1 | /* $NetBSD: rf_disks.c,v 1.87 2014/10/18 08:33:28 snj 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 | * |
18 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
19 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
20 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
21 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
22 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
23 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
24 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
25 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
26 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
27 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
28 | * POSSIBILITY OF SUCH DAMAGE. |
29 | */ |
30 | |
31 | /* |
32 | * Copyright (c) 1995 Carnegie-Mellon University. |
33 | * All rights reserved. |
34 | * |
35 | * Author: Mark Holland |
36 | * |
37 | * Permission to use, copy, modify and distribute this software and |
38 | * its documentation is hereby granted, provided that both the copyright |
39 | * notice and this permission notice appear in all copies of the |
40 | * software, derivative works or modified versions, and any portions |
41 | * thereof, and that both notices appear in supporting documentation. |
42 | * |
43 | * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" |
44 | * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND |
45 | * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. |
46 | * |
47 | * Carnegie Mellon requests users of this software to return to |
48 | * |
49 | * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU |
50 | * School of Computer Science |
51 | * Carnegie Mellon University |
52 | * Pittsburgh PA 15213-3890 |
53 | * |
54 | * any improvements or extensions that they make and grant Carnegie the |
55 | * rights to redistribute these changes. |
56 | */ |
57 | |
58 | /*************************************************************** |
59 | * rf_disks.c -- code to perform operations on the actual disks |
60 | ***************************************************************/ |
61 | |
62 | #include <sys/cdefs.h> |
63 | __KERNEL_RCSID(0, "$NetBSD: rf_disks.c,v 1.87 2014/10/18 08:33:28 snj Exp $" ); |
64 | |
65 | #include <dev/raidframe/raidframevar.h> |
66 | |
67 | #include "rf_raid.h" |
68 | #include "rf_alloclist.h" |
69 | #include "rf_utils.h" |
70 | #include "rf_general.h" |
71 | #include "rf_options.h" |
72 | #include "rf_kintf.h" |
73 | #include "rf_netbsd.h" |
74 | |
75 | #include <sys/param.h> |
76 | #include <sys/systm.h> |
77 | #include <sys/proc.h> |
78 | #include <sys/ioctl.h> |
79 | #include <sys/fcntl.h> |
80 | #include <sys/vnode.h> |
81 | #include <sys/namei.h> /* for pathbuf */ |
82 | #include <sys/kauth.h> |
83 | #include <miscfs/specfs/specdev.h> /* for v_rdev */ |
84 | |
85 | static int rf_AllocDiskStructures(RF_Raid_t *, RF_Config_t *); |
86 | static void rf_print_label_status( RF_Raid_t *, int, char *, |
87 | RF_ComponentLabel_t *); |
88 | static int rf_check_label_vitals( RF_Raid_t *, int, int, char *, |
89 | RF_ComponentLabel_t *, int, int ); |
90 | |
91 | #define DPRINTF6(a,b,c,d,e,f) if (rf_diskDebug) printf(a,b,c,d,e,f) |
92 | #define DPRINTF7(a,b,c,d,e,f,g) if (rf_diskDebug) printf(a,b,c,d,e,f,g) |
93 | |
94 | /************************************************************************** |
95 | * |
96 | * initialize the disks comprising the array |
97 | * |
98 | * We want the spare disks to have regular row,col numbers so that we can |
99 | * easily substitue a spare for a failed disk. But, the driver code assumes |
100 | * throughout that the array contains numRow by numCol _non-spare_ disks, so |
101 | * it's not clear how to fit in the spares. This is an unfortunate holdover |
102 | * from raidSim. The quick and dirty fix is to make row zero bigger than the |
103 | * rest, and put all the spares in it. This probably needs to get changed |
104 | * eventually. |
105 | * |
106 | **************************************************************************/ |
107 | |
108 | int |
109 | rf_ConfigureDisks(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr, |
110 | RF_Config_t *cfgPtr) |
111 | { |
112 | RF_RaidDisk_t *disks; |
113 | RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL; |
114 | RF_RowCol_t c; |
115 | int bs, ret; |
116 | unsigned i, count, foundone = 0, numFailuresThisRow; |
117 | int force; |
118 | |
119 | force = cfgPtr->force; |
120 | |
121 | ret = rf_AllocDiskStructures(raidPtr, cfgPtr); |
122 | if (ret) |
123 | goto fail; |
124 | |
125 | disks = raidPtr->Disks; |
126 | |
127 | numFailuresThisRow = 0; |
128 | for (c = 0; c < raidPtr->numCol; c++) { |
129 | ret = rf_ConfigureDisk(raidPtr, |
130 | &cfgPtr->devnames[0][c][0], |
131 | &disks[c], c); |
132 | |
133 | if (ret) |
134 | goto fail; |
135 | |
136 | if (disks[c].status == rf_ds_optimal) { |
137 | ret = raidfetch_component_label(raidPtr, c); |
138 | if (ret) |
139 | goto fail; |
140 | |
141 | /* mark it as failed if the label looks bogus... */ |
142 | if (!rf_reasonable_label(&raidPtr->raid_cinfo[c].ci_label,0) && !force) { |
143 | disks[c].status = rf_ds_failed; |
144 | } |
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 %" PRIu64 " blockSize %d (%ld MB)\n" , |
153 | c, disks[c].devname, |
154 | 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 %" PRIu64 " blockSize %d (%ld MB)\n" , i, |
263 | disks[i].devname, |
264 | 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 %" PRIu64 " blocks)\n" , |
281 | disks[i].devname, disks[i].blockSize, |
282 | raidPtr->sectorsPerDisk); |
283 | ret = EINVAL; |
284 | goto fail; |
285 | } else |
286 | if (disks[i].numBlocks > raidPtr->sectorsPerDisk) { |
287 | RF_ERRORMSG3("Warning: truncating spare disk %s to %" PRIu64 " blocks (from %" PRIu64 ")\n" , |
288 | disks[i].devname, |
289 | raidPtr->sectorsPerDisk, |
290 | disks[i].numBlocks); |
291 | |
292 | disks[i].numBlocks = raidPtr->sectorsPerDisk; |
293 | } |
294 | } |
295 | |
296 | return (0); |
297 | |
298 | fail: |
299 | |
300 | /* Release the hold on the main components. We've failed to allocate |
301 | * a spare, and since we're failing, we need to free things.. |
302 | |
303 | XXX failing to allocate a spare is *not* that big of a deal... |
304 | We *can* survive without it, if need be, esp. if we get hot |
305 | adding working. |
306 | |
307 | If we don't fail out here, then we need a way to remove this spare... |
308 | that should be easier to do here than if we are "live"... |
309 | |
310 | */ |
311 | |
312 | rf_UnconfigureVnodes( raidPtr ); |
313 | |
314 | return (ret); |
315 | } |
316 | |
317 | static int |
318 | rf_AllocDiskStructures(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr) |
319 | { |
320 | int ret; |
321 | |
322 | /* We allocate RF_MAXSPARE on the first row so that we |
323 | have room to do hot-swapping of spares */ |
324 | RF_MallocAndAdd(raidPtr->Disks, (raidPtr->numCol + RF_MAXSPARE) * |
325 | sizeof(RF_RaidDisk_t), (RF_RaidDisk_t *), |
326 | raidPtr->cleanupList); |
327 | if (raidPtr->Disks == NULL) { |
328 | ret = ENOMEM; |
329 | goto fail; |
330 | } |
331 | |
332 | /* get space for device specific stuff.. */ |
333 | RF_MallocAndAdd(raidPtr->raid_cinfo, |
334 | (raidPtr->numCol + RF_MAXSPARE) * |
335 | sizeof(struct raidcinfo), (struct raidcinfo *), |
336 | raidPtr->cleanupList); |
337 | |
338 | if (raidPtr->raid_cinfo == NULL) { |
339 | ret = ENOMEM; |
340 | goto fail; |
341 | } |
342 | |
343 | return(0); |
344 | fail: |
345 | rf_UnconfigureVnodes( raidPtr ); |
346 | |
347 | return(ret); |
348 | } |
349 | |
350 | |
351 | /* configure a single disk during auto-configuration at boot */ |
352 | int |
353 | rf_AutoConfigureDisks(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, |
354 | RF_AutoConfig_t *auto_config) |
355 | { |
356 | RF_RaidDisk_t *disks; |
357 | RF_RaidDisk_t *diskPtr; |
358 | RF_RowCol_t c; |
359 | RF_SectorCount_t min_numblks = (RF_SectorCount_t) 0x7FFFFFFFFFFFLL; |
360 | int bs, ret; |
361 | int numFailuresThisRow; |
362 | RF_AutoConfig_t *ac; |
363 | int parity_good; |
364 | int mod_counter; |
365 | int mod_counter_found; |
366 | |
367 | #if DEBUG |
368 | printf("Starting autoconfiguration of RAID set...\n" ); |
369 | #endif |
370 | |
371 | ret = rf_AllocDiskStructures(raidPtr, cfgPtr); |
372 | if (ret) |
373 | goto fail; |
374 | |
375 | disks = raidPtr->Disks; |
376 | |
377 | /* assume the parity will be fine.. */ |
378 | parity_good = RF_RAID_CLEAN; |
379 | |
380 | /* Check for mod_counters that are too low */ |
381 | mod_counter_found = 0; |
382 | mod_counter = 0; |
383 | ac = auto_config; |
384 | while(ac!=NULL) { |
385 | if (mod_counter_found==0) { |
386 | mod_counter = ac->clabel->mod_counter; |
387 | mod_counter_found = 1; |
388 | } else { |
389 | if (ac->clabel->mod_counter > mod_counter) { |
390 | mod_counter = ac->clabel->mod_counter; |
391 | } |
392 | } |
393 | ac->flag = 0; /* clear the general purpose flag */ |
394 | ac = ac->next; |
395 | } |
396 | |
397 | bs = 0; |
398 | |
399 | numFailuresThisRow = 0; |
400 | for (c = 0; c < raidPtr->numCol; c++) { |
401 | diskPtr = &disks[c]; |
402 | |
403 | /* find this row/col in the autoconfig */ |
404 | #if DEBUG |
405 | printf("Looking for %d in autoconfig\n" ,c); |
406 | #endif |
407 | ac = auto_config; |
408 | while(ac!=NULL) { |
409 | if (ac->clabel==NULL) { |
410 | /* big-time bad news. */ |
411 | goto fail; |
412 | } |
413 | if ((ac->clabel->column == c) && |
414 | (ac->clabel->mod_counter == mod_counter)) { |
415 | /* it's this one... */ |
416 | /* flag it as 'used', so we don't |
417 | free it later. */ |
418 | ac->flag = 1; |
419 | #if DEBUG |
420 | printf("Found: %s at %d\n" , |
421 | ac->devname,c); |
422 | #endif |
423 | |
424 | break; |
425 | } |
426 | ac=ac->next; |
427 | } |
428 | |
429 | if (ac==NULL) { |
430 | /* we didn't find an exact match with a |
431 | correct mod_counter above... can we find |
432 | one with an incorrect mod_counter to use |
433 | instead? (this one, if we find it, will be |
434 | marked as failed once the set configures) |
435 | */ |
436 | |
437 | ac = auto_config; |
438 | while(ac!=NULL) { |
439 | if (ac->clabel==NULL) { |
440 | /* big-time bad news. */ |
441 | goto fail; |
442 | } |
443 | if (ac->clabel->column == c) { |
444 | /* it's this one... |
445 | flag it as 'used', so we |
446 | don't free it later. */ |
447 | ac->flag = 1; |
448 | #if DEBUG |
449 | printf("Found(low mod_counter): %s at %d\n" , |
450 | ac->devname,c); |
451 | #endif |
452 | |
453 | break; |
454 | } |
455 | ac=ac->next; |
456 | } |
457 | } |
458 | |
459 | |
460 | |
461 | if (ac!=NULL) { |
462 | /* Found it. Configure it.. */ |
463 | diskPtr->blockSize = ac->clabel->blockSize; |
464 | diskPtr->numBlocks = |
465 | rf_component_label_numblocks(ac->clabel); |
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(raidget_component_label(raidPtr, c), |
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 its 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 pathbuf *pb; |
579 | struct vnode *vp; |
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 | snprintf(diskPtr->devname, sizeof(diskPtr->devname), |
598 | "component%d" , col); |
599 | diskPtr->status = rf_ds_failed; |
600 | return (0); |
601 | } |
602 | |
603 | pb = pathbuf_create(diskPtr->devname); |
604 | if (pb == NULL) { |
605 | printf("pathbuf_create for device: %s failed!\n" , |
606 | diskPtr->devname); |
607 | return ENOMEM; |
608 | } |
609 | error = dk_lookup(pb, curlwp, &vp); |
610 | pathbuf_destroy(pb); |
611 | if (error) { |
612 | printf("dk_lookup on device: %s failed!\n" , diskPtr->devname); |
613 | if (error == ENXIO) { |
614 | /* the component isn't there... must be dead :-( */ |
615 | diskPtr->status = rf_ds_failed; |
616 | return 0; |
617 | } else { |
618 | return (error); |
619 | } |
620 | } |
621 | |
622 | if ((error = rf_getdisksize(vp, diskPtr)) != 0) |
623 | return (error); |
624 | |
625 | /* |
626 | * If this raidPtr's bytesPerSector is zero, fill it in with this |
627 | * components blockSize. This will give us something to work with |
628 | * initially, and if it is wrong, we'll get errors later. |
629 | */ |
630 | if (raidPtr->bytesPerSector == 0) |
631 | raidPtr->bytesPerSector = diskPtr->blockSize; |
632 | |
633 | if (diskPtr->status == rf_ds_optimal) { |
634 | raidPtr->raid_cinfo[col].ci_vp = vp; |
635 | raidPtr->raid_cinfo[col].ci_dev = vp->v_rdev; |
636 | |
637 | /* This component was not automatically configured */ |
638 | diskPtr->auto_configured = 0; |
639 | diskPtr->dev = vp->v_rdev; |
640 | |
641 | /* we allow the user to specify that only a fraction of the |
642 | * disks should be used this is just for debug: it speeds up |
643 | * the parity scan */ |
644 | diskPtr->numBlocks = diskPtr->numBlocks * |
645 | rf_sizePercentage / 100; |
646 | } |
647 | return (0); |
648 | } |
649 | |
650 | static void |
651 | rf_print_label_status(RF_Raid_t *raidPtr, int column, char *dev_name, |
652 | RF_ComponentLabel_t *ci_label) |
653 | { |
654 | |
655 | printf("raid%d: Component %s being configured at col: %d\n" , |
656 | raidPtr->raidid, dev_name, column ); |
657 | printf(" Column: %d Num Columns: %d\n" , |
658 | ci_label->column, |
659 | ci_label->num_columns); |
660 | printf(" Version: %d Serial Number: %d Mod Counter: %d\n" , |
661 | ci_label->version, ci_label->serial_number, |
662 | ci_label->mod_counter); |
663 | printf(" Clean: %s Status: %d\n" , |
664 | ci_label->clean ? "Yes" : "No" , ci_label->status ); |
665 | } |
666 | |
667 | static int rf_check_label_vitals(RF_Raid_t *raidPtr, int row, int column, |
668 | char *dev_name, RF_ComponentLabel_t *ci_label, |
669 | int serial_number, int mod_counter) |
670 | { |
671 | int fatal_error = 0; |
672 | |
673 | if (serial_number != ci_label->serial_number) { |
674 | printf("%s has a different serial number: %d %d\n" , |
675 | dev_name, serial_number, ci_label->serial_number); |
676 | fatal_error = 1; |
677 | } |
678 | if (mod_counter != ci_label->mod_counter) { |
679 | printf("%s has a different modification count: %d %d\n" , |
680 | dev_name, mod_counter, ci_label->mod_counter); |
681 | } |
682 | |
683 | if (row != ci_label->row) { |
684 | printf("Row out of alignment for: %s\n" , dev_name); |
685 | fatal_error = 1; |
686 | } |
687 | if (column != ci_label->column) { |
688 | printf("Column out of alignment for: %s\n" , dev_name); |
689 | fatal_error = 1; |
690 | } |
691 | if (raidPtr->numCol != ci_label->num_columns) { |
692 | printf("Number of columns do not match for: %s\n" , dev_name); |
693 | fatal_error = 1; |
694 | } |
695 | if (ci_label->clean == 0) { |
696 | /* it's not clean, but that's not fatal */ |
697 | printf("%s is not clean!\n" , dev_name); |
698 | } |
699 | return(fatal_error); |
700 | } |
701 | |
702 | |
703 | /* |
704 | |
705 | rf_CheckLabels() - check all the component labels for consistency. |
706 | Return an error if there is anything major amiss. |
707 | |
708 | */ |
709 | |
710 | int |
711 | rf_CheckLabels(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr) |
712 | { |
713 | int c; |
714 | char *dev_name; |
715 | RF_ComponentLabel_t *ci_label; |
716 | int serial_number = 0; |
717 | int mod_number = 0; |
718 | int fatal_error = 0; |
719 | int mod_values[4]; |
720 | int mod_count[4]; |
721 | int ser_values[4]; |
722 | int ser_count[4]; |
723 | int num_ser; |
724 | int num_mod; |
725 | int i; |
726 | int found; |
727 | int hosed_column; |
728 | int too_fatal; |
729 | int parity_good; |
730 | int force; |
731 | |
732 | hosed_column = -1; |
733 | too_fatal = 0; |
734 | force = cfgPtr->force; |
735 | |
736 | /* |
737 | We're going to try to be a little intelligent here. If one |
738 | component's label is bogus, and we can identify that it's the |
739 | *only* one that's gone, we'll mark it as "failed" and allow |
740 | the configuration to proceed. This will be the *only* case |
741 | that we'll proceed if there would be (otherwise) fatal errors. |
742 | |
743 | Basically we simply keep a count of how many components had |
744 | what serial number. If all but one agree, we simply mark |
745 | the disagreeing component as being failed, and allow |
746 | things to come up "normally". |
747 | |
748 | We do this first for serial numbers, and then for "mod_counter". |
749 | |
750 | */ |
751 | |
752 | num_ser = 0; |
753 | num_mod = 0; |
754 | |
755 | ser_values[0] = ser_values[1] = ser_values[2] = ser_values[3] = 0; |
756 | ser_count[0] = ser_count[1] = ser_count[2] = ser_count[3] = 0; |
757 | mod_values[0] = mod_values[1] = mod_values[2] = mod_values[3] = 0; |
758 | mod_count[0] = mod_count[1] = mod_count[2] = mod_count[3] = 0; |
759 | |
760 | for (c = 0; c < raidPtr->numCol; c++) { |
761 | if (raidPtr->Disks[c].status != rf_ds_optimal) |
762 | continue; |
763 | ci_label = raidget_component_label(raidPtr, c); |
764 | found=0; |
765 | for(i=0;i<num_ser;i++) { |
766 | if (ser_values[i] == ci_label->serial_number) { |
767 | ser_count[i]++; |
768 | found=1; |
769 | break; |
770 | } |
771 | } |
772 | if (!found) { |
773 | ser_values[num_ser] = ci_label->serial_number; |
774 | ser_count[num_ser] = 1; |
775 | num_ser++; |
776 | if (num_ser>2) { |
777 | fatal_error = 1; |
778 | break; |
779 | } |
780 | } |
781 | found=0; |
782 | for(i=0;i<num_mod;i++) { |
783 | if (mod_values[i] == ci_label->mod_counter) { |
784 | mod_count[i]++; |
785 | found=1; |
786 | break; |
787 | } |
788 | } |
789 | if (!found) { |
790 | mod_values[num_mod] = ci_label->mod_counter; |
791 | mod_count[num_mod] = 1; |
792 | num_mod++; |
793 | if (num_mod>2) { |
794 | fatal_error = 1; |
795 | break; |
796 | } |
797 | } |
798 | } |
799 | #if DEBUG |
800 | printf("raid%d: Summary of serial numbers:\n" , raidPtr->raidid); |
801 | for(i=0;i<num_ser;i++) { |
802 | printf("%d %d\n" , ser_values[i], ser_count[i]); |
803 | } |
804 | printf("raid%d: Summary of mod counters:\n" , raidPtr->raidid); |
805 | for(i=0;i<num_mod;i++) { |
806 | printf("%d %d\n" , mod_values[i], mod_count[i]); |
807 | } |
808 | #endif |
809 | serial_number = ser_values[0]; |
810 | if (num_ser == 2) { |
811 | if ((ser_count[0] == 1) || (ser_count[1] == 1)) { |
812 | /* Locate the maverick component */ |
813 | if (ser_count[1] > ser_count[0]) { |
814 | serial_number = ser_values[1]; |
815 | } |
816 | |
817 | for (c = 0; c < raidPtr->numCol; c++) { |
818 | if (raidPtr->Disks[c].status != rf_ds_optimal) |
819 | continue; |
820 | ci_label = raidget_component_label(raidPtr, c); |
821 | if (serial_number != ci_label->serial_number) { |
822 | hosed_column = c; |
823 | break; |
824 | } |
825 | } |
826 | printf("Hosed component: %s\n" , |
827 | &cfgPtr->devnames[0][hosed_column][0]); |
828 | if (!force) { |
829 | /* we'll fail this component, as if there are |
830 | other major errors, we arn't forcing things |
831 | and we'll abort the config anyways */ |
832 | raidPtr->Disks[hosed_column].status |
833 | = rf_ds_failed; |
834 | raidPtr->numFailures++; |
835 | raidPtr->status = rf_rs_degraded; |
836 | } |
837 | } else { |
838 | too_fatal = 1; |
839 | } |
840 | if (cfgPtr->parityConfig == '0') { |
841 | /* We've identified two different serial numbers. |
842 | RAID 0 can't cope with that, so we'll punt */ |
843 | too_fatal = 1; |
844 | } |
845 | |
846 | } |
847 | |
848 | /* record the serial number for later. If we bail later, setting |
849 | this doesn't matter, otherwise we've got the best guess at the |
850 | correct serial number */ |
851 | raidPtr->serial_number = serial_number; |
852 | |
853 | mod_number = mod_values[0]; |
854 | if (num_mod == 2) { |
855 | if ((mod_count[0] == 1) || (mod_count[1] == 1)) { |
856 | /* Locate the maverick component */ |
857 | if (mod_count[1] > mod_count[0]) { |
858 | mod_number = mod_values[1]; |
859 | } else if (mod_count[1] < mod_count[0]) { |
860 | mod_number = mod_values[0]; |
861 | } else { |
862 | /* counts of different modification values |
863 | are the same. Assume greater value is |
864 | the correct one, all other things |
865 | considered */ |
866 | if (mod_values[0] > mod_values[1]) { |
867 | mod_number = mod_values[0]; |
868 | } else { |
869 | mod_number = mod_values[1]; |
870 | } |
871 | |
872 | } |
873 | |
874 | for (c = 0; c < raidPtr->numCol; c++) { |
875 | if (raidPtr->Disks[c].status != rf_ds_optimal) |
876 | continue; |
877 | |
878 | ci_label = raidget_component_label(raidPtr, c); |
879 | if (mod_number != ci_label->mod_counter) { |
880 | if (hosed_column == c) { |
881 | /* same one. Can |
882 | deal with it. */ |
883 | } else { |
884 | hosed_column = c; |
885 | if (num_ser != 1) { |
886 | too_fatal = 1; |
887 | break; |
888 | } |
889 | } |
890 | } |
891 | } |
892 | printf("Hosed component: %s\n" , |
893 | &cfgPtr->devnames[0][hosed_column][0]); |
894 | if (!force) { |
895 | /* we'll fail this component, as if there are |
896 | other major errors, we arn't forcing things |
897 | and we'll abort the config anyways */ |
898 | if (raidPtr->Disks[hosed_column].status != rf_ds_failed) { |
899 | raidPtr->Disks[hosed_column].status |
900 | = rf_ds_failed; |
901 | raidPtr->numFailures++; |
902 | raidPtr->status = rf_rs_degraded; |
903 | } |
904 | } |
905 | } else { |
906 | too_fatal = 1; |
907 | } |
908 | if (cfgPtr->parityConfig == '0') { |
909 | /* We've identified two different mod counters. |
910 | RAID 0 can't cope with that, so we'll punt */ |
911 | too_fatal = 1; |
912 | } |
913 | } |
914 | |
915 | raidPtr->mod_counter = mod_number; |
916 | |
917 | if (too_fatal) { |
918 | /* we've had both a serial number mismatch, and a mod_counter |
919 | mismatch -- and they involved two different components!! |
920 | Bail -- make things fail so that the user must force |
921 | the issue... */ |
922 | hosed_column = -1; |
923 | fatal_error = 1; |
924 | } |
925 | |
926 | if (num_ser > 2) { |
927 | printf("raid%d: Too many different serial numbers!\n" , |
928 | raidPtr->raidid); |
929 | fatal_error = 1; |
930 | } |
931 | |
932 | if (num_mod > 2) { |
933 | printf("raid%d: Too many different mod counters!\n" , |
934 | raidPtr->raidid); |
935 | fatal_error = 1; |
936 | } |
937 | |
938 | for (c = 0; c < raidPtr->numCol; c++) { |
939 | if (raidPtr->Disks[c].status != rf_ds_optimal) { |
940 | hosed_column = c; |
941 | break; |
942 | } |
943 | } |
944 | |
945 | /* we start by assuming the parity will be good, and flee from |
946 | that notion at the slightest sign of trouble */ |
947 | |
948 | parity_good = RF_RAID_CLEAN; |
949 | |
950 | for (c = 0; c < raidPtr->numCol; c++) { |
951 | dev_name = &cfgPtr->devnames[0][c][0]; |
952 | ci_label = raidget_component_label(raidPtr, c); |
953 | |
954 | if (c == hosed_column) { |
955 | printf("raid%d: Ignoring %s\n" , |
956 | raidPtr->raidid, dev_name); |
957 | } else { |
958 | rf_print_label_status( raidPtr, c, dev_name, ci_label); |
959 | if (rf_check_label_vitals( raidPtr, 0, c, |
960 | dev_name, ci_label, |
961 | serial_number, |
962 | mod_number )) { |
963 | fatal_error = 1; |
964 | } |
965 | if (ci_label->clean != RF_RAID_CLEAN) { |
966 | parity_good = RF_RAID_DIRTY; |
967 | } |
968 | } |
969 | } |
970 | |
971 | if (fatal_error) { |
972 | parity_good = RF_RAID_DIRTY; |
973 | } |
974 | |
975 | /* we note the state of the parity */ |
976 | raidPtr->parity_good = parity_good; |
977 | |
978 | return(fatal_error); |
979 | } |
980 | |
981 | int |
982 | rf_add_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr) |
983 | { |
984 | RF_RaidDisk_t *disks; |
985 | RF_DiskQueue_t *spareQueues; |
986 | int ret; |
987 | unsigned int bs; |
988 | int spare_number; |
989 | |
990 | ret=0; |
991 | |
992 | if (raidPtr->numSpare >= RF_MAXSPARE) { |
993 | RF_ERRORMSG1("Too many spares: %d\n" , raidPtr->numSpare); |
994 | return(EINVAL); |
995 | } |
996 | |
997 | rf_lock_mutex2(raidPtr->mutex); |
998 | while (raidPtr->adding_hot_spare == 1) { |
999 | rf_wait_cond2(raidPtr->adding_hot_spare_cv, raidPtr->mutex); |
1000 | } |
1001 | raidPtr->adding_hot_spare = 1; |
1002 | rf_unlock_mutex2(raidPtr->mutex); |
1003 | |
1004 | /* the beginning of the spares... */ |
1005 | disks = &raidPtr->Disks[raidPtr->numCol]; |
1006 | |
1007 | spare_number = raidPtr->numSpare; |
1008 | |
1009 | ret = rf_ConfigureDisk(raidPtr, sparePtr->component_name, |
1010 | &disks[spare_number], |
1011 | raidPtr->numCol + spare_number); |
1012 | |
1013 | if (ret) |
1014 | goto fail; |
1015 | if (disks[spare_number].status != rf_ds_optimal) { |
1016 | RF_ERRORMSG1("Warning: spare disk %s failed TUR\n" , |
1017 | sparePtr->component_name); |
1018 | rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); |
1019 | ret=EINVAL; |
1020 | goto fail; |
1021 | } else { |
1022 | disks[spare_number].status = rf_ds_spare; |
1023 | DPRINTF6("Spare Disk %d: dev %s numBlocks %" PRIu64 " blockSize %d (%ld MB)\n" , |
1024 | spare_number, |
1025 | disks[spare_number].devname, |
1026 | disks[spare_number].numBlocks, |
1027 | disks[spare_number].blockSize, |
1028 | (long int) disks[spare_number].numBlocks * |
1029 | disks[spare_number].blockSize / 1024 / 1024); |
1030 | } |
1031 | |
1032 | |
1033 | /* check sizes and block sizes on the spare disk */ |
1034 | bs = 1 << raidPtr->logBytesPerSector; |
1035 | if (disks[spare_number].blockSize != bs) { |
1036 | 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); |
1037 | rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); |
1038 | ret = EINVAL; |
1039 | goto fail; |
1040 | } |
1041 | if (disks[spare_number].numBlocks < raidPtr->sectorsPerDisk) { |
1042 | RF_ERRORMSG3("Spare disk %s (%d blocks) is too small to serve as a spare (need %" PRIu64 " blocks)\n" , |
1043 | disks[spare_number].devname, |
1044 | disks[spare_number].blockSize, |
1045 | raidPtr->sectorsPerDisk); |
1046 | rf_close_component(raidPtr, raidPtr->raid_cinfo[raidPtr->numCol+spare_number].ci_vp, 0); |
1047 | ret = EINVAL; |
1048 | goto fail; |
1049 | } else { |
1050 | if (disks[spare_number].numBlocks > |
1051 | raidPtr->sectorsPerDisk) { |
1052 | RF_ERRORMSG3("Warning: truncating spare disk %s to %" PRIu64 " blocks (from %" PRIu64 ")\n" , |
1053 | disks[spare_number].devname, |
1054 | raidPtr->sectorsPerDisk, |
1055 | disks[spare_number].numBlocks); |
1056 | |
1057 | disks[spare_number].numBlocks = raidPtr->sectorsPerDisk; |
1058 | } |
1059 | } |
1060 | |
1061 | spareQueues = &raidPtr->Queues[raidPtr->numCol]; |
1062 | ret = rf_ConfigureDiskQueue( raidPtr, &spareQueues[spare_number], |
1063 | raidPtr->numCol + spare_number, |
1064 | raidPtr->qType, |
1065 | raidPtr->sectorsPerDisk, |
1066 | raidPtr->Disks[raidPtr->numCol + |
1067 | spare_number].dev, |
1068 | raidPtr->maxOutstanding, |
1069 | &raidPtr->shutdownList, |
1070 | raidPtr->cleanupList); |
1071 | |
1072 | rf_lock_mutex2(raidPtr->mutex); |
1073 | raidPtr->numSpare++; |
1074 | rf_unlock_mutex2(raidPtr->mutex); |
1075 | |
1076 | fail: |
1077 | rf_lock_mutex2(raidPtr->mutex); |
1078 | raidPtr->adding_hot_spare = 0; |
1079 | rf_signal_cond2(raidPtr->adding_hot_spare_cv); |
1080 | rf_unlock_mutex2(raidPtr->mutex); |
1081 | |
1082 | return(ret); |
1083 | } |
1084 | |
1085 | int |
1086 | rf_remove_hot_spare(RF_Raid_t *raidPtr, RF_SingleComponent_t *sparePtr) |
1087 | { |
1088 | #if 0 |
1089 | int spare_number; |
1090 | #endif |
1091 | |
1092 | if (raidPtr->numSpare==0) { |
1093 | printf("No spares to remove!\n" ); |
1094 | return(EINVAL); |
1095 | } |
1096 | |
1097 | return(EINVAL); /* XXX not implemented yet */ |
1098 | #if 0 |
1099 | spare_number = sparePtr->column; |
1100 | |
1101 | if (spare_number < 0 || spare_number > raidPtr->numSpare) { |
1102 | return(EINVAL); |
1103 | } |
1104 | |
1105 | /* verify that this spare isn't in use... */ |
1106 | |
1107 | |
1108 | |
1109 | |
1110 | /* it's gone.. */ |
1111 | |
1112 | raidPtr->numSpare--; |
1113 | |
1114 | return(0); |
1115 | #endif |
1116 | } |
1117 | |
1118 | |
1119 | int |
1120 | rf_delete_component(RF_Raid_t *raidPtr, RF_SingleComponent_t *component) |
1121 | { |
1122 | #if 0 |
1123 | RF_RaidDisk_t *disks; |
1124 | #endif |
1125 | |
1126 | if ((component->column < 0) || |
1127 | (component->column >= raidPtr->numCol)) { |
1128 | return(EINVAL); |
1129 | } |
1130 | |
1131 | #if 0 |
1132 | disks = &raidPtr->Disks[component->column]; |
1133 | #endif |
1134 | |
1135 | /* 1. This component must be marked as 'failed' */ |
1136 | |
1137 | return(EINVAL); /* Not implemented yet. */ |
1138 | } |
1139 | |
1140 | int |
1141 | rf_incorporate_hot_spare(RF_Raid_t *raidPtr, |
1142 | RF_SingleComponent_t *component) |
1143 | { |
1144 | |
1145 | /* Issues here include how to 'move' this in if there is IO |
1146 | taking place (e.g. component queues and such) */ |
1147 | |
1148 | return(EINVAL); /* Not implemented yet. */ |
1149 | } |
1150 | |