]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
5 | * | |
6 | * RAID-5 management functions. | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License as published by | |
10 | * the Free Software Foundation; either version 2, or (at your option) | |
11 | * any later version. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License | |
14 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
15 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
16 | */ | |
17 | ||
18 | ||
19 | #include <linux/config.h> | |
20 | #include <linux/module.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/raid/raid5.h> | |
23 | #include <linux/highmem.h> | |
24 | #include <linux/bitops.h> | |
25 | #include <asm/atomic.h> | |
26 | ||
72626685 N |
27 | #include <linux/raid/bitmap.h> |
28 | ||
1da177e4 LT |
29 | /* |
30 | * Stripe cache | |
31 | */ | |
32 | ||
33 | #define NR_STRIPES 256 | |
34 | #define STRIPE_SIZE PAGE_SIZE | |
35 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
36 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
37 | #define IO_THRESHOLD 1 | |
fccddba0 | 38 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
39 | #define HASH_MASK (NR_HASH - 1) |
40 | ||
fccddba0 | 41 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
42 | |
43 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
44 | * order without overlap. There may be several bio's per stripe+device, and | |
45 | * a bio could span several devices. | |
46 | * When walking this list for a particular stripe+device, we must never proceed | |
47 | * beyond a bio that extends past this device, as the next bio might no longer | |
48 | * be valid. | |
49 | * This macro is used to determine the 'next' bio in the list, given the sector | |
50 | * of the current stripe+device | |
51 | */ | |
52 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
53 | /* | |
54 | * The following can be used to debug the driver | |
55 | */ | |
56 | #define RAID5_DEBUG 0 | |
57 | #define RAID5_PARANOIA 1 | |
58 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
59 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
60 | #else | |
61 | # define CHECK_DEVLOCK() | |
62 | #endif | |
63 | ||
64 | #define PRINTK(x...) ((void)(RAID5_DEBUG && printk(x))) | |
65 | #if RAID5_DEBUG | |
66 | #define inline | |
67 | #define __inline__ | |
68 | #endif | |
69 | ||
70 | static void print_raid5_conf (raid5_conf_t *conf); | |
71 | ||
858119e1 | 72 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
73 | { |
74 | if (atomic_dec_and_test(&sh->count)) { | |
75 | if (!list_empty(&sh->lru)) | |
76 | BUG(); | |
77 | if (atomic_read(&conf->active_stripes)==0) | |
78 | BUG(); | |
79 | if (test_bit(STRIPE_HANDLE, &sh->state)) { | |
80 | if (test_bit(STRIPE_DELAYED, &sh->state)) | |
81 | list_add_tail(&sh->lru, &conf->delayed_list); | |
72626685 N |
82 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
83 | conf->seq_write == sh->bm_seq) | |
84 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
85 | else { | |
86 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
1da177e4 | 87 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 88 | } |
1da177e4 LT |
89 | md_wakeup_thread(conf->mddev->thread); |
90 | } else { | |
91 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
92 | atomic_dec(&conf->preread_active_stripes); | |
93 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
94 | md_wakeup_thread(conf->mddev->thread); | |
95 | } | |
96 | list_add_tail(&sh->lru, &conf->inactive_list); | |
97 | atomic_dec(&conf->active_stripes); | |
98 | if (!conf->inactive_blocked || | |
5036805b | 99 | atomic_read(&conf->active_stripes) < (conf->max_nr_stripes*3/4)) |
1da177e4 LT |
100 | wake_up(&conf->wait_for_stripe); |
101 | } | |
102 | } | |
103 | } | |
104 | static void release_stripe(struct stripe_head *sh) | |
105 | { | |
106 | raid5_conf_t *conf = sh->raid_conf; | |
107 | unsigned long flags; | |
108 | ||
109 | spin_lock_irqsave(&conf->device_lock, flags); | |
110 | __release_stripe(conf, sh); | |
111 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
112 | } | |
113 | ||
fccddba0 | 114 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 LT |
115 | { |
116 | PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh->sector); | |
117 | ||
fccddba0 | 118 | hlist_del_init(&sh->hash); |
1da177e4 LT |
119 | } |
120 | ||
858119e1 | 121 | static void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 122 | { |
fccddba0 | 123 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 LT |
124 | |
125 | PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh->sector); | |
126 | ||
127 | CHECK_DEVLOCK(); | |
fccddba0 | 128 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
129 | } |
130 | ||
131 | ||
132 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
133 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
134 | { | |
135 | struct stripe_head *sh = NULL; | |
136 | struct list_head *first; | |
137 | ||
138 | CHECK_DEVLOCK(); | |
139 | if (list_empty(&conf->inactive_list)) | |
140 | goto out; | |
141 | first = conf->inactive_list.next; | |
142 | sh = list_entry(first, struct stripe_head, lru); | |
143 | list_del_init(first); | |
144 | remove_hash(sh); | |
145 | atomic_inc(&conf->active_stripes); | |
146 | out: | |
147 | return sh; | |
148 | } | |
149 | ||
150 | static void shrink_buffers(struct stripe_head *sh, int num) | |
151 | { | |
152 | struct page *p; | |
153 | int i; | |
154 | ||
155 | for (i=0; i<num ; i++) { | |
156 | p = sh->dev[i].page; | |
157 | if (!p) | |
158 | continue; | |
159 | sh->dev[i].page = NULL; | |
2d1f3b5d | 160 | put_page(p); |
1da177e4 LT |
161 | } |
162 | } | |
163 | ||
164 | static int grow_buffers(struct stripe_head *sh, int num) | |
165 | { | |
166 | int i; | |
167 | ||
168 | for (i=0; i<num; i++) { | |
169 | struct page *page; | |
170 | ||
171 | if (!(page = alloc_page(GFP_KERNEL))) { | |
172 | return 1; | |
173 | } | |
174 | sh->dev[i].page = page; | |
175 | } | |
176 | return 0; | |
177 | } | |
178 | ||
179 | static void raid5_build_block (struct stripe_head *sh, int i); | |
180 | ||
858119e1 | 181 | static void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx) |
1da177e4 LT |
182 | { |
183 | raid5_conf_t *conf = sh->raid_conf; | |
184 | int disks = conf->raid_disks, i; | |
185 | ||
186 | if (atomic_read(&sh->count) != 0) | |
187 | BUG(); | |
188 | if (test_bit(STRIPE_HANDLE, &sh->state)) | |
189 | BUG(); | |
190 | ||
191 | CHECK_DEVLOCK(); | |
192 | PRINTK("init_stripe called, stripe %llu\n", | |
193 | (unsigned long long)sh->sector); | |
194 | ||
195 | remove_hash(sh); | |
196 | ||
197 | sh->sector = sector; | |
198 | sh->pd_idx = pd_idx; | |
199 | sh->state = 0; | |
200 | ||
201 | for (i=disks; i--; ) { | |
202 | struct r5dev *dev = &sh->dev[i]; | |
203 | ||
204 | if (dev->toread || dev->towrite || dev->written || | |
205 | test_bit(R5_LOCKED, &dev->flags)) { | |
206 | printk("sector=%llx i=%d %p %p %p %d\n", | |
207 | (unsigned long long)sh->sector, i, dev->toread, | |
208 | dev->towrite, dev->written, | |
209 | test_bit(R5_LOCKED, &dev->flags)); | |
210 | BUG(); | |
211 | } | |
212 | dev->flags = 0; | |
213 | raid5_build_block(sh, i); | |
214 | } | |
215 | insert_hash(conf, sh); | |
216 | } | |
217 | ||
218 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector) | |
219 | { | |
220 | struct stripe_head *sh; | |
fccddba0 | 221 | struct hlist_node *hn; |
1da177e4 LT |
222 | |
223 | CHECK_DEVLOCK(); | |
224 | PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector); | |
fccddba0 | 225 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
1da177e4 LT |
226 | if (sh->sector == sector) |
227 | return sh; | |
228 | PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector); | |
229 | return NULL; | |
230 | } | |
231 | ||
232 | static void unplug_slaves(mddev_t *mddev); | |
233 | static void raid5_unplug_device(request_queue_t *q); | |
234 | ||
235 | static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, | |
236 | int pd_idx, int noblock) | |
237 | { | |
238 | struct stripe_head *sh; | |
239 | ||
240 | PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector); | |
241 | ||
242 | spin_lock_irq(&conf->device_lock); | |
243 | ||
244 | do { | |
72626685 N |
245 | wait_event_lock_irq(conf->wait_for_stripe, |
246 | conf->quiesce == 0, | |
247 | conf->device_lock, /* nothing */); | |
1da177e4 LT |
248 | sh = __find_stripe(conf, sector); |
249 | if (!sh) { | |
250 | if (!conf->inactive_blocked) | |
251 | sh = get_free_stripe(conf); | |
252 | if (noblock && sh == NULL) | |
253 | break; | |
254 | if (!sh) { | |
255 | conf->inactive_blocked = 1; | |
256 | wait_event_lock_irq(conf->wait_for_stripe, | |
257 | !list_empty(&conf->inactive_list) && | |
5036805b N |
258 | (atomic_read(&conf->active_stripes) |
259 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
260 | || !conf->inactive_blocked), |
261 | conf->device_lock, | |
262 | unplug_slaves(conf->mddev); | |
263 | ); | |
264 | conf->inactive_blocked = 0; | |
265 | } else | |
266 | init_stripe(sh, sector, pd_idx); | |
267 | } else { | |
268 | if (atomic_read(&sh->count)) { | |
269 | if (!list_empty(&sh->lru)) | |
270 | BUG(); | |
271 | } else { | |
272 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
273 | atomic_inc(&conf->active_stripes); | |
274 | if (list_empty(&sh->lru)) | |
275 | BUG(); | |
276 | list_del_init(&sh->lru); | |
277 | } | |
278 | } | |
279 | } while (sh == NULL); | |
280 | ||
281 | if (sh) | |
282 | atomic_inc(&sh->count); | |
283 | ||
284 | spin_unlock_irq(&conf->device_lock); | |
285 | return sh; | |
286 | } | |
287 | ||
3f294f4f | 288 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
289 | { |
290 | struct stripe_head *sh; | |
3f294f4f N |
291 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
292 | if (!sh) | |
293 | return 0; | |
294 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
295 | sh->raid_conf = conf; | |
296 | spin_lock_init(&sh->lock); | |
297 | ||
298 | if (grow_buffers(sh, conf->raid_disks)) { | |
299 | shrink_buffers(sh, conf->raid_disks); | |
300 | kmem_cache_free(conf->slab_cache, sh); | |
301 | return 0; | |
302 | } | |
303 | /* we just created an active stripe so... */ | |
304 | atomic_set(&sh->count, 1); | |
305 | atomic_inc(&conf->active_stripes); | |
306 | INIT_LIST_HEAD(&sh->lru); | |
307 | release_stripe(sh); | |
308 | return 1; | |
309 | } | |
310 | ||
311 | static int grow_stripes(raid5_conf_t *conf, int num) | |
312 | { | |
1da177e4 LT |
313 | kmem_cache_t *sc; |
314 | int devs = conf->raid_disks; | |
315 | ||
ad01c9e3 N |
316 | sprintf(conf->cache_name[0], "raid5/%s", mdname(conf->mddev)); |
317 | sprintf(conf->cache_name[1], "raid5/%s-alt", mdname(conf->mddev)); | |
318 | conf->active_name = 0; | |
319 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 LT |
320 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
321 | 0, 0, NULL, NULL); | |
322 | if (!sc) | |
323 | return 1; | |
324 | conf->slab_cache = sc; | |
ad01c9e3 | 325 | conf->pool_size = devs; |
1da177e4 | 326 | while (num--) { |
3f294f4f | 327 | if (!grow_one_stripe(conf)) |
1da177e4 | 328 | return 1; |
1da177e4 LT |
329 | } |
330 | return 0; | |
331 | } | |
ad01c9e3 N |
332 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
333 | { | |
334 | /* Make all the stripes able to hold 'newsize' devices. | |
335 | * New slots in each stripe get 'page' set to a new page. | |
336 | * | |
337 | * This happens in stages: | |
338 | * 1/ create a new kmem_cache and allocate the required number of | |
339 | * stripe_heads. | |
340 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
341 | * to the new stripe_heads. This will have the side effect of | |
342 | * freezing the array as once all stripe_heads have been collected, | |
343 | * no IO will be possible. Old stripe heads are freed once their | |
344 | * pages have been transferred over, and the old kmem_cache is | |
345 | * freed when all stripes are done. | |
346 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
347 | * we simple return a failre status - no need to clean anything up. | |
348 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
349 | * If this fails, we don't bother trying the shrink the | |
350 | * stripe_heads down again, we just leave them as they are. | |
351 | * As each stripe_head is processed the new one is released into | |
352 | * active service. | |
353 | * | |
354 | * Once step2 is started, we cannot afford to wait for a write, | |
355 | * so we use GFP_NOIO allocations. | |
356 | */ | |
357 | struct stripe_head *osh, *nsh; | |
358 | LIST_HEAD(newstripes); | |
359 | struct disk_info *ndisks; | |
360 | int err = 0; | |
361 | kmem_cache_t *sc; | |
362 | int i; | |
363 | ||
364 | if (newsize <= conf->pool_size) | |
365 | return 0; /* never bother to shrink */ | |
366 | ||
367 | /* Step 1 */ | |
368 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
369 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
370 | 0, 0, NULL, NULL); | |
371 | if (!sc) | |
372 | return -ENOMEM; | |
373 | ||
374 | for (i = conf->max_nr_stripes; i; i--) { | |
375 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
376 | if (!nsh) | |
377 | break; | |
378 | ||
379 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
380 | ||
381 | nsh->raid_conf = conf; | |
382 | spin_lock_init(&nsh->lock); | |
383 | ||
384 | list_add(&nsh->lru, &newstripes); | |
385 | } | |
386 | if (i) { | |
387 | /* didn't get enough, give up */ | |
388 | while (!list_empty(&newstripes)) { | |
389 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
390 | list_del(&nsh->lru); | |
391 | kmem_cache_free(sc, nsh); | |
392 | } | |
393 | kmem_cache_destroy(sc); | |
394 | return -ENOMEM; | |
395 | } | |
396 | /* Step 2 - Must use GFP_NOIO now. | |
397 | * OK, we have enough stripes, start collecting inactive | |
398 | * stripes and copying them over | |
399 | */ | |
400 | list_for_each_entry(nsh, &newstripes, lru) { | |
401 | spin_lock_irq(&conf->device_lock); | |
402 | wait_event_lock_irq(conf->wait_for_stripe, | |
403 | !list_empty(&conf->inactive_list), | |
404 | conf->device_lock, | |
405 | unplug_slaves(conf->mddev); | |
406 | ); | |
407 | osh = get_free_stripe(conf); | |
408 | spin_unlock_irq(&conf->device_lock); | |
409 | atomic_set(&nsh->count, 1); | |
410 | for(i=0; i<conf->pool_size; i++) | |
411 | nsh->dev[i].page = osh->dev[i].page; | |
412 | for( ; i<newsize; i++) | |
413 | nsh->dev[i].page = NULL; | |
414 | kmem_cache_free(conf->slab_cache, osh); | |
415 | } | |
416 | kmem_cache_destroy(conf->slab_cache); | |
417 | ||
418 | /* Step 3. | |
419 | * At this point, we are holding all the stripes so the array | |
420 | * is completely stalled, so now is a good time to resize | |
421 | * conf->disks. | |
422 | */ | |
423 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
424 | if (ndisks) { | |
425 | for (i=0; i<conf->raid_disks; i++) | |
426 | ndisks[i] = conf->disks[i]; | |
427 | kfree(conf->disks); | |
428 | conf->disks = ndisks; | |
429 | } else | |
430 | err = -ENOMEM; | |
431 | ||
432 | /* Step 4, return new stripes to service */ | |
433 | while(!list_empty(&newstripes)) { | |
434 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
435 | list_del_init(&nsh->lru); | |
436 | for (i=conf->raid_disks; i < newsize; i++) | |
437 | if (nsh->dev[i].page == NULL) { | |
438 | struct page *p = alloc_page(GFP_NOIO); | |
439 | nsh->dev[i].page = p; | |
440 | if (!p) | |
441 | err = -ENOMEM; | |
442 | } | |
443 | release_stripe(nsh); | |
444 | } | |
445 | /* critical section pass, GFP_NOIO no longer needed */ | |
446 | ||
447 | conf->slab_cache = sc; | |
448 | conf->active_name = 1-conf->active_name; | |
449 | conf->pool_size = newsize; | |
450 | return err; | |
451 | } | |
452 | ||
1da177e4 | 453 | |
3f294f4f | 454 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
455 | { |
456 | struct stripe_head *sh; | |
457 | ||
3f294f4f N |
458 | spin_lock_irq(&conf->device_lock); |
459 | sh = get_free_stripe(conf); | |
460 | spin_unlock_irq(&conf->device_lock); | |
461 | if (!sh) | |
462 | return 0; | |
463 | if (atomic_read(&sh->count)) | |
464 | BUG(); | |
ad01c9e3 | 465 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
466 | kmem_cache_free(conf->slab_cache, sh); |
467 | atomic_dec(&conf->active_stripes); | |
468 | return 1; | |
469 | } | |
470 | ||
471 | static void shrink_stripes(raid5_conf_t *conf) | |
472 | { | |
473 | while (drop_one_stripe(conf)) | |
474 | ; | |
475 | ||
29fc7e3e N |
476 | if (conf->slab_cache) |
477 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
478 | conf->slab_cache = NULL; |
479 | } | |
480 | ||
4e5314b5 | 481 | static int raid5_end_read_request(struct bio * bi, unsigned int bytes_done, |
1da177e4 LT |
482 | int error) |
483 | { | |
484 | struct stripe_head *sh = bi->bi_private; | |
485 | raid5_conf_t *conf = sh->raid_conf; | |
486 | int disks = conf->raid_disks, i; | |
487 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
488 | ||
489 | if (bi->bi_size) | |
490 | return 1; | |
491 | ||
492 | for (i=0 ; i<disks; i++) | |
493 | if (bi == &sh->dev[i].req) | |
494 | break; | |
495 | ||
496 | PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n", | |
497 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
498 | uptodate); | |
499 | if (i == disks) { | |
500 | BUG(); | |
501 | return 0; | |
502 | } | |
503 | ||
504 | if (uptodate) { | |
505 | #if 0 | |
506 | struct bio *bio; | |
507 | unsigned long flags; | |
508 | spin_lock_irqsave(&conf->device_lock, flags); | |
509 | /* we can return a buffer if we bypassed the cache or | |
510 | * if the top buffer is not in highmem. If there are | |
511 | * multiple buffers, leave the extra work to | |
512 | * handle_stripe | |
513 | */ | |
514 | buffer = sh->bh_read[i]; | |
515 | if (buffer && | |
516 | (!PageHighMem(buffer->b_page) | |
517 | || buffer->b_page == bh->b_page ) | |
518 | ) { | |
519 | sh->bh_read[i] = buffer->b_reqnext; | |
520 | buffer->b_reqnext = NULL; | |
521 | } else | |
522 | buffer = NULL; | |
523 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
524 | if (sh->bh_page[i]==bh->b_page) | |
525 | set_buffer_uptodate(bh); | |
526 | if (buffer) { | |
527 | if (buffer->b_page != bh->b_page) | |
528 | memcpy(buffer->b_data, bh->b_data, bh->b_size); | |
529 | buffer->b_end_io(buffer, 1); | |
530 | } | |
531 | #else | |
532 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
4e5314b5 N |
533 | #endif |
534 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
14f8d26b | 535 | printk(KERN_INFO "raid5: read error corrected!!\n"); |
4e5314b5 N |
536 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
537 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
538 | } | |
ba22dcbf N |
539 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
540 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 541 | } else { |
ba22dcbf | 542 | int retry = 0; |
1da177e4 | 543 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
ba22dcbf N |
544 | atomic_inc(&conf->disks[i].rdev->read_errors); |
545 | if (conf->mddev->degraded) | |
14f8d26b | 546 | printk(KERN_WARNING "raid5: read error not correctable.\n"); |
ba22dcbf | 547 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 548 | /* Oh, no!!! */ |
14f8d26b | 549 | printk(KERN_WARNING "raid5: read error NOT corrected!!\n"); |
ba22dcbf N |
550 | else if (atomic_read(&conf->disks[i].rdev->read_errors) |
551 | > conf->max_nr_stripes) | |
14f8d26b N |
552 | printk(KERN_WARNING |
553 | "raid5: Too many read errors, failing device.\n"); | |
ba22dcbf N |
554 | else |
555 | retry = 1; | |
556 | if (retry) | |
557 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
558 | else { | |
4e5314b5 N |
559 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
560 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
561 | md_error(conf->mddev, conf->disks[i].rdev); | |
ba22dcbf | 562 | } |
1da177e4 LT |
563 | } |
564 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
565 | #if 0 | |
566 | /* must restore b_page before unlocking buffer... */ | |
567 | if (sh->bh_page[i] != bh->b_page) { | |
568 | bh->b_page = sh->bh_page[i]; | |
569 | bh->b_data = page_address(bh->b_page); | |
570 | clear_buffer_uptodate(bh); | |
571 | } | |
572 | #endif | |
573 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
574 | set_bit(STRIPE_HANDLE, &sh->state); | |
575 | release_stripe(sh); | |
576 | return 0; | |
577 | } | |
578 | ||
579 | static int raid5_end_write_request (struct bio *bi, unsigned int bytes_done, | |
580 | int error) | |
581 | { | |
582 | struct stripe_head *sh = bi->bi_private; | |
583 | raid5_conf_t *conf = sh->raid_conf; | |
584 | int disks = conf->raid_disks, i; | |
585 | unsigned long flags; | |
586 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
587 | ||
588 | if (bi->bi_size) | |
589 | return 1; | |
590 | ||
591 | for (i=0 ; i<disks; i++) | |
592 | if (bi == &sh->dev[i].req) | |
593 | break; | |
594 | ||
595 | PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n", | |
596 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
597 | uptodate); | |
598 | if (i == disks) { | |
599 | BUG(); | |
600 | return 0; | |
601 | } | |
602 | ||
603 | spin_lock_irqsave(&conf->device_lock, flags); | |
604 | if (!uptodate) | |
605 | md_error(conf->mddev, conf->disks[i].rdev); | |
606 | ||
607 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
608 | ||
609 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
610 | set_bit(STRIPE_HANDLE, &sh->state); | |
611 | __release_stripe(conf, sh); | |
612 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
613 | return 0; | |
614 | } | |
615 | ||
616 | ||
617 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
618 | ||
619 | static void raid5_build_block (struct stripe_head *sh, int i) | |
620 | { | |
621 | struct r5dev *dev = &sh->dev[i]; | |
622 | ||
623 | bio_init(&dev->req); | |
624 | dev->req.bi_io_vec = &dev->vec; | |
625 | dev->req.bi_vcnt++; | |
626 | dev->req.bi_max_vecs++; | |
627 | dev->vec.bv_page = dev->page; | |
628 | dev->vec.bv_len = STRIPE_SIZE; | |
629 | dev->vec.bv_offset = 0; | |
630 | ||
631 | dev->req.bi_sector = sh->sector; | |
632 | dev->req.bi_private = sh; | |
633 | ||
634 | dev->flags = 0; | |
635 | if (i != sh->pd_idx) | |
636 | dev->sector = compute_blocknr(sh, i); | |
637 | } | |
638 | ||
639 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
640 | { | |
641 | char b[BDEVNAME_SIZE]; | |
642 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
643 | PRINTK("raid5: error called\n"); | |
644 | ||
b2d444d7 | 645 | if (!test_bit(Faulty, &rdev->flags)) { |
1da177e4 | 646 | mddev->sb_dirty = 1; |
b2d444d7 | 647 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
648 | conf->working_disks--; |
649 | mddev->degraded++; | |
650 | conf->failed_disks++; | |
b2d444d7 | 651 | clear_bit(In_sync, &rdev->flags); |
1da177e4 LT |
652 | /* |
653 | * if recovery was running, make sure it aborts. | |
654 | */ | |
655 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
656 | } | |
b2d444d7 | 657 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
658 | printk (KERN_ALERT |
659 | "raid5: Disk failure on %s, disabling device." | |
660 | " Operation continuing on %d devices\n", | |
661 | bdevname(rdev->bdev,b), conf->working_disks); | |
662 | } | |
663 | } | |
664 | ||
665 | /* | |
666 | * Input: a 'big' sector number, | |
667 | * Output: index of the data and parity disk, and the sector # in them. | |
668 | */ | |
669 | static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks, | |
670 | unsigned int data_disks, unsigned int * dd_idx, | |
671 | unsigned int * pd_idx, raid5_conf_t *conf) | |
672 | { | |
673 | long stripe; | |
674 | unsigned long chunk_number; | |
675 | unsigned int chunk_offset; | |
676 | sector_t new_sector; | |
677 | int sectors_per_chunk = conf->chunk_size >> 9; | |
678 | ||
679 | /* First compute the information on this sector */ | |
680 | ||
681 | /* | |
682 | * Compute the chunk number and the sector offset inside the chunk | |
683 | */ | |
684 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
685 | chunk_number = r_sector; | |
686 | BUG_ON(r_sector != chunk_number); | |
687 | ||
688 | /* | |
689 | * Compute the stripe number | |
690 | */ | |
691 | stripe = chunk_number / data_disks; | |
692 | ||
693 | /* | |
694 | * Compute the data disk and parity disk indexes inside the stripe | |
695 | */ | |
696 | *dd_idx = chunk_number % data_disks; | |
697 | ||
698 | /* | |
699 | * Select the parity disk based on the user selected algorithm. | |
700 | */ | |
701 | if (conf->level == 4) | |
702 | *pd_idx = data_disks; | |
703 | else switch (conf->algorithm) { | |
704 | case ALGORITHM_LEFT_ASYMMETRIC: | |
705 | *pd_idx = data_disks - stripe % raid_disks; | |
706 | if (*dd_idx >= *pd_idx) | |
707 | (*dd_idx)++; | |
708 | break; | |
709 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
710 | *pd_idx = stripe % raid_disks; | |
711 | if (*dd_idx >= *pd_idx) | |
712 | (*dd_idx)++; | |
713 | break; | |
714 | case ALGORITHM_LEFT_SYMMETRIC: | |
715 | *pd_idx = data_disks - stripe % raid_disks; | |
716 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
717 | break; | |
718 | case ALGORITHM_RIGHT_SYMMETRIC: | |
719 | *pd_idx = stripe % raid_disks; | |
720 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
721 | break; | |
722 | default: | |
14f8d26b | 723 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 LT |
724 | conf->algorithm); |
725 | } | |
726 | ||
727 | /* | |
728 | * Finally, compute the new sector number | |
729 | */ | |
730 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
731 | return new_sector; | |
732 | } | |
733 | ||
734 | ||
735 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
736 | { | |
737 | raid5_conf_t *conf = sh->raid_conf; | |
738 | int raid_disks = conf->raid_disks, data_disks = raid_disks - 1; | |
739 | sector_t new_sector = sh->sector, check; | |
740 | int sectors_per_chunk = conf->chunk_size >> 9; | |
741 | sector_t stripe; | |
742 | int chunk_offset; | |
743 | int chunk_number, dummy1, dummy2, dd_idx = i; | |
744 | sector_t r_sector; | |
745 | ||
746 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
747 | stripe = new_sector; | |
748 | BUG_ON(new_sector != stripe); | |
749 | ||
750 | ||
751 | switch (conf->algorithm) { | |
752 | case ALGORITHM_LEFT_ASYMMETRIC: | |
753 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
754 | if (i > sh->pd_idx) | |
755 | i--; | |
756 | break; | |
757 | case ALGORITHM_LEFT_SYMMETRIC: | |
758 | case ALGORITHM_RIGHT_SYMMETRIC: | |
759 | if (i < sh->pd_idx) | |
760 | i += raid_disks; | |
761 | i -= (sh->pd_idx + 1); | |
762 | break; | |
763 | default: | |
14f8d26b | 764 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 LT |
765 | conf->algorithm); |
766 | } | |
767 | ||
768 | chunk_number = stripe * data_disks + i; | |
769 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
770 | ||
771 | check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf); | |
772 | if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { | |
14f8d26b | 773 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
774 | return 0; |
775 | } | |
776 | return r_sector; | |
777 | } | |
778 | ||
779 | ||
780 | ||
781 | /* | |
782 | * Copy data between a page in the stripe cache, and a bio. | |
783 | * There are no alignment or size guarantees between the page or the | |
784 | * bio except that there is some overlap. | |
785 | * All iovecs in the bio must be considered. | |
786 | */ | |
787 | static void copy_data(int frombio, struct bio *bio, | |
788 | struct page *page, | |
789 | sector_t sector) | |
790 | { | |
791 | char *pa = page_address(page); | |
792 | struct bio_vec *bvl; | |
793 | int i; | |
794 | int page_offset; | |
795 | ||
796 | if (bio->bi_sector >= sector) | |
797 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
798 | else | |
799 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
800 | bio_for_each_segment(bvl, bio, i) { | |
801 | int len = bio_iovec_idx(bio,i)->bv_len; | |
802 | int clen; | |
803 | int b_offset = 0; | |
804 | ||
805 | if (page_offset < 0) { | |
806 | b_offset = -page_offset; | |
807 | page_offset += b_offset; | |
808 | len -= b_offset; | |
809 | } | |
810 | ||
811 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
812 | clen = STRIPE_SIZE - page_offset; | |
813 | else clen = len; | |
814 | ||
815 | if (clen > 0) { | |
816 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
817 | if (frombio) | |
818 | memcpy(pa+page_offset, ba+b_offset, clen); | |
819 | else | |
820 | memcpy(ba+b_offset, pa+page_offset, clen); | |
821 | __bio_kunmap_atomic(ba, KM_USER0); | |
822 | } | |
823 | if (clen < len) /* hit end of page */ | |
824 | break; | |
825 | page_offset += len; | |
826 | } | |
827 | } | |
828 | ||
829 | #define check_xor() do { \ | |
830 | if (count == MAX_XOR_BLOCKS) { \ | |
831 | xor_block(count, STRIPE_SIZE, ptr); \ | |
832 | count = 1; \ | |
833 | } \ | |
834 | } while(0) | |
835 | ||
836 | ||
837 | static void compute_block(struct stripe_head *sh, int dd_idx) | |
838 | { | |
839 | raid5_conf_t *conf = sh->raid_conf; | |
840 | int i, count, disks = conf->raid_disks; | |
841 | void *ptr[MAX_XOR_BLOCKS], *p; | |
842 | ||
843 | PRINTK("compute_block, stripe %llu, idx %d\n", | |
844 | (unsigned long long)sh->sector, dd_idx); | |
845 | ||
846 | ptr[0] = page_address(sh->dev[dd_idx].page); | |
847 | memset(ptr[0], 0, STRIPE_SIZE); | |
848 | count = 1; | |
849 | for (i = disks ; i--; ) { | |
850 | if (i == dd_idx) | |
851 | continue; | |
852 | p = page_address(sh->dev[i].page); | |
853 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
854 | ptr[count++] = p; | |
855 | else | |
14f8d26b | 856 | printk(KERN_ERR "compute_block() %d, stripe %llu, %d" |
1da177e4 LT |
857 | " not present\n", dd_idx, |
858 | (unsigned long long)sh->sector, i); | |
859 | ||
860 | check_xor(); | |
861 | } | |
862 | if (count != 1) | |
863 | xor_block(count, STRIPE_SIZE, ptr); | |
864 | set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
865 | } | |
866 | ||
867 | static void compute_parity(struct stripe_head *sh, int method) | |
868 | { | |
869 | raid5_conf_t *conf = sh->raid_conf; | |
870 | int i, pd_idx = sh->pd_idx, disks = conf->raid_disks, count; | |
871 | void *ptr[MAX_XOR_BLOCKS]; | |
872 | struct bio *chosen; | |
873 | ||
874 | PRINTK("compute_parity, stripe %llu, method %d\n", | |
875 | (unsigned long long)sh->sector, method); | |
876 | ||
877 | count = 1; | |
878 | ptr[0] = page_address(sh->dev[pd_idx].page); | |
879 | switch(method) { | |
880 | case READ_MODIFY_WRITE: | |
881 | if (!test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags)) | |
882 | BUG(); | |
883 | for (i=disks ; i-- ;) { | |
884 | if (i==pd_idx) | |
885 | continue; | |
886 | if (sh->dev[i].towrite && | |
887 | test_bit(R5_UPTODATE, &sh->dev[i].flags)) { | |
888 | ptr[count++] = page_address(sh->dev[i].page); | |
889 | chosen = sh->dev[i].towrite; | |
890 | sh->dev[i].towrite = NULL; | |
891 | ||
892 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
893 | wake_up(&conf->wait_for_overlap); | |
894 | ||
895 | if (sh->dev[i].written) BUG(); | |
896 | sh->dev[i].written = chosen; | |
897 | check_xor(); | |
898 | } | |
899 | } | |
900 | break; | |
901 | case RECONSTRUCT_WRITE: | |
902 | memset(ptr[0], 0, STRIPE_SIZE); | |
903 | for (i= disks; i-- ;) | |
904 | if (i!=pd_idx && sh->dev[i].towrite) { | |
905 | chosen = sh->dev[i].towrite; | |
906 | sh->dev[i].towrite = NULL; | |
907 | ||
908 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
909 | wake_up(&conf->wait_for_overlap); | |
910 | ||
911 | if (sh->dev[i].written) BUG(); | |
912 | sh->dev[i].written = chosen; | |
913 | } | |
914 | break; | |
915 | case CHECK_PARITY: | |
916 | break; | |
917 | } | |
918 | if (count>1) { | |
919 | xor_block(count, STRIPE_SIZE, ptr); | |
920 | count = 1; | |
921 | } | |
922 | ||
923 | for (i = disks; i--;) | |
924 | if (sh->dev[i].written) { | |
925 | sector_t sector = sh->dev[i].sector; | |
926 | struct bio *wbi = sh->dev[i].written; | |
927 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
928 | copy_data(1, wbi, sh->dev[i].page, sector); | |
929 | wbi = r5_next_bio(wbi, sector); | |
930 | } | |
931 | ||
932 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
933 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
934 | } | |
935 | ||
936 | switch(method) { | |
937 | case RECONSTRUCT_WRITE: | |
938 | case CHECK_PARITY: | |
939 | for (i=disks; i--;) | |
940 | if (i != pd_idx) { | |
941 | ptr[count++] = page_address(sh->dev[i].page); | |
942 | check_xor(); | |
943 | } | |
944 | break; | |
945 | case READ_MODIFY_WRITE: | |
946 | for (i = disks; i--;) | |
947 | if (sh->dev[i].written) { | |
948 | ptr[count++] = page_address(sh->dev[i].page); | |
949 | check_xor(); | |
950 | } | |
951 | } | |
952 | if (count != 1) | |
953 | xor_block(count, STRIPE_SIZE, ptr); | |
954 | ||
955 | if (method != CHECK_PARITY) { | |
956 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
957 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
958 | } else | |
959 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
960 | } | |
961 | ||
962 | /* | |
963 | * Each stripe/dev can have one or more bion attached. | |
964 | * toread/towrite point to the first in a chain. | |
965 | * The bi_next chain must be in order. | |
966 | */ | |
967 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
968 | { | |
969 | struct bio **bip; | |
970 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 971 | int firstwrite=0; |
1da177e4 LT |
972 | |
973 | PRINTK("adding bh b#%llu to stripe s#%llu\n", | |
974 | (unsigned long long)bi->bi_sector, | |
975 | (unsigned long long)sh->sector); | |
976 | ||
977 | ||
978 | spin_lock(&sh->lock); | |
979 | spin_lock_irq(&conf->device_lock); | |
72626685 | 980 | if (forwrite) { |
1da177e4 | 981 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
982 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
983 | firstwrite = 1; | |
984 | } else | |
1da177e4 LT |
985 | bip = &sh->dev[dd_idx].toread; |
986 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
987 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
988 | goto overlap; | |
989 | bip = & (*bip)->bi_next; | |
990 | } | |
991 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
992 | goto overlap; | |
993 | ||
994 | if (*bip && bi->bi_next && (*bip) != bi->bi_next) | |
995 | BUG(); | |
996 | if (*bip) | |
997 | bi->bi_next = *bip; | |
998 | *bip = bi; | |
999 | bi->bi_phys_segments ++; | |
1000 | spin_unlock_irq(&conf->device_lock); | |
1001 | spin_unlock(&sh->lock); | |
1002 | ||
1003 | PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
1004 | (unsigned long long)bi->bi_sector, | |
1005 | (unsigned long long)sh->sector, dd_idx); | |
1006 | ||
72626685 N |
1007 | if (conf->mddev->bitmap && firstwrite) { |
1008 | sh->bm_seq = conf->seq_write; | |
1009 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
1010 | STRIPE_SECTORS, 0); | |
1011 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
1012 | } | |
1013 | ||
1da177e4 LT |
1014 | if (forwrite) { |
1015 | /* check if page is covered */ | |
1016 | sector_t sector = sh->dev[dd_idx].sector; | |
1017 | for (bi=sh->dev[dd_idx].towrite; | |
1018 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1019 | bi && bi->bi_sector <= sector; | |
1020 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1021 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1022 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1023 | } | |
1024 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1025 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1026 | } | |
1027 | return 1; | |
1028 | ||
1029 | overlap: | |
1030 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1031 | spin_unlock_irq(&conf->device_lock); | |
1032 | spin_unlock(&sh->lock); | |
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | ||
1037 | /* | |
1038 | * handle_stripe - do things to a stripe. | |
1039 | * | |
1040 | * We lock the stripe and then examine the state of various bits | |
1041 | * to see what needs to be done. | |
1042 | * Possible results: | |
1043 | * return some read request which now have data | |
1044 | * return some write requests which are safely on disc | |
1045 | * schedule a read on some buffers | |
1046 | * schedule a write of some buffers | |
1047 | * return confirmation of parity correctness | |
1048 | * | |
1049 | * Parity calculations are done inside the stripe lock | |
1050 | * buffers are taken off read_list or write_list, and bh_cache buffers | |
1051 | * get BH_Lock set before the stripe lock is released. | |
1052 | * | |
1053 | */ | |
1054 | ||
1055 | static void handle_stripe(struct stripe_head *sh) | |
1056 | { | |
1057 | raid5_conf_t *conf = sh->raid_conf; | |
1058 | int disks = conf->raid_disks; | |
1059 | struct bio *return_bi= NULL; | |
1060 | struct bio *bi; | |
1061 | int i; | |
1062 | int syncing; | |
1063 | int locked=0, uptodate=0, to_read=0, to_write=0, failed=0, written=0; | |
1064 | int non_overwrite = 0; | |
1065 | int failed_num=0; | |
1066 | struct r5dev *dev; | |
1067 | ||
1068 | PRINTK("handling stripe %llu, cnt=%d, pd_idx=%d\n", | |
1069 | (unsigned long long)sh->sector, atomic_read(&sh->count), | |
1070 | sh->pd_idx); | |
1071 | ||
1072 | spin_lock(&sh->lock); | |
1073 | clear_bit(STRIPE_HANDLE, &sh->state); | |
1074 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1075 | ||
1076 | syncing = test_bit(STRIPE_SYNCING, &sh->state); | |
1077 | /* Now to look around and see what can be done */ | |
1078 | ||
9910f16a | 1079 | rcu_read_lock(); |
1da177e4 LT |
1080 | for (i=disks; i--; ) { |
1081 | mdk_rdev_t *rdev; | |
1082 | dev = &sh->dev[i]; | |
1083 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
1084 | |
1085 | PRINTK("check %d: state 0x%lx read %p write %p written %p\n", | |
1086 | i, dev->flags, dev->toread, dev->towrite, dev->written); | |
1087 | /* maybe we can reply to a read */ | |
1088 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
1089 | struct bio *rbi, *rbi2; | |
1090 | PRINTK("Return read for disc %d\n", i); | |
1091 | spin_lock_irq(&conf->device_lock); | |
1092 | rbi = dev->toread; | |
1093 | dev->toread = NULL; | |
1094 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1095 | wake_up(&conf->wait_for_overlap); | |
1096 | spin_unlock_irq(&conf->device_lock); | |
1097 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
1098 | copy_data(0, rbi, dev->page, dev->sector); | |
1099 | rbi2 = r5_next_bio(rbi, dev->sector); | |
1100 | spin_lock_irq(&conf->device_lock); | |
1101 | if (--rbi->bi_phys_segments == 0) { | |
1102 | rbi->bi_next = return_bi; | |
1103 | return_bi = rbi; | |
1104 | } | |
1105 | spin_unlock_irq(&conf->device_lock); | |
1106 | rbi = rbi2; | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | /* now count some things */ | |
1111 | if (test_bit(R5_LOCKED, &dev->flags)) locked++; | |
1112 | if (test_bit(R5_UPTODATE, &dev->flags)) uptodate++; | |
1113 | ||
1114 | ||
1115 | if (dev->toread) to_read++; | |
1116 | if (dev->towrite) { | |
1117 | to_write++; | |
1118 | if (!test_bit(R5_OVERWRITE, &dev->flags)) | |
1119 | non_overwrite++; | |
1120 | } | |
1121 | if (dev->written) written++; | |
9910f16a | 1122 | rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 1123 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 1124 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
1125 | clear_bit(R5_ReadError, &dev->flags); |
1126 | clear_bit(R5_ReWrite, &dev->flags); | |
1127 | } | |
b2d444d7 | 1128 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 1129 | || test_bit(R5_ReadError, &dev->flags)) { |
1da177e4 LT |
1130 | failed++; |
1131 | failed_num = i; | |
1132 | } else | |
1133 | set_bit(R5_Insync, &dev->flags); | |
1134 | } | |
9910f16a | 1135 | rcu_read_unlock(); |
1da177e4 LT |
1136 | PRINTK("locked=%d uptodate=%d to_read=%d" |
1137 | " to_write=%d failed=%d failed_num=%d\n", | |
1138 | locked, uptodate, to_read, to_write, failed, failed_num); | |
1139 | /* check if the array has lost two devices and, if so, some requests might | |
1140 | * need to be failed | |
1141 | */ | |
1142 | if (failed > 1 && to_read+to_write+written) { | |
1da177e4 | 1143 | for (i=disks; i--; ) { |
72626685 | 1144 | int bitmap_end = 0; |
4e5314b5 N |
1145 | |
1146 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
9910f16a N |
1147 | mdk_rdev_t *rdev; |
1148 | rcu_read_lock(); | |
1149 | rdev = rcu_dereference(conf->disks[i].rdev); | |
b2d444d7 | 1150 | if (rdev && test_bit(In_sync, &rdev->flags)) |
4e5314b5 N |
1151 | /* multiple read failures in one stripe */ |
1152 | md_error(conf->mddev, rdev); | |
9910f16a | 1153 | rcu_read_unlock(); |
4e5314b5 N |
1154 | } |
1155 | ||
72626685 | 1156 | spin_lock_irq(&conf->device_lock); |
1da177e4 LT |
1157 | /* fail all writes first */ |
1158 | bi = sh->dev[i].towrite; | |
1159 | sh->dev[i].towrite = NULL; | |
72626685 | 1160 | if (bi) { to_write--; bitmap_end = 1; } |
1da177e4 LT |
1161 | |
1162 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1163 | wake_up(&conf->wait_for_overlap); | |
1164 | ||
1165 | while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ | |
1166 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
1167 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1168 | if (--bi->bi_phys_segments == 0) { | |
1169 | md_write_end(conf->mddev); | |
1170 | bi->bi_next = return_bi; | |
1171 | return_bi = bi; | |
1172 | } | |
1173 | bi = nextbi; | |
1174 | } | |
1175 | /* and fail all 'written' */ | |
1176 | bi = sh->dev[i].written; | |
1177 | sh->dev[i].written = NULL; | |
72626685 | 1178 | if (bi) bitmap_end = 1; |
1da177e4 LT |
1179 | while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS) { |
1180 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
1181 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1182 | if (--bi->bi_phys_segments == 0) { | |
1183 | md_write_end(conf->mddev); | |
1184 | bi->bi_next = return_bi; | |
1185 | return_bi = bi; | |
1186 | } | |
1187 | bi = bi2; | |
1188 | } | |
1189 | ||
1190 | /* fail any reads if this device is non-operational */ | |
4e5314b5 N |
1191 | if (!test_bit(R5_Insync, &sh->dev[i].flags) || |
1192 | test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1da177e4 LT |
1193 | bi = sh->dev[i].toread; |
1194 | sh->dev[i].toread = NULL; | |
1195 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1196 | wake_up(&conf->wait_for_overlap); | |
1197 | if (bi) to_read--; | |
1198 | while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ | |
1199 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
1200 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1201 | if (--bi->bi_phys_segments == 0) { | |
1202 | bi->bi_next = return_bi; | |
1203 | return_bi = bi; | |
1204 | } | |
1205 | bi = nextbi; | |
1206 | } | |
1207 | } | |
72626685 N |
1208 | spin_unlock_irq(&conf->device_lock); |
1209 | if (bitmap_end) | |
1210 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
1211 | STRIPE_SECTORS, 0, 0); | |
1da177e4 | 1212 | } |
1da177e4 LT |
1213 | } |
1214 | if (failed > 1 && syncing) { | |
1215 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); | |
1216 | clear_bit(STRIPE_SYNCING, &sh->state); | |
1217 | syncing = 0; | |
1218 | } | |
1219 | ||
1220 | /* might be able to return some write requests if the parity block | |
1221 | * is safe, or on a failed drive | |
1222 | */ | |
1223 | dev = &sh->dev[sh->pd_idx]; | |
1224 | if ( written && | |
1225 | ( (test_bit(R5_Insync, &dev->flags) && !test_bit(R5_LOCKED, &dev->flags) && | |
1226 | test_bit(R5_UPTODATE, &dev->flags)) | |
1227 | || (failed == 1 && failed_num == sh->pd_idx)) | |
1228 | ) { | |
1229 | /* any written block on an uptodate or failed drive can be returned. | |
1230 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
1231 | * never LOCKED, so we don't need to test 'failed' directly. | |
1232 | */ | |
1233 | for (i=disks; i--; ) | |
1234 | if (sh->dev[i].written) { | |
1235 | dev = &sh->dev[i]; | |
1236 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1237 | test_bit(R5_UPTODATE, &dev->flags) ) { | |
1238 | /* We can return any write requests */ | |
1239 | struct bio *wbi, *wbi2; | |
72626685 | 1240 | int bitmap_end = 0; |
1da177e4 LT |
1241 | PRINTK("Return write for disc %d\n", i); |
1242 | spin_lock_irq(&conf->device_lock); | |
1243 | wbi = dev->written; | |
1244 | dev->written = NULL; | |
1245 | while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
1246 | wbi2 = r5_next_bio(wbi, dev->sector); | |
1247 | if (--wbi->bi_phys_segments == 0) { | |
1248 | md_write_end(conf->mddev); | |
1249 | wbi->bi_next = return_bi; | |
1250 | return_bi = wbi; | |
1251 | } | |
1252 | wbi = wbi2; | |
1253 | } | |
72626685 N |
1254 | if (dev->towrite == NULL) |
1255 | bitmap_end = 1; | |
1da177e4 | 1256 | spin_unlock_irq(&conf->device_lock); |
72626685 N |
1257 | if (bitmap_end) |
1258 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
1259 | STRIPE_SECTORS, | |
1260 | !test_bit(STRIPE_DEGRADED, &sh->state), 0); | |
1da177e4 LT |
1261 | } |
1262 | } | |
1263 | } | |
1264 | ||
1265 | /* Now we might consider reading some blocks, either to check/generate | |
1266 | * parity, or to satisfy requests | |
1267 | * or to load a block that is being partially written. | |
1268 | */ | |
1269 | if (to_read || non_overwrite || (syncing && (uptodate < disks))) { | |
1270 | for (i=disks; i--;) { | |
1271 | dev = &sh->dev[i]; | |
1272 | if (!test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && | |
1273 | (dev->toread || | |
1274 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
1275 | syncing || | |
1276 | (failed && (sh->dev[failed_num].toread || | |
1277 | (sh->dev[failed_num].towrite && !test_bit(R5_OVERWRITE, &sh->dev[failed_num].flags)))) | |
1278 | ) | |
1279 | ) { | |
1280 | /* we would like to get this block, possibly | |
1281 | * by computing it, but we might not be able to | |
1282 | */ | |
1283 | if (uptodate == disks-1) { | |
1284 | PRINTK("Computing block %d\n", i); | |
1285 | compute_block(sh, i); | |
1286 | uptodate++; | |
1287 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
1288 | set_bit(R5_LOCKED, &dev->flags); | |
1289 | set_bit(R5_Wantread, &dev->flags); | |
1290 | #if 0 | |
1291 | /* if I am just reading this block and we don't have | |
1292 | a failed drive, or any pending writes then sidestep the cache */ | |
1293 | if (sh->bh_read[i] && !sh->bh_read[i]->b_reqnext && | |
1294 | ! syncing && !failed && !to_write) { | |
1295 | sh->bh_cache[i]->b_page = sh->bh_read[i]->b_page; | |
1296 | sh->bh_cache[i]->b_data = sh->bh_read[i]->b_data; | |
1297 | } | |
1298 | #endif | |
1299 | locked++; | |
1300 | PRINTK("Reading block %d (sync=%d)\n", | |
1301 | i, syncing); | |
1da177e4 LT |
1302 | } |
1303 | } | |
1304 | } | |
1305 | set_bit(STRIPE_HANDLE, &sh->state); | |
1306 | } | |
1307 | ||
1308 | /* now to consider writing and what else, if anything should be read */ | |
1309 | if (to_write) { | |
1310 | int rmw=0, rcw=0; | |
1311 | for (i=disks ; i--;) { | |
1312 | /* would I have to read this buffer for read_modify_write */ | |
1313 | dev = &sh->dev[i]; | |
1314 | if ((dev->towrite || i == sh->pd_idx) && | |
1315 | (!test_bit(R5_LOCKED, &dev->flags) | |
1316 | #if 0 | |
1317 | || sh->bh_page[i]!=bh->b_page | |
1318 | #endif | |
1319 | ) && | |
1320 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
1321 | if (test_bit(R5_Insync, &dev->flags) | |
1322 | /* && !(!mddev->insync && i == sh->pd_idx) */ | |
1323 | ) | |
1324 | rmw++; | |
1325 | else rmw += 2*disks; /* cannot read it */ | |
1326 | } | |
1327 | /* Would I have to read this buffer for reconstruct_write */ | |
1328 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
1329 | (!test_bit(R5_LOCKED, &dev->flags) | |
1330 | #if 0 | |
1331 | || sh->bh_page[i] != bh->b_page | |
1332 | #endif | |
1333 | ) && | |
1334 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
1335 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
1336 | else rcw += 2*disks; | |
1337 | } | |
1338 | } | |
1339 | PRINTK("for sector %llu, rmw=%d rcw=%d\n", | |
1340 | (unsigned long long)sh->sector, rmw, rcw); | |
1341 | set_bit(STRIPE_HANDLE, &sh->state); | |
1342 | if (rmw < rcw && rmw > 0) | |
1343 | /* prefer read-modify-write, but need to get some data */ | |
1344 | for (i=disks; i--;) { | |
1345 | dev = &sh->dev[i]; | |
1346 | if ((dev->towrite || i == sh->pd_idx) && | |
1347 | !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && | |
1348 | test_bit(R5_Insync, &dev->flags)) { | |
1349 | if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
1350 | { | |
1351 | PRINTK("Read_old block %d for r-m-w\n", i); | |
1352 | set_bit(R5_LOCKED, &dev->flags); | |
1353 | set_bit(R5_Wantread, &dev->flags); | |
1354 | locked++; | |
1355 | } else { | |
1356 | set_bit(STRIPE_DELAYED, &sh->state); | |
1357 | set_bit(STRIPE_HANDLE, &sh->state); | |
1358 | } | |
1359 | } | |
1360 | } | |
1361 | if (rcw <= rmw && rcw > 0) | |
1362 | /* want reconstruct write, but need to get some data */ | |
1363 | for (i=disks; i--;) { | |
1364 | dev = &sh->dev[i]; | |
1365 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
1366 | !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && | |
1367 | test_bit(R5_Insync, &dev->flags)) { | |
1368 | if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
1369 | { | |
1370 | PRINTK("Read_old block %d for Reconstruct\n", i); | |
1371 | set_bit(R5_LOCKED, &dev->flags); | |
1372 | set_bit(R5_Wantread, &dev->flags); | |
1373 | locked++; | |
1374 | } else { | |
1375 | set_bit(STRIPE_DELAYED, &sh->state); | |
1376 | set_bit(STRIPE_HANDLE, &sh->state); | |
1377 | } | |
1378 | } | |
1379 | } | |
1380 | /* now if nothing is locked, and if we have enough data, we can start a write request */ | |
72626685 N |
1381 | if (locked == 0 && (rcw == 0 ||rmw == 0) && |
1382 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
1da177e4 LT |
1383 | PRINTK("Computing parity...\n"); |
1384 | compute_parity(sh, rcw==0 ? RECONSTRUCT_WRITE : READ_MODIFY_WRITE); | |
1385 | /* now every locked buffer is ready to be written */ | |
1386 | for (i=disks; i--;) | |
1387 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
1388 | PRINTK("Writing block %d\n", i); | |
1389 | locked++; | |
1390 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
1391 | if (!test_bit(R5_Insync, &sh->dev[i].flags) | |
1392 | || (i==sh->pd_idx && failed == 0)) | |
1393 | set_bit(STRIPE_INSYNC, &sh->state); | |
1394 | } | |
1395 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1396 | atomic_dec(&conf->preread_active_stripes); | |
1397 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
1398 | md_wakeup_thread(conf->mddev->thread); | |
1399 | } | |
1400 | } | |
1401 | } | |
1402 | ||
1403 | /* maybe we need to check and possibly fix the parity for this stripe | |
1404 | * Any reads will already have been scheduled, so we just see if enough data | |
1405 | * is available | |
1406 | */ | |
1407 | if (syncing && locked == 0 && | |
14f8d26b | 1408 | !test_bit(STRIPE_INSYNC, &sh->state)) { |
1da177e4 LT |
1409 | set_bit(STRIPE_HANDLE, &sh->state); |
1410 | if (failed == 0) { | |
1411 | char *pagea; | |
1412 | if (uptodate != disks) | |
1413 | BUG(); | |
1414 | compute_parity(sh, CHECK_PARITY); | |
1415 | uptodate--; | |
1416 | pagea = page_address(sh->dev[sh->pd_idx].page); | |
1417 | if ((*(u32*)pagea) == 0 && | |
1418 | !memcmp(pagea, pagea+4, STRIPE_SIZE-4)) { | |
1419 | /* parity is correct (on disc, not in buffer any more) */ | |
1420 | set_bit(STRIPE_INSYNC, &sh->state); | |
9d88883e N |
1421 | } else { |
1422 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
1423 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
1424 | /* don't try to repair!! */ | |
1425 | set_bit(STRIPE_INSYNC, &sh->state); | |
14f8d26b N |
1426 | else { |
1427 | compute_block(sh, sh->pd_idx); | |
1428 | uptodate++; | |
1429 | } | |
1da177e4 LT |
1430 | } |
1431 | } | |
1432 | if (!test_bit(STRIPE_INSYNC, &sh->state)) { | |
14f8d26b | 1433 | /* either failed parity check, or recovery is happening */ |
1da177e4 LT |
1434 | if (failed==0) |
1435 | failed_num = sh->pd_idx; | |
1da177e4 | 1436 | dev = &sh->dev[failed_num]; |
14f8d26b N |
1437 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
1438 | BUG_ON(uptodate != disks); | |
1439 | ||
1da177e4 LT |
1440 | set_bit(R5_LOCKED, &dev->flags); |
1441 | set_bit(R5_Wantwrite, &dev->flags); | |
72626685 | 1442 | clear_bit(STRIPE_DEGRADED, &sh->state); |
1da177e4 LT |
1443 | locked++; |
1444 | set_bit(STRIPE_INSYNC, &sh->state); | |
1da177e4 LT |
1445 | } |
1446 | } | |
1447 | if (syncing && locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { | |
1448 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); | |
1449 | clear_bit(STRIPE_SYNCING, &sh->state); | |
1450 | } | |
4e5314b5 N |
1451 | |
1452 | /* If the failed drive is just a ReadError, then we might need to progress | |
1453 | * the repair/check process | |
1454 | */ | |
ba22dcbf N |
1455 | if (failed == 1 && ! conf->mddev->ro && |
1456 | test_bit(R5_ReadError, &sh->dev[failed_num].flags) | |
4e5314b5 N |
1457 | && !test_bit(R5_LOCKED, &sh->dev[failed_num].flags) |
1458 | && test_bit(R5_UPTODATE, &sh->dev[failed_num].flags) | |
1459 | ) { | |
1460 | dev = &sh->dev[failed_num]; | |
1461 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
1462 | set_bit(R5_Wantwrite, &dev->flags); | |
1463 | set_bit(R5_ReWrite, &dev->flags); | |
1464 | set_bit(R5_LOCKED, &dev->flags); | |
1465 | } else { | |
1466 | /* let's read it back */ | |
1467 | set_bit(R5_Wantread, &dev->flags); | |
1468 | set_bit(R5_LOCKED, &dev->flags); | |
1469 | } | |
1470 | } | |
1471 | ||
1da177e4 LT |
1472 | spin_unlock(&sh->lock); |
1473 | ||
1474 | while ((bi=return_bi)) { | |
1475 | int bytes = bi->bi_size; | |
1476 | ||
1477 | return_bi = bi->bi_next; | |
1478 | bi->bi_next = NULL; | |
1479 | bi->bi_size = 0; | |
1480 | bi->bi_end_io(bi, bytes, 0); | |
1481 | } | |
1482 | for (i=disks; i-- ;) { | |
1483 | int rw; | |
1484 | struct bio *bi; | |
1485 | mdk_rdev_t *rdev; | |
1486 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
1487 | rw = 1; | |
1488 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) | |
1489 | rw = 0; | |
1490 | else | |
1491 | continue; | |
1492 | ||
1493 | bi = &sh->dev[i].req; | |
1494 | ||
1495 | bi->bi_rw = rw; | |
1496 | if (rw) | |
1497 | bi->bi_end_io = raid5_end_write_request; | |
1498 | else | |
1499 | bi->bi_end_io = raid5_end_read_request; | |
1500 | ||
1501 | rcu_read_lock(); | |
d6065f7b | 1502 | rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 1503 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
1504 | rdev = NULL; |
1505 | if (rdev) | |
1506 | atomic_inc(&rdev->nr_pending); | |
1507 | rcu_read_unlock(); | |
1508 | ||
1509 | if (rdev) { | |
9910f16a | 1510 | if (syncing) |
1da177e4 LT |
1511 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1512 | ||
1513 | bi->bi_bdev = rdev->bdev; | |
1514 | PRINTK("for %llu schedule op %ld on disc %d\n", | |
1515 | (unsigned long long)sh->sector, bi->bi_rw, i); | |
1516 | atomic_inc(&sh->count); | |
1517 | bi->bi_sector = sh->sector + rdev->data_offset; | |
1518 | bi->bi_flags = 1 << BIO_UPTODATE; | |
1519 | bi->bi_vcnt = 1; | |
1520 | bi->bi_max_vecs = 1; | |
1521 | bi->bi_idx = 0; | |
1522 | bi->bi_io_vec = &sh->dev[i].vec; | |
1523 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
1524 | bi->bi_io_vec[0].bv_offset = 0; | |
1525 | bi->bi_size = STRIPE_SIZE; | |
1526 | bi->bi_next = NULL; | |
4dbcdc75 N |
1527 | if (rw == WRITE && |
1528 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
1529 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); | |
1da177e4 LT |
1530 | generic_make_request(bi); |
1531 | } else { | |
72626685 N |
1532 | if (rw == 1) |
1533 | set_bit(STRIPE_DEGRADED, &sh->state); | |
1da177e4 LT |
1534 | PRINTK("skip op %ld on disc %d for sector %llu\n", |
1535 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
1536 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1537 | set_bit(STRIPE_HANDLE, &sh->state); | |
1538 | } | |
1539 | } | |
1540 | } | |
1541 | ||
858119e1 | 1542 | static void raid5_activate_delayed(raid5_conf_t *conf) |
1da177e4 LT |
1543 | { |
1544 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
1545 | while (!list_empty(&conf->delayed_list)) { | |
1546 | struct list_head *l = conf->delayed_list.next; | |
1547 | struct stripe_head *sh; | |
1548 | sh = list_entry(l, struct stripe_head, lru); | |
1549 | list_del_init(l); | |
1550 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1551 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
1552 | atomic_inc(&conf->preread_active_stripes); | |
1553 | list_add_tail(&sh->lru, &conf->handle_list); | |
1554 | } | |
1555 | } | |
1556 | } | |
1557 | ||
858119e1 | 1558 | static void activate_bit_delay(raid5_conf_t *conf) |
72626685 N |
1559 | { |
1560 | /* device_lock is held */ | |
1561 | struct list_head head; | |
1562 | list_add(&head, &conf->bitmap_list); | |
1563 | list_del_init(&conf->bitmap_list); | |
1564 | while (!list_empty(&head)) { | |
1565 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
1566 | list_del_init(&sh->lru); | |
1567 | atomic_inc(&sh->count); | |
1568 | __release_stripe(conf, sh); | |
1569 | } | |
1570 | } | |
1571 | ||
1da177e4 LT |
1572 | static void unplug_slaves(mddev_t *mddev) |
1573 | { | |
1574 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1575 | int i; | |
1576 | ||
1577 | rcu_read_lock(); | |
1578 | for (i=0; i<mddev->raid_disks; i++) { | |
d6065f7b | 1579 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 1580 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { |
1da177e4 LT |
1581 | request_queue_t *r_queue = bdev_get_queue(rdev->bdev); |
1582 | ||
1583 | atomic_inc(&rdev->nr_pending); | |
1584 | rcu_read_unlock(); | |
1585 | ||
1586 | if (r_queue->unplug_fn) | |
1587 | r_queue->unplug_fn(r_queue); | |
1588 | ||
1589 | rdev_dec_pending(rdev, mddev); | |
1590 | rcu_read_lock(); | |
1591 | } | |
1592 | } | |
1593 | rcu_read_unlock(); | |
1594 | } | |
1595 | ||
1596 | static void raid5_unplug_device(request_queue_t *q) | |
1597 | { | |
1598 | mddev_t *mddev = q->queuedata; | |
1599 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1600 | unsigned long flags; | |
1601 | ||
1602 | spin_lock_irqsave(&conf->device_lock, flags); | |
1603 | ||
72626685 N |
1604 | if (blk_remove_plug(q)) { |
1605 | conf->seq_flush++; | |
1da177e4 | 1606 | raid5_activate_delayed(conf); |
72626685 | 1607 | } |
1da177e4 LT |
1608 | md_wakeup_thread(mddev->thread); |
1609 | ||
1610 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1611 | ||
1612 | unplug_slaves(mddev); | |
1613 | } | |
1614 | ||
1615 | static int raid5_issue_flush(request_queue_t *q, struct gendisk *disk, | |
1616 | sector_t *error_sector) | |
1617 | { | |
1618 | mddev_t *mddev = q->queuedata; | |
1619 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1620 | int i, ret = 0; | |
1621 | ||
1622 | rcu_read_lock(); | |
1623 | for (i=0; i<mddev->raid_disks && ret == 0; i++) { | |
d6065f7b | 1624 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 1625 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
1626 | struct block_device *bdev = rdev->bdev; |
1627 | request_queue_t *r_queue = bdev_get_queue(bdev); | |
1628 | ||
1629 | if (!r_queue->issue_flush_fn) | |
1630 | ret = -EOPNOTSUPP; | |
1631 | else { | |
1632 | atomic_inc(&rdev->nr_pending); | |
1633 | rcu_read_unlock(); | |
1634 | ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, | |
1635 | error_sector); | |
1636 | rdev_dec_pending(rdev, mddev); | |
1637 | rcu_read_lock(); | |
1638 | } | |
1639 | } | |
1640 | } | |
1641 | rcu_read_unlock(); | |
1642 | return ret; | |
1643 | } | |
1644 | ||
1645 | static inline void raid5_plug_device(raid5_conf_t *conf) | |
1646 | { | |
1647 | spin_lock_irq(&conf->device_lock); | |
1648 | blk_plug_device(conf->mddev->queue); | |
1649 | spin_unlock_irq(&conf->device_lock); | |
1650 | } | |
1651 | ||
1652 | static int make_request (request_queue_t *q, struct bio * bi) | |
1653 | { | |
1654 | mddev_t *mddev = q->queuedata; | |
1655 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1656 | const unsigned int raid_disks = conf->raid_disks; | |
1657 | const unsigned int data_disks = raid_disks - 1; | |
1658 | unsigned int dd_idx, pd_idx; | |
1659 | sector_t new_sector; | |
1660 | sector_t logical_sector, last_sector; | |
1661 | struct stripe_head *sh; | |
a362357b | 1662 | const int rw = bio_data_dir(bi); |
1da177e4 | 1663 | |
e5dcdd80 N |
1664 | if (unlikely(bio_barrier(bi))) { |
1665 | bio_endio(bi, bi->bi_size, -EOPNOTSUPP); | |
1666 | return 0; | |
1667 | } | |
1668 | ||
3d310eb7 | 1669 | md_write_start(mddev, bi); |
06d91a5f | 1670 | |
a362357b JA |
1671 | disk_stat_inc(mddev->gendisk, ios[rw]); |
1672 | disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bi)); | |
1da177e4 LT |
1673 | |
1674 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
1675 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
1676 | bi->bi_next = NULL; | |
1677 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 1678 | |
1da177e4 LT |
1679 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
1680 | DEFINE_WAIT(w); | |
1681 | ||
1682 | new_sector = raid5_compute_sector(logical_sector, | |
1683 | raid_disks, data_disks, &dd_idx, &pd_idx, conf); | |
1684 | ||
1685 | PRINTK("raid5: make_request, sector %llu logical %llu\n", | |
1686 | (unsigned long long)new_sector, | |
1687 | (unsigned long long)logical_sector); | |
1688 | ||
1689 | retry: | |
1690 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); | |
1691 | sh = get_active_stripe(conf, new_sector, pd_idx, (bi->bi_rw&RWA_MASK)); | |
1692 | if (sh) { | |
1693 | if (!add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
1694 | /* Add failed due to overlap. Flush everything | |
1695 | * and wait a while | |
1696 | */ | |
1697 | raid5_unplug_device(mddev->queue); | |
1698 | release_stripe(sh); | |
1699 | schedule(); | |
1700 | goto retry; | |
1701 | } | |
1702 | finish_wait(&conf->wait_for_overlap, &w); | |
1703 | raid5_plug_device(conf); | |
1704 | handle_stripe(sh); | |
1705 | release_stripe(sh); | |
1706 | ||
1707 | } else { | |
1708 | /* cannot get stripe for read-ahead, just give-up */ | |
1709 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1710 | finish_wait(&conf->wait_for_overlap, &w); | |
1711 | break; | |
1712 | } | |
1713 | ||
1714 | } | |
1715 | spin_lock_irq(&conf->device_lock); | |
1716 | if (--bi->bi_phys_segments == 0) { | |
1717 | int bytes = bi->bi_size; | |
1718 | ||
1719 | if ( bio_data_dir(bi) == WRITE ) | |
1720 | md_write_end(mddev); | |
1721 | bi->bi_size = 0; | |
1722 | bi->bi_end_io(bi, bytes, 0); | |
1723 | } | |
1724 | spin_unlock_irq(&conf->device_lock); | |
1725 | return 0; | |
1726 | } | |
1727 | ||
1728 | /* FIXME go_faster isn't used */ | |
57afd89f | 1729 | static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) |
1da177e4 LT |
1730 | { |
1731 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
1732 | struct stripe_head *sh; | |
1733 | int sectors_per_chunk = conf->chunk_size >> 9; | |
1734 | sector_t x; | |
1735 | unsigned long stripe; | |
1736 | int chunk_offset; | |
1737 | int dd_idx, pd_idx; | |
1738 | sector_t first_sector; | |
1739 | int raid_disks = conf->raid_disks; | |
1740 | int data_disks = raid_disks-1; | |
72626685 N |
1741 | sector_t max_sector = mddev->size << 1; |
1742 | int sync_blocks; | |
1da177e4 | 1743 | |
72626685 | 1744 | if (sector_nr >= max_sector) { |
1da177e4 LT |
1745 | /* just being told to finish up .. nothing much to do */ |
1746 | unplug_slaves(mddev); | |
72626685 N |
1747 | |
1748 | if (mddev->curr_resync < max_sector) /* aborted */ | |
1749 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
1750 | &sync_blocks, 1); | |
1751 | else /* compelted sync */ | |
1752 | conf->fullsync = 0; | |
1753 | bitmap_close_sync(mddev->bitmap); | |
1754 | ||
1da177e4 LT |
1755 | return 0; |
1756 | } | |
1757 | /* if there is 1 or more failed drives and we are trying | |
1758 | * to resync, then assert that we are finished, because there is | |
1759 | * nothing we can do. | |
1760 | */ | |
1761 | if (mddev->degraded >= 1 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
57afd89f N |
1762 | sector_t rv = (mddev->size << 1) - sector_nr; |
1763 | *skipped = 1; | |
1da177e4 LT |
1764 | return rv; |
1765 | } | |
72626685 | 1766 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 1767 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
1768 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
1769 | /* we can skip this block, and probably more */ | |
1770 | sync_blocks /= STRIPE_SECTORS; | |
1771 | *skipped = 1; | |
1772 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
1773 | } | |
1da177e4 LT |
1774 | |
1775 | x = sector_nr; | |
1776 | chunk_offset = sector_div(x, sectors_per_chunk); | |
1777 | stripe = x; | |
1778 | BUG_ON(x != stripe); | |
1779 | ||
1780 | first_sector = raid5_compute_sector((sector_t)stripe*data_disks*sectors_per_chunk | |
1781 | + chunk_offset, raid_disks, data_disks, &dd_idx, &pd_idx, conf); | |
1782 | sh = get_active_stripe(conf, sector_nr, pd_idx, 1); | |
1783 | if (sh == NULL) { | |
1784 | sh = get_active_stripe(conf, sector_nr, pd_idx, 0); | |
1785 | /* make sure we don't swamp the stripe cache if someone else | |
1786 | * is trying to get access | |
1787 | */ | |
66c006a5 | 1788 | schedule_timeout_uninterruptible(1); |
1da177e4 | 1789 | } |
72626685 | 1790 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 0); |
1da177e4 LT |
1791 | spin_lock(&sh->lock); |
1792 | set_bit(STRIPE_SYNCING, &sh->state); | |
1793 | clear_bit(STRIPE_INSYNC, &sh->state); | |
1794 | spin_unlock(&sh->lock); | |
1795 | ||
1796 | handle_stripe(sh); | |
1797 | release_stripe(sh); | |
1798 | ||
1799 | return STRIPE_SECTORS; | |
1800 | } | |
1801 | ||
1802 | /* | |
1803 | * This is our raid5 kernel thread. | |
1804 | * | |
1805 | * We scan the hash table for stripes which can be handled now. | |
1806 | * During the scan, completed stripes are saved for us by the interrupt | |
1807 | * handler, so that they will not have to wait for our next wakeup. | |
1808 | */ | |
1809 | static void raid5d (mddev_t *mddev) | |
1810 | { | |
1811 | struct stripe_head *sh; | |
1812 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1813 | int handled; | |
1814 | ||
1815 | PRINTK("+++ raid5d active\n"); | |
1816 | ||
1817 | md_check_recovery(mddev); | |
1da177e4 LT |
1818 | |
1819 | handled = 0; | |
1820 | spin_lock_irq(&conf->device_lock); | |
1821 | while (1) { | |
1822 | struct list_head *first; | |
1823 | ||
72626685 N |
1824 | if (conf->seq_flush - conf->seq_write > 0) { |
1825 | int seq = conf->seq_flush; | |
700e432d | 1826 | spin_unlock_irq(&conf->device_lock); |
72626685 | 1827 | bitmap_unplug(mddev->bitmap); |
700e432d | 1828 | spin_lock_irq(&conf->device_lock); |
72626685 N |
1829 | conf->seq_write = seq; |
1830 | activate_bit_delay(conf); | |
1831 | } | |
1832 | ||
1da177e4 LT |
1833 | if (list_empty(&conf->handle_list) && |
1834 | atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD && | |
1835 | !blk_queue_plugged(mddev->queue) && | |
1836 | !list_empty(&conf->delayed_list)) | |
1837 | raid5_activate_delayed(conf); | |
1838 | ||
1839 | if (list_empty(&conf->handle_list)) | |
1840 | break; | |
1841 | ||
1842 | first = conf->handle_list.next; | |
1843 | sh = list_entry(first, struct stripe_head, lru); | |
1844 | ||
1845 | list_del_init(first); | |
1846 | atomic_inc(&sh->count); | |
1847 | if (atomic_read(&sh->count)!= 1) | |
1848 | BUG(); | |
1849 | spin_unlock_irq(&conf->device_lock); | |
1850 | ||
1851 | handled++; | |
1852 | handle_stripe(sh); | |
1853 | release_stripe(sh); | |
1854 | ||
1855 | spin_lock_irq(&conf->device_lock); | |
1856 | } | |
1857 | PRINTK("%d stripes handled\n", handled); | |
1858 | ||
1859 | spin_unlock_irq(&conf->device_lock); | |
1860 | ||
1861 | unplug_slaves(mddev); | |
1862 | ||
1863 | PRINTK("--- raid5d inactive\n"); | |
1864 | } | |
1865 | ||
3f294f4f | 1866 | static ssize_t |
007583c9 | 1867 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 1868 | { |
007583c9 | 1869 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
1870 | if (conf) |
1871 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
1872 | else | |
1873 | return 0; | |
3f294f4f N |
1874 | } |
1875 | ||
1876 | static ssize_t | |
007583c9 | 1877 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 1878 | { |
007583c9 | 1879 | raid5_conf_t *conf = mddev_to_conf(mddev); |
3f294f4f N |
1880 | char *end; |
1881 | int new; | |
1882 | if (len >= PAGE_SIZE) | |
1883 | return -EINVAL; | |
96de1e66 N |
1884 | if (!conf) |
1885 | return -ENODEV; | |
3f294f4f N |
1886 | |
1887 | new = simple_strtoul(page, &end, 10); | |
1888 | if (!*page || (*end && *end != '\n') ) | |
1889 | return -EINVAL; | |
1890 | if (new <= 16 || new > 32768) | |
1891 | return -EINVAL; | |
1892 | while (new < conf->max_nr_stripes) { | |
1893 | if (drop_one_stripe(conf)) | |
1894 | conf->max_nr_stripes--; | |
1895 | else | |
1896 | break; | |
1897 | } | |
1898 | while (new > conf->max_nr_stripes) { | |
1899 | if (grow_one_stripe(conf)) | |
1900 | conf->max_nr_stripes++; | |
1901 | else break; | |
1902 | } | |
1903 | return len; | |
1904 | } | |
007583c9 | 1905 | |
96de1e66 N |
1906 | static struct md_sysfs_entry |
1907 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
1908 | raid5_show_stripe_cache_size, | |
1909 | raid5_store_stripe_cache_size); | |
3f294f4f N |
1910 | |
1911 | static ssize_t | |
96de1e66 | 1912 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 1913 | { |
007583c9 | 1914 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
1915 | if (conf) |
1916 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
1917 | else | |
1918 | return 0; | |
3f294f4f N |
1919 | } |
1920 | ||
96de1e66 N |
1921 | static struct md_sysfs_entry |
1922 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 1923 | |
007583c9 | 1924 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
1925 | &raid5_stripecache_size.attr, |
1926 | &raid5_stripecache_active.attr, | |
1927 | NULL, | |
1928 | }; | |
007583c9 N |
1929 | static struct attribute_group raid5_attrs_group = { |
1930 | .name = NULL, | |
1931 | .attrs = raid5_attrs, | |
3f294f4f N |
1932 | }; |
1933 | ||
72626685 | 1934 | static int run(mddev_t *mddev) |
1da177e4 LT |
1935 | { |
1936 | raid5_conf_t *conf; | |
1937 | int raid_disk, memory; | |
1938 | mdk_rdev_t *rdev; | |
1939 | struct disk_info *disk; | |
1940 | struct list_head *tmp; | |
1941 | ||
1942 | if (mddev->level != 5 && mddev->level != 4) { | |
14f8d26b N |
1943 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5 (%d)\n", |
1944 | mdname(mddev), mddev->level); | |
1da177e4 LT |
1945 | return -EIO; |
1946 | } | |
1947 | ||
b55e6bfc | 1948 | mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL); |
1da177e4 LT |
1949 | if ((conf = mddev->private) == NULL) |
1950 | goto abort; | |
b55e6bfc N |
1951 | conf->disks = kzalloc(mddev->raid_disks * sizeof(struct disk_info), |
1952 | GFP_KERNEL); | |
1953 | if (!conf->disks) | |
1954 | goto abort; | |
9ffae0cf | 1955 | |
1da177e4 LT |
1956 | conf->mddev = mddev; |
1957 | ||
fccddba0 | 1958 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 1959 | goto abort; |
1da177e4 LT |
1960 | |
1961 | spin_lock_init(&conf->device_lock); | |
1962 | init_waitqueue_head(&conf->wait_for_stripe); | |
1963 | init_waitqueue_head(&conf->wait_for_overlap); | |
1964 | INIT_LIST_HEAD(&conf->handle_list); | |
1965 | INIT_LIST_HEAD(&conf->delayed_list); | |
72626685 | 1966 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
1967 | INIT_LIST_HEAD(&conf->inactive_list); |
1968 | atomic_set(&conf->active_stripes, 0); | |
1969 | atomic_set(&conf->preread_active_stripes, 0); | |
1970 | ||
1da177e4 LT |
1971 | PRINTK("raid5: run(%s) called.\n", mdname(mddev)); |
1972 | ||
1973 | ITERATE_RDEV(mddev,rdev,tmp) { | |
1974 | raid_disk = rdev->raid_disk; | |
1975 | if (raid_disk >= mddev->raid_disks | |
1976 | || raid_disk < 0) | |
1977 | continue; | |
1978 | disk = conf->disks + raid_disk; | |
1979 | ||
1980 | disk->rdev = rdev; | |
1981 | ||
b2d444d7 | 1982 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
1983 | char b[BDEVNAME_SIZE]; |
1984 | printk(KERN_INFO "raid5: device %s operational as raid" | |
1985 | " disk %d\n", bdevname(rdev->bdev,b), | |
1986 | raid_disk); | |
1987 | conf->working_disks++; | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | conf->raid_disks = mddev->raid_disks; | |
1992 | /* | |
1993 | * 0 for a fully functional array, 1 for a degraded array. | |
1994 | */ | |
1995 | mddev->degraded = conf->failed_disks = conf->raid_disks - conf->working_disks; | |
1996 | conf->mddev = mddev; | |
1997 | conf->chunk_size = mddev->chunk_size; | |
1998 | conf->level = mddev->level; | |
1999 | conf->algorithm = mddev->layout; | |
2000 | conf->max_nr_stripes = NR_STRIPES; | |
2001 | ||
2002 | /* device size must be a multiple of chunk size */ | |
2003 | mddev->size &= ~(mddev->chunk_size/1024 -1); | |
b1581566 | 2004 | mddev->resync_max_sectors = mddev->size << 1; |
1da177e4 LT |
2005 | |
2006 | if (!conf->chunk_size || conf->chunk_size % 4) { | |
2007 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
2008 | conf->chunk_size, mdname(mddev)); | |
2009 | goto abort; | |
2010 | } | |
2011 | if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { | |
2012 | printk(KERN_ERR | |
2013 | "raid5: unsupported parity algorithm %d for %s\n", | |
2014 | conf->algorithm, mdname(mddev)); | |
2015 | goto abort; | |
2016 | } | |
2017 | if (mddev->degraded > 1) { | |
2018 | printk(KERN_ERR "raid5: not enough operational devices for %s" | |
2019 | " (%d/%d failed)\n", | |
2020 | mdname(mddev), conf->failed_disks, conf->raid_disks); | |
2021 | goto abort; | |
2022 | } | |
2023 | ||
2024 | if (mddev->degraded == 1 && | |
2025 | mddev->recovery_cp != MaxSector) { | |
6ff8d8ec N |
2026 | if (mddev->ok_start_degraded) |
2027 | printk(KERN_WARNING | |
2028 | "raid5: starting dirty degraded array: %s" | |
2029 | "- data corruption possible.\n", | |
2030 | mdname(mddev)); | |
2031 | else { | |
2032 | printk(KERN_ERR | |
2033 | "raid5: cannot start dirty degraded array for %s\n", | |
2034 | mdname(mddev)); | |
2035 | goto abort; | |
2036 | } | |
1da177e4 LT |
2037 | } |
2038 | ||
2039 | { | |
2040 | mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); | |
2041 | if (!mddev->thread) { | |
2042 | printk(KERN_ERR | |
2043 | "raid5: couldn't allocate thread for %s\n", | |
2044 | mdname(mddev)); | |
2045 | goto abort; | |
2046 | } | |
2047 | } | |
5036805b | 2048 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
1da177e4 LT |
2049 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
2050 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
2051 | printk(KERN_ERR | |
2052 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
2053 | shrink_stripes(conf); | |
2054 | md_unregister_thread(mddev->thread); | |
2055 | goto abort; | |
2056 | } else | |
2057 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
2058 | memory, mdname(mddev)); | |
2059 | ||
2060 | if (mddev->degraded == 0) | |
2061 | printk("raid5: raid level %d set %s active with %d out of %d" | |
2062 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
2063 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
2064 | conf->algorithm); | |
2065 | else | |
2066 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
2067 | " out of %d devices, algorithm %d\n", conf->level, | |
2068 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
2069 | mddev->raid_disks, conf->algorithm); | |
2070 | ||
2071 | print_raid5_conf(conf); | |
2072 | ||
2073 | /* read-ahead size must cover two whole stripes, which is | |
2074 | * 2 * (n-1) * chunksize where 'n' is the number of raid devices | |
2075 | */ | |
2076 | { | |
2077 | int stripe = (mddev->raid_disks-1) * mddev->chunk_size | |
2d1f3b5d | 2078 | / PAGE_SIZE; |
1da177e4 LT |
2079 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
2080 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
2081 | } | |
2082 | ||
2083 | /* Ok, everything is just fine now */ | |
007583c9 | 2084 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group); |
7a5febe9 N |
2085 | |
2086 | mddev->queue->unplug_fn = raid5_unplug_device; | |
2087 | mddev->queue->issue_flush_fn = raid5_issue_flush; | |
2088 | ||
1da177e4 LT |
2089 | mddev->array_size = mddev->size * (mddev->raid_disks - 1); |
2090 | return 0; | |
2091 | abort: | |
2092 | if (conf) { | |
2093 | print_raid5_conf(conf); | |
b55e6bfc | 2094 | kfree(conf->disks); |
fccddba0 | 2095 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
2096 | kfree(conf); |
2097 | } | |
2098 | mddev->private = NULL; | |
2099 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
2100 | return -EIO; | |
2101 | } | |
2102 | ||
2103 | ||
2104 | ||
3f294f4f | 2105 | static int stop(mddev_t *mddev) |
1da177e4 LT |
2106 | { |
2107 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
2108 | ||
2109 | md_unregister_thread(mddev->thread); | |
2110 | mddev->thread = NULL; | |
2111 | shrink_stripes(conf); | |
fccddba0 | 2112 | kfree(conf->stripe_hashtbl); |
1da177e4 | 2113 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 2114 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 2115 | kfree(conf->disks); |
96de1e66 | 2116 | kfree(conf); |
1da177e4 LT |
2117 | mddev->private = NULL; |
2118 | return 0; | |
2119 | } | |
2120 | ||
2121 | #if RAID5_DEBUG | |
2122 | static void print_sh (struct stripe_head *sh) | |
2123 | { | |
2124 | int i; | |
2125 | ||
2126 | printk("sh %llu, pd_idx %d, state %ld.\n", | |
2127 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
2128 | printk("sh %llu, count %d.\n", | |
2129 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
2130 | printk("sh %llu, ", (unsigned long long)sh->sector); | |
2131 | for (i = 0; i < sh->raid_conf->raid_disks; i++) { | |
2132 | printk("(cache%d: %p %ld) ", | |
2133 | i, sh->dev[i].page, sh->dev[i].flags); | |
2134 | } | |
2135 | printk("\n"); | |
2136 | } | |
2137 | ||
2138 | static void printall (raid5_conf_t *conf) | |
2139 | { | |
2140 | struct stripe_head *sh; | |
fccddba0 | 2141 | struct hlist_node *hn; |
1da177e4 LT |
2142 | int i; |
2143 | ||
2144 | spin_lock_irq(&conf->device_lock); | |
2145 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 2146 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
2147 | if (sh->raid_conf != conf) |
2148 | continue; | |
2149 | print_sh(sh); | |
2150 | } | |
2151 | } | |
2152 | spin_unlock_irq(&conf->device_lock); | |
2153 | } | |
2154 | #endif | |
2155 | ||
2156 | static void status (struct seq_file *seq, mddev_t *mddev) | |
2157 | { | |
2158 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
2159 | int i; | |
2160 | ||
2161 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
2162 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->working_disks); | |
2163 | for (i = 0; i < conf->raid_disks; i++) | |
2164 | seq_printf (seq, "%s", | |
2165 | conf->disks[i].rdev && | |
b2d444d7 | 2166 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
2167 | seq_printf (seq, "]"); |
2168 | #if RAID5_DEBUG | |
2169 | #define D(x) \ | |
2170 | seq_printf (seq, "<"#x":%d>", atomic_read(&conf->x)) | |
2171 | printall(conf); | |
2172 | #endif | |
2173 | } | |
2174 | ||
2175 | static void print_raid5_conf (raid5_conf_t *conf) | |
2176 | { | |
2177 | int i; | |
2178 | struct disk_info *tmp; | |
2179 | ||
2180 | printk("RAID5 conf printout:\n"); | |
2181 | if (!conf) { | |
2182 | printk("(conf==NULL)\n"); | |
2183 | return; | |
2184 | } | |
2185 | printk(" --- rd:%d wd:%d fd:%d\n", conf->raid_disks, | |
2186 | conf->working_disks, conf->failed_disks); | |
2187 | ||
2188 | for (i = 0; i < conf->raid_disks; i++) { | |
2189 | char b[BDEVNAME_SIZE]; | |
2190 | tmp = conf->disks + i; | |
2191 | if (tmp->rdev) | |
2192 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 2193 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
2194 | bdevname(tmp->rdev->bdev,b)); |
2195 | } | |
2196 | } | |
2197 | ||
2198 | static int raid5_spare_active(mddev_t *mddev) | |
2199 | { | |
2200 | int i; | |
2201 | raid5_conf_t *conf = mddev->private; | |
2202 | struct disk_info *tmp; | |
2203 | ||
2204 | for (i = 0; i < conf->raid_disks; i++) { | |
2205 | tmp = conf->disks + i; | |
2206 | if (tmp->rdev | |
b2d444d7 N |
2207 | && !test_bit(Faulty, &tmp->rdev->flags) |
2208 | && !test_bit(In_sync, &tmp->rdev->flags)) { | |
1da177e4 LT |
2209 | mddev->degraded--; |
2210 | conf->failed_disks--; | |
2211 | conf->working_disks++; | |
b2d444d7 | 2212 | set_bit(In_sync, &tmp->rdev->flags); |
1da177e4 LT |
2213 | } |
2214 | } | |
2215 | print_raid5_conf(conf); | |
2216 | return 0; | |
2217 | } | |
2218 | ||
2219 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
2220 | { | |
2221 | raid5_conf_t *conf = mddev->private; | |
2222 | int err = 0; | |
2223 | mdk_rdev_t *rdev; | |
2224 | struct disk_info *p = conf->disks + number; | |
2225 | ||
2226 | print_raid5_conf(conf); | |
2227 | rdev = p->rdev; | |
2228 | if (rdev) { | |
b2d444d7 | 2229 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
2230 | atomic_read(&rdev->nr_pending)) { |
2231 | err = -EBUSY; | |
2232 | goto abort; | |
2233 | } | |
2234 | p->rdev = NULL; | |
fbd568a3 | 2235 | synchronize_rcu(); |
1da177e4 LT |
2236 | if (atomic_read(&rdev->nr_pending)) { |
2237 | /* lost the race, try later */ | |
2238 | err = -EBUSY; | |
2239 | p->rdev = rdev; | |
2240 | } | |
2241 | } | |
2242 | abort: | |
2243 | ||
2244 | print_raid5_conf(conf); | |
2245 | return err; | |
2246 | } | |
2247 | ||
2248 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
2249 | { | |
2250 | raid5_conf_t *conf = mddev->private; | |
2251 | int found = 0; | |
2252 | int disk; | |
2253 | struct disk_info *p; | |
2254 | ||
2255 | if (mddev->degraded > 1) | |
2256 | /* no point adding a device */ | |
2257 | return 0; | |
2258 | ||
2259 | /* | |
2260 | * find the disk ... | |
2261 | */ | |
2262 | for (disk=0; disk < mddev->raid_disks; disk++) | |
2263 | if ((p=conf->disks + disk)->rdev == NULL) { | |
b2d444d7 | 2264 | clear_bit(In_sync, &rdev->flags); |
1da177e4 LT |
2265 | rdev->raid_disk = disk; |
2266 | found = 1; | |
72626685 N |
2267 | if (rdev->saved_raid_disk != disk) |
2268 | conf->fullsync = 1; | |
d6065f7b | 2269 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
2270 | break; |
2271 | } | |
2272 | print_raid5_conf(conf); | |
2273 | return found; | |
2274 | } | |
2275 | ||
2276 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
2277 | { | |
2278 | /* no resync is happening, and there is enough space | |
2279 | * on all devices, so we can resize. | |
2280 | * We need to make sure resync covers any new space. | |
2281 | * If the array is shrinking we should possibly wait until | |
2282 | * any io in the removed space completes, but it hardly seems | |
2283 | * worth it. | |
2284 | */ | |
2285 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); | |
2286 | mddev->array_size = (sectors * (mddev->raid_disks-1))>>1; | |
2287 | set_capacity(mddev->gendisk, mddev->array_size << 1); | |
2288 | mddev->changed = 1; | |
2289 | if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) { | |
2290 | mddev->recovery_cp = mddev->size << 1; | |
2291 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
2292 | } | |
2293 | mddev->size = sectors /2; | |
4b5c7ae8 | 2294 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
2295 | return 0; |
2296 | } | |
2297 | ||
72626685 N |
2298 | static void raid5_quiesce(mddev_t *mddev, int state) |
2299 | { | |
2300 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2301 | ||
2302 | switch(state) { | |
2303 | case 1: /* stop all writes */ | |
2304 | spin_lock_irq(&conf->device_lock); | |
2305 | conf->quiesce = 1; | |
2306 | wait_event_lock_irq(conf->wait_for_stripe, | |
2307 | atomic_read(&conf->active_stripes) == 0, | |
2308 | conf->device_lock, /* nothing */); | |
2309 | spin_unlock_irq(&conf->device_lock); | |
2310 | break; | |
2311 | ||
2312 | case 0: /* re-enable writes */ | |
2313 | spin_lock_irq(&conf->device_lock); | |
2314 | conf->quiesce = 0; | |
2315 | wake_up(&conf->wait_for_stripe); | |
2316 | spin_unlock_irq(&conf->device_lock); | |
2317 | break; | |
2318 | } | |
72626685 | 2319 | } |
b15c2e57 | 2320 | |
2604b703 | 2321 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
2322 | { |
2323 | .name = "raid5", | |
2604b703 | 2324 | .level = 5, |
1da177e4 LT |
2325 | .owner = THIS_MODULE, |
2326 | .make_request = make_request, | |
2327 | .run = run, | |
2328 | .stop = stop, | |
2329 | .status = status, | |
2330 | .error_handler = error, | |
2331 | .hot_add_disk = raid5_add_disk, | |
2332 | .hot_remove_disk= raid5_remove_disk, | |
2333 | .spare_active = raid5_spare_active, | |
2334 | .sync_request = sync_request, | |
2335 | .resize = raid5_resize, | |
72626685 | 2336 | .quiesce = raid5_quiesce, |
1da177e4 LT |
2337 | }; |
2338 | ||
2604b703 | 2339 | static struct mdk_personality raid4_personality = |
1da177e4 | 2340 | { |
2604b703 N |
2341 | .name = "raid4", |
2342 | .level = 4, | |
2343 | .owner = THIS_MODULE, | |
2344 | .make_request = make_request, | |
2345 | .run = run, | |
2346 | .stop = stop, | |
2347 | .status = status, | |
2348 | .error_handler = error, | |
2349 | .hot_add_disk = raid5_add_disk, | |
2350 | .hot_remove_disk= raid5_remove_disk, | |
2351 | .spare_active = raid5_spare_active, | |
2352 | .sync_request = sync_request, | |
2353 | .resize = raid5_resize, | |
2354 | .quiesce = raid5_quiesce, | |
2355 | }; | |
2356 | ||
2357 | static int __init raid5_init(void) | |
2358 | { | |
2359 | register_md_personality(&raid5_personality); | |
2360 | register_md_personality(&raid4_personality); | |
2361 | return 0; | |
1da177e4 LT |
2362 | } |
2363 | ||
2604b703 | 2364 | static void raid5_exit(void) |
1da177e4 | 2365 | { |
2604b703 N |
2366 | unregister_md_personality(&raid5_personality); |
2367 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
2368 | } |
2369 | ||
2370 | module_init(raid5_init); | |
2371 | module_exit(raid5_exit); | |
2372 | MODULE_LICENSE("GPL"); | |
2373 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
2374 | MODULE_ALIAS("md-raid5"); |
2375 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
2376 | MODULE_ALIAS("md-level-5"); |
2377 | MODULE_ALIAS("md-level-4"); |