rf_paritymap.c revision 1.6 1 /* $NetBSD: rf_paritymap.c,v 1.6 2011/03/01 22:51:14 riz Exp $ */
2
3 /*-
4 * Copyright (c) 2009 Jed Davis.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 * POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: rf_paritymap.c,v 1.6 2011/03/01 22:51:14 riz Exp $");
31
32 #include <sys/param.h>
33 #include <sys/callout.h>
34 #include <sys/kmem.h>
35 #include <sys/mutex.h>
36 #include <sys/rwlock.h>
37 #include <sys/systm.h>
38 #include <sys/types.h>
39
40 #include <dev/raidframe/rf_paritymap.h>
41 #include <dev/raidframe/rf_stripelocks.h>
42 #include <dev/raidframe/rf_layout.h>
43 #include <dev/raidframe/rf_raid.h>
44 #include <dev/raidframe/rf_parityscan.h>
45 #include <dev/raidframe/rf_kintf.h>
46
47 /* Important parameters: */
48 #define REGION_MINSIZE (25ULL << 20)
49 #define DFL_TICKMS 40000
50 #define DFL_COOLDOWN 8 /* 7-8 intervals of 40s = 5min +/- 20s */
51
52 /* Internal-use flag bits. */
53 #define TICKING 1
54 #define TICKED 2
55
56 /* Prototypes! */
57 static void rf_paritymap_write_locked(struct rf_paritymap *);
58 static void rf_paritymap_tick(void *);
59 static u_int rf_paritymap_nreg(RF_Raid_t *);
60
61 /* Extract the current status of the parity map. */
62 void
63 rf_paritymap_status(struct rf_paritymap *pm, struct rf_pmstat *ps)
64 {
65 memset(ps, 0, sizeof(*ps));
66 if (pm == NULL)
67 ps->enabled = 0;
68 else {
69 ps->enabled = 1;
70 ps->region_size = pm->region_size;
71 mutex_enter(&pm->lock);
72 memcpy(&ps->params, &pm->params, sizeof(ps->params));
73 memcpy(ps->dirty, pm->disk_now, sizeof(ps->dirty));
74 memcpy(&ps->ctrs, &pm->ctrs, sizeof(ps->ctrs));
75 mutex_exit(&pm->lock);
76 }
77 }
78
79 /*
80 * Test whether parity in a given sector is suspected of being inconsistent
81 * on disk (assuming that any pending I/O to it is allowed to complete).
82 * This may be of interest to future work on parity scrubbing.
83 */
84 int
85 rf_paritymap_test(struct rf_paritymap *pm, daddr_t sector)
86 {
87 unsigned region = sector / pm->region_size;
88 int retval;
89
90 mutex_enter(&pm->lock);
91 retval = isset(pm->disk_boot->bits, region) ? 1 : 0;
92 mutex_exit(&pm->lock);
93 return retval;
94 }
95
96 /* To be called before a write to the RAID is submitted. */
97 void
98 rf_paritymap_begin(struct rf_paritymap *pm, daddr_t offset, daddr_t size)
99 {
100 unsigned i, b, e;
101
102 b = offset / pm->region_size;
103 e = (offset + size - 1) / pm->region_size;
104
105 for (i = b; i <= e; i++)
106 rf_paritymap_begin_region(pm, i);
107 }
108
109 /* To be called after a write to the RAID completes. */
110 void
111 rf_paritymap_end(struct rf_paritymap *pm, daddr_t offset, daddr_t size)
112 {
113 unsigned i, b, e;
114
115 b = offset / pm->region_size;
116 e = (offset + size - 1) / pm->region_size;
117
118 for (i = b; i <= e; i++)
119 rf_paritymap_end_region(pm, i);
120 }
121
122 void
123 rf_paritymap_begin_region(struct rf_paritymap *pm, unsigned region)
124 {
125 int needs_write;
126
127 KASSERT(region < RF_PARITYMAP_NREG);
128 pm->ctrs.nwrite++;
129
130 /* If it was being kept warm, deal with that. */
131 mutex_enter(&pm->lock);
132 if (pm->current->state[region] < 0)
133 pm->current->state[region] = 0;
134
135 /* This shouldn't happen unless RAIDOUTSTANDING is set too high. */
136 KASSERT(pm->current->state[region] < 127);
137 pm->current->state[region]++;
138
139 needs_write = isclr(pm->disk_now->bits, region);
140
141 if (needs_write) {
142 KASSERT(pm->current->state[region] == 1);
143 rf_paritymap_write_locked(pm);
144 }
145
146 mutex_exit(&pm->lock);
147 }
148
149 void
150 rf_paritymap_end_region(struct rf_paritymap *pm, unsigned region)
151 {
152 KASSERT(region < RF_PARITYMAP_NREG);
153
154 mutex_enter(&pm->lock);
155 KASSERT(pm->current->state[region] > 0);
156 --pm->current->state[region];
157
158 if (pm->current->state[region] <= 0) {
159 pm->current->state[region] = -pm->params.cooldown;
160 KASSERT(pm->current->state[region] <= 0);
161 mutex_enter(&pm->lk_flags);
162 if (!(pm->flags & TICKING)) {
163 pm->flags |= TICKING;
164 mutex_exit(&pm->lk_flags);
165 callout_schedule(&pm->ticker,
166 mstohz(pm->params.tickms));
167 } else
168 mutex_exit(&pm->lk_flags);
169 }
170 mutex_exit(&pm->lock);
171 }
172
173 /*
174 * Updates the parity map to account for any changes in current activity
175 * and/or an ongoing parity scan, then writes it to disk with appropriate
176 * synchronization.
177 */
178 void
179 rf_paritymap_write(struct rf_paritymap *pm)
180 {
181 mutex_enter(&pm->lock);
182 rf_paritymap_write_locked(pm);
183 mutex_exit(&pm->lock);
184 }
185
186 /* As above, but to be used when pm->lock is already held. */
187 static void
188 rf_paritymap_write_locked(struct rf_paritymap *pm)
189 {
190 char w, w0;
191 int i, j, setting, clearing;
192
193 setting = clearing = 0;
194 for (i = 0; i < RF_PARITYMAP_NBYTE; i++) {
195 w0 = pm->disk_now->bits[i];
196 w = pm->disk_boot->bits[i];
197
198 for (j = 0; j < NBBY; j++)
199 if (pm->current->state[i * NBBY + j] != 0)
200 w |= 1 << j;
201
202 if (w & ~w0)
203 setting = 1;
204 if (w0 & ~w)
205 clearing = 1;
206
207 pm->disk_now->bits[i] = w;
208 }
209 pm->ctrs.ncachesync += setting + clearing;
210 pm->ctrs.nclearing += clearing;
211
212 /*
213 * If bits are being set in the parity map, then a sync is
214 * required afterwards, so that the regions are marked dirty
215 * on disk before any writes to them take place. If bits are
216 * being cleared, then a sync is required before the write, so
217 * that any writes to those regions are processed before the
218 * region is marked clean. (Synchronization is somewhat
219 * overkill; a write ordering barrier would suffice, but we
220 * currently have no way to express that directly.)
221 */
222 if (clearing)
223 rf_sync_component_caches(pm->raid);
224 rf_paritymap_kern_write(pm->raid, pm->disk_now);
225 if (setting)
226 rf_sync_component_caches(pm->raid);
227 }
228
229 /* Mark all parity as being in need of rewrite. */
230 void
231 rf_paritymap_invalidate(struct rf_paritymap *pm)
232 {
233 mutex_enter(&pm->lock);
234 memset(pm->disk_boot, ~(unsigned char)0,
235 sizeof(struct rf_paritymap_ondisk));
236 mutex_exit(&pm->lock);
237 }
238
239 /* Mark all parity as being correct. */
240 void
241 rf_paritymap_forceclean(struct rf_paritymap *pm)
242 {
243 mutex_enter(&pm->lock);
244 memset(pm->disk_boot, (unsigned char)0,
245 sizeof(struct rf_paritymap_ondisk));
246 mutex_exit(&pm->lock);
247 }
248
249 /*
250 * The cooldown callout routine just defers its work to a thread; it can't do
251 * the parity map write itself as it would block, and although mutex-induced
252 * blocking is permitted it seems wise to avoid tying up the softint.
253 */
254 static void
255 rf_paritymap_tick(void *arg)
256 {
257 struct rf_paritymap *pm = arg;
258
259 mutex_enter(&pm->lk_flags);
260 pm->flags |= TICKED;
261 mutex_exit(&pm->lk_flags);
262 wakeup(&(pm->raid->iodone)); /* XXX */
263 }
264
265 /*
266 * This is where the parity cooling work (and rearming the callout if needed)
267 * is done; the raidio thread calls it when woken up, as by the above.
268 */
269 void
270 rf_paritymap_checkwork(struct rf_paritymap *pm)
271 {
272 int i, zerop, progressp;
273
274 mutex_enter(&pm->lk_flags);
275 if (pm->flags & TICKED) {
276 zerop = progressp = 0;
277
278 pm->flags &= ~TICKED;
279 mutex_exit(&pm->lk_flags);
280
281 mutex_enter(&pm->lock);
282 for (i = 0; i < RF_PARITYMAP_NREG; i++) {
283 if (pm->current->state[i] < 0) {
284 progressp = 1;
285 pm->current->state[i]++;
286 if (pm->current->state[i] == 0)
287 zerop = 1;
288 }
289 }
290
291 if (progressp)
292 callout_schedule(&pm->ticker,
293 mstohz(pm->params.tickms));
294 else {
295 mutex_enter(&pm->lk_flags);
296 pm->flags &= ~TICKING;
297 mutex_exit(&pm->lk_flags);
298 }
299
300 if (zerop)
301 rf_paritymap_write_locked(pm);
302 mutex_exit(&pm->lock);
303 } else
304 mutex_exit(&pm->lk_flags);
305 }
306
307 /*
308 * Set parity map parameters; used both to alter parameters on the fly and to
309 * establish their initial values. Note that setting a parameter to 0 means
310 * to leave the previous setting unchanged, and that if this is done for the
311 * initial setting of "regions", then a default value will be computed based
312 * on the RAID component size.
313 */
314 int
315 rf_paritymap_set_params(struct rf_paritymap *pm,
316 const struct rf_pmparams *params, int todisk)
317 {
318 int cooldown, tickms;
319 u_int regions;
320 RF_RowCol_t col;
321 RF_ComponentLabel_t *clabel;
322 RF_Raid_t *raidPtr;
323
324 cooldown = params->cooldown != 0
325 ? params->cooldown : pm->params.cooldown;
326 tickms = params->tickms != 0
327 ? params->tickms : pm->params.tickms;
328 regions = params->regions != 0
329 ? params->regions : pm->params.regions;
330
331 if (cooldown < 1 || cooldown > 128) {
332 printf("raid%d: cooldown %d out of range\n", pm->raid->raidid,
333 cooldown);
334 return (-1);
335 }
336 if (tickms < 10) {
337 printf("raid%d: tick time %dms out of range\n",
338 pm->raid->raidid, tickms);
339 return (-1);
340 }
341 if (regions == 0) {
342 regions = rf_paritymap_nreg(pm->raid);
343 } else if (regions > RF_PARITYMAP_NREG) {
344 printf("raid%d: region count %u too large (more than %u)\n",
345 pm->raid->raidid, regions, RF_PARITYMAP_NREG);
346 return (-1);
347 }
348
349 /* XXX any currently warm parity will be used with the new tickms! */
350 pm->params.cooldown = cooldown;
351 pm->params.tickms = tickms;
352 /* Apply the initial region count, but do not change it after that. */
353 if (pm->params.regions == 0)
354 pm->params.regions = regions;
355
356 /* So that the newly set parameters can be tested: */
357 pm->ctrs.nwrite = pm->ctrs.ncachesync = pm->ctrs.nclearing = 0;
358
359 if (todisk) {
360 raidPtr = pm->raid;
361 for (col = 0; col < raidPtr->numCol; col++) {
362 if (RF_DEAD_DISK(raidPtr->Disks[col].status))
363 continue;
364
365 clabel = raidget_component_label(raidPtr, col);
366 clabel->parity_map_ntick = cooldown;
367 clabel->parity_map_tickms = tickms;
368 clabel->parity_map_regions = regions;
369
370 /* Don't touch the disk if it's been spared */
371 if (clabel->status == rf_ds_spared)
372 continue;
373
374 raidflush_component_label(raidPtr, col);
375 }
376
377 /* handle the spares too... */
378 for (col = 0; col < raidPtr->numSpare; col++) {
379 if (raidPtr->Disks[raidPtr->numCol+col].status == rf_ds_used_spare) {
380 clabel = raidget_component_label(raidPtr, raidPtr->numCol+col);
381 clabel->parity_map_ntick = cooldown;
382 clabel->parity_map_tickms = tickms;
383 clabel->parity_map_regions = regions;
384 raidflush_component_label(raidPtr, raidPtr->numCol+col);
385 }
386 }
387 }
388 return 0;
389 }
390
391 /*
392 * The number of regions may not be as many as can fit into the map, because
393 * when regions are too small, the overhead of setting parity map bits
394 * becomes significant in comparison to the actual I/O, while the
395 * corresponding gains in parity verification time become negligible. Thus,
396 * a minimum region size (defined above) is imposed.
397 *
398 * Note that, if the number of regions is less than the maximum, then some of
399 * the regions will be "fictional", corresponding to no actual disk; some
400 * parts of the code may process them as normal, but they can not ever be
401 * written to.
402 */
403 static u_int
404 rf_paritymap_nreg(RF_Raid_t *raid)
405 {
406 daddr_t bytes_per_disk, nreg;
407
408 bytes_per_disk = raid->sectorsPerDisk << raid->logBytesPerSector;
409 nreg = bytes_per_disk / REGION_MINSIZE;
410 if (nreg > RF_PARITYMAP_NREG)
411 nreg = RF_PARITYMAP_NREG;
412 if (nreg < 1)
413 nreg = 1;
414
415 return (u_int)nreg;
416 }
417
418 /*
419 * Initialize a parity map given specific parameters. This neither reads nor
420 * writes the parity map config in the component labels; for that, see below.
421 */
422 int
423 rf_paritymap_init(struct rf_paritymap *pm, RF_Raid_t *raid,
424 const struct rf_pmparams *params)
425 {
426 daddr_t rstripes;
427 struct rf_pmparams safe;
428
429 pm->raid = raid;
430 pm->params.regions = 0;
431 if (0 != rf_paritymap_set_params(pm, params, 0)) {
432 /*
433 * If the parameters are out-of-range, then bring the
434 * parity map up with something reasonable, so that
435 * the admin can at least go and fix it (or ignore it
436 * entirely).
437 */
438 safe.cooldown = DFL_COOLDOWN;
439 safe.tickms = DFL_TICKMS;
440 safe.regions = 0;
441
442 if (0 != rf_paritymap_set_params(pm, &safe, 0))
443 return (-1);
444 }
445
446 rstripes = howmany(raid->Layout.numStripe, pm->params.regions);
447 pm->region_size = rstripes * raid->Layout.dataSectorsPerStripe;
448
449 callout_init(&pm->ticker, CALLOUT_MPSAFE);
450 callout_setfunc(&pm->ticker, rf_paritymap_tick, pm);
451 pm->flags = 0;
452
453 pm->disk_boot = kmem_alloc(sizeof(struct rf_paritymap_ondisk),
454 KM_SLEEP);
455 pm->disk_now = kmem_alloc(sizeof(struct rf_paritymap_ondisk),
456 KM_SLEEP);
457 pm->current = kmem_zalloc(sizeof(struct rf_paritymap_current),
458 KM_SLEEP);
459
460 rf_paritymap_kern_read(pm->raid, pm->disk_boot);
461 memcpy(pm->disk_now, pm->disk_boot, sizeof(*pm->disk_now));
462
463 mutex_init(&pm->lock, MUTEX_DEFAULT, IPL_NONE);
464 mutex_init(&pm->lk_flags, MUTEX_DEFAULT, IPL_SOFTCLOCK);
465
466 return 0;
467 }
468
469 /*
470 * Destroys a parity map; unless "force" is set, also cleans parity for any
471 * regions which were still in cooldown (but are not dirty on disk).
472 */
473 void
474 rf_paritymap_destroy(struct rf_paritymap *pm, int force)
475 {
476 int i;
477
478 callout_halt(&pm->ticker, NULL); /* XXX stop? halt? */
479 callout_destroy(&pm->ticker);
480
481 if (!force) {
482 for (i = 0; i < RF_PARITYMAP_NREG; i++) {
483 /* XXX check for > 0 ? */
484 if (pm->current->state[i] < 0)
485 pm->current->state[i] = 0;
486 }
487
488 rf_paritymap_write_locked(pm);
489 }
490
491 mutex_destroy(&pm->lock);
492 mutex_destroy(&pm->lk_flags);
493
494 kmem_free(pm->disk_boot, sizeof(struct rf_paritymap_ondisk));
495 kmem_free(pm->disk_now, sizeof(struct rf_paritymap_ondisk));
496 kmem_free(pm->current, sizeof(struct rf_paritymap_current));
497 }
498
499 /*
500 * Rewrite parity, taking parity map into account; this is the equivalent of
501 * the old rf_RewriteParity, and is likewise to be called from a suitable
502 * thread and shouldn't have multiple copies running in parallel and so on.
503 *
504 * Note that the fictional regions are "cleaned" in one shot, so that very
505 * small RAIDs (useful for testing) will not experience potentially severe
506 * regressions in rewrite time.
507 */
508 int
509 rf_paritymap_rewrite(struct rf_paritymap *pm)
510 {
511 int i, ret_val = 0;
512 daddr_t reg_b, reg_e;
513
514 /* Process only the actual regions. */
515 for (i = 0; i < pm->params.regions; i++) {
516 mutex_enter(&pm->lock);
517 if (isset(pm->disk_boot->bits, i)) {
518 mutex_exit(&pm->lock);
519
520 reg_b = i * pm->region_size;
521 reg_e = reg_b + pm->region_size;
522 if (reg_e > pm->raid->totalSectors)
523 reg_e = pm->raid->totalSectors;
524
525 if (rf_RewriteParityRange(pm->raid, reg_b,
526 reg_e - reg_b)) {
527 ret_val = 1;
528 if (pm->raid->waitShutdown)
529 return ret_val;
530 } else {
531 mutex_enter(&pm->lock);
532 clrbit(pm->disk_boot->bits, i);
533 rf_paritymap_write_locked(pm);
534 mutex_exit(&pm->lock);
535 }
536 } else {
537 mutex_exit(&pm->lock);
538 }
539 }
540
541 /* Now, clear the fictional regions, if any. */
542 rf_paritymap_forceclean(pm);
543 rf_paritymap_write(pm);
544
545 return ret_val;
546 }
547
548 /*
549 * How to merge the on-disk parity maps when reading them in from the
550 * various components; returns whether they differ. In the case that
551 * they do differ, sets *dst to the union of *dst and *src.
552 *
553 * In theory, it should be safe to take the intersection (or just pick
554 * a single component arbitrarily), but the paranoid approach costs
555 * little.
556 *
557 * Appropriate locking, if any, is the responsibility of the caller.
558 */
559 int
560 rf_paritymap_merge(struct rf_paritymap_ondisk *dst,
561 struct rf_paritymap_ondisk *src)
562 {
563 int i, discrep = 0;
564
565 for (i = 0; i < RF_PARITYMAP_NBYTE; i++) {
566 if (dst->bits[i] != src->bits[i])
567 discrep = 1;
568 dst->bits[i] |= src->bits[i];
569 }
570
571 return discrep;
572 }
573
574 /*
575 * Detach a parity map from its RAID. This is not meant to be applied except
576 * when unconfiguring the RAID after all I/O has been resolved, as otherwise
577 * an out-of-date parity map could be treated as current.
578 */
579 void
580 rf_paritymap_detach(RF_Raid_t *raidPtr)
581 {
582 if (raidPtr->parity_map == NULL)
583 return;
584
585 simple_lock(&(raidPtr->iodone_lock));
586 struct rf_paritymap *pm = raidPtr->parity_map;
587 raidPtr->parity_map = NULL;
588 simple_unlock(&(raidPtr->iodone_lock));
589 /* XXXjld is that enough locking? Or too much? */
590 rf_paritymap_destroy(pm, 0);
591 kmem_free(pm, sizeof(*pm));
592 }
593
594 /*
595 * Is this RAID set ineligible for parity-map use due to not actually
596 * having any parity? (If so, rf_paritymap_attach is a no-op, but
597 * rf_paritymap_{get,set}_disable will still pointlessly act on the
598 * component labels.)
599 */
600 int
601 rf_paritymap_ineligible(RF_Raid_t *raidPtr)
602 {
603 return raidPtr->Layout.map->faultsTolerated == 0;
604 }
605
606 /*
607 * Attach a parity map to a RAID set if appropriate. Includes
608 * configure-time processing of parity-map fields of component label.
609 */
610 void
611 rf_paritymap_attach(RF_Raid_t *raidPtr, int force)
612 {
613 RF_RowCol_t col;
614 int pm_use, pm_zap;
615 int g_tickms, g_ntick, g_regions;
616 int good;
617 RF_ComponentLabel_t *clabel;
618 u_int flags, regions;
619 struct rf_pmparams params;
620
621 if (rf_paritymap_ineligible(raidPtr)) {
622 /* There isn't any parity. */
623 return;
624 }
625
626 pm_use = 1;
627 pm_zap = 0;
628 g_tickms = DFL_TICKMS;
629 g_ntick = DFL_COOLDOWN;
630 g_regions = 0;
631
632 /*
633 * Collect opinions on the set config. If this is the initial
634 * config (raidctl -C), treat all labels as invalid, since
635 * there may be random data present.
636 */
637 if (!force) {
638 for (col = 0; col < raidPtr->numCol; col++) {
639 if (RF_DEAD_DISK(raidPtr->Disks[col].status))
640 continue;
641 clabel = raidget_component_label(raidPtr, col);
642 flags = clabel->parity_map_flags;
643 /* Check for use by non-parity-map kernel. */
644 if (clabel->parity_map_modcount
645 != clabel->mod_counter) {
646 flags &= ~RF_PMLABEL_WASUSED;
647 }
648
649 if (flags & RF_PMLABEL_VALID) {
650 g_tickms = clabel->parity_map_tickms;
651 g_ntick = clabel->parity_map_ntick;
652 regions = clabel->parity_map_regions;
653 if (g_regions == 0)
654 g_regions = regions;
655 else if (g_regions != regions) {
656 pm_zap = 1; /* important! */
657 }
658
659 if (flags & RF_PMLABEL_DISABLE) {
660 pm_use = 0;
661 }
662 if (!(flags & RF_PMLABEL_WASUSED)) {
663 pm_zap = 1;
664 }
665 } else {
666 pm_zap = 1;
667 }
668 }
669 } else {
670 pm_zap = 1;
671 }
672
673 /* Finally, create and attach the parity map. */
674 if (pm_use) {
675 params.cooldown = g_ntick;
676 params.tickms = g_tickms;
677 params.regions = g_regions;
678
679 raidPtr->parity_map = kmem_alloc(sizeof(struct rf_paritymap),
680 KM_SLEEP);
681 if (0 != rf_paritymap_init(raidPtr->parity_map, raidPtr,
682 ¶ms)) {
683 /* It failed; do without. */
684 kmem_free(raidPtr->parity_map,
685 sizeof(struct rf_paritymap));
686 raidPtr->parity_map = NULL;
687 return;
688 }
689
690 if (g_regions == 0)
691 /* Pick up the autoconfigured region count. */
692 g_regions = raidPtr->parity_map->params.regions;
693
694 if (pm_zap) {
695 good = raidPtr->parity_good && !force;
696
697 if (good)
698 rf_paritymap_forceclean(raidPtr->parity_map);
699 else
700 rf_paritymap_invalidate(raidPtr->parity_map);
701 /* This needs to be on disk before WASUSED is set. */
702 rf_paritymap_write(raidPtr->parity_map);
703 }
704 }
705
706 /* Alter labels in-core to reflect the current view of things. */
707 for (col = 0; col < raidPtr->numCol; col++) {
708 if (RF_DEAD_DISK(raidPtr->Disks[col].status))
709 continue;
710 clabel = raidget_component_label(raidPtr, col);
711
712 if (pm_use)
713 flags = RF_PMLABEL_VALID | RF_PMLABEL_WASUSED;
714 else
715 flags = RF_PMLABEL_VALID | RF_PMLABEL_DISABLE;
716
717 clabel->parity_map_flags = flags;
718 clabel->parity_map_tickms = g_tickms;
719 clabel->parity_map_ntick = g_ntick;
720 clabel->parity_map_regions = g_regions;
721 raidflush_component_label(raidPtr, col);
722 }
723 /* Note that we're just in 'attach' here, and there won't
724 be any spare disks at this point. */
725 }
726
727 /*
728 * For initializing the parity-map fields of a component label, both on
729 * initial creation and on reconstruct/copyback/etc. */
730 void
731 rf_paritymap_init_label(struct rf_paritymap *pm, RF_ComponentLabel_t *clabel)
732 {
733 if (pm != NULL) {
734 clabel->parity_map_flags =
735 RF_PMLABEL_VALID | RF_PMLABEL_WASUSED;
736 clabel->parity_map_tickms = pm->params.tickms;
737 clabel->parity_map_ntick = pm->params.cooldown;
738 /*
739 * XXXjld: If the number of regions is changed on disk, and
740 * then a new component is labeled before the next configure,
741 * then it will get the old value and they will conflict on
742 * the next boot (and the default will be used instead).
743 */
744 clabel->parity_map_regions = pm->params.regions;
745 } else {
746 /*
747 * XXXjld: if the map is disabled, and all the components are
748 * replaced without an intervening unconfigure/reconfigure,
749 * then it will become enabled on the next unconfig/reconfig.
750 */
751 }
752 }
753
754
755 /* Will the parity map be disabled next time? */
756 int
757 rf_paritymap_get_disable(RF_Raid_t *raidPtr)
758 {
759 RF_ComponentLabel_t *clabel;
760 RF_RowCol_t col;
761 int dis;
762
763 dis = 0;
764 for (col = 0; col < raidPtr->numCol; col++) {
765 if (RF_DEAD_DISK(raidPtr->Disks[col].status))
766 continue;
767 clabel = raidget_component_label(raidPtr, col);
768 if (clabel->parity_map_flags & RF_PMLABEL_DISABLE)
769 dis = 1;
770 }
771 for (col = 0; col < raidPtr->numSpare; col++) {
772 if (raidPtr->Disks[raidPtr->numCol+col].status != rf_ds_used_spare)
773 continue;
774 clabel = raidget_component_label(raidPtr, raidPtr->numCol+col);
775 if (clabel->parity_map_flags & RF_PMLABEL_DISABLE)
776 dis = 1;
777 }
778
779 return dis;
780 }
781
782 /* Set whether the parity map will be disabled next time. */
783 void
784 rf_paritymap_set_disable(RF_Raid_t *raidPtr, int dis)
785 {
786 RF_ComponentLabel_t *clabel;
787 RF_RowCol_t col;
788
789 for (col = 0; col < raidPtr->numCol; col++) {
790 if (RF_DEAD_DISK(raidPtr->Disks[col].status))
791 continue;
792 clabel = raidget_component_label(raidPtr, col);
793 if (dis)
794 clabel->parity_map_flags |= RF_PMLABEL_DISABLE;
795 else
796 clabel->parity_map_flags &= ~RF_PMLABEL_DISABLE;
797 raidflush_component_label(raidPtr, col);
798 }
799
800 /* update any used spares as well */
801 for (col = 0; col < raidPtr->numSpare; col++) {
802 if (raidPtr->Disks[raidPtr->numCol+col].status != rf_ds_used_spare)
803 continue;
804
805 clabel = raidget_component_label(raidPtr, raidPtr->numCol+col);
806 if (dis)
807 clabel->parity_map_flags |= RF_PMLABEL_DISABLE;
808 else
809 clabel->parity_map_flags &= ~RF_PMLABEL_DISABLE;
810 raidflush_component_label(raidPtr, raidPtr->numCol+col);
811 }
812 }
813