			disks[c].status = rf_ds_failed;
524 			snprintf(disks[c].devname, sizeof(disks[c].devname),
525 			    "component%d", c);
526 			numFailuresThisRow++;
527 		}
528 	}
529 	/* XXX fix for n-fault tolerant */
530 	/* XXX this should probably check to see how many failures
531 	   we can handle for this configuration! */
532 	if (numFailuresThisRow > 0) {
533 		raidPtr->status = rf_rs_degraded;
534 		raidPtr->numFailures = numFailuresThisRow;
535 	}
536 
537 	/* close the device for the ones that didn't get used */
538 
539 	ac = auto_config;
540 	while(ac!=NULL) {
541 		if (ac->flag == 0) {
542 			vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
543 			VOP_CLOSE(ac->vp, FREAD | FWRITE, NOCRED);
544 			vput(ac->vp);
545 			ac->vp = NULL;
546 #if DEBUG
547 			printf("Released %s from auto-config set.\n",
548 			       ac->devname);
549 #endif
550 		}
551 		ac = ac->next;
552 	}
553 
554 	raidPtr->mod_counter = mod_counter;
555 
556 	/* note the state of the parity, if any */
557 	raidPtr->parity_good = parity_good;
558 	raidPtr->sectorsPerDisk = min_numblks;
559 	raidPtr->logBytesPerSector = ffs(bs) - 1;
560 	raidPtr->bytesPerSector = bs;
561 	raidPtr->sectorMask = bs - 1;
562 	return (0);
563 
564 fail:
565 
566 	rf_UnconfigureVnodes( raidPtr );
567 
568 	return (ret);
569 
570 }
571 
572 /* configure a single disk in the array */
573 int
574 rf_ConfigureDisk(RF_Raid_t *raidPtr, char *bf, RF_RaidDisk_t *diskPtr,
575 		 RF_RowCol_t col)
576 {
577 	char   *p;
578 	struct vnode *vp;
579 	struct vattr va;
580 	int     error;
581 
582 	p = rf_find_non_white(bf);
583 	if (p[strlen(p) - 1] == '\n') {
584 		/* strip off the newline */
585 		p[strlen(p) - 1] = '\0';
586 	}
587 	(void) strcpy(diskPtr->devname, p);
588 
589 	/* Let's start by claiming the component is fine and well... */
590 	diskPtr->status = rf_ds_optimal;
591 
592 	raidPtr->raid_cinfo[col].ci_vp = NULL;
593 	raidPtr->raid_cinfo[col].ci_dev = 0;
594 
595 	if (!strcmp("absent", diskPtr->devname)) {
596 		printf("Ignoring missing component at column %d\n", col);
597 		sprintf(diskPtr->devname, "component%d", col);
598 		diskPtr->status = rf_ds_failed;
599 		return (0);
600 	}
601 
602 	error = dk_lookup(diskPtr->devname, curlwp, &vp, UIO_SYSSPACE);
603 	if (error) {
604 		printf("dk_lookup on device: %s failed!\n", diskPtr->devname);
605 		if (error == ENXIO) {
606 			/* the component isn't there... must be dead :-( */
607 			diskPtr->status = rf_ds_failed;
608 		} else {
609 			return (error);
610 		}
611 	}
612 	if (diskPtr->status == rf_ds_optimal) {
613 
614 		if ((error = VOP_GETATTR(vp, &va, curlwp->l_cred)) != 0)
615 			return (error);
616 		if ((error = rf_getdisksize(vp, curlwp, diskPtr)) != 0)
617 			return (error);
618 
619 		raidPtr->raid_cinfo[col].ci_vp = vp;
620 		raidPtr->raid_cinfo[col].ci_dev = va.va_rdev;
621 
622 		/* This component was not automatically configured */
623 		diskPtr->auto_configured = 0;
624 		diskPtr->dev = va.va_rdev;
625 
626 		/* we allow the user to specify that only a fraction of the
627 		 * disks should be used this is just for debug:  it speeds up
628 		 * the parity scan */
629 		diskPtr->numBlocks = diskPtr->numBlocks *
630 			rf_sizePercentage / 100;
631 	}
632 	return (0);
633 }
634 
635 static void
636 rf_print_label_status(RF_Raid_t *raidPtr, int column, char *dev_name,
637 		      RF_ComponentLabel_t *ci_label)
638 {
639 
640 	printf("raid%d: Component %s being configured at col: %d\n",
641 	       raidPtr->raidid, dev_name, column );
642 	printf("         Column: %d Num Columns: %d\n",
643 	       ci_label->column,
644 	       ci_label->num_columns);
645 	printf("         Version: %d Serial Number: %d Mod Counter: %d\n",
646 	       ci_label->version, ci_label->serial_number,
647 	       ci_label->mod_counter);
648 	printf("         Clean: %s Status: %d\n",
649 	       ci_label->clean ? "Yes" : "No", ci_label->status );
650 }
651 
652 static int rf_check_label_vitals(RF_Raid_t *raidPtr, int row, int column,
653 				 char *dev_name, RF_ComponentLabel_t *ci_label,
654 				 int serial_number, int mod_counter)
655 {
656 	int fatal_error = 0;
657 
658 	if (serial_number != ci_label->serial_number) {
659 		printf("%s has a different serial number: %d %d\n",
660 		       dev_name, serial_number, ci_label->serial_number);
661 		fatal_error = 1;
662 	}
663 	if (mod_counter != ci_label->mod_counter) {
664 		printf("%s has a different modification count: %d %d\n",
665 		       dev_name, mod_counter, ci_label->mod_counter);
666 	}
667 
668 	if (row != ci_label->row) {
669 		printf("Row out of alignment for: %s\n", dev_name);
670 		fatal_error = 1;
671 	}
672 	if (column != ci_label->column) {
673 		printf("Column out of alignment for: %s\n", dev_name);
674 		fatal_error = 1;
675 	}
676 	if (raidPtr->numCol != ci_label->num_columns) {
677 		printf("Number of columns do not match for: %s\n", dev_name);
678 		fatal_error = 1;
679 	}
680 	if (ci_label->clean == 0) {
681 		/* it's not clean, but that's not fatal */
682 		printf("%s is not clean!\n", dev_name);
683 	}
684 	return(fatal_error);
685 }
686 
687 
688 /*
689 
690    rf_CheckLabels() - check all the component labels for consistency.
691    Return an error if there is anything major amiss.
692 
693  */
694 
695 int
696 rf_CheckLabels(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr)
697 {
698 	int c;
699 	char *dev_name;
700 	RF_ComponentLabel_t *ci_label;
701 	int serial_number = 0;
702 	int mod_number = 0;
703 	int fatal_error = 0;
704 	int mod_values[4];
705 	int mod_count[4];
706 	int ser_values[4];
707 	int ser_count[4];
708 	int num_ser;
709 	int num_mod;
710 	int i;
711 	int found;
712 	int hosed_column;
713 	int too_fatal;
714 	int parity_good;
715 	int force;
716 
717 	hosed_column = -1;
718 	too_fatal = 0;
719 	force = cfgPtr->force;
720 
721 	/*
722 	   We're going to try to be a little intelligent here.  If one
723 	   component's label is bogus, and we can identify that it's the
724 	   *only* one that's gone, we'll mark it as "failed" and allow
725 	   the configuration to proceed.  This will be the *only* case
726 	   that we'll proceed if there would be (otherwise) fatal errors.
727 
728 	   Basically we simply keep a count of how many components had
729 	   what serial number.  If all but one agree, we simply mark
730 	   the disagreeing component as being failed, and allow
731 	   things to come up "normally".
732 
733 	   We do this first for serial numbers, and then for "mod_counter".
734 
735 	 */
736 
737 	num_ser = 0;
738 	num_mod = 0;
739 
740 	for (c = 0; c < raidPtr->numCol; c++) {
741 		ci_label = &raidPtr->raid_cinfo[c].ci_label;
742 		found=0;
743 		for(i=0;i<num_ser;i++) {
744 			if (ser_values[i] == ci_label->serial_number) {
745 				ser_count[i]++;
746 				found=1;
747 				break;
748 			}
749 		}
750 		if (!found) {
751 			ser_values[num_ser] = ci_label->serial_number;
752 			ser_count[num_ser] = 1;
753 			num_ser++;
754 			if (num_ser>2) {
755 				fatal_error = 1;
756 				break;
757 			}
758 		}
759 		found=0;
760 		for(i=0;i<num_mod;i++) {
761 			if (mod_values[i] == ci_label->mod_counter) {
762 				mod_count[i]++;
763 				found=1;
764 				break;
765 			}
766 		}
767 		if (!found) {
768 			mod_values[num_mod] = ci_label->mod_counter;
769 			mod_count[num_mod] = 1;
770 			num_mod++;
771 			if (num_mod>2) {
772 				fatal_error = 1;
773 				break;
774 			}
775 		}
776 	}
777 #if DEBUG
778 	printf("raid%d: Summary of serial numbers:\n", raidPtr->raidid);
779 	for(i=0;i<num_ser;i++) {
780 		printf("%d %d\n", ser_values[i], ser_count[i]);
781 	}
782 	printf("raid%d: Summary of mod counters:\n", raidPtr->raidid);
783 	for(i=0;i<num_mod;i++) {
784 		printf("%d %d\n", mod_values[i], mod_count[i]);
785 	}
786 #endif
787 	serial_number = ser_values[0];
788 	if (num_ser == 2) {
789 		if ((ser_count[0] == 1) || (ser_count[1] == 1)) {
790 			/* Locate the maverick component */
791 			if (ser_count[1] > ser_count[0]) {
792 				serial_number = ser_values[1];
793 			}
794 
795 			for (c = 0; c < raidPtr->numCol; c++) {
796 				ci_label = &raidPtr->raid_cinfo[c].ci_label;
797 				if (serial_number != ci_label->serial_number) {
798 					hosed_column = c;
799 					break;
800 				}
801 			}
802 			printf("Hosed component: %s\n",
803 			       &cfgPtr->devnames[0][hosed_column][0]);
804 			if (!force) {
805 				/* we'll fail this component, as if there are
806 				   other major errors, we arn't forcing things
807 				   and we'll abort the config anyways */
808 				raidPtr->Disks[hosed_column].status
809 					= rf_ds_failed;
810 				raidPtr->numFailures++;
811 				raidPtr->status = rf_rs_degraded;
812 			}
813 		} else {
814 			too_fatal = 1;
815 		}
816 		if (cfgPtr->parityConfig == '0') {
817 			/* We've identified two different serial numbers.
818 			   RAID 0 can't cope with that, so we'll punt */
819 			too_fatal = 1;
820 		}
821 
822 	}
823 
824 	/* record the serial number for later.  If we bail later, setting
825 	   this doesn't matter, otherwise we've got the best guess at the
826 	   correct serial number */
827 	raidPtr->serial_number = serial_number;
828 
829 	mod_number = mod_values[0];
830 	if (num_mod == 2) {
831 		if ((mod_count[0] == 1) || (mod_count[1] == 1)) {
832 			/* Locate the maverick component */
833 			if (mod_count[1] > mod_count[0]) {
834 				mod_number = mod_values[1];
835 			} else if (mod_count[1] < mod_count[0]) {
836 				mod_number = mod_values[0];
837 			} else {
838 				/* counts of different modification values
839 				   are the same.   Assume greater value is
840 				   the correct one, all other things
841 				   considered */
842 				if (mod_values[0] > mod_values[1]) {
843 					mod_number = mod_values[0];
844 				} else {
845 					mod_number = mod_values[1];
846 				}
847 
848 			}
849 
850 			for (c = 0; c < raidPtr->numCol; c++) {
851 				ci_label = &raidPtr->raid_cinfo[c].ci_label;
852 				if (mod_number != ci_label->mod_counter) {
853 					if (hosed_column == c) {
854 						/* same one.  Can
855 						   deal with it.  */
856 					} else {
857 						hosed_column = c;
858 						if (num_ser != 1) {
859 							too_fatal = 1;
860 							break;
861 						}
862 					}
863 				}
864 			}
865 			printf("Hosed component: %s\n",
866 			       &cfgPtr->devnames[0][hosed_column][0]);
867 			if (!force) {
868 				/* we'll fail this component, as if there are
869 				   other major errors, we arn't forcing things
870 				   and we'll abort the config anyways */
871 				if (raidPtr->Disks[hosed_column].status != rf_ds_failed) {
872 					raidPtr->Disks[hosed_column].status
873 						= rf_ds_failed;
874 					raidPtr->numFailures++;
875 					raidPtr->status = rf_rs_degraded;
876 				}
877 			}
878 		} else {
879 			too_fatal = 1;
880 		}
881 		if (cfgPtr->parityConfig == '0') {
882 			/* We've identified two different mod counters.
883 			   RAID 0 can't cope with that, so we'll punt */
884 			too_fatal = 1;
885 		}
886 	}
887 
888 	raidPtr->mod_counter = mod_number;
889 
890 	if (too_fatal) {
891 		/* we've had both a serial number mismatch, and a mod_counter
892 		   mismatch -- and they involved two different components!!
893 		   Bail -- make things fail so that the user must force
894 		   the issue... */
895 		hosed_column = -1;
896 		fatal_error = 1;
897 	}
898 
899 	if (num_ser > 2) {
900 		printf("raid%d: Too many different serial numbers!\n",
901 		       raidPtr->raidid);
902 		fatal_error = 1;
903 	}
904 
905 	if (num_mod > 2) {
906 		printf("raid%d: Too many different mod counters!\n",
907 		       raidPtr->raidid);
908 		fatal_error = 1;
909 	}
910 
911 	/* we start by assuming the parity will be good, and flee from
912 	   that notion at the slightest sign of trouble */
913 
914 	parity_good = RF_RAID_CLEAN;
915 
916 	for (c = 0; c < raidPtr->numCol; c++) {
917 		dev_name = &cfgPtr->devnames[0][c][0];
918 		ci_label = &raidPtr->raid_cinfo[c].ci_label;
919 
920 		if (c == hosed_column) {
921 			printf("raid%d: Ignoring %s\n",
922 			       raidPtr->raidid, dev_name);
923 		} else {
924 			rf_print_label_status( raidPtr, c, dev_name, ci_label);
925 			if (rf_check_label_vitals( raidPtr, 0, c,
926 						   dev_name, ci_label,
927 						   serial_number,
928 						   mod_number )) {
929 				fatal_error = 1;
930 			}
931 			if (ci_label->clean != RF_RAID_CLEAN) {
932 				parity_good = RF_RAID_DIRTY;
933 			}
934 		}
935 	}
936 
937 	if (fatal_error) {
938 		parity_good = RF_RAID_DIRTY;
939 	}
940 
941 	/* we note the state of the parity */
942 	raidPtr->parity_good = parity_good;
943 
944 	return(fatal_error);
945 }
946 
947 int
948 rf_add_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
949 {
950 	RF_RaidDisk_t *disks;
951 	RF_DiskQueue_t *spareQueues;
952 	int ret;
953 	unsigned int bs;
954 	int spare_number;
955 
956 	ret=0;
957 
958 	if (raidPtr->numSpare >= RF_MAXSPARE) {
959 		RF_ERRORMSG1("Too many spares: %d\n", raidPtr->numSpare);
960 		return(EINVAL);
961 	}
962 
963 	RF_LOCK_MUTEX(raidPtr->mutex);
964 	while (raidPtr->adding_hot_spare==1) {
965 		ltsleep(&(raidPtr->adding_hot_spare), PRIBIO, "raidhs", 0,
966 			&(raidPtr->mutex));
967 	}
968 	raidPtr->adding_hot_spare=1;
969 	RF_UNLOCK_MUTEX(raidPtr->mutex);
970 
971 	/* the beginning of the spares... */
972 	disks = &raidPtr->Disks[raidPtr->numCol];
973 
974 	spare_number = raidPtr->numSpare;
975 
976 	ret = rf_ConfigureDisk(raidPtr, sparePtr->component_name,
977 			       &disks[spare_number],
978 			       raidPtr->numCol + spare_number);
979 
980 	if (ret)
981 		goto fail;
982 	if (disks[spare_number].status != rf_ds_optimal) {
983 		RF_ERRORMSG1("Warning: spare disk %s failed TUR\n",
984 			     sparePtr->component_name);
985 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
986 		ret=EINVAL;
987 		goto fail;
988 	} else {
989 		disks[spare_number].status = rf_ds_spare;
990 		DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", spare_number,
991 			 disks[spare_number].devname,
992 			 (long int) disks[spare_number].numBlocks,
993 			 disks[spare_number].blockSize,
994 			 (long int) disks[spare_number].numBlocks *
995 			 disks[spare_number].blockSize / 1024 / 1024);
996 	}
997 
998 
999 	/* check sizes and block sizes on the spare disk */
1000 	bs = 1 << raidPtr->logBytesPerSector;
1001 	if (disks[spare_number].blockSize != bs) {
1002 		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);
1003 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1004 		ret = EINVAL;
1005 		goto fail;
1006 	}
1007 	if (disks[spare_number].numBlocks < raidPtr->sectorsPerDisk) {
1008 		RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
1009 			     disks[spare_number].devname,
1010 			     disks[spare_number].blockSize,
1011 			     (long int) raidPtr->sectorsPerDisk);
1012 		rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1013 		ret = EINVAL;
1014 		goto fail;
1015 	} else {
1016 		if (disks[spare_number].numBlocks >
1017 		    raidPtr->sectorsPerDisk) {
1018 			RF_ERRORMSG3("Warning: truncating spare disk %s to %ld blocks (from %ld)\n",
1019 			    disks[spare_number].devname,
1020 			    (long int) raidPtr->sectorsPerDisk,
1021 			    (long int) disks[spare_number].numBlocks);
1022 
1023 			disks[spare_number].numBlocks = raidPtr->sectorsPerDisk;
1024 		}
1025 	}
1026 
1027 	spareQueues = &raidPtr->Queues[raidPtr->numCol];
1028 	ret = rf_ConfigureDiskQueue( raidPtr, &spareQueues[spare_number],
1029 				 raidPtr->numCol + spare_number,
1030 				 raidPtr->qType,
1031 				 raidPtr->sectorsPerDisk,
1032 				 raidPtr->Disks[raidPtr->numCol +
1033 						  spare_number].dev,
1034 				 raidPtr->maxOutstanding,
1035 				 &raidPtr->shutdownList,
1036 				 raidPtr->cleanupList);
1037 
1038 	RF_LOCK_MUTEX(raidPtr->mutex);
1039 	raidPtr->numSpare++;
1040 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1041 
1042 fail:
1043 	RF_LOCK_MUTEX(raidPtr->mutex);
1044 	raidPtr->adding_hot_spare=0;
1045 	wakeup(&(raidPtr->adding_hot_spare));
1046 	RF_UNLOCK_MUTEX(raidPtr->mutex);
1047 
1048 	return(ret);
1049 }
1050 
1051 int
1052 rf_remove_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
1053 {
1054 	int spare_number;
1055 
1056 
1057 	if (raidPtr->numSpare==0) {
1058 		printf("No spares to remove!\n");
1059 		return(EINVAL);
1060 	}
1061 
1062 	spare_number = sparePtr->column;
1063 
1064 	return(EINVAL); /* XXX not implemented yet */
1065 #if 0
1066 	if (spare_number < 0 || spare_number > raidPtr->numSpare) {
1067 		return(EINVAL);
1068 	}
1069 
1070 	/* verify that this spare isn't in use... */
1071 
1072 
1073 
1074 
1075 	/* it's gone.. */
1076 
1077 	raidPtr->numSpare--;
1078 
1079 	return(0);
1080 #endif
1081 }
1082 
1083 
1084 int
1085 rf_delete_component(RF_Raid_t *raidPtr, RF_SingleComponent_t *component)
1086 {
1087 	RF_RaidDisk_t *disks;
1088 
1089 	if ((component->column < 0) ||
1090 	    (component->column >= raidPtr->numCol)) {
1091 		return(EINVAL);
1092 	}
1093 
1094 	disks = &raidPtr->Disks[component->column];
1095 
1096 	/* 1. This component must be marked as 'failed' */
1097 
1098 	return(EINVAL); /* Not implemented yet. */
1099 }
1100 
1101 int
1102 rf_incorporate_hot_spare(RF_Raid_t *raidPtr,
1103     RF_SingleComponent_t *component)
1104 {
1105 
1106 	/* Issues here include how to 'move' this in if there is IO
1107 	   taking place (e.g. component queues and such) */
1108 
1109 	return(EINVAL); /* Not implemented yet. */
1110 }
1111