1 /* $NetBSD: rf_disks.c,v 1.51.2.2 2004/08/30 08:42:48 tron 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 1.51.2.2 2004/08/30 08:42:48 tron 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 + RF_MAXSPARE) *
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 raidPtr->numFailures = numFailuresThisRow;
530 }
531
532 /* close the device for the ones that didn't get used */
533
534 ac = auto_config;
535 while(ac!=NULL) {
536 if (ac->flag == 0) {
537 vn_lock(ac->vp, LK_EXCLUSIVE | LK_RETRY);
538 VOP_CLOSE(ac->vp, FREAD | FWRITE, NOCRED, 0);
539 vput(ac->vp);
540 ac->vp = NULL;
541 #if DEBUG
542 printf("Released %s from auto-config set.\n",
543 ac->devname);
544 #endif
545 }
546 ac = ac->next;
547 }
548
549 raidPtr->mod_counter = mod_counter;
550
551 /* note the state of the parity, if any */
552 raidPtr->parity_good = parity_good;
553 raidPtr->sectorsPerDisk = min_numblks;
554 raidPtr->logBytesPerSector = ffs(bs) - 1;
555 raidPtr->bytesPerSector = bs;
556 raidPtr->sectorMask = bs - 1;
557 return (0);
558
559 fail:
560
561 rf_UnconfigureVnodes( raidPtr );
562
563 return (ret);
564
565 }
566
567 /* configure a single disk in the array */
568 int
569 rf_ConfigureDisk(RF_Raid_t *raidPtr, char *buf, RF_RaidDisk_t *diskPtr,
570 RF_RowCol_t col)
571 {
572 char *p;
573 struct partinfo dpart;
574 struct vnode *vp;
575 struct vattr va;
576 struct proc *proc;
577 int error;
578
579 p = rf_find_non_white(buf);
580 if (p[strlen(p) - 1] == '\n') {
581 /* strip off the newline */
582 p[strlen(p) - 1] = '\0';
583 }
584 (void) strcpy(diskPtr->devname, p);
585
586 proc = raidPtr->engine_thread;
587
588 /* Let's start by claiming the component is fine and well... */
589 diskPtr->status = rf_ds_optimal;
590
591 raidPtr->raid_cinfo[col].ci_vp = NULL;
592 raidPtr->raid_cinfo[col].ci_dev = 0;
593
594 if (!strcmp("absent", diskPtr->devname)) {
595 printf("Ignoring missing component at column %d\n", col);
596 sprintf(diskPtr->devname, "component%d", col);
597 diskPtr->status = rf_ds_failed;
598 return (0);
599 }
600
601 error = raidlookup(diskPtr->devname, proc, &vp);
602 if (error) {
603 printf("raidlookup on device: %s failed!\n", diskPtr->devname);
604 if (error == ENXIO) {
605 /* the component isn't there... must be dead :-( */
606 diskPtr->status = rf_ds_failed;
607 } else {
608 return (error);
609 }
610 }
611 if (diskPtr->status == rf_ds_optimal) {
612
613 if ((error = VOP_GETATTR(vp, &va, proc->p_ucred, proc)) != 0) {
614 return (error);
615 }
616 error = VOP_IOCTL(vp, DIOCGPART, &dpart,
617 FREAD, proc->p_ucred, proc);
618 if (error) {
619 return (error);
620 }
621
622 diskPtr->blockSize = dpart.disklab->d_secsize;
623
624 diskPtr->numBlocks = dpart.part->p_size - rf_protectedSectors;
625 diskPtr->partitionSize = dpart.part->p_size;
626
627 raidPtr->raid_cinfo[col].ci_vp = vp;
628 raidPtr->raid_cinfo[col].ci_dev = va.va_rdev;
629
630 /* This component was not automatically configured */
631 diskPtr->auto_configured = 0;
632 diskPtr->dev = va.va_rdev;
633
634 /* we allow the user to specify that only a fraction of the
635 * disks should be used this is just for debug: it speeds up
636 * the parity scan */
637 diskPtr->numBlocks = diskPtr->numBlocks *
638 rf_sizePercentage / 100;
639 }
640 return (0);
641 }
642
643 static void
644 rf_print_label_status(RF_Raid_t *raidPtr, int column, char *dev_name,
645 RF_ComponentLabel_t *ci_label)
646 {
647
648 printf("raid%d: Component %s being configured at col: %d\n",
649 raidPtr->raidid, dev_name, column );
650 printf(" Column: %d Num Columns: %d\n",
651 ci_label->column,
652 ci_label->num_columns);
653 printf(" Version: %d Serial Number: %d Mod Counter: %d\n",
654 ci_label->version, ci_label->serial_number,
655 ci_label->mod_counter);
656 printf(" Clean: %s Status: %d\n",
657 ci_label->clean ? "Yes" : "No", ci_label->status );
658 }
659
660 static int rf_check_label_vitals(RF_Raid_t *raidPtr, int row, int column,
661 char *dev_name, RF_ComponentLabel_t *ci_label,
662 int serial_number, int mod_counter)
663 {
664 int fatal_error = 0;
665
666 if (serial_number != ci_label->serial_number) {
667 printf("%s has a different serial number: %d %d\n",
668 dev_name, serial_number, ci_label->serial_number);
669 fatal_error = 1;
670 }
671 if (mod_counter != ci_label->mod_counter) {
672 printf("%s has a different modfication count: %d %d\n",
673 dev_name, mod_counter, ci_label->mod_counter);
674 }
675
676 if (row != ci_label->row) {
677 printf("Row out of alignment for: %s\n", dev_name);
678 fatal_error = 1;
679 }
680 if (column != ci_label->column) {
681 printf("Column out of alignment for: %s\n", dev_name);
682 fatal_error = 1;
683 }
684 if (raidPtr->numCol != ci_label->num_columns) {
685 printf("Number of columns do not match for: %s\n", dev_name);
686 fatal_error = 1;
687 }
688 if (ci_label->clean == 0) {
689 /* it's not clean, but that's not fatal */
690 printf("%s is not clean!\n", dev_name);
691 }
692 return(fatal_error);
693 }
694
695
696 /*
697
698 rf_CheckLabels() - check all the component labels for consistency.
699 Return an error if there is anything major amiss.
700
701 */
702
703 int
704 rf_CheckLabels(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr)
705 {
706 int c;
707 char *dev_name;
708 RF_ComponentLabel_t *ci_label;
709 int serial_number = 0;
710 int mod_number = 0;
711 int fatal_error = 0;
712 int mod_values[4];
713 int mod_count[4];
714 int ser_values[4];
715 int ser_count[4];
716 int num_ser;
717 int num_mod;
718 int i;
719 int found;
720 int hosed_column;
721 int too_fatal;
722 int parity_good;
723 int force;
724
725 hosed_column = -1;
726 too_fatal = 0;
727 force = cfgPtr->force;
728
729 /*
730 We're going to try to be a little intelligent here. If one
731 component's label is bogus, and we can identify that it's the
732 *only* one that's gone, we'll mark it as "failed" and allow
733 the configuration to proceed. This will be the *only* case
734 that we'll proceed if there would be (otherwise) fatal errors.
735
736 Basically we simply keep a count of how many components had
737 what serial number. If all but one agree, we simply mark
738 the disagreeing component as being failed, and allow
739 things to come up "normally".
740
741 We do this first for serial numbers, and then for "mod_counter".
742
743 */
744
745 num_ser = 0;
746 num_mod = 0;
747
748 for (c = 0; c < raidPtr->numCol; c++) {
749 ci_label = &raidPtr->raid_cinfo[c].ci_label;
750 found=0;
751 for(i=0;i<num_ser;i++) {
752 if (ser_values[i] == ci_label->serial_number) {
753 ser_count[i]++;
754 found=1;
755 break;
756 }
757 }
758 if (!found) {
759 ser_values[num_ser] = ci_label->serial_number;
760 ser_count[num_ser] = 1;
761 num_ser++;
762 if (num_ser>2) {
763 fatal_error = 1;
764 break;
765 }
766 }
767 found=0;
768 for(i=0;i<num_mod;i++) {
769 if (mod_values[i] == ci_label->mod_counter) {
770 mod_count[i]++;
771 found=1;
772 break;
773 }
774 }
775 if (!found) {
776 mod_values[num_mod] = ci_label->mod_counter;
777 mod_count[num_mod] = 1;
778 num_mod++;
779 if (num_mod>2) {
780 fatal_error = 1;
781 break;
782 }
783 }
784 }
785 #if DEBUG
786 printf("raid%d: Summary of serial numbers:\n", raidPtr->raidid);
787 for(i=0;i<num_ser;i++) {
788 printf("%d %d\n", ser_values[i], ser_count[i]);
789 }
790 printf("raid%d: Summary of mod counters:\n", raidPtr->raidid);
791 for(i=0;i<num_mod;i++) {
792 printf("%d %d\n", mod_values[i], mod_count[i]);
793 }
794 #endif
795 serial_number = ser_values[0];
796 if (num_ser == 2) {
797 if ((ser_count[0] == 1) || (ser_count[1] == 1)) {
798 /* Locate the maverick component */
799 if (ser_count[1] > ser_count[0]) {
800 serial_number = ser_values[1];
801 }
802
803 for (c = 0; c < raidPtr->numCol; c++) {
804 ci_label = &raidPtr->raid_cinfo[c].ci_label;
805 if (serial_number != ci_label->serial_number) {
806 hosed_column = c;
807 break;
808 }
809 }
810 printf("Hosed component: %s\n",
811 &cfgPtr->devnames[0][hosed_column][0]);
812 if (!force) {
813 /* we'll fail this component, as if there are
814 other major errors, we arn't forcing things
815 and we'll abort the config anyways */
816 raidPtr->Disks[hosed_column].status
817 = rf_ds_failed;
818 raidPtr->numFailures++;
819 raidPtr->status = rf_rs_degraded;
820 }
821 } else {
822 too_fatal = 1;
823 }
824 if (cfgPtr->parityConfig == '') {
825 /* We've identified two different serial numbers.
826 RAID 0 can't cope with that, so we'll punt */
827 too_fatal = 1;
828 }
829
830 }
831
832 /* record the serial number for later. If we bail later, setting
833 this doesn't matter, otherwise we've got the best guess at the
834 correct serial number */
835 raidPtr->serial_number = serial_number;
836
837 mod_number = mod_values[0];
838 if (num_mod == 2) {
839 if ((mod_count[0] == 1) || (mod_count[1] == 1)) {
840 /* Locate the maverick component */
841 if (mod_count[1] > mod_count[0]) {
842 mod_number = mod_values[1];
843 } else if (mod_count[1] < mod_count[0]) {
844 mod_number = mod_values[0];
845 } else {
846 /* counts of different modification values
847 are the same. Assume greater value is
848 the correct one, all other things
849 considered */
850 if (mod_values[0] > mod_values[1]) {
851 mod_number = mod_values[0];
852 } else {
853 mod_number = mod_values[1];
854 }
855
856 }
857
858 for (c = 0; c < raidPtr->numCol; c++) {
859 ci_label = &raidPtr->raid_cinfo[c].ci_label;
860 if (mod_number != ci_label->mod_counter) {
861 if (hosed_column == c) {
862 /* same one. Can
863 deal with it. */
864 } else {
865 hosed_column = c;
866 if (num_ser != 1) {
867 too_fatal = 1;
868 break;
869 }
870 }
871 }
872 }
873 printf("Hosed component: %s\n",
874 &cfgPtr->devnames[0][hosed_column][0]);
875 if (!force) {
876 /* we'll fail this component, as if there are
877 other major errors, we arn't forcing things
878 and we'll abort the config anyways */
879 if (raidPtr->Disks[hosed_column].status != rf_ds_failed) {
880 raidPtr->Disks[hosed_column].status
881 = rf_ds_failed;
882 raidPtr->numFailures++;
883 raidPtr->status = rf_rs_degraded;
884 }
885 }
886 } else {
887 too_fatal = 1;
888 }
889 if (cfgPtr->parityConfig == '') {
890 /* We've identified two different mod counters.
891 RAID 0 can't cope with that, so we'll punt */
892 too_fatal = 1;
893 }
894 }
895
896 raidPtr->mod_counter = mod_number;
897
898 if (too_fatal) {
899 /* we've had both a serial number mismatch, and a mod_counter
900 mismatch -- and they involved two different components!!
901 Bail -- make things fail so that the user must force
902 the issue... */
903 hosed_column = -1;
904 fatal_error = 1;
905 }
906
907 if (num_ser > 2) {
908 printf("raid%d: Too many different serial numbers!\n",
909 raidPtr->raidid);
910 fatal_error = 1;
911 }
912
913 if (num_mod > 2) {
914 printf("raid%d: Too many different mod counters!\n",
915 raidPtr->raidid);
916 fatal_error = 1;
917 }
918
919 /* we start by assuming the parity will be good, and flee from
920 that notion at the slightest sign of trouble */
921
922 parity_good = RF_RAID_CLEAN;
923
924 for (c = 0; c < raidPtr->numCol; c++) {
925 dev_name = &cfgPtr->devnames[0][c][0];
926 ci_label = &raidPtr->raid_cinfo[c].ci_label;
927
928 if (c == hosed_column) {
929 printf("raid%d: Ignoring %s\n",
930 raidPtr->raidid, dev_name);
931 } else {
932 rf_print_label_status( raidPtr, c, dev_name, ci_label);
933 if (rf_check_label_vitals( raidPtr, 0, c,
934 dev_name, ci_label,
935 serial_number,
936 mod_number )) {
937 fatal_error = 1;
938 }
939 if (ci_label->clean != RF_RAID_CLEAN) {
940 parity_good = RF_RAID_DIRTY;
941 }
942 }
943 }
944
945 if (fatal_error) {
946 parity_good = RF_RAID_DIRTY;
947 }
948
949 /* we note the state of the parity */
950 raidPtr->parity_good = parity_good;
951
952 return(fatal_error);
953 }
954
955 int
956 rf_add_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
957 {
958 RF_RaidDisk_t *disks;
959 RF_DiskQueue_t *spareQueues;
960 int ret;
961 unsigned int bs;
962 int spare_number;
963
964 ret=0;
965
966 if (raidPtr->numSpare >= RF_MAXSPARE) {
967 RF_ERRORMSG1("Too many spares: %d\n", raidPtr->numSpare);
968 return(EINVAL);
969 }
970
971 RF_LOCK_MUTEX(raidPtr->mutex);
972 while (raidPtr->adding_hot_spare==1) {
973 ltsleep(&(raidPtr->adding_hot_spare), PRIBIO, "raidhs", 0,
974 &(raidPtr->mutex));
975 }
976 raidPtr->adding_hot_spare=1;
977 RF_UNLOCK_MUTEX(raidPtr->mutex);
978
979 /* the beginning of the spares... */
980 disks = &raidPtr->Disks[raidPtr->numCol];
981
982 spare_number = raidPtr->numSpare;
983
984 ret = rf_ConfigureDisk(raidPtr, sparePtr->component_name,
985 &disks[spare_number],
986 raidPtr->numCol + spare_number);
987
988 if (ret)
989 goto fail;
990 if (disks[spare_number].status != rf_ds_optimal) {
991 RF_ERRORMSG1("Warning: spare disk %s failed TUR\n",
992 sparePtr->component_name);
993 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
994 ret=EINVAL;
995 goto fail;
996 } else {
997 disks[spare_number].status = rf_ds_spare;
998 DPRINTF6("Spare Disk %d: dev %s numBlocks %ld blockSize %d (%ld MB)\n", spare_number,
999 disks[spare_number].devname,
1000 (long int) disks[spare_number].numBlocks,
1001 disks[spare_number].blockSize,
1002 (long int) disks[spare_number].numBlocks *
1003 disks[spare_number].blockSize / 1024 / 1024);
1004 }
1005
1006
1007 /* check sizes and block sizes on the spare disk */
1008 bs = 1 << raidPtr->logBytesPerSector;
1009 if (disks[spare_number].blockSize != bs) {
1010 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);
1011 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1012 ret = EINVAL;
1013 goto fail;
1014 }
1015 if (disks[spare_number].numBlocks < raidPtr->sectorsPerDisk) {
1016 RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %ld blocks)\n",
1017 disks[spare_number].devname,
1018 disks[spare_number].blockSize,
1019 (long int) raidPtr->sectorsPerDisk);
1020 rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0);
1021 ret = EINVAL;
1022 goto fail;
1023 } else {
1024 if (disks[spare_number].numBlocks >
1025 raidPtr->sectorsPerDisk) {
1026 RF_ERRORMSG2("Warning: truncating spare disk %s to %ld blocks\n", disks[spare_number].devname,
1027 (long int) raidPtr->sectorsPerDisk);
1028
1029 disks[spare_number].numBlocks = raidPtr->sectorsPerDisk;
1030 }
1031 }
1032
1033 spareQueues = &raidPtr->Queues[raidPtr->numCol];
1034 ret = rf_ConfigureDiskQueue( raidPtr, &spareQueues[spare_number],
1035 raidPtr->numCol + spare_number,
1036 raidPtr->qType,
1037 raidPtr->sectorsPerDisk,
1038 raidPtr->Disks[raidPtr->numCol +
1039 spare_number].dev,
1040 raidPtr->maxOutstanding,
1041 &raidPtr->shutdownList,
1042 raidPtr->cleanupList);
1043
1044 RF_LOCK_MUTEX(raidPtr->mutex);
1045 raidPtr->numSpare++;
1046 RF_UNLOCK_MUTEX(raidPtr->mutex);
1047
1048 fail:
1049 RF_LOCK_MUTEX(raidPtr->mutex);
1050 raidPtr->adding_hot_spare=0;
1051 wakeup(&(raidPtr->adding_hot_spare));
1052 RF_UNLOCK_MUTEX(raidPtr->mutex);
1053
1054 return(ret);
1055 }
1056
1057 int
1058 rf_remove_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr)
1059 {
1060 int spare_number;
1061
1062
1063 if (raidPtr->numSpare==0) {
1064 printf("No spares to remove!\n");
1065 return(EINVAL);
1066 }
1067
1068 spare_number = sparePtr->column;
1069
1070 return(EINVAL); /* XXX not implemented yet */
1071 #if 0
1072 if (spare_number < 0 || spare_number > raidPtr->numSpare) {
1073 return(EINVAL);
1074 }
1075
1076 /* verify that this spare isn't in use... */
1077
1078
1079
1080
1081 /* it's gone.. */
1082
1083 raidPtr->numSpare--;
1084
1085 return(0);
1086 #endif
1087 }
1088
1089
1090 int
1091 rf_delete_component(RF_Raid_t *raidPtr, RF_SingleComponent_t *component)
1092 {
1093 RF_RaidDisk_t *disks;
1094
1095 if ((component->column < 0) ||
1096 (component->column >= raidPtr->numCol)) {
1097 return(EINVAL);
1098 }
1099
1100 disks = &raidPtr->Disks[component->column];
1101
1102 /* 1. This component must be marked as 'failed' */
1103
1104 return(EINVAL); /* Not implemented yet. */
1105 }
1106
1107 int
1108 rf_incorporate_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *component)
1109 {
1110
1111 /* Issues here include how to 'move' this in if there is IO
1112 taking place (e.g. component queues and such) */
1113
1114 return(EINVAL); /* Not implemented yet. */
1115 }
Cache object: e1bdd17abe5578c3f574a8cc3d83b344
|