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 | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
30 | * conf->bm_write is the number of the last batch successfully written. | |
31 | * conf->bm_flush is the number of the last batch that was closed to | |
32 | * new additions. | |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
35 | * the number of the batch it will be in. This is bm_flush+1. | |
36 | * When we are ready to do a write, if that batch hasn't been written yet, | |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <linux/module.h> |
47 | #include <linux/slab.h> | |
1da177e4 LT |
48 | #include <linux/highmem.h> |
49 | #include <linux/bitops.h> | |
f6705578 | 50 | #include <linux/kthread.h> |
1da177e4 | 51 | #include <asm/atomic.h> |
16a53ecc | 52 | #include "raid6.h" |
1da177e4 | 53 | |
72626685 N |
54 | #include <linux/raid/bitmap.h> |
55 | ||
1da177e4 LT |
56 | /* |
57 | * Stripe cache | |
58 | */ | |
59 | ||
60 | #define NR_STRIPES 256 | |
61 | #define STRIPE_SIZE PAGE_SIZE | |
62 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
63 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
64 | #define IO_THRESHOLD 1 | |
fccddba0 | 65 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 LT |
66 | #define HASH_MASK (NR_HASH - 1) |
67 | ||
fccddba0 | 68 | #define stripe_hash(conf, sect) (&((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK])) |
1da177e4 LT |
69 | |
70 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector | |
71 | * order without overlap. There may be several bio's per stripe+device, and | |
72 | * a bio could span several devices. | |
73 | * When walking this list for a particular stripe+device, we must never proceed | |
74 | * beyond a bio that extends past this device, as the next bio might no longer | |
75 | * be valid. | |
76 | * This macro is used to determine the 'next' bio in the list, given the sector | |
77 | * of the current stripe+device | |
78 | */ | |
79 | #define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) | |
80 | /* | |
81 | * The following can be used to debug the driver | |
82 | */ | |
83 | #define RAID5_DEBUG 0 | |
84 | #define RAID5_PARANOIA 1 | |
85 | #if RAID5_PARANOIA && defined(CONFIG_SMP) | |
86 | # define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) | |
87 | #else | |
88 | # define CHECK_DEVLOCK() | |
89 | #endif | |
90 | ||
91 | #define PRINTK(x...) ((void)(RAID5_DEBUG && printk(x))) | |
92 | #if RAID5_DEBUG | |
93 | #define inline | |
94 | #define __inline__ | |
95 | #endif | |
96 | ||
16a53ecc N |
97 | #if !RAID6_USE_EMPTY_ZERO_PAGE |
98 | /* In .bss so it's zeroed */ | |
99 | const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); | |
100 | #endif | |
101 | ||
102 | static inline int raid6_next_disk(int disk, int raid_disks) | |
103 | { | |
104 | disk++; | |
105 | return (disk < raid_disks) ? disk : 0; | |
106 | } | |
a4456856 DW |
107 | |
108 | static void return_io(struct bio *return_bi) | |
109 | { | |
110 | struct bio *bi = return_bi; | |
111 | while (bi) { | |
112 | int bytes = bi->bi_size; | |
113 | ||
114 | return_bi = bi->bi_next; | |
115 | bi->bi_next = NULL; | |
116 | bi->bi_size = 0; | |
117 | bi->bi_end_io(bi, bytes, | |
118 | test_bit(BIO_UPTODATE, &bi->bi_flags) | |
119 | ? 0 : -EIO); | |
120 | bi = return_bi; | |
121 | } | |
122 | } | |
123 | ||
1da177e4 LT |
124 | static void print_raid5_conf (raid5_conf_t *conf); |
125 | ||
858119e1 | 126 | static void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 LT |
127 | { |
128 | if (atomic_dec_and_test(&sh->count)) { | |
78bafebd ES |
129 | BUG_ON(!list_empty(&sh->lru)); |
130 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1da177e4 | 131 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
7c785b7a | 132 | if (test_bit(STRIPE_DELAYED, &sh->state)) { |
1da177e4 | 133 | list_add_tail(&sh->lru, &conf->delayed_list); |
7c785b7a N |
134 | blk_plug_device(conf->mddev->queue); |
135 | } else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && | |
ae3c20cc | 136 | sh->bm_seq - conf->seq_write > 0) { |
72626685 | 137 | list_add_tail(&sh->lru, &conf->bitmap_list); |
7c785b7a N |
138 | blk_plug_device(conf->mddev->queue); |
139 | } else { | |
72626685 | 140 | clear_bit(STRIPE_BIT_DELAY, &sh->state); |
1da177e4 | 141 | list_add_tail(&sh->lru, &conf->handle_list); |
72626685 | 142 | } |
1da177e4 LT |
143 | md_wakeup_thread(conf->mddev->thread); |
144 | } else { | |
145 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
146 | atomic_dec(&conf->preread_active_stripes); | |
147 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) | |
148 | md_wakeup_thread(conf->mddev->thread); | |
149 | } | |
1da177e4 | 150 | atomic_dec(&conf->active_stripes); |
ccfcc3c1 N |
151 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
152 | list_add_tail(&sh->lru, &conf->inactive_list); | |
1da177e4 | 153 | wake_up(&conf->wait_for_stripe); |
46031f9a RBJ |
154 | if (conf->retry_read_aligned) |
155 | md_wakeup_thread(conf->mddev->thread); | |
ccfcc3c1 | 156 | } |
1da177e4 LT |
157 | } |
158 | } | |
159 | } | |
160 | static void release_stripe(struct stripe_head *sh) | |
161 | { | |
162 | raid5_conf_t *conf = sh->raid_conf; | |
163 | unsigned long flags; | |
16a53ecc | 164 | |
1da177e4 LT |
165 | spin_lock_irqsave(&conf->device_lock, flags); |
166 | __release_stripe(conf, sh); | |
167 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
168 | } | |
169 | ||
fccddba0 | 170 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 LT |
171 | { |
172 | PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh->sector); | |
173 | ||
fccddba0 | 174 | hlist_del_init(&sh->hash); |
1da177e4 LT |
175 | } |
176 | ||
16a53ecc | 177 | static inline void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) |
1da177e4 | 178 | { |
fccddba0 | 179 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 LT |
180 | |
181 | PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh->sector); | |
182 | ||
183 | CHECK_DEVLOCK(); | |
fccddba0 | 184 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
185 | } |
186 | ||
187 | ||
188 | /* find an idle stripe, make sure it is unhashed, and return it. */ | |
189 | static struct stripe_head *get_free_stripe(raid5_conf_t *conf) | |
190 | { | |
191 | struct stripe_head *sh = NULL; | |
192 | struct list_head *first; | |
193 | ||
194 | CHECK_DEVLOCK(); | |
195 | if (list_empty(&conf->inactive_list)) | |
196 | goto out; | |
197 | first = conf->inactive_list.next; | |
198 | sh = list_entry(first, struct stripe_head, lru); | |
199 | list_del_init(first); | |
200 | remove_hash(sh); | |
201 | atomic_inc(&conf->active_stripes); | |
202 | out: | |
203 | return sh; | |
204 | } | |
205 | ||
206 | static void shrink_buffers(struct stripe_head *sh, int num) | |
207 | { | |
208 | struct page *p; | |
209 | int i; | |
210 | ||
211 | for (i=0; i<num ; i++) { | |
212 | p = sh->dev[i].page; | |
213 | if (!p) | |
214 | continue; | |
215 | sh->dev[i].page = NULL; | |
2d1f3b5d | 216 | put_page(p); |
1da177e4 LT |
217 | } |
218 | } | |
219 | ||
220 | static int grow_buffers(struct stripe_head *sh, int num) | |
221 | { | |
222 | int i; | |
223 | ||
224 | for (i=0; i<num; i++) { | |
225 | struct page *page; | |
226 | ||
227 | if (!(page = alloc_page(GFP_KERNEL))) { | |
228 | return 1; | |
229 | } | |
230 | sh->dev[i].page = page; | |
231 | } | |
232 | return 0; | |
233 | } | |
234 | ||
235 | static void raid5_build_block (struct stripe_head *sh, int i); | |
236 | ||
7ecaa1e6 | 237 | static void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx, int disks) |
1da177e4 LT |
238 | { |
239 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 240 | int i; |
1da177e4 | 241 | |
78bafebd ES |
242 | BUG_ON(atomic_read(&sh->count) != 0); |
243 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
1da177e4 LT |
244 | |
245 | CHECK_DEVLOCK(); | |
246 | PRINTK("init_stripe called, stripe %llu\n", | |
247 | (unsigned long long)sh->sector); | |
248 | ||
249 | remove_hash(sh); | |
16a53ecc | 250 | |
1da177e4 LT |
251 | sh->sector = sector; |
252 | sh->pd_idx = pd_idx; | |
253 | sh->state = 0; | |
254 | ||
7ecaa1e6 N |
255 | sh->disks = disks; |
256 | ||
257 | for (i = sh->disks; i--; ) { | |
1da177e4 LT |
258 | struct r5dev *dev = &sh->dev[i]; |
259 | ||
260 | if (dev->toread || dev->towrite || dev->written || | |
261 | test_bit(R5_LOCKED, &dev->flags)) { | |
262 | printk("sector=%llx i=%d %p %p %p %d\n", | |
263 | (unsigned long long)sh->sector, i, dev->toread, | |
264 | dev->towrite, dev->written, | |
265 | test_bit(R5_LOCKED, &dev->flags)); | |
266 | BUG(); | |
267 | } | |
268 | dev->flags = 0; | |
269 | raid5_build_block(sh, i); | |
270 | } | |
271 | insert_hash(conf, sh); | |
272 | } | |
273 | ||
7ecaa1e6 | 274 | static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector, int disks) |
1da177e4 LT |
275 | { |
276 | struct stripe_head *sh; | |
fccddba0 | 277 | struct hlist_node *hn; |
1da177e4 LT |
278 | |
279 | CHECK_DEVLOCK(); | |
280 | PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector); | |
fccddba0 | 281 | hlist_for_each_entry(sh, hn, stripe_hash(conf, sector), hash) |
7ecaa1e6 | 282 | if (sh->sector == sector && sh->disks == disks) |
1da177e4 LT |
283 | return sh; |
284 | PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector); | |
285 | return NULL; | |
286 | } | |
287 | ||
288 | static void unplug_slaves(mddev_t *mddev); | |
289 | static void raid5_unplug_device(request_queue_t *q); | |
290 | ||
7ecaa1e6 N |
291 | static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, int disks, |
292 | int pd_idx, int noblock) | |
1da177e4 LT |
293 | { |
294 | struct stripe_head *sh; | |
295 | ||
296 | PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector); | |
297 | ||
298 | spin_lock_irq(&conf->device_lock); | |
299 | ||
300 | do { | |
72626685 N |
301 | wait_event_lock_irq(conf->wait_for_stripe, |
302 | conf->quiesce == 0, | |
303 | conf->device_lock, /* nothing */); | |
7ecaa1e6 | 304 | sh = __find_stripe(conf, sector, disks); |
1da177e4 LT |
305 | if (!sh) { |
306 | if (!conf->inactive_blocked) | |
307 | sh = get_free_stripe(conf); | |
308 | if (noblock && sh == NULL) | |
309 | break; | |
310 | if (!sh) { | |
311 | conf->inactive_blocked = 1; | |
312 | wait_event_lock_irq(conf->wait_for_stripe, | |
313 | !list_empty(&conf->inactive_list) && | |
5036805b N |
314 | (atomic_read(&conf->active_stripes) |
315 | < (conf->max_nr_stripes *3/4) | |
1da177e4 LT |
316 | || !conf->inactive_blocked), |
317 | conf->device_lock, | |
f4370781 | 318 | raid5_unplug_device(conf->mddev->queue) |
1da177e4 LT |
319 | ); |
320 | conf->inactive_blocked = 0; | |
321 | } else | |
7ecaa1e6 | 322 | init_stripe(sh, sector, pd_idx, disks); |
1da177e4 LT |
323 | } else { |
324 | if (atomic_read(&sh->count)) { | |
78bafebd | 325 | BUG_ON(!list_empty(&sh->lru)); |
1da177e4 LT |
326 | } else { |
327 | if (!test_bit(STRIPE_HANDLE, &sh->state)) | |
328 | atomic_inc(&conf->active_stripes); | |
ff4e8d9a N |
329 | if (list_empty(&sh->lru) && |
330 | !test_bit(STRIPE_EXPANDING, &sh->state)) | |
16a53ecc N |
331 | BUG(); |
332 | list_del_init(&sh->lru); | |
1da177e4 LT |
333 | } |
334 | } | |
335 | } while (sh == NULL); | |
336 | ||
337 | if (sh) | |
338 | atomic_inc(&sh->count); | |
339 | ||
340 | spin_unlock_irq(&conf->device_lock); | |
341 | return sh; | |
342 | } | |
343 | ||
3f294f4f | 344 | static int grow_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
345 | { |
346 | struct stripe_head *sh; | |
3f294f4f N |
347 | sh = kmem_cache_alloc(conf->slab_cache, GFP_KERNEL); |
348 | if (!sh) | |
349 | return 0; | |
350 | memset(sh, 0, sizeof(*sh) + (conf->raid_disks-1)*sizeof(struct r5dev)); | |
351 | sh->raid_conf = conf; | |
352 | spin_lock_init(&sh->lock); | |
353 | ||
354 | if (grow_buffers(sh, conf->raid_disks)) { | |
355 | shrink_buffers(sh, conf->raid_disks); | |
356 | kmem_cache_free(conf->slab_cache, sh); | |
357 | return 0; | |
358 | } | |
7ecaa1e6 | 359 | sh->disks = conf->raid_disks; |
3f294f4f N |
360 | /* we just created an active stripe so... */ |
361 | atomic_set(&sh->count, 1); | |
362 | atomic_inc(&conf->active_stripes); | |
363 | INIT_LIST_HEAD(&sh->lru); | |
364 | release_stripe(sh); | |
365 | return 1; | |
366 | } | |
367 | ||
368 | static int grow_stripes(raid5_conf_t *conf, int num) | |
369 | { | |
e18b890b | 370 | struct kmem_cache *sc; |
1da177e4 LT |
371 | int devs = conf->raid_disks; |
372 | ||
42b9bebe N |
373 | sprintf(conf->cache_name[0], "raid5-%s", mdname(conf->mddev)); |
374 | sprintf(conf->cache_name[1], "raid5-%s-alt", mdname(conf->mddev)); | |
ad01c9e3 N |
375 | conf->active_name = 0; |
376 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 LT |
377 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
378 | 0, 0, NULL, NULL); | |
379 | if (!sc) | |
380 | return 1; | |
381 | conf->slab_cache = sc; | |
ad01c9e3 | 382 | conf->pool_size = devs; |
16a53ecc | 383 | while (num--) |
3f294f4f | 384 | if (!grow_one_stripe(conf)) |
1da177e4 | 385 | return 1; |
1da177e4 LT |
386 | return 0; |
387 | } | |
29269553 N |
388 | |
389 | #ifdef CONFIG_MD_RAID5_RESHAPE | |
ad01c9e3 N |
390 | static int resize_stripes(raid5_conf_t *conf, int newsize) |
391 | { | |
392 | /* Make all the stripes able to hold 'newsize' devices. | |
393 | * New slots in each stripe get 'page' set to a new page. | |
394 | * | |
395 | * This happens in stages: | |
396 | * 1/ create a new kmem_cache and allocate the required number of | |
397 | * stripe_heads. | |
398 | * 2/ gather all the old stripe_heads and tranfer the pages across | |
399 | * to the new stripe_heads. This will have the side effect of | |
400 | * freezing the array as once all stripe_heads have been collected, | |
401 | * no IO will be possible. Old stripe heads are freed once their | |
402 | * pages have been transferred over, and the old kmem_cache is | |
403 | * freed when all stripes are done. | |
404 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
405 | * we simple return a failre status - no need to clean anything up. | |
406 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
407 | * If this fails, we don't bother trying the shrink the | |
408 | * stripe_heads down again, we just leave them as they are. | |
409 | * As each stripe_head is processed the new one is released into | |
410 | * active service. | |
411 | * | |
412 | * Once step2 is started, we cannot afford to wait for a write, | |
413 | * so we use GFP_NOIO allocations. | |
414 | */ | |
415 | struct stripe_head *osh, *nsh; | |
416 | LIST_HEAD(newstripes); | |
417 | struct disk_info *ndisks; | |
418 | int err = 0; | |
e18b890b | 419 | struct kmem_cache *sc; |
ad01c9e3 N |
420 | int i; |
421 | ||
422 | if (newsize <= conf->pool_size) | |
423 | return 0; /* never bother to shrink */ | |
424 | ||
2a2275d6 N |
425 | md_allow_write(conf->mddev); |
426 | ||
ad01c9e3 N |
427 | /* Step 1 */ |
428 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
429 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
430 | 0, 0, NULL, NULL); | |
431 | if (!sc) | |
432 | return -ENOMEM; | |
433 | ||
434 | for (i = conf->max_nr_stripes; i; i--) { | |
435 | nsh = kmem_cache_alloc(sc, GFP_KERNEL); | |
436 | if (!nsh) | |
437 | break; | |
438 | ||
439 | memset(nsh, 0, sizeof(*nsh) + (newsize-1)*sizeof(struct r5dev)); | |
440 | ||
441 | nsh->raid_conf = conf; | |
442 | spin_lock_init(&nsh->lock); | |
443 | ||
444 | list_add(&nsh->lru, &newstripes); | |
445 | } | |
446 | if (i) { | |
447 | /* didn't get enough, give up */ | |
448 | while (!list_empty(&newstripes)) { | |
449 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
450 | list_del(&nsh->lru); | |
451 | kmem_cache_free(sc, nsh); | |
452 | } | |
453 | kmem_cache_destroy(sc); | |
454 | return -ENOMEM; | |
455 | } | |
456 | /* Step 2 - Must use GFP_NOIO now. | |
457 | * OK, we have enough stripes, start collecting inactive | |
458 | * stripes and copying them over | |
459 | */ | |
460 | list_for_each_entry(nsh, &newstripes, lru) { | |
461 | spin_lock_irq(&conf->device_lock); | |
462 | wait_event_lock_irq(conf->wait_for_stripe, | |
463 | !list_empty(&conf->inactive_list), | |
464 | conf->device_lock, | |
b3b46be3 | 465 | unplug_slaves(conf->mddev) |
ad01c9e3 N |
466 | ); |
467 | osh = get_free_stripe(conf); | |
468 | spin_unlock_irq(&conf->device_lock); | |
469 | atomic_set(&nsh->count, 1); | |
470 | for(i=0; i<conf->pool_size; i++) | |
471 | nsh->dev[i].page = osh->dev[i].page; | |
472 | for( ; i<newsize; i++) | |
473 | nsh->dev[i].page = NULL; | |
474 | kmem_cache_free(conf->slab_cache, osh); | |
475 | } | |
476 | kmem_cache_destroy(conf->slab_cache); | |
477 | ||
478 | /* Step 3. | |
479 | * At this point, we are holding all the stripes so the array | |
480 | * is completely stalled, so now is a good time to resize | |
481 | * conf->disks. | |
482 | */ | |
483 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
484 | if (ndisks) { | |
485 | for (i=0; i<conf->raid_disks; i++) | |
486 | ndisks[i] = conf->disks[i]; | |
487 | kfree(conf->disks); | |
488 | conf->disks = ndisks; | |
489 | } else | |
490 | err = -ENOMEM; | |
491 | ||
492 | /* Step 4, return new stripes to service */ | |
493 | while(!list_empty(&newstripes)) { | |
494 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
495 | list_del_init(&nsh->lru); | |
496 | for (i=conf->raid_disks; i < newsize; i++) | |
497 | if (nsh->dev[i].page == NULL) { | |
498 | struct page *p = alloc_page(GFP_NOIO); | |
499 | nsh->dev[i].page = p; | |
500 | if (!p) | |
501 | err = -ENOMEM; | |
502 | } | |
503 | release_stripe(nsh); | |
504 | } | |
505 | /* critical section pass, GFP_NOIO no longer needed */ | |
506 | ||
507 | conf->slab_cache = sc; | |
508 | conf->active_name = 1-conf->active_name; | |
509 | conf->pool_size = newsize; | |
510 | return err; | |
511 | } | |
29269553 | 512 | #endif |
1da177e4 | 513 | |
3f294f4f | 514 | static int drop_one_stripe(raid5_conf_t *conf) |
1da177e4 LT |
515 | { |
516 | struct stripe_head *sh; | |
517 | ||
3f294f4f N |
518 | spin_lock_irq(&conf->device_lock); |
519 | sh = get_free_stripe(conf); | |
520 | spin_unlock_irq(&conf->device_lock); | |
521 | if (!sh) | |
522 | return 0; | |
78bafebd | 523 | BUG_ON(atomic_read(&sh->count)); |
ad01c9e3 | 524 | shrink_buffers(sh, conf->pool_size); |
3f294f4f N |
525 | kmem_cache_free(conf->slab_cache, sh); |
526 | atomic_dec(&conf->active_stripes); | |
527 | return 1; | |
528 | } | |
529 | ||
530 | static void shrink_stripes(raid5_conf_t *conf) | |
531 | { | |
532 | while (drop_one_stripe(conf)) | |
533 | ; | |
534 | ||
29fc7e3e N |
535 | if (conf->slab_cache) |
536 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
537 | conf->slab_cache = NULL; |
538 | } | |
539 | ||
4e5314b5 | 540 | static int raid5_end_read_request(struct bio * bi, unsigned int bytes_done, |
1da177e4 LT |
541 | int error) |
542 | { | |
543 | struct stripe_head *sh = bi->bi_private; | |
544 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 545 | int disks = sh->disks, i; |
1da177e4 | 546 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 N |
547 | char b[BDEVNAME_SIZE]; |
548 | mdk_rdev_t *rdev; | |
1da177e4 LT |
549 | |
550 | if (bi->bi_size) | |
551 | return 1; | |
552 | ||
553 | for (i=0 ; i<disks; i++) | |
554 | if (bi == &sh->dev[i].req) | |
555 | break; | |
556 | ||
557 | PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n", | |
558 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
559 | uptodate); | |
560 | if (i == disks) { | |
561 | BUG(); | |
562 | return 0; | |
563 | } | |
564 | ||
565 | if (uptodate) { | |
1da177e4 | 566 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 567 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
d6950432 N |
568 | rdev = conf->disks[i].rdev; |
569 | printk(KERN_INFO "raid5:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
570 | mdname(conf->mddev), STRIPE_SECTORS, | |
571 | (unsigned long long)sh->sector + rdev->data_offset, | |
572 | bdevname(rdev->bdev, b)); | |
4e5314b5 N |
573 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
574 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
575 | } | |
ba22dcbf N |
576 | if (atomic_read(&conf->disks[i].rdev->read_errors)) |
577 | atomic_set(&conf->disks[i].rdev->read_errors, 0); | |
1da177e4 | 578 | } else { |
d6950432 | 579 | const char *bdn = bdevname(conf->disks[i].rdev->bdev, b); |
ba22dcbf | 580 | int retry = 0; |
d6950432 N |
581 | rdev = conf->disks[i].rdev; |
582 | ||
1da177e4 | 583 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 584 | atomic_inc(&rdev->read_errors); |
ba22dcbf | 585 | if (conf->mddev->degraded) |
d6950432 N |
586 | printk(KERN_WARNING "raid5:%s: read error not correctable (sector %llu on %s).\n", |
587 | mdname(conf->mddev), | |
588 | (unsigned long long)sh->sector + rdev->data_offset, | |
589 | bdn); | |
ba22dcbf | 590 | else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) |
4e5314b5 | 591 | /* Oh, no!!! */ |
d6950432 N |
592 | printk(KERN_WARNING "raid5:%s: read error NOT corrected!! (sector %llu on %s).\n", |
593 | mdname(conf->mddev), | |
594 | (unsigned long long)sh->sector + rdev->data_offset, | |
595 | bdn); | |
596 | else if (atomic_read(&rdev->read_errors) | |
ba22dcbf | 597 | > conf->max_nr_stripes) |
14f8d26b | 598 | printk(KERN_WARNING |
d6950432 N |
599 | "raid5:%s: Too many read errors, failing device %s.\n", |
600 | mdname(conf->mddev), bdn); | |
ba22dcbf N |
601 | else |
602 | retry = 1; | |
603 | if (retry) | |
604 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
605 | else { | |
4e5314b5 N |
606 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
607 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
d6950432 | 608 | md_error(conf->mddev, rdev); |
ba22dcbf | 609 | } |
1da177e4 LT |
610 | } |
611 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
1da177e4 LT |
612 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
613 | set_bit(STRIPE_HANDLE, &sh->state); | |
614 | release_stripe(sh); | |
615 | return 0; | |
616 | } | |
617 | ||
618 | static int raid5_end_write_request (struct bio *bi, unsigned int bytes_done, | |
619 | int error) | |
620 | { | |
621 | struct stripe_head *sh = bi->bi_private; | |
622 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 623 | int disks = sh->disks, i; |
1da177e4 LT |
624 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
625 | ||
626 | if (bi->bi_size) | |
627 | return 1; | |
628 | ||
629 | for (i=0 ; i<disks; i++) | |
630 | if (bi == &sh->dev[i].req) | |
631 | break; | |
632 | ||
633 | PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n", | |
634 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
635 | uptodate); | |
636 | if (i == disks) { | |
637 | BUG(); | |
638 | return 0; | |
639 | } | |
640 | ||
1da177e4 LT |
641 | if (!uptodate) |
642 | md_error(conf->mddev, conf->disks[i].rdev); | |
643 | ||
644 | rdev_dec_pending(conf->disks[i].rdev, conf->mddev); | |
645 | ||
646 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
647 | set_bit(STRIPE_HANDLE, &sh->state); | |
c04be0aa | 648 | release_stripe(sh); |
1da177e4 LT |
649 | return 0; |
650 | } | |
651 | ||
652 | ||
653 | static sector_t compute_blocknr(struct stripe_head *sh, int i); | |
654 | ||
655 | static void raid5_build_block (struct stripe_head *sh, int i) | |
656 | { | |
657 | struct r5dev *dev = &sh->dev[i]; | |
658 | ||
659 | bio_init(&dev->req); | |
660 | dev->req.bi_io_vec = &dev->vec; | |
661 | dev->req.bi_vcnt++; | |
662 | dev->req.bi_max_vecs++; | |
663 | dev->vec.bv_page = dev->page; | |
664 | dev->vec.bv_len = STRIPE_SIZE; | |
665 | dev->vec.bv_offset = 0; | |
666 | ||
667 | dev->req.bi_sector = sh->sector; | |
668 | dev->req.bi_private = sh; | |
669 | ||
670 | dev->flags = 0; | |
16a53ecc | 671 | dev->sector = compute_blocknr(sh, i); |
1da177e4 LT |
672 | } |
673 | ||
674 | static void error(mddev_t *mddev, mdk_rdev_t *rdev) | |
675 | { | |
676 | char b[BDEVNAME_SIZE]; | |
677 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
678 | PRINTK("raid5: error called\n"); | |
679 | ||
b2d444d7 | 680 | if (!test_bit(Faulty, &rdev->flags)) { |
850b2b42 | 681 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
c04be0aa N |
682 | if (test_and_clear_bit(In_sync, &rdev->flags)) { |
683 | unsigned long flags; | |
684 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 685 | mddev->degraded++; |
c04be0aa | 686 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
687 | /* |
688 | * if recovery was running, make sure it aborts. | |
689 | */ | |
690 | set_bit(MD_RECOVERY_ERR, &mddev->recovery); | |
691 | } | |
b2d444d7 | 692 | set_bit(Faulty, &rdev->flags); |
1da177e4 LT |
693 | printk (KERN_ALERT |
694 | "raid5: Disk failure on %s, disabling device." | |
695 | " Operation continuing on %d devices\n", | |
02c2de8c | 696 | bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded); |
1da177e4 | 697 | } |
16a53ecc | 698 | } |
1da177e4 LT |
699 | |
700 | /* | |
701 | * Input: a 'big' sector number, | |
702 | * Output: index of the data and parity disk, and the sector # in them. | |
703 | */ | |
704 | static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks, | |
705 | unsigned int data_disks, unsigned int * dd_idx, | |
706 | unsigned int * pd_idx, raid5_conf_t *conf) | |
707 | { | |
708 | long stripe; | |
709 | unsigned long chunk_number; | |
710 | unsigned int chunk_offset; | |
711 | sector_t new_sector; | |
712 | int sectors_per_chunk = conf->chunk_size >> 9; | |
713 | ||
714 | /* First compute the information on this sector */ | |
715 | ||
716 | /* | |
717 | * Compute the chunk number and the sector offset inside the chunk | |
718 | */ | |
719 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
720 | chunk_number = r_sector; | |
721 | BUG_ON(r_sector != chunk_number); | |
722 | ||
723 | /* | |
724 | * Compute the stripe number | |
725 | */ | |
726 | stripe = chunk_number / data_disks; | |
727 | ||
728 | /* | |
729 | * Compute the data disk and parity disk indexes inside the stripe | |
730 | */ | |
731 | *dd_idx = chunk_number % data_disks; | |
732 | ||
733 | /* | |
734 | * Select the parity disk based on the user selected algorithm. | |
735 | */ | |
16a53ecc N |
736 | switch(conf->level) { |
737 | case 4: | |
1da177e4 | 738 | *pd_idx = data_disks; |
16a53ecc N |
739 | break; |
740 | case 5: | |
741 | switch (conf->algorithm) { | |
1da177e4 LT |
742 | case ALGORITHM_LEFT_ASYMMETRIC: |
743 | *pd_idx = data_disks - stripe % raid_disks; | |
744 | if (*dd_idx >= *pd_idx) | |
745 | (*dd_idx)++; | |
746 | break; | |
747 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
748 | *pd_idx = stripe % raid_disks; | |
749 | if (*dd_idx >= *pd_idx) | |
750 | (*dd_idx)++; | |
751 | break; | |
752 | case ALGORITHM_LEFT_SYMMETRIC: | |
753 | *pd_idx = data_disks - stripe % raid_disks; | |
754 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
755 | break; | |
756 | case ALGORITHM_RIGHT_SYMMETRIC: | |
757 | *pd_idx = stripe % raid_disks; | |
758 | *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; | |
759 | break; | |
760 | default: | |
14f8d26b | 761 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
1da177e4 | 762 | conf->algorithm); |
16a53ecc N |
763 | } |
764 | break; | |
765 | case 6: | |
766 | ||
767 | /**** FIX THIS ****/ | |
768 | switch (conf->algorithm) { | |
769 | case ALGORITHM_LEFT_ASYMMETRIC: | |
770 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
771 | if (*pd_idx == raid_disks-1) | |
772 | (*dd_idx)++; /* Q D D D P */ | |
773 | else if (*dd_idx >= *pd_idx) | |
774 | (*dd_idx) += 2; /* D D P Q D */ | |
775 | break; | |
776 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
777 | *pd_idx = stripe % raid_disks; | |
778 | if (*pd_idx == raid_disks-1) | |
779 | (*dd_idx)++; /* Q D D D P */ | |
780 | else if (*dd_idx >= *pd_idx) | |
781 | (*dd_idx) += 2; /* D D P Q D */ | |
782 | break; | |
783 | case ALGORITHM_LEFT_SYMMETRIC: | |
784 | *pd_idx = raid_disks - 1 - (stripe % raid_disks); | |
785 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
786 | break; | |
787 | case ALGORITHM_RIGHT_SYMMETRIC: | |
788 | *pd_idx = stripe % raid_disks; | |
789 | *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; | |
790 | break; | |
791 | default: | |
792 | printk (KERN_CRIT "raid6: unsupported algorithm %d\n", | |
793 | conf->algorithm); | |
794 | } | |
795 | break; | |
1da177e4 LT |
796 | } |
797 | ||
798 | /* | |
799 | * Finally, compute the new sector number | |
800 | */ | |
801 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
802 | return new_sector; | |
803 | } | |
804 | ||
805 | ||
806 | static sector_t compute_blocknr(struct stripe_head *sh, int i) | |
807 | { | |
808 | raid5_conf_t *conf = sh->raid_conf; | |
b875e531 N |
809 | int raid_disks = sh->disks; |
810 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
811 | sector_t new_sector = sh->sector, check; |
812 | int sectors_per_chunk = conf->chunk_size >> 9; | |
813 | sector_t stripe; | |
814 | int chunk_offset; | |
815 | int chunk_number, dummy1, dummy2, dd_idx = i; | |
816 | sector_t r_sector; | |
817 | ||
16a53ecc | 818 | |
1da177e4 LT |
819 | chunk_offset = sector_div(new_sector, sectors_per_chunk); |
820 | stripe = new_sector; | |
821 | BUG_ON(new_sector != stripe); | |
822 | ||
16a53ecc N |
823 | if (i == sh->pd_idx) |
824 | return 0; | |
825 | switch(conf->level) { | |
826 | case 4: break; | |
827 | case 5: | |
828 | switch (conf->algorithm) { | |
1da177e4 LT |
829 | case ALGORITHM_LEFT_ASYMMETRIC: |
830 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
831 | if (i > sh->pd_idx) | |
832 | i--; | |
833 | break; | |
834 | case ALGORITHM_LEFT_SYMMETRIC: | |
835 | case ALGORITHM_RIGHT_SYMMETRIC: | |
836 | if (i < sh->pd_idx) | |
837 | i += raid_disks; | |
838 | i -= (sh->pd_idx + 1); | |
839 | break; | |
840 | default: | |
14f8d26b | 841 | printk(KERN_ERR "raid5: unsupported algorithm %d\n", |
16a53ecc N |
842 | conf->algorithm); |
843 | } | |
844 | break; | |
845 | case 6: | |
16a53ecc N |
846 | if (i == raid6_next_disk(sh->pd_idx, raid_disks)) |
847 | return 0; /* It is the Q disk */ | |
848 | switch (conf->algorithm) { | |
849 | case ALGORITHM_LEFT_ASYMMETRIC: | |
850 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
851 | if (sh->pd_idx == raid_disks-1) | |
852 | i--; /* Q D D D P */ | |
853 | else if (i > sh->pd_idx) | |
854 | i -= 2; /* D D P Q D */ | |
855 | break; | |
856 | case ALGORITHM_LEFT_SYMMETRIC: | |
857 | case ALGORITHM_RIGHT_SYMMETRIC: | |
858 | if (sh->pd_idx == raid_disks-1) | |
859 | i--; /* Q D D D P */ | |
860 | else { | |
861 | /* D D P Q D */ | |
862 | if (i < sh->pd_idx) | |
863 | i += raid_disks; | |
864 | i -= (sh->pd_idx + 2); | |
865 | } | |
866 | break; | |
867 | default: | |
868 | printk (KERN_CRIT "raid6: unsupported algorithm %d\n", | |
1da177e4 | 869 | conf->algorithm); |
16a53ecc N |
870 | } |
871 | break; | |
1da177e4 LT |
872 | } |
873 | ||
874 | chunk_number = stripe * data_disks + i; | |
875 | r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; | |
876 | ||
877 | check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf); | |
878 | if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { | |
14f8d26b | 879 | printk(KERN_ERR "compute_blocknr: map not correct\n"); |
1da177e4 LT |
880 | return 0; |
881 | } | |
882 | return r_sector; | |
883 | } | |
884 | ||
885 | ||
886 | ||
887 | /* | |
16a53ecc N |
888 | * Copy data between a page in the stripe cache, and one or more bion |
889 | * The page could align with the middle of the bio, or there could be | |
890 | * several bion, each with several bio_vecs, which cover part of the page | |
891 | * Multiple bion are linked together on bi_next. There may be extras | |
892 | * at the end of this list. We ignore them. | |
1da177e4 LT |
893 | */ |
894 | static void copy_data(int frombio, struct bio *bio, | |
895 | struct page *page, | |
896 | sector_t sector) | |
897 | { | |
898 | char *pa = page_address(page); | |
899 | struct bio_vec *bvl; | |
900 | int i; | |
901 | int page_offset; | |
902 | ||
903 | if (bio->bi_sector >= sector) | |
904 | page_offset = (signed)(bio->bi_sector - sector) * 512; | |
905 | else | |
906 | page_offset = (signed)(sector - bio->bi_sector) * -512; | |
907 | bio_for_each_segment(bvl, bio, i) { | |
908 | int len = bio_iovec_idx(bio,i)->bv_len; | |
909 | int clen; | |
910 | int b_offset = 0; | |
911 | ||
912 | if (page_offset < 0) { | |
913 | b_offset = -page_offset; | |
914 | page_offset += b_offset; | |
915 | len -= b_offset; | |
916 | } | |
917 | ||
918 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
919 | clen = STRIPE_SIZE - page_offset; | |
920 | else clen = len; | |
16a53ecc | 921 | |
1da177e4 LT |
922 | if (clen > 0) { |
923 | char *ba = __bio_kmap_atomic(bio, i, KM_USER0); | |
924 | if (frombio) | |
925 | memcpy(pa+page_offset, ba+b_offset, clen); | |
926 | else | |
927 | memcpy(ba+b_offset, pa+page_offset, clen); | |
928 | __bio_kunmap_atomic(ba, KM_USER0); | |
929 | } | |
930 | if (clen < len) /* hit end of page */ | |
931 | break; | |
932 | page_offset += len; | |
933 | } | |
934 | } | |
935 | ||
9bc89cd8 DW |
936 | #define check_xor() do { \ |
937 | if (count == MAX_XOR_BLOCKS) { \ | |
938 | xor_blocks(count, STRIPE_SIZE, dest, ptr);\ | |
939 | count = 0; \ | |
940 | } \ | |
1da177e4 LT |
941 | } while(0) |
942 | ||
943 | ||
944 | static void compute_block(struct stripe_head *sh, int dd_idx) | |
945 | { | |
7ecaa1e6 | 946 | int i, count, disks = sh->disks; |
9bc89cd8 | 947 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
1da177e4 LT |
948 | |
949 | PRINTK("compute_block, stripe %llu, idx %d\n", | |
950 | (unsigned long long)sh->sector, dd_idx); | |
951 | ||
9bc89cd8 DW |
952 | dest = page_address(sh->dev[dd_idx].page); |
953 | memset(dest, 0, STRIPE_SIZE); | |
954 | count = 0; | |
1da177e4 LT |
955 | for (i = disks ; i--; ) { |
956 | if (i == dd_idx) | |
957 | continue; | |
958 | p = page_address(sh->dev[i].page); | |
959 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
960 | ptr[count++] = p; | |
961 | else | |
14f8d26b | 962 | printk(KERN_ERR "compute_block() %d, stripe %llu, %d" |
1da177e4 LT |
963 | " not present\n", dd_idx, |
964 | (unsigned long long)sh->sector, i); | |
965 | ||
966 | check_xor(); | |
967 | } | |
9bc89cd8 DW |
968 | if (count) |
969 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
1da177e4 LT |
970 | set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
971 | } | |
972 | ||
16a53ecc | 973 | static void compute_parity5(struct stripe_head *sh, int method) |
1da177e4 LT |
974 | { |
975 | raid5_conf_t *conf = sh->raid_conf; | |
7ecaa1e6 | 976 | int i, pd_idx = sh->pd_idx, disks = sh->disks, count; |
9bc89cd8 | 977 | void *ptr[MAX_XOR_BLOCKS], *dest; |
1da177e4 LT |
978 | struct bio *chosen; |
979 | ||
16a53ecc | 980 | PRINTK("compute_parity5, stripe %llu, method %d\n", |
1da177e4 LT |
981 | (unsigned long long)sh->sector, method); |
982 | ||
9bc89cd8 DW |
983 | count = 0; |
984 | dest = page_address(sh->dev[pd_idx].page); | |
1da177e4 LT |
985 | switch(method) { |
986 | case READ_MODIFY_WRITE: | |
78bafebd | 987 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags)); |
1da177e4 LT |
988 | for (i=disks ; i-- ;) { |
989 | if (i==pd_idx) | |
990 | continue; | |
991 | if (sh->dev[i].towrite && | |
992 | test_bit(R5_UPTODATE, &sh->dev[i].flags)) { | |
993 | ptr[count++] = page_address(sh->dev[i].page); | |
994 | chosen = sh->dev[i].towrite; | |
995 | sh->dev[i].towrite = NULL; | |
996 | ||
997 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
998 | wake_up(&conf->wait_for_overlap); | |
999 | ||
78bafebd | 1000 | BUG_ON(sh->dev[i].written); |
1da177e4 LT |
1001 | sh->dev[i].written = chosen; |
1002 | check_xor(); | |
1003 | } | |
1004 | } | |
1005 | break; | |
1006 | case RECONSTRUCT_WRITE: | |
9bc89cd8 | 1007 | memset(dest, 0, STRIPE_SIZE); |
1da177e4 LT |
1008 | for (i= disks; i-- ;) |
1009 | if (i!=pd_idx && sh->dev[i].towrite) { | |
1010 | chosen = sh->dev[i].towrite; | |
1011 | sh->dev[i].towrite = NULL; | |
1012 | ||
1013 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1014 | wake_up(&conf->wait_for_overlap); | |
1015 | ||
78bafebd | 1016 | BUG_ON(sh->dev[i].written); |
1da177e4 LT |
1017 | sh->dev[i].written = chosen; |
1018 | } | |
1019 | break; | |
1020 | case CHECK_PARITY: | |
1021 | break; | |
1022 | } | |
9bc89cd8 DW |
1023 | if (count) { |
1024 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
1025 | count = 0; | |
1da177e4 LT |
1026 | } |
1027 | ||
1028 | for (i = disks; i--;) | |
1029 | if (sh->dev[i].written) { | |
1030 | sector_t sector = sh->dev[i].sector; | |
1031 | struct bio *wbi = sh->dev[i].written; | |
1032 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1033 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1034 | wbi = r5_next_bio(wbi, sector); | |
1035 | } | |
1036 | ||
1037 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1038 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1039 | } | |
1040 | ||
1041 | switch(method) { | |
1042 | case RECONSTRUCT_WRITE: | |
1043 | case CHECK_PARITY: | |
1044 | for (i=disks; i--;) | |
1045 | if (i != pd_idx) { | |
1046 | ptr[count++] = page_address(sh->dev[i].page); | |
1047 | check_xor(); | |
1048 | } | |
1049 | break; | |
1050 | case READ_MODIFY_WRITE: | |
1051 | for (i = disks; i--;) | |
1052 | if (sh->dev[i].written) { | |
1053 | ptr[count++] = page_address(sh->dev[i].page); | |
1054 | check_xor(); | |
1055 | } | |
1056 | } | |
9bc89cd8 DW |
1057 | if (count) |
1058 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
1059 | ||
1da177e4 LT |
1060 | if (method != CHECK_PARITY) { |
1061 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1062 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1063 | } else | |
1064 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1065 | } | |
1066 | ||
16a53ecc N |
1067 | static void compute_parity6(struct stripe_head *sh, int method) |
1068 | { | |
1069 | raid6_conf_t *conf = sh->raid_conf; | |
f416885e | 1070 | int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = sh->disks, count; |
16a53ecc N |
1071 | struct bio *chosen; |
1072 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1073 | void *ptrs[disks]; | |
1074 | ||
1075 | qd_idx = raid6_next_disk(pd_idx, disks); | |
1076 | d0_idx = raid6_next_disk(qd_idx, disks); | |
1077 | ||
1078 | PRINTK("compute_parity, stripe %llu, method %d\n", | |
1079 | (unsigned long long)sh->sector, method); | |
1080 | ||
1081 | switch(method) { | |
1082 | case READ_MODIFY_WRITE: | |
1083 | BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ | |
1084 | case RECONSTRUCT_WRITE: | |
1085 | for (i= disks; i-- ;) | |
1086 | if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { | |
1087 | chosen = sh->dev[i].towrite; | |
1088 | sh->dev[i].towrite = NULL; | |
1089 | ||
1090 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1091 | wake_up(&conf->wait_for_overlap); | |
1092 | ||
52e5f9d1 | 1093 | BUG_ON(sh->dev[i].written); |
16a53ecc N |
1094 | sh->dev[i].written = chosen; |
1095 | } | |
1096 | break; | |
1097 | case CHECK_PARITY: | |
1098 | BUG(); /* Not implemented yet */ | |
1099 | } | |
1100 | ||
1101 | for (i = disks; i--;) | |
1102 | if (sh->dev[i].written) { | |
1103 | sector_t sector = sh->dev[i].sector; | |
1104 | struct bio *wbi = sh->dev[i].written; | |
1105 | while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { | |
1106 | copy_data(1, wbi, sh->dev[i].page, sector); | |
1107 | wbi = r5_next_bio(wbi, sector); | |
1108 | } | |
1109 | ||
1110 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
1111 | set_bit(R5_UPTODATE, &sh->dev[i].flags); | |
1112 | } | |
1113 | ||
1114 | // switch(method) { | |
1115 | // case RECONSTRUCT_WRITE: | |
1116 | // case CHECK_PARITY: | |
1117 | // case UPDATE_PARITY: | |
1118 | /* Note that unlike RAID-5, the ordering of the disks matters greatly. */ | |
1119 | /* FIX: Is this ordering of drives even remotely optimal? */ | |
1120 | count = 0; | |
1121 | i = d0_idx; | |
1122 | do { | |
1123 | ptrs[count++] = page_address(sh->dev[i].page); | |
1124 | if (count <= disks-2 && !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1125 | printk("block %d/%d not uptodate on parity calc\n", i,count); | |
1126 | i = raid6_next_disk(i, disks); | |
1127 | } while ( i != d0_idx ); | |
1128 | // break; | |
1129 | // } | |
1130 | ||
1131 | raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs); | |
1132 | ||
1133 | switch(method) { | |
1134 | case RECONSTRUCT_WRITE: | |
1135 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1136 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1137 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
1138 | set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); | |
1139 | break; | |
1140 | case UPDATE_PARITY: | |
1141 | set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
1142 | set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); | |
1143 | break; | |
1144 | } | |
1145 | } | |
1146 | ||
1147 | ||
1148 | /* Compute one missing block */ | |
1149 | static void compute_block_1(struct stripe_head *sh, int dd_idx, int nozero) | |
1150 | { | |
f416885e | 1151 | int i, count, disks = sh->disks; |
9bc89cd8 | 1152 | void *ptr[MAX_XOR_BLOCKS], *dest, *p; |
16a53ecc N |
1153 | int pd_idx = sh->pd_idx; |
1154 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1155 | ||
1156 | PRINTK("compute_block_1, stripe %llu, idx %d\n", | |
1157 | (unsigned long long)sh->sector, dd_idx); | |
1158 | ||
1159 | if ( dd_idx == qd_idx ) { | |
1160 | /* We're actually computing the Q drive */ | |
1161 | compute_parity6(sh, UPDATE_PARITY); | |
1162 | } else { | |
9bc89cd8 DW |
1163 | dest = page_address(sh->dev[dd_idx].page); |
1164 | if (!nozero) memset(dest, 0, STRIPE_SIZE); | |
1165 | count = 0; | |
16a53ecc N |
1166 | for (i = disks ; i--; ) { |
1167 | if (i == dd_idx || i == qd_idx) | |
1168 | continue; | |
1169 | p = page_address(sh->dev[i].page); | |
1170 | if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1171 | ptr[count++] = p; | |
1172 | else | |
1173 | printk("compute_block() %d, stripe %llu, %d" | |
1174 | " not present\n", dd_idx, | |
1175 | (unsigned long long)sh->sector, i); | |
1176 | ||
1177 | check_xor(); | |
1178 | } | |
9bc89cd8 DW |
1179 | if (count) |
1180 | xor_blocks(count, STRIPE_SIZE, dest, ptr); | |
16a53ecc N |
1181 | if (!nozero) set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); |
1182 | else clear_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | /* Compute two missing blocks */ | |
1187 | static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) | |
1188 | { | |
f416885e | 1189 | int i, count, disks = sh->disks; |
16a53ecc N |
1190 | int pd_idx = sh->pd_idx; |
1191 | int qd_idx = raid6_next_disk(pd_idx, disks); | |
1192 | int d0_idx = raid6_next_disk(qd_idx, disks); | |
1193 | int faila, failb; | |
1194 | ||
1195 | /* faila and failb are disk numbers relative to d0_idx */ | |
1196 | /* pd_idx become disks-2 and qd_idx become disks-1 */ | |
1197 | faila = (dd_idx1 < d0_idx) ? dd_idx1+(disks-d0_idx) : dd_idx1-d0_idx; | |
1198 | failb = (dd_idx2 < d0_idx) ? dd_idx2+(disks-d0_idx) : dd_idx2-d0_idx; | |
1199 | ||
1200 | BUG_ON(faila == failb); | |
1201 | if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } | |
1202 | ||
1203 | PRINTK("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", | |
1204 | (unsigned long long)sh->sector, dd_idx1, dd_idx2, faila, failb); | |
1205 | ||
1206 | if ( failb == disks-1 ) { | |
1207 | /* Q disk is one of the missing disks */ | |
1208 | if ( faila == disks-2 ) { | |
1209 | /* Missing P+Q, just recompute */ | |
1210 | compute_parity6(sh, UPDATE_PARITY); | |
1211 | return; | |
1212 | } else { | |
1213 | /* We're missing D+Q; recompute D from P */ | |
1214 | compute_block_1(sh, (dd_idx1 == qd_idx) ? dd_idx2 : dd_idx1, 0); | |
1215 | compute_parity6(sh, UPDATE_PARITY); /* Is this necessary? */ | |
1216 | return; | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | /* We're missing D+P or D+D; build pointer table */ | |
1221 | { | |
1222 | /**** FIX THIS: This could be very bad if disks is close to 256 ****/ | |
1223 | void *ptrs[disks]; | |
1224 | ||
1225 | count = 0; | |
1226 | i = d0_idx; | |
1227 | do { | |
1228 | ptrs[count++] = page_address(sh->dev[i].page); | |
1229 | i = raid6_next_disk(i, disks); | |
1230 | if (i != dd_idx1 && i != dd_idx2 && | |
1231 | !test_bit(R5_UPTODATE, &sh->dev[i].flags)) | |
1232 | printk("compute_2 with missing block %d/%d\n", count, i); | |
1233 | } while ( i != d0_idx ); | |
1234 | ||
1235 | if ( failb == disks-2 ) { | |
1236 | /* We're missing D+P. */ | |
1237 | raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs); | |
1238 | } else { | |
1239 | /* We're missing D+D. */ | |
1240 | raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs); | |
1241 | } | |
1242 | ||
1243 | /* Both the above update both missing blocks */ | |
1244 | set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); | |
1245 | set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | ||
1250 | ||
1da177e4 LT |
1251 | /* |
1252 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 1253 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
1254 | * The bi_next chain must be in order. |
1255 | */ | |
1256 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) | |
1257 | { | |
1258 | struct bio **bip; | |
1259 | raid5_conf_t *conf = sh->raid_conf; | |
72626685 | 1260 | int firstwrite=0; |
1da177e4 LT |
1261 | |
1262 | PRINTK("adding bh b#%llu to stripe s#%llu\n", | |
1263 | (unsigned long long)bi->bi_sector, | |
1264 | (unsigned long long)sh->sector); | |
1265 | ||
1266 | ||
1267 | spin_lock(&sh->lock); | |
1268 | spin_lock_irq(&conf->device_lock); | |
72626685 | 1269 | if (forwrite) { |
1da177e4 | 1270 | bip = &sh->dev[dd_idx].towrite; |
72626685 N |
1271 | if (*bip == NULL && sh->dev[dd_idx].written == NULL) |
1272 | firstwrite = 1; | |
1273 | } else | |
1da177e4 LT |
1274 | bip = &sh->dev[dd_idx].toread; |
1275 | while (*bip && (*bip)->bi_sector < bi->bi_sector) { | |
1276 | if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) | |
1277 | goto overlap; | |
1278 | bip = & (*bip)->bi_next; | |
1279 | } | |
1280 | if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) | |
1281 | goto overlap; | |
1282 | ||
78bafebd | 1283 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
1284 | if (*bip) |
1285 | bi->bi_next = *bip; | |
1286 | *bip = bi; | |
1287 | bi->bi_phys_segments ++; | |
1288 | spin_unlock_irq(&conf->device_lock); | |
1289 | spin_unlock(&sh->lock); | |
1290 | ||
1291 | PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
1292 | (unsigned long long)bi->bi_sector, | |
1293 | (unsigned long long)sh->sector, dd_idx); | |
1294 | ||
72626685 | 1295 | if (conf->mddev->bitmap && firstwrite) { |
72626685 N |
1296 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
1297 | STRIPE_SECTORS, 0); | |
ae3c20cc | 1298 | sh->bm_seq = conf->seq_flush+1; |
72626685 N |
1299 | set_bit(STRIPE_BIT_DELAY, &sh->state); |
1300 | } | |
1301 | ||
1da177e4 LT |
1302 | if (forwrite) { |
1303 | /* check if page is covered */ | |
1304 | sector_t sector = sh->dev[dd_idx].sector; | |
1305 | for (bi=sh->dev[dd_idx].towrite; | |
1306 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
1307 | bi && bi->bi_sector <= sector; | |
1308 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { | |
1309 | if (bi->bi_sector + (bi->bi_size>>9) >= sector) | |
1310 | sector = bi->bi_sector + (bi->bi_size>>9); | |
1311 | } | |
1312 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
1313 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
1314 | } | |
1315 | return 1; | |
1316 | ||
1317 | overlap: | |
1318 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
1319 | spin_unlock_irq(&conf->device_lock); | |
1320 | spin_unlock(&sh->lock); | |
1321 | return 0; | |
1322 | } | |
1323 | ||
29269553 N |
1324 | static void end_reshape(raid5_conf_t *conf); |
1325 | ||
16a53ecc N |
1326 | static int page_is_zero(struct page *p) |
1327 | { | |
1328 | char *a = page_address(p); | |
1329 | return ((*(u32*)a) == 0 && | |
1330 | memcmp(a, a+4, STRIPE_SIZE-4)==0); | |
1331 | } | |
1332 | ||
ccfcc3c1 N |
1333 | static int stripe_to_pdidx(sector_t stripe, raid5_conf_t *conf, int disks) |
1334 | { | |
1335 | int sectors_per_chunk = conf->chunk_size >> 9; | |
ccfcc3c1 | 1336 | int pd_idx, dd_idx; |
2d2063ce CQH |
1337 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
1338 | ||
b875e531 N |
1339 | raid5_compute_sector(stripe * (disks - conf->max_degraded) |
1340 | *sectors_per_chunk + chunk_offset, | |
1341 | disks, disks - conf->max_degraded, | |
1342 | &dd_idx, &pd_idx, conf); | |
ccfcc3c1 N |
1343 | return pd_idx; |
1344 | } | |
1345 | ||
a4456856 DW |
1346 | static void |
1347 | handle_requests_to_failed_array(raid5_conf_t *conf, struct stripe_head *sh, | |
1348 | struct stripe_head_state *s, int disks, | |
1349 | struct bio **return_bi) | |
1350 | { | |
1351 | int i; | |
1352 | for (i = disks; i--; ) { | |
1353 | struct bio *bi; | |
1354 | int bitmap_end = 0; | |
1355 | ||
1356 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1357 | mdk_rdev_t *rdev; | |
1358 | rcu_read_lock(); | |
1359 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1360 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
1361 | /* multiple read failures in one stripe */ | |
1362 | md_error(conf->mddev, rdev); | |
1363 | rcu_read_unlock(); | |
1364 | } | |
1365 | spin_lock_irq(&conf->device_lock); | |
1366 | /* fail all writes first */ | |
1367 | bi = sh->dev[i].towrite; | |
1368 | sh->dev[i].towrite = NULL; | |
1369 | if (bi) { | |
1370 | s->to_write--; | |
1371 | bitmap_end = 1; | |
1372 | } | |
1373 | ||
1374 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1375 | wake_up(&conf->wait_for_overlap); | |
1376 | ||
1377 | while (bi && bi->bi_sector < | |
1378 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1379 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
1380 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1381 | if (--bi->bi_phys_segments == 0) { | |
1382 | md_write_end(conf->mddev); | |
1383 | bi->bi_next = *return_bi; | |
1384 | *return_bi = bi; | |
1385 | } | |
1386 | bi = nextbi; | |
1387 | } | |
1388 | /* and fail all 'written' */ | |
1389 | bi = sh->dev[i].written; | |
1390 | sh->dev[i].written = NULL; | |
1391 | if (bi) bitmap_end = 1; | |
1392 | while (bi && bi->bi_sector < | |
1393 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1394 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
1395 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1396 | if (--bi->bi_phys_segments == 0) { | |
1397 | md_write_end(conf->mddev); | |
1398 | bi->bi_next = *return_bi; | |
1399 | *return_bi = bi; | |
1400 | } | |
1401 | bi = bi2; | |
1402 | } | |
1403 | ||
1404 | /* fail any reads if this device is non-operational */ | |
1405 | if (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
1406 | test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
1407 | bi = sh->dev[i].toread; | |
1408 | sh->dev[i].toread = NULL; | |
1409 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
1410 | wake_up(&conf->wait_for_overlap); | |
1411 | if (bi) s->to_read--; | |
1412 | while (bi && bi->bi_sector < | |
1413 | sh->dev[i].sector + STRIPE_SECTORS) { | |
1414 | struct bio *nextbi = | |
1415 | r5_next_bio(bi, sh->dev[i].sector); | |
1416 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1417 | if (--bi->bi_phys_segments == 0) { | |
1418 | bi->bi_next = *return_bi; | |
1419 | *return_bi = bi; | |
1420 | } | |
1421 | bi = nextbi; | |
1422 | } | |
1423 | } | |
1424 | spin_unlock_irq(&conf->device_lock); | |
1425 | if (bitmap_end) | |
1426 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
1427 | STRIPE_SECTORS, 0, 0); | |
1428 | } | |
1429 | ||
1430 | } | |
1431 | ||
1432 | static void handle_issuing_new_read_requests5(struct stripe_head *sh, | |
1433 | struct stripe_head_state *s, int disks) | |
1434 | { | |
1435 | int i; | |
1436 | for (i = disks; i--; ) { | |
1437 | struct r5dev *dev = &sh->dev[i]; | |
1438 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1439 | !test_bit(R5_UPTODATE, &dev->flags) && | |
1440 | (dev->toread || | |
1441 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || | |
1442 | s->syncing || s->expanding || | |
1443 | (s->failed && (sh->dev[s->failed_num].toread || | |
1444 | (sh->dev[s->failed_num].towrite && | |
1445 | !test_bit(R5_OVERWRITE, &sh->dev[s->failed_num].flags)) | |
1446 | )))) { | |
1447 | /* we would like to get this block, possibly | |
1448 | * by computing it, but we might not be able to | |
1449 | */ | |
1450 | if (s->uptodate == disks-1) { | |
1451 | PRINTK("Computing block %d\n", i); | |
1452 | compute_block(sh, i); | |
1453 | s->uptodate++; | |
1454 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
1455 | set_bit(R5_LOCKED, &dev->flags); | |
1456 | set_bit(R5_Wantread, &dev->flags); | |
1457 | s->locked++; | |
1458 | PRINTK("Reading block %d (sync=%d)\n", | |
1459 | i, s->syncing); | |
1460 | } | |
1461 | } | |
1462 | } | |
1463 | set_bit(STRIPE_HANDLE, &sh->state); | |
1464 | } | |
1465 | ||
1466 | static void handle_issuing_new_read_requests6(struct stripe_head *sh, | |
1467 | struct stripe_head_state *s, struct r6_state *r6s, | |
1468 | int disks) | |
1469 | { | |
1470 | int i; | |
1471 | for (i = disks; i--; ) { | |
1472 | struct r5dev *dev = &sh->dev[i]; | |
1473 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1474 | !test_bit(R5_UPTODATE, &dev->flags) && | |
1475 | (dev->toread || (dev->towrite && | |
1476 | !test_bit(R5_OVERWRITE, &dev->flags)) || | |
1477 | s->syncing || s->expanding || | |
1478 | (s->failed >= 1 && | |
1479 | (sh->dev[r6s->failed_num[0]].toread || | |
1480 | s->to_write)) || | |
1481 | (s->failed >= 2 && | |
1482 | (sh->dev[r6s->failed_num[1]].toread || | |
1483 | s->to_write)))) { | |
1484 | /* we would like to get this block, possibly | |
1485 | * by computing it, but we might not be able to | |
1486 | */ | |
1487 | if (s->uptodate == disks-1) { | |
1488 | PRINTK("Computing stripe %llu block %d\n", | |
1489 | (unsigned long long)sh->sector, i); | |
1490 | compute_block_1(sh, i, 0); | |
1491 | s->uptodate++; | |
1492 | } else if ( s->uptodate == disks-2 && s->failed >= 2 ) { | |
1493 | /* Computing 2-failure is *very* expensive; only | |
1494 | * do it if failed >= 2 | |
1495 | */ | |
1496 | int other; | |
1497 | for (other = disks; other--; ) { | |
1498 | if (other == i) | |
1499 | continue; | |
1500 | if (!test_bit(R5_UPTODATE, | |
1501 | &sh->dev[other].flags)) | |
1502 | break; | |
1503 | } | |
1504 | BUG_ON(other < 0); | |
1505 | PRINTK("Computing stripe %llu blocks %d,%d\n", | |
1506 | (unsigned long long)sh->sector, | |
1507 | i, other); | |
1508 | compute_block_2(sh, i, other); | |
1509 | s->uptodate += 2; | |
1510 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
1511 | set_bit(R5_LOCKED, &dev->flags); | |
1512 | set_bit(R5_Wantread, &dev->flags); | |
1513 | s->locked++; | |
1514 | PRINTK("Reading block %d (sync=%d)\n", | |
1515 | i, s->syncing); | |
1516 | } | |
1517 | } | |
1518 | } | |
1519 | set_bit(STRIPE_HANDLE, &sh->state); | |
1520 | } | |
1521 | ||
1522 | ||
1523 | /* handle_completed_write_requests | |
1524 | * any written block on an uptodate or failed drive can be returned. | |
1525 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
1526 | * never LOCKED, so we don't need to test 'failed' directly. | |
1527 | */ | |
1528 | static void handle_completed_write_requests(raid5_conf_t *conf, | |
1529 | struct stripe_head *sh, int disks, struct bio **return_bi) | |
1530 | { | |
1531 | int i; | |
1532 | struct r5dev *dev; | |
1533 | ||
1534 | for (i = disks; i--; ) | |
1535 | if (sh->dev[i].written) { | |
1536 | dev = &sh->dev[i]; | |
1537 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
1538 | test_bit(R5_UPTODATE, &dev->flags)) { | |
1539 | /* We can return any write requests */ | |
1540 | struct bio *wbi, *wbi2; | |
1541 | int bitmap_end = 0; | |
1542 | PRINTK("Return write for disc %d\n", i); | |
1543 | spin_lock_irq(&conf->device_lock); | |
1544 | wbi = dev->written; | |
1545 | dev->written = NULL; | |
1546 | while (wbi && wbi->bi_sector < | |
1547 | dev->sector + STRIPE_SECTORS) { | |
1548 | wbi2 = r5_next_bio(wbi, dev->sector); | |
1549 | if (--wbi->bi_phys_segments == 0) { | |
1550 | md_write_end(conf->mddev); | |
1551 | wbi->bi_next = *return_bi; | |
1552 | *return_bi = wbi; | |
1553 | } | |
1554 | wbi = wbi2; | |
1555 | } | |
1556 | if (dev->towrite == NULL) | |
1557 | bitmap_end = 1; | |
1558 | spin_unlock_irq(&conf->device_lock); | |
1559 | if (bitmap_end) | |
1560 | bitmap_endwrite(conf->mddev->bitmap, | |
1561 | sh->sector, | |
1562 | STRIPE_SECTORS, | |
1563 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
1564 | 0); | |
1565 | } | |
1566 | } | |
1567 | } | |
1568 | ||
1569 | static void handle_issuing_new_write_requests5(raid5_conf_t *conf, | |
1570 | struct stripe_head *sh, struct stripe_head_state *s, int disks) | |
1571 | { | |
1572 | int rmw = 0, rcw = 0, i; | |
1573 | for (i = disks; i--; ) { | |
1574 | /* would I have to read this buffer for read_modify_write */ | |
1575 | struct r5dev *dev = &sh->dev[i]; | |
1576 | if ((dev->towrite || i == sh->pd_idx) && | |
1577 | !test_bit(R5_LOCKED, &dev->flags) && | |
1578 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
1579 | if (test_bit(R5_Insync, &dev->flags)) | |
1580 | rmw++; | |
1581 | else | |
1582 | rmw += 2*disks; /* cannot read it */ | |
1583 | } | |
1584 | /* Would I have to read this buffer for reconstruct_write */ | |
1585 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
1586 | !test_bit(R5_LOCKED, &dev->flags) && | |
1587 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
1588 | if (test_bit(R5_Insync, &dev->flags)) | |
1589 | rcw++; | |
1590 | else | |
1591 | rcw += 2*disks; | |
1592 | } | |
1593 | } | |
1594 | PRINTK("for sector %llu, rmw=%d rcw=%d\n", | |
1595 | (unsigned long long)sh->sector, rmw, rcw); | |
1596 | set_bit(STRIPE_HANDLE, &sh->state); | |
1597 | if (rmw < rcw && rmw > 0) | |
1598 | /* prefer read-modify-write, but need to get some data */ | |
1599 | for (i = disks; i--; ) { | |
1600 | struct r5dev *dev = &sh->dev[i]; | |
1601 | if ((dev->towrite || i == sh->pd_idx) && | |
1602 | !test_bit(R5_LOCKED, &dev->flags) && | |
1603 | !test_bit(R5_UPTODATE, &dev->flags) && | |
1604 | test_bit(R5_Insync, &dev->flags)) { | |
1605 | if ( | |
1606 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1607 | PRINTK("Read_old block " | |
1608 | "%d for r-m-w\n", i); | |
1609 | set_bit(R5_LOCKED, &dev->flags); | |
1610 | set_bit(R5_Wantread, &dev->flags); | |
1611 | s->locked++; | |
1612 | } else { | |
1613 | set_bit(STRIPE_DELAYED, &sh->state); | |
1614 | set_bit(STRIPE_HANDLE, &sh->state); | |
1615 | } | |
1616 | } | |
1617 | } | |
1618 | if (rcw <= rmw && rcw > 0) | |
1619 | /* want reconstruct write, but need to get some data */ | |
1620 | for (i = disks; i--; ) { | |
1621 | struct r5dev *dev = &sh->dev[i]; | |
1622 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
1623 | i != sh->pd_idx && | |
1624 | !test_bit(R5_LOCKED, &dev->flags) && | |
1625 | !test_bit(R5_UPTODATE, &dev->flags) && | |
1626 | test_bit(R5_Insync, &dev->flags)) { | |
1627 | if ( | |
1628 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1629 | PRINTK("Read_old block " | |
1630 | "%d for Reconstruct\n", i); | |
1631 | set_bit(R5_LOCKED, &dev->flags); | |
1632 | set_bit(R5_Wantread, &dev->flags); | |
1633 | s->locked++; | |
1634 | } else { | |
1635 | set_bit(STRIPE_DELAYED, &sh->state); | |
1636 | set_bit(STRIPE_HANDLE, &sh->state); | |
1637 | } | |
1638 | } | |
1639 | } | |
1640 | /* now if nothing is locked, and if we have enough data, | |
1641 | * we can start a write request | |
1642 | */ | |
1643 | if (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1644 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
1645 | PRINTK("Computing parity...\n"); | |
1646 | compute_parity5(sh, rcw == 0 ? | |
1647 | RECONSTRUCT_WRITE : READ_MODIFY_WRITE); | |
1648 | /* now every locked buffer is ready to be written */ | |
1649 | for (i = disks; i--; ) | |
1650 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
1651 | PRINTK("Writing block %d\n", i); | |
1652 | s->locked++; | |
1653 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
1654 | if (!test_bit(R5_Insync, &sh->dev[i].flags) | |
1655 | || (i == sh->pd_idx && s->failed == 0)) | |
1656 | set_bit(STRIPE_INSYNC, &sh->state); | |
1657 | } | |
1658 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1659 | atomic_dec(&conf->preread_active_stripes); | |
1660 | if (atomic_read(&conf->preread_active_stripes) < | |
1661 | IO_THRESHOLD) | |
1662 | md_wakeup_thread(conf->mddev->thread); | |
1663 | } | |
1664 | } | |
1665 | } | |
1666 | ||
1667 | static void handle_issuing_new_write_requests6(raid5_conf_t *conf, | |
1668 | struct stripe_head *sh, struct stripe_head_state *s, | |
1669 | struct r6_state *r6s, int disks) | |
1670 | { | |
1671 | int rcw = 0, must_compute = 0, pd_idx = sh->pd_idx, i; | |
1672 | int qd_idx = r6s->qd_idx; | |
1673 | for (i = disks; i--; ) { | |
1674 | struct r5dev *dev = &sh->dev[i]; | |
1675 | /* Would I have to read this buffer for reconstruct_write */ | |
1676 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
1677 | && i != pd_idx && i != qd_idx | |
1678 | && (!test_bit(R5_LOCKED, &dev->flags) | |
1679 | ) && | |
1680 | !test_bit(R5_UPTODATE, &dev->flags)) { | |
1681 | if (test_bit(R5_Insync, &dev->flags)) rcw++; | |
1682 | else { | |
1683 | PRINTK("raid6: must_compute: " | |
1684 | "disk %d flags=%#lx\n", i, dev->flags); | |
1685 | must_compute++; | |
1686 | } | |
1687 | } | |
1688 | } | |
1689 | PRINTK("for sector %llu, rcw=%d, must_compute=%d\n", | |
1690 | (unsigned long long)sh->sector, rcw, must_compute); | |
1691 | set_bit(STRIPE_HANDLE, &sh->state); | |
1692 | ||
1693 | if (rcw > 0) | |
1694 | /* want reconstruct write, but need to get some data */ | |
1695 | for (i = disks; i--; ) { | |
1696 | struct r5dev *dev = &sh->dev[i]; | |
1697 | if (!test_bit(R5_OVERWRITE, &dev->flags) | |
1698 | && !(s->failed == 0 && (i == pd_idx || i == qd_idx)) | |
1699 | && !test_bit(R5_LOCKED, &dev->flags) && | |
1700 | !test_bit(R5_UPTODATE, &dev->flags) && | |
1701 | test_bit(R5_Insync, &dev->flags)) { | |
1702 | if ( | |
1703 | test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1704 | PRINTK("Read_old stripe %llu " | |
1705 | "block %d for Reconstruct\n", | |
1706 | (unsigned long long)sh->sector, i); | |
1707 | set_bit(R5_LOCKED, &dev->flags); | |
1708 | set_bit(R5_Wantread, &dev->flags); | |
1709 | s->locked++; | |
1710 | } else { | |
1711 | PRINTK("Request delayed stripe %llu " | |
1712 | "block %d for Reconstruct\n", | |
1713 | (unsigned long long)sh->sector, i); | |
1714 | set_bit(STRIPE_DELAYED, &sh->state); | |
1715 | set_bit(STRIPE_HANDLE, &sh->state); | |
1716 | } | |
1717 | } | |
1718 | } | |
1719 | /* now if nothing is locked, and if we have enough data, we can start a | |
1720 | * write request | |
1721 | */ | |
1722 | if (s->locked == 0 && rcw == 0 && | |
1723 | !test_bit(STRIPE_BIT_DELAY, &sh->state)) { | |
1724 | if (must_compute > 0) { | |
1725 | /* We have failed blocks and need to compute them */ | |
1726 | switch (s->failed) { | |
1727 | case 0: | |
1728 | BUG(); | |
1729 | case 1: | |
1730 | compute_block_1(sh, r6s->failed_num[0], 0); | |
1731 | break; | |
1732 | case 2: | |
1733 | compute_block_2(sh, r6s->failed_num[0], | |
1734 | r6s->failed_num[1]); | |
1735 | break; | |
1736 | default: /* This request should have been failed? */ | |
1737 | BUG(); | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | PRINTK("Computing parity for stripe %llu\n", | |
1742 | (unsigned long long)sh->sector); | |
1743 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
1744 | /* now every locked buffer is ready to be written */ | |
1745 | for (i = disks; i--; ) | |
1746 | if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { | |
1747 | PRINTK("Writing stripe %llu block %d\n", | |
1748 | (unsigned long long)sh->sector, i); | |
1749 | s->locked++; | |
1750 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
1751 | } | |
1752 | /* after a RECONSTRUCT_WRITE, the stripe MUST be in-sync */ | |
1753 | set_bit(STRIPE_INSYNC, &sh->state); | |
1754 | ||
1755 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { | |
1756 | atomic_dec(&conf->preread_active_stripes); | |
1757 | if (atomic_read(&conf->preread_active_stripes) < | |
1758 | IO_THRESHOLD) | |
1759 | md_wakeup_thread(conf->mddev->thread); | |
1760 | } | |
1761 | } | |
1762 | } | |
1763 | ||
1764 | static void handle_parity_checks5(raid5_conf_t *conf, struct stripe_head *sh, | |
1765 | struct stripe_head_state *s, int disks) | |
1766 | { | |
1767 | set_bit(STRIPE_HANDLE, &sh->state); | |
1768 | if (s->failed == 0) { | |
1769 | BUG_ON(s->uptodate != disks); | |
1770 | compute_parity5(sh, CHECK_PARITY); | |
1771 | s->uptodate--; | |
1772 | if (page_is_zero(sh->dev[sh->pd_idx].page)) { | |
1773 | /* parity is correct (on disc, not in buffer any more) | |
1774 | */ | |
1775 | set_bit(STRIPE_INSYNC, &sh->state); | |
1776 | } else { | |
1777 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
1778 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
1779 | /* don't try to repair!! */ | |
1780 | set_bit(STRIPE_INSYNC, &sh->state); | |
1781 | else { | |
1782 | compute_block(sh, sh->pd_idx); | |
1783 | s->uptodate++; | |
1784 | } | |
1785 | } | |
1786 | } | |
1787 | if (!test_bit(STRIPE_INSYNC, &sh->state)) { | |
1788 | struct r5dev *dev; | |
1789 | /* either failed parity check, or recovery is happening */ | |
1790 | if (s->failed == 0) | |
1791 | s->failed_num = sh->pd_idx; | |
1792 | dev = &sh->dev[s->failed_num]; | |
1793 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); | |
1794 | BUG_ON(s->uptodate != disks); | |
1795 | ||
1796 | set_bit(R5_LOCKED, &dev->flags); | |
1797 | set_bit(R5_Wantwrite, &dev->flags); | |
1798 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
1799 | s->locked++; | |
1800 | set_bit(STRIPE_INSYNC, &sh->state); | |
1801 | } | |
1802 | } | |
1803 | ||
1804 | ||
1805 | static void handle_parity_checks6(raid5_conf_t *conf, struct stripe_head *sh, | |
1806 | struct stripe_head_state *s, | |
1807 | struct r6_state *r6s, struct page *tmp_page, | |
1808 | int disks) | |
1809 | { | |
1810 | int update_p = 0, update_q = 0; | |
1811 | struct r5dev *dev; | |
1812 | int pd_idx = sh->pd_idx; | |
1813 | int qd_idx = r6s->qd_idx; | |
1814 | ||
1815 | set_bit(STRIPE_HANDLE, &sh->state); | |
1816 | ||
1817 | BUG_ON(s->failed > 2); | |
1818 | BUG_ON(s->uptodate < disks); | |
1819 | /* Want to check and possibly repair P and Q. | |
1820 | * However there could be one 'failed' device, in which | |
1821 | * case we can only check one of them, possibly using the | |
1822 | * other to generate missing data | |
1823 | */ | |
1824 | ||
1825 | /* If !tmp_page, we cannot do the calculations, | |
1826 | * but as we have set STRIPE_HANDLE, we will soon be called | |
1827 | * by stripe_handle with a tmp_page - just wait until then. | |
1828 | */ | |
1829 | if (tmp_page) { | |
1830 | if (s->failed == r6s->q_failed) { | |
1831 | /* The only possible failed device holds 'Q', so it | |
1832 | * makes sense to check P (If anything else were failed, | |
1833 | * we would have used P to recreate it). | |
1834 | */ | |
1835 | compute_block_1(sh, pd_idx, 1); | |
1836 | if (!page_is_zero(sh->dev[pd_idx].page)) { | |
1837 | compute_block_1(sh, pd_idx, 0); | |
1838 | update_p = 1; | |
1839 | } | |
1840 | } | |
1841 | if (!r6s->q_failed && s->failed < 2) { | |
1842 | /* q is not failed, and we didn't use it to generate | |
1843 | * anything, so it makes sense to check it | |
1844 | */ | |
1845 | memcpy(page_address(tmp_page), | |
1846 | page_address(sh->dev[qd_idx].page), | |
1847 | STRIPE_SIZE); | |
1848 | compute_parity6(sh, UPDATE_PARITY); | |
1849 | if (memcmp(page_address(tmp_page), | |
1850 | page_address(sh->dev[qd_idx].page), | |
1851 | STRIPE_SIZE) != 0) { | |
1852 | clear_bit(STRIPE_INSYNC, &sh->state); | |
1853 | update_q = 1; | |
1854 | } | |
1855 | } | |
1856 | if (update_p || update_q) { | |
1857 | conf->mddev->resync_mismatches += STRIPE_SECTORS; | |
1858 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) | |
1859 | /* don't try to repair!! */ | |
1860 | update_p = update_q = 0; | |
1861 | } | |
1862 | ||
1863 | /* now write out any block on a failed drive, | |
1864 | * or P or Q if they need it | |
1865 | */ | |
1866 | ||
1867 | if (s->failed == 2) { | |
1868 | dev = &sh->dev[r6s->failed_num[1]]; | |
1869 | s->locked++; | |
1870 | set_bit(R5_LOCKED, &dev->flags); | |
1871 | set_bit(R5_Wantwrite, &dev->flags); | |
1872 | } | |
1873 | if (s->failed >= 1) { | |
1874 | dev = &sh->dev[r6s->failed_num[0]]; | |
1875 | s->locked++; | |
1876 | set_bit(R5_LOCKED, &dev->flags); | |
1877 | set_bit(R5_Wantwrite, &dev->flags); | |
1878 | } | |
1879 | ||
1880 | if (update_p) { | |
1881 | dev = &sh->dev[pd_idx]; | |
1882 | s->locked++; | |
1883 | set_bit(R5_LOCKED, &dev->flags); | |
1884 | set_bit(R5_Wantwrite, &dev->flags); | |
1885 | } | |
1886 | if (update_q) { | |
1887 | dev = &sh->dev[qd_idx]; | |
1888 | s->locked++; | |
1889 | set_bit(R5_LOCKED, &dev->flags); | |
1890 | set_bit(R5_Wantwrite, &dev->flags); | |
1891 | } | |
1892 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
1893 | ||
1894 | set_bit(STRIPE_INSYNC, &sh->state); | |
1895 | } | |
1896 | } | |
1897 | ||
1898 | static void handle_stripe_expansion(raid5_conf_t *conf, struct stripe_head *sh, | |
1899 | struct r6_state *r6s) | |
1900 | { | |
1901 | int i; | |
1902 | ||
1903 | /* We have read all the blocks in this stripe and now we need to | |
1904 | * copy some of them into a target stripe for expand. | |
1905 | */ | |
1906 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
1907 | for (i = 0; i < sh->disks; i++) | |
1908 | if (i != sh->pd_idx && (r6s && i != r6s->qd_idx)) { | |
1909 | int dd_idx, pd_idx, j; | |
1910 | struct stripe_head *sh2; | |
1911 | ||
1912 | sector_t bn = compute_blocknr(sh, i); | |
1913 | sector_t s = raid5_compute_sector(bn, conf->raid_disks, | |
1914 | conf->raid_disks - | |
1915 | conf->max_degraded, &dd_idx, | |
1916 | &pd_idx, conf); | |
1917 | sh2 = get_active_stripe(conf, s, conf->raid_disks, | |
1918 | pd_idx, 1); | |
1919 | if (sh2 == NULL) | |
1920 | /* so far only the early blocks of this stripe | |
1921 | * have been requested. When later blocks | |
1922 | * get requested, we will try again | |
1923 | */ | |
1924 | continue; | |
1925 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
1926 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
1927 | /* must have already done this block */ | |
1928 | release_stripe(sh2); | |
1929 | continue; | |
1930 | } | |
1931 | memcpy(page_address(sh2->dev[dd_idx].page), | |
1932 | page_address(sh->dev[i].page), | |
1933 | STRIPE_SIZE); | |
1934 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); | |
1935 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
1936 | for (j = 0; j < conf->raid_disks; j++) | |
1937 | if (j != sh2->pd_idx && | |
1938 | (r6s && j != r6s->qd_idx) && | |
1939 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) | |
1940 | break; | |
1941 | if (j == conf->raid_disks) { | |
1942 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
1943 | set_bit(STRIPE_HANDLE, &sh2->state); | |
1944 | } | |
1945 | release_stripe(sh2); | |
1946 | } | |
1947 | } | |
1da177e4 LT |
1948 | |
1949 | /* | |
1950 | * handle_stripe - do things to a stripe. | |
1951 | * | |
1952 | * We lock the stripe and then examine the state of various bits | |
1953 | * to see what needs to be done. | |
1954 | * Possible results: | |
1955 | * return some read request which now have data | |
1956 | * return some write requests which are safely on disc | |
1957 | * schedule a read on some buffers | |
1958 | * schedule a write of some buffers | |
1959 | * return confirmation of parity correctness | |
1960 | * | |
1961 | * Parity calculations are done inside the stripe lock | |
1962 | * buffers are taken off read_list or write_list, and bh_cache buffers | |
1963 | * get BH_Lock set before the stripe lock is released. | |
1964 | * | |
1965 | */ | |
a4456856 | 1966 | |
16a53ecc | 1967 | static void handle_stripe5(struct stripe_head *sh) |
1da177e4 LT |
1968 | { |
1969 | raid5_conf_t *conf = sh->raid_conf; | |
a4456856 DW |
1970 | int disks = sh->disks, i; |
1971 | struct bio *return_bi = NULL; | |
1972 | struct stripe_head_state s; | |
1da177e4 LT |
1973 | struct r5dev *dev; |
1974 | ||
a4456856 | 1975 | memset(&s, 0, sizeof(s)); |
1da177e4 LT |
1976 | PRINTK("handling stripe %llu, cnt=%d, pd_idx=%d\n", |
1977 | (unsigned long long)sh->sector, atomic_read(&sh->count), | |
1978 | sh->pd_idx); | |
1979 | ||
1980 | spin_lock(&sh->lock); | |
1981 | clear_bit(STRIPE_HANDLE, &sh->state); | |
1982 | clear_bit(STRIPE_DELAYED, &sh->state); | |
1983 | ||
a4456856 DW |
1984 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
1985 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
1986 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
1da177e4 LT |
1987 | /* Now to look around and see what can be done */ |
1988 | ||
9910f16a | 1989 | rcu_read_lock(); |
1da177e4 LT |
1990 | for (i=disks; i--; ) { |
1991 | mdk_rdev_t *rdev; | |
a4456856 | 1992 | struct r5dev *dev = &sh->dev[i]; |
1da177e4 | 1993 | clear_bit(R5_Insync, &dev->flags); |
1da177e4 LT |
1994 | |
1995 | PRINTK("check %d: state 0x%lx read %p write %p written %p\n", | |
1996 | i, dev->flags, dev->toread, dev->towrite, dev->written); | |
1997 | /* maybe we can reply to a read */ | |
1998 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
1999 | struct bio *rbi, *rbi2; | |
2000 | PRINTK("Return read for disc %d\n", i); | |
2001 | spin_lock_irq(&conf->device_lock); | |
2002 | rbi = dev->toread; | |
2003 | dev->toread = NULL; | |
2004 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2005 | wake_up(&conf->wait_for_overlap); | |
2006 | spin_unlock_irq(&conf->device_lock); | |
2007 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
2008 | copy_data(0, rbi, dev->page, dev->sector); | |
2009 | rbi2 = r5_next_bio(rbi, dev->sector); | |
2010 | spin_lock_irq(&conf->device_lock); | |
2011 | if (--rbi->bi_phys_segments == 0) { | |
2012 | rbi->bi_next = return_bi; | |
2013 | return_bi = rbi; | |
2014 | } | |
2015 | spin_unlock_irq(&conf->device_lock); | |
2016 | rbi = rbi2; | |
2017 | } | |
2018 | } | |
2019 | ||
2020 | /* now count some things */ | |
a4456856 DW |
2021 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2022 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 2023 | |
a4456856 DW |
2024 | if (dev->toread) |
2025 | s.to_read++; | |
1da177e4 | 2026 | if (dev->towrite) { |
a4456856 | 2027 | s.to_write++; |
1da177e4 | 2028 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2029 | s.non_overwrite++; |
1da177e4 | 2030 | } |
a4456856 DW |
2031 | if (dev->written) |
2032 | s.written++; | |
9910f16a | 2033 | rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 2034 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { |
14f8d26b | 2035 | /* The ReadError flag will just be confusing now */ |
4e5314b5 N |
2036 | clear_bit(R5_ReadError, &dev->flags); |
2037 | clear_bit(R5_ReWrite, &dev->flags); | |
2038 | } | |
b2d444d7 | 2039 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
4e5314b5 | 2040 | || test_bit(R5_ReadError, &dev->flags)) { |
a4456856 DW |
2041 | s.failed++; |
2042 | s.failed_num = i; | |
1da177e4 LT |
2043 | } else |
2044 | set_bit(R5_Insync, &dev->flags); | |
2045 | } | |
9910f16a | 2046 | rcu_read_unlock(); |
1da177e4 LT |
2047 | PRINTK("locked=%d uptodate=%d to_read=%d" |
2048 | " to_write=%d failed=%d failed_num=%d\n", | |
a4456856 DW |
2049 | s.locked, s.uptodate, s.to_read, s.to_write, |
2050 | s.failed, s.failed_num); | |
1da177e4 LT |
2051 | /* check if the array has lost two devices and, if so, some requests might |
2052 | * need to be failed | |
2053 | */ | |
a4456856 DW |
2054 | if (s.failed > 1 && s.to_read+s.to_write+s.written) |
2055 | handle_requests_to_failed_array(conf, sh, &s, disks, | |
2056 | &return_bi); | |
2057 | if (s.failed > 1 && s.syncing) { | |
1da177e4 LT |
2058 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2059 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2060 | s.syncing = 0; |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | /* might be able to return some write requests if the parity block | |
2064 | * is safe, or on a failed drive | |
2065 | */ | |
2066 | dev = &sh->dev[sh->pd_idx]; | |
a4456856 DW |
2067 | if ( s.written && |
2068 | ((test_bit(R5_Insync, &dev->flags) && | |
2069 | !test_bit(R5_LOCKED, &dev->flags) && | |
2070 | test_bit(R5_UPTODATE, &dev->flags)) || | |
2071 | (s.failed == 1 && s.failed_num == sh->pd_idx))) | |
2072 | handle_completed_write_requests(conf, sh, disks, &return_bi); | |
1da177e4 LT |
2073 | |
2074 | /* Now we might consider reading some blocks, either to check/generate | |
2075 | * parity, or to satisfy requests | |
2076 | * or to load a block that is being partially written. | |
2077 | */ | |
a4456856 DW |
2078 | if (s.to_read || s.non_overwrite || |
2079 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
2080 | handle_issuing_new_read_requests5(sh, &s, disks); | |
1da177e4 LT |
2081 | |
2082 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 DW |
2083 | if (s.to_write) |
2084 | handle_issuing_new_write_requests5(conf, sh, &s, disks); | |
1da177e4 LT |
2085 | |
2086 | /* maybe we need to check and possibly fix the parity for this stripe | |
2087 | * Any reads will already have been scheduled, so we just see if enough data | |
2088 | * is available | |
2089 | */ | |
a4456856 DW |
2090 | if (s.syncing && s.locked == 0 && |
2091 | !test_bit(STRIPE_INSYNC, &sh->state)) | |
2092 | handle_parity_checks5(conf, sh, &s, disks); | |
2093 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { | |
1da177e4 LT |
2094 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2095 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2096 | } | |
4e5314b5 N |
2097 | |
2098 | /* If the failed drive is just a ReadError, then we might need to progress | |
2099 | * the repair/check process | |
2100 | */ | |
a4456856 DW |
2101 | if (s.failed == 1 && !conf->mddev->ro && |
2102 | test_bit(R5_ReadError, &sh->dev[s.failed_num].flags) | |
2103 | && !test_bit(R5_LOCKED, &sh->dev[s.failed_num].flags) | |
2104 | && test_bit(R5_UPTODATE, &sh->dev[s.failed_num].flags) | |
4e5314b5 | 2105 | ) { |
a4456856 | 2106 | dev = &sh->dev[s.failed_num]; |
4e5314b5 N |
2107 | if (!test_bit(R5_ReWrite, &dev->flags)) { |
2108 | set_bit(R5_Wantwrite, &dev->flags); | |
2109 | set_bit(R5_ReWrite, &dev->flags); | |
2110 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2111 | s.locked++; |
4e5314b5 N |
2112 | } else { |
2113 | /* let's read it back */ | |
2114 | set_bit(R5_Wantread, &dev->flags); | |
2115 | set_bit(R5_LOCKED, &dev->flags); | |
a4456856 | 2116 | s.locked++; |
4e5314b5 N |
2117 | } |
2118 | } | |
2119 | ||
a4456856 | 2120 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
ccfcc3c1 N |
2121 | /* Need to write out all blocks after computing parity */ |
2122 | sh->disks = conf->raid_disks; | |
2123 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, conf->raid_disks); | |
16a53ecc | 2124 | compute_parity5(sh, RECONSTRUCT_WRITE); |
a4456856 | 2125 | for (i = conf->raid_disks; i--; ) { |
ccfcc3c1 | 2126 | set_bit(R5_LOCKED, &sh->dev[i].flags); |
a4456856 | 2127 | s.locked++; |
ccfcc3c1 N |
2128 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
2129 | } | |
2130 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 2131 | } else if (s.expanded) { |
ccfcc3c1 | 2132 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
f6705578 | 2133 | atomic_dec(&conf->reshape_stripes); |
ccfcc3c1 N |
2134 | wake_up(&conf->wait_for_overlap); |
2135 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2136 | } | |
2137 | ||
a4456856 DW |
2138 | if (s.expanding && s.locked == 0) |
2139 | handle_stripe_expansion(conf, sh, NULL); | |
ccfcc3c1 | 2140 | |
1da177e4 LT |
2141 | spin_unlock(&sh->lock); |
2142 | ||
a4456856 | 2143 | return_io(return_bi); |
1da177e4 | 2144 | |
1da177e4 LT |
2145 | for (i=disks; i-- ;) { |
2146 | int rw; | |
2147 | struct bio *bi; | |
2148 | mdk_rdev_t *rdev; | |
2149 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
802ba064 | 2150 | rw = WRITE; |
1da177e4 | 2151 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
802ba064 | 2152 | rw = READ; |
1da177e4 LT |
2153 | else |
2154 | continue; | |
2155 | ||
2156 | bi = &sh->dev[i].req; | |
2157 | ||
2158 | bi->bi_rw = rw; | |
802ba064 | 2159 | if (rw == WRITE) |
1da177e4 LT |
2160 | bi->bi_end_io = raid5_end_write_request; |
2161 | else | |
2162 | bi->bi_end_io = raid5_end_read_request; | |
2163 | ||
2164 | rcu_read_lock(); | |
d6065f7b | 2165 | rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 2166 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1da177e4 LT |
2167 | rdev = NULL; |
2168 | if (rdev) | |
2169 | atomic_inc(&rdev->nr_pending); | |
2170 | rcu_read_unlock(); | |
2171 | ||
2172 | if (rdev) { | |
a4456856 | 2173 | if (s.syncing || s.expanding || s.expanded) |
1da177e4 LT |
2174 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
2175 | ||
2176 | bi->bi_bdev = rdev->bdev; | |
2177 | PRINTK("for %llu schedule op %ld on disc %d\n", | |
2178 | (unsigned long long)sh->sector, bi->bi_rw, i); | |
2179 | atomic_inc(&sh->count); | |
2180 | bi->bi_sector = sh->sector + rdev->data_offset; | |
2181 | bi->bi_flags = 1 << BIO_UPTODATE; | |
2182 | bi->bi_vcnt = 1; | |
2183 | bi->bi_max_vecs = 1; | |
2184 | bi->bi_idx = 0; | |
2185 | bi->bi_io_vec = &sh->dev[i].vec; | |
2186 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
2187 | bi->bi_io_vec[0].bv_offset = 0; | |
2188 | bi->bi_size = STRIPE_SIZE; | |
2189 | bi->bi_next = NULL; | |
4dbcdc75 N |
2190 | if (rw == WRITE && |
2191 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
2192 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); | |
1da177e4 LT |
2193 | generic_make_request(bi); |
2194 | } else { | |
802ba064 | 2195 | if (rw == WRITE) |
72626685 | 2196 | set_bit(STRIPE_DEGRADED, &sh->state); |
1da177e4 LT |
2197 | PRINTK("skip op %ld on disc %d for sector %llu\n", |
2198 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
2199 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
2200 | set_bit(STRIPE_HANDLE, &sh->state); | |
2201 | } | |
2202 | } | |
2203 | } | |
2204 | ||
16a53ecc | 2205 | static void handle_stripe6(struct stripe_head *sh, struct page *tmp_page) |
1da177e4 | 2206 | { |
16a53ecc | 2207 | raid6_conf_t *conf = sh->raid_conf; |
f416885e | 2208 | int disks = sh->disks; |
a4456856 DW |
2209 | struct bio *return_bi = NULL; |
2210 | int i, pd_idx = sh->pd_idx; | |
2211 | struct stripe_head_state s; | |
2212 | struct r6_state r6s; | |
16a53ecc | 2213 | struct r5dev *dev, *pdev, *qdev; |
1da177e4 | 2214 | |
a4456856 DW |
2215 | r6s.qd_idx = raid6_next_disk(pd_idx, disks); |
2216 | PRINTK("handling stripe %llu, state=%#lx cnt=%d, " | |
2217 | "pd_idx=%d, qd_idx=%d\n", | |
2218 | (unsigned long long)sh->sector, sh->state, | |
2219 | atomic_read(&sh->count), pd_idx, r6s.qd_idx); | |
2220 | memset(&s, 0, sizeof(s)); | |
72626685 | 2221 | |
16a53ecc N |
2222 | spin_lock(&sh->lock); |
2223 | clear_bit(STRIPE_HANDLE, &sh->state); | |
2224 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2225 | ||
a4456856 DW |
2226 | s.syncing = test_bit(STRIPE_SYNCING, &sh->state); |
2227 | s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
2228 | s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
16a53ecc | 2229 | /* Now to look around and see what can be done */ |
1da177e4 LT |
2230 | |
2231 | rcu_read_lock(); | |
16a53ecc N |
2232 | for (i=disks; i--; ) { |
2233 | mdk_rdev_t *rdev; | |
2234 | dev = &sh->dev[i]; | |
2235 | clear_bit(R5_Insync, &dev->flags); | |
1da177e4 | 2236 | |
16a53ecc N |
2237 | PRINTK("check %d: state 0x%lx read %p write %p written %p\n", |
2238 | i, dev->flags, dev->toread, dev->towrite, dev->written); | |
2239 | /* maybe we can reply to a read */ | |
2240 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { | |
2241 | struct bio *rbi, *rbi2; | |
2242 | PRINTK("Return read for disc %d\n", i); | |
2243 | spin_lock_irq(&conf->device_lock); | |
2244 | rbi = dev->toread; | |
2245 | dev->toread = NULL; | |
2246 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2247 | wake_up(&conf->wait_for_overlap); | |
2248 | spin_unlock_irq(&conf->device_lock); | |
2249 | while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { | |
2250 | copy_data(0, rbi, dev->page, dev->sector); | |
2251 | rbi2 = r5_next_bio(rbi, dev->sector); | |
2252 | spin_lock_irq(&conf->device_lock); | |
2253 | if (--rbi->bi_phys_segments == 0) { | |
2254 | rbi->bi_next = return_bi; | |
2255 | return_bi = rbi; | |
2256 | } | |
2257 | spin_unlock_irq(&conf->device_lock); | |
2258 | rbi = rbi2; | |
2259 | } | |
2260 | } | |
1da177e4 | 2261 | |
16a53ecc | 2262 | /* now count some things */ |
a4456856 DW |
2263 | if (test_bit(R5_LOCKED, &dev->flags)) s.locked++; |
2264 | if (test_bit(R5_UPTODATE, &dev->flags)) s.uptodate++; | |
1da177e4 | 2265 | |
16a53ecc | 2266 | |
a4456856 DW |
2267 | if (dev->toread) |
2268 | s.to_read++; | |
16a53ecc | 2269 | if (dev->towrite) { |
a4456856 | 2270 | s.to_write++; |
16a53ecc | 2271 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
a4456856 | 2272 | s.non_overwrite++; |
16a53ecc | 2273 | } |
a4456856 DW |
2274 | if (dev->written) |
2275 | s.written++; | |
16a53ecc N |
2276 | rdev = rcu_dereference(conf->disks[i].rdev); |
2277 | if (!rdev || !test_bit(In_sync, &rdev->flags)) { | |
2278 | /* The ReadError flag will just be confusing now */ | |
2279 | clear_bit(R5_ReadError, &dev->flags); | |
2280 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 2281 | } |
16a53ecc N |
2282 | if (!rdev || !test_bit(In_sync, &rdev->flags) |
2283 | || test_bit(R5_ReadError, &dev->flags)) { | |
a4456856 DW |
2284 | if (s.failed < 2) |
2285 | r6s.failed_num[s.failed] = i; | |
2286 | s.failed++; | |
16a53ecc N |
2287 | } else |
2288 | set_bit(R5_Insync, &dev->flags); | |
1da177e4 LT |
2289 | } |
2290 | rcu_read_unlock(); | |
16a53ecc N |
2291 | PRINTK("locked=%d uptodate=%d to_read=%d" |
2292 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
a4456856 DW |
2293 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, |
2294 | r6s.failed_num[0], r6s.failed_num[1]); | |
2295 | /* check if the array has lost >2 devices and, if so, some requests | |
2296 | * might need to be failed | |
16a53ecc | 2297 | */ |
a4456856 DW |
2298 | if (s.failed > 2 && s.to_read+s.to_write+s.written) |
2299 | handle_requests_to_failed_array(conf, sh, &s, disks, | |
2300 | &return_bi); | |
2301 | if (s.failed > 2 && s.syncing) { | |
16a53ecc N |
2302 | md_done_sync(conf->mddev, STRIPE_SECTORS,0); |
2303 | clear_bit(STRIPE_SYNCING, &sh->state); | |
a4456856 | 2304 | s.syncing = 0; |
16a53ecc N |
2305 | } |
2306 | ||
2307 | /* | |
2308 | * might be able to return some write requests if the parity blocks | |
2309 | * are safe, or on a failed drive | |
2310 | */ | |
2311 | pdev = &sh->dev[pd_idx]; | |
a4456856 DW |
2312 | r6s.p_failed = (s.failed >= 1 && r6s.failed_num[0] == pd_idx) |
2313 | || (s.failed >= 2 && r6s.failed_num[1] == pd_idx); | |
2314 | qdev = &sh->dev[r6s.qd_idx]; | |
2315 | r6s.q_failed = (s.failed >= 1 && r6s.failed_num[0] == r6s.qd_idx) | |
2316 | || (s.failed >= 2 && r6s.failed_num[1] == r6s.qd_idx); | |
2317 | ||
2318 | if ( s.written && | |
2319 | ( r6s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
16a53ecc | 2320 | && !test_bit(R5_LOCKED, &pdev->flags) |
a4456856 DW |
2321 | && test_bit(R5_UPTODATE, &pdev->flags)))) && |
2322 | ( r6s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
16a53ecc | 2323 | && !test_bit(R5_LOCKED, &qdev->flags) |
a4456856 DW |
2324 | && test_bit(R5_UPTODATE, &qdev->flags))))) |
2325 | handle_completed_write_requests(conf, sh, disks, &return_bi); | |
16a53ecc N |
2326 | |
2327 | /* Now we might consider reading some blocks, either to check/generate | |
2328 | * parity, or to satisfy requests | |
2329 | * or to load a block that is being partially written. | |
2330 | */ | |
a4456856 DW |
2331 | if (s.to_read || s.non_overwrite || (s.to_write && s.failed) || |
2332 | (s.syncing && (s.uptodate < disks)) || s.expanding) | |
2333 | handle_issuing_new_read_requests6(sh, &s, &r6s, disks); | |
16a53ecc N |
2334 | |
2335 | /* now to consider writing and what else, if anything should be read */ | |
a4456856 DW |
2336 | if (s.to_write) |
2337 | handle_issuing_new_write_requests6(conf, sh, &s, &r6s, disks); | |
16a53ecc N |
2338 | |
2339 | /* maybe we need to check and possibly fix the parity for this stripe | |
a4456856 DW |
2340 | * Any reads will already have been scheduled, so we just see if enough |
2341 | * data is available | |
16a53ecc | 2342 | */ |
a4456856 DW |
2343 | if (s.syncing && s.locked == 0 && !test_bit(STRIPE_INSYNC, &sh->state)) |
2344 | handle_parity_checks6(conf, sh, &s, &r6s, tmp_page, disks); | |
16a53ecc | 2345 | |
a4456856 | 2346 | if (s.syncing && s.locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { |
16a53ecc N |
2347 | md_done_sync(conf->mddev, STRIPE_SECTORS,1); |
2348 | clear_bit(STRIPE_SYNCING, &sh->state); | |
2349 | } | |
2350 | ||
2351 | /* If the failed drives are just a ReadError, then we might need | |
2352 | * to progress the repair/check process | |
2353 | */ | |
a4456856 DW |
2354 | if (s.failed <= 2 && !conf->mddev->ro) |
2355 | for (i = 0; i < s.failed; i++) { | |
2356 | dev = &sh->dev[r6s.failed_num[i]]; | |
16a53ecc N |
2357 | if (test_bit(R5_ReadError, &dev->flags) |
2358 | && !test_bit(R5_LOCKED, &dev->flags) | |
2359 | && test_bit(R5_UPTODATE, &dev->flags) | |
2360 | ) { | |
2361 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
2362 | set_bit(R5_Wantwrite, &dev->flags); | |
2363 | set_bit(R5_ReWrite, &dev->flags); | |
2364 | set_bit(R5_LOCKED, &dev->flags); | |
2365 | } else { | |
2366 | /* let's read it back */ | |
2367 | set_bit(R5_Wantread, &dev->flags); | |
2368 | set_bit(R5_LOCKED, &dev->flags); | |
2369 | } | |
2370 | } | |
2371 | } | |
f416885e | 2372 | |
a4456856 | 2373 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state)) { |
f416885e N |
2374 | /* Need to write out all blocks after computing P&Q */ |
2375 | sh->disks = conf->raid_disks; | |
2376 | sh->pd_idx = stripe_to_pdidx(sh->sector, conf, | |
2377 | conf->raid_disks); | |
2378 | compute_parity6(sh, RECONSTRUCT_WRITE); | |
2379 | for (i = conf->raid_disks ; i-- ; ) { | |
2380 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 2381 | s.locked++; |
f416885e N |
2382 | set_bit(R5_Wantwrite, &sh->dev[i].flags); |
2383 | } | |
2384 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
a4456856 | 2385 | } else if (s.expanded) { |
f416885e N |
2386 | clear_bit(STRIPE_EXPAND_READY, &sh->state); |
2387 | atomic_dec(&conf->reshape_stripes); | |
2388 | wake_up(&conf->wait_for_overlap); | |
2389 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
2390 | } | |
2391 | ||
a4456856 DW |
2392 | if (s.expanding && s.locked == 0) |
2393 | handle_stripe_expansion(conf, sh, &r6s); | |
f416885e | 2394 | |
16a53ecc N |
2395 | spin_unlock(&sh->lock); |
2396 | ||
a4456856 | 2397 | return_io(return_bi); |
16a53ecc | 2398 | |
16a53ecc N |
2399 | for (i=disks; i-- ;) { |
2400 | int rw; | |
2401 | struct bio *bi; | |
2402 | mdk_rdev_t *rdev; | |
2403 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) | |
802ba064 | 2404 | rw = WRITE; |
16a53ecc | 2405 | else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
802ba064 | 2406 | rw = READ; |
16a53ecc N |
2407 | else |
2408 | continue; | |
2409 | ||
2410 | bi = &sh->dev[i].req; | |
2411 | ||
2412 | bi->bi_rw = rw; | |
802ba064 | 2413 | if (rw == WRITE) |
16a53ecc N |
2414 | bi->bi_end_io = raid5_end_write_request; |
2415 | else | |
2416 | bi->bi_end_io = raid5_end_read_request; | |
2417 | ||
2418 | rcu_read_lock(); | |
2419 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2420 | if (rdev && test_bit(Faulty, &rdev->flags)) | |
2421 | rdev = NULL; | |
2422 | if (rdev) | |
2423 | atomic_inc(&rdev->nr_pending); | |
2424 | rcu_read_unlock(); | |
2425 | ||
2426 | if (rdev) { | |
a4456856 | 2427 | if (s.syncing || s.expanding || s.expanded) |
16a53ecc N |
2428 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
2429 | ||
2430 | bi->bi_bdev = rdev->bdev; | |
2431 | PRINTK("for %llu schedule op %ld on disc %d\n", | |
2432 | (unsigned long long)sh->sector, bi->bi_rw, i); | |
2433 | atomic_inc(&sh->count); | |
2434 | bi->bi_sector = sh->sector + rdev->data_offset; | |
2435 | bi->bi_flags = 1 << BIO_UPTODATE; | |
2436 | bi->bi_vcnt = 1; | |
2437 | bi->bi_max_vecs = 1; | |
2438 | bi->bi_idx = 0; | |
2439 | bi->bi_io_vec = &sh->dev[i].vec; | |
2440 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; | |
2441 | bi->bi_io_vec[0].bv_offset = 0; | |
2442 | bi->bi_size = STRIPE_SIZE; | |
2443 | bi->bi_next = NULL; | |
2444 | if (rw == WRITE && | |
2445 | test_bit(R5_ReWrite, &sh->dev[i].flags)) | |
2446 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); | |
2447 | generic_make_request(bi); | |
2448 | } else { | |
802ba064 | 2449 | if (rw == WRITE) |
16a53ecc N |
2450 | set_bit(STRIPE_DEGRADED, &sh->state); |
2451 | PRINTK("skip op %ld on disc %d for sector %llu\n", | |
2452 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
2453 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
2454 | set_bit(STRIPE_HANDLE, &sh->state); | |
2455 | } | |
2456 | } | |
2457 | } | |
2458 | ||
2459 | static void handle_stripe(struct stripe_head *sh, struct page *tmp_page) | |
2460 | { | |
2461 | if (sh->raid_conf->level == 6) | |
2462 | handle_stripe6(sh, tmp_page); | |
2463 | else | |
2464 | handle_stripe5(sh); | |
2465 | } | |
2466 | ||
2467 | ||
2468 | ||
2469 | static void raid5_activate_delayed(raid5_conf_t *conf) | |
2470 | { | |
2471 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
2472 | while (!list_empty(&conf->delayed_list)) { | |
2473 | struct list_head *l = conf->delayed_list.next; | |
2474 | struct stripe_head *sh; | |
2475 | sh = list_entry(l, struct stripe_head, lru); | |
2476 | list_del_init(l); | |
2477 | clear_bit(STRIPE_DELAYED, &sh->state); | |
2478 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
2479 | atomic_inc(&conf->preread_active_stripes); | |
2480 | list_add_tail(&sh->lru, &conf->handle_list); | |
2481 | } | |
2482 | } | |
2483 | } | |
2484 | ||
2485 | static void activate_bit_delay(raid5_conf_t *conf) | |
2486 | { | |
2487 | /* device_lock is held */ | |
2488 | struct list_head head; | |
2489 | list_add(&head, &conf->bitmap_list); | |
2490 | list_del_init(&conf->bitmap_list); | |
2491 | while (!list_empty(&head)) { | |
2492 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
2493 | list_del_init(&sh->lru); | |
2494 | atomic_inc(&sh->count); | |
2495 | __release_stripe(conf, sh); | |
2496 | } | |
2497 | } | |
2498 | ||
2499 | static void unplug_slaves(mddev_t *mddev) | |
2500 | { | |
2501 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2502 | int i; | |
2503 | ||
2504 | rcu_read_lock(); | |
2505 | for (i=0; i<mddev->raid_disks; i++) { | |
2506 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); | |
2507 | if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) { | |
2508 | request_queue_t *r_queue = bdev_get_queue(rdev->bdev); | |
2509 | ||
2510 | atomic_inc(&rdev->nr_pending); | |
2511 | rcu_read_unlock(); | |
2512 | ||
2513 | if (r_queue->unplug_fn) | |
2514 | r_queue->unplug_fn(r_queue); | |
2515 | ||
2516 | rdev_dec_pending(rdev, mddev); | |
2517 | rcu_read_lock(); | |
2518 | } | |
2519 | } | |
2520 | rcu_read_unlock(); | |
2521 | } | |
2522 | ||
2523 | static void raid5_unplug_device(request_queue_t *q) | |
2524 | { | |
2525 | mddev_t *mddev = q->queuedata; | |
2526 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2527 | unsigned long flags; | |
2528 | ||
2529 | spin_lock_irqsave(&conf->device_lock, flags); | |
2530 | ||
2531 | if (blk_remove_plug(q)) { | |
2532 | conf->seq_flush++; | |
2533 | raid5_activate_delayed(conf); | |
72626685 | 2534 | } |
1da177e4 LT |
2535 | md_wakeup_thread(mddev->thread); |
2536 | ||
2537 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2538 | ||
2539 | unplug_slaves(mddev); | |
2540 | } | |
2541 | ||
2542 | static int raid5_issue_flush(request_queue_t *q, struct gendisk *disk, | |
2543 | sector_t *error_sector) | |
2544 | { | |
2545 | mddev_t *mddev = q->queuedata; | |
2546 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2547 | int i, ret = 0; | |
2548 | ||
2549 | rcu_read_lock(); | |
2550 | for (i=0; i<mddev->raid_disks && ret == 0; i++) { | |
d6065f7b | 2551 | mdk_rdev_t *rdev = rcu_dereference(conf->disks[i].rdev); |
b2d444d7 | 2552 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
1da177e4 LT |
2553 | struct block_device *bdev = rdev->bdev; |
2554 | request_queue_t *r_queue = bdev_get_queue(bdev); | |
2555 | ||
2556 | if (!r_queue->issue_flush_fn) | |
2557 | ret = -EOPNOTSUPP; | |
2558 | else { | |
2559 | atomic_inc(&rdev->nr_pending); | |
2560 | rcu_read_unlock(); | |
2561 | ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, | |
2562 | error_sector); | |
2563 | rdev_dec_pending(rdev, mddev); | |
2564 | rcu_read_lock(); | |
2565 | } | |
2566 | } | |
2567 | } | |
2568 | rcu_read_unlock(); | |
2569 | return ret; | |
2570 | } | |
2571 | ||
f022b2fd N |
2572 | static int raid5_congested(void *data, int bits) |
2573 | { | |
2574 | mddev_t *mddev = data; | |
2575 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2576 | ||
2577 | /* No difference between reads and writes. Just check | |
2578 | * how busy the stripe_cache is | |
2579 | */ | |
2580 | if (conf->inactive_blocked) | |
2581 | return 1; | |
2582 | if (conf->quiesce) | |
2583 | return 1; | |
2584 | if (list_empty_careful(&conf->inactive_list)) | |
2585 | return 1; | |
2586 | ||
2587 | return 0; | |
2588 | } | |
2589 | ||
23032a0e RBJ |
2590 | /* We want read requests to align with chunks where possible, |
2591 | * but write requests don't need to. | |
2592 | */ | |
2593 | static int raid5_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec) | |
2594 | { | |
2595 | mddev_t *mddev = q->queuedata; | |
2596 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
2597 | int max; | |
2598 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
2599 | unsigned int bio_sectors = bio->bi_size >> 9; | |
2600 | ||
802ba064 | 2601 | if (bio_data_dir(bio) == WRITE) |
23032a0e RBJ |
2602 | return biovec->bv_len; /* always allow writes to be mergeable */ |
2603 | ||
2604 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; | |
2605 | if (max < 0) max = 0; | |
2606 | if (max <= biovec->bv_len && bio_sectors == 0) | |
2607 | return biovec->bv_len; | |
2608 | else | |
2609 | return max; | |
2610 | } | |
2611 | ||
f679623f RBJ |
2612 | |
2613 | static int in_chunk_boundary(mddev_t *mddev, struct bio *bio) | |
2614 | { | |
2615 | sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); | |
2616 | unsigned int chunk_sectors = mddev->chunk_size >> 9; | |
2617 | unsigned int bio_sectors = bio->bi_size >> 9; | |
2618 | ||
2619 | return chunk_sectors >= | |
2620 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
2621 | } | |
2622 | ||
46031f9a RBJ |
2623 | /* |
2624 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
2625 | * later sampled by raid5d. | |
2626 | */ | |
2627 | static void add_bio_to_retry(struct bio *bi,raid5_conf_t *conf) | |
2628 | { | |
2629 | unsigned long flags; | |
2630 | ||
2631 | spin_lock_irqsave(&conf->device_lock, flags); | |
2632 | ||
2633 | bi->bi_next = conf->retry_read_aligned_list; | |
2634 | conf->retry_read_aligned_list = bi; | |
2635 | ||
2636 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2637 | md_wakeup_thread(conf->mddev->thread); | |
2638 | } | |
2639 | ||
2640 | ||
2641 | static struct bio *remove_bio_from_retry(raid5_conf_t *conf) | |
2642 | { | |
2643 | struct bio *bi; | |
2644 | ||
2645 | bi = conf->retry_read_aligned; | |
2646 | if (bi) { | |
2647 | conf->retry_read_aligned = NULL; | |
2648 | return bi; | |
2649 | } | |
2650 | bi = conf->retry_read_aligned_list; | |
2651 | if(bi) { | |
387bb173 | 2652 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a RBJ |
2653 | bi->bi_next = NULL; |
2654 | bi->bi_phys_segments = 1; /* biased count of active stripes */ | |
2655 | bi->bi_hw_segments = 0; /* count of processed stripes */ | |
2656 | } | |
2657 | ||
2658 | return bi; | |
2659 | } | |
2660 | ||
2661 | ||
f679623f RBJ |
2662 | /* |
2663 | * The "raid5_align_endio" should check if the read succeeded and if it | |
2664 | * did, call bio_endio on the original bio (having bio_put the new bio | |
2665 | * first). | |
2666 | * If the read failed.. | |
2667 | */ | |
46031f9a | 2668 | static int raid5_align_endio(struct bio *bi, unsigned int bytes, int error) |
f679623f RBJ |
2669 | { |
2670 | struct bio* raid_bi = bi->bi_private; | |
46031f9a RBJ |
2671 | mddev_t *mddev; |
2672 | raid5_conf_t *conf; | |
2673 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); | |
2674 | mdk_rdev_t *rdev; | |
2675 | ||
f679623f RBJ |
2676 | if (bi->bi_size) |
2677 | return 1; | |
2678 | bio_put(bi); | |
46031f9a RBJ |
2679 | |
2680 | mddev = raid_bi->bi_bdev->bd_disk->queue->queuedata; | |
2681 | conf = mddev_to_conf(mddev); | |
2682 | rdev = (void*)raid_bi->bi_next; | |
2683 | raid_bi->bi_next = NULL; | |
2684 | ||
2685 | rdev_dec_pending(rdev, conf->mddev); | |
2686 | ||
2687 | if (!error && uptodate) { | |
2688 | bio_endio(raid_bi, bytes, 0); | |
2689 | if (atomic_dec_and_test(&conf->active_aligned_reads)) | |
2690 | wake_up(&conf->wait_for_stripe); | |
2691 | return 0; | |
2692 | } | |
2693 | ||
2694 | ||
2695 | PRINTK("raid5_align_endio : io error...handing IO for a retry\n"); | |
2696 | ||
2697 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
2698 | return 0; |
2699 | } | |
2700 | ||
387bb173 NB |
2701 | static int bio_fits_rdev(struct bio *bi) |
2702 | { | |
2703 | request_queue_t *q = bdev_get_queue(bi->bi_bdev); | |
2704 | ||
2705 | if ((bi->bi_size>>9) > q->max_sectors) | |
2706 | return 0; | |
2707 | blk_recount_segments(q, bi); | |
2708 | if (bi->bi_phys_segments > q->max_phys_segments || | |
2709 | bi->bi_hw_segments > q->max_hw_segments) | |
2710 | return 0; | |
2711 | ||
2712 | if (q->merge_bvec_fn) | |
2713 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
2714 | * just just give up | |
2715 | */ | |
2716 | return 0; | |
2717 | ||
2718 | return 1; | |
2719 | } | |
2720 | ||
2721 | ||
f679623f RBJ |
2722 | static int chunk_aligned_read(request_queue_t *q, struct bio * raid_bio) |
2723 | { | |
2724 | mddev_t *mddev = q->queuedata; | |
2725 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
2726 | const unsigned int raid_disks = conf->raid_disks; | |
46031f9a | 2727 | const unsigned int data_disks = raid_disks - conf->max_degraded; |
f679623f RBJ |
2728 | unsigned int dd_idx, pd_idx; |
2729 | struct bio* align_bi; | |
2730 | mdk_rdev_t *rdev; | |
2731 | ||
2732 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
c20086de | 2733 | PRINTK("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
2734 | return 0; |
2735 | } | |
2736 | /* | |
2737 | * use bio_clone to make a copy of the bio | |
2738 | */ | |
2739 | align_bi = bio_clone(raid_bio, GFP_NOIO); | |
2740 | if (!align_bi) | |
2741 | return 0; | |
2742 | /* | |
2743 | * set bi_end_io to a new function, and set bi_private to the | |
2744 | * original bio. | |
2745 | */ | |
2746 | align_bi->bi_end_io = raid5_align_endio; | |
2747 | align_bi->bi_private = raid_bio; | |
2748 | /* | |
2749 | * compute position | |
2750 | */ | |
2751 | align_bi->bi_sector = raid5_compute_sector(raid_bio->bi_sector, | |
2752 | raid_disks, | |
2753 | data_disks, | |
2754 | &dd_idx, | |
2755 | &pd_idx, | |
2756 | conf); | |
2757 | ||
2758 | rcu_read_lock(); | |
2759 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
2760 | if (rdev && test_bit(In_sync, &rdev->flags)) { | |
f679623f RBJ |
2761 | atomic_inc(&rdev->nr_pending); |
2762 | rcu_read_unlock(); | |
46031f9a RBJ |
2763 | raid_bio->bi_next = (void*)rdev; |
2764 | align_bi->bi_bdev = rdev->bdev; | |
2765 | align_bi->bi_flags &= ~(1 << BIO_SEG_VALID); | |
2766 | align_bi->bi_sector += rdev->data_offset; | |
2767 | ||
387bb173 NB |
2768 | if (!bio_fits_rdev(align_bi)) { |
2769 | /* too big in some way */ | |
2770 | bio_put(align_bi); | |
2771 | rdev_dec_pending(rdev, mddev); | |
2772 | return 0; | |
2773 | } | |
2774 | ||
46031f9a RBJ |
2775 | spin_lock_irq(&conf->device_lock); |
2776 | wait_event_lock_irq(conf->wait_for_stripe, | |
2777 | conf->quiesce == 0, | |
2778 | conf->device_lock, /* nothing */); | |
2779 | atomic_inc(&conf->active_aligned_reads); | |
2780 | spin_unlock_irq(&conf->device_lock); | |
2781 | ||
f679623f RBJ |
2782 | generic_make_request(align_bi); |
2783 | return 1; | |
2784 | } else { | |
2785 | rcu_read_unlock(); | |
46031f9a | 2786 | bio_put(align_bi); |
f679623f RBJ |
2787 | return 0; |
2788 | } | |
2789 | } | |
2790 | ||
2791 | ||
7ecaa1e6 | 2792 | static int make_request(request_queue_t *q, struct bio * bi) |
1da177e4 LT |
2793 | { |
2794 | mddev_t *mddev = q->queuedata; | |
2795 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
1da177e4 LT |
2796 | unsigned int dd_idx, pd_idx; |
2797 | sector_t new_sector; | |
2798 | sector_t logical_sector, last_sector; | |
2799 | struct stripe_head *sh; | |
a362357b | 2800 | const int rw = bio_data_dir(bi); |
f6344757 | 2801 | int remaining; |
1da177e4 | 2802 | |
e5dcdd80 N |
2803 | if (unlikely(bio_barrier(bi))) { |
2804 | bio_endio(bi, bi->bi_size, -EOPNOTSUPP); | |
2805 | return 0; | |
2806 | } | |
2807 | ||
3d310eb7 | 2808 | md_write_start(mddev, bi); |
06d91a5f | 2809 | |
a362357b JA |
2810 | disk_stat_inc(mddev->gendisk, ios[rw]); |
2811 | disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bi)); | |
1da177e4 | 2812 | |
802ba064 | 2813 | if (rw == READ && |
52488615 RBJ |
2814 | mddev->reshape_position == MaxSector && |
2815 | chunk_aligned_read(q,bi)) | |
2816 | return 0; | |
2817 | ||
1da177e4 LT |
2818 | logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
2819 | last_sector = bi->bi_sector + (bi->bi_size>>9); | |
2820 | bi->bi_next = NULL; | |
2821 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 2822 | |
1da177e4 LT |
2823 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
2824 | DEFINE_WAIT(w); | |
16a53ecc | 2825 | int disks, data_disks; |
b578d55f | 2826 | |
7ecaa1e6 | 2827 | retry: |
b578d55f | 2828 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
7ecaa1e6 N |
2829 | if (likely(conf->expand_progress == MaxSector)) |
2830 | disks = conf->raid_disks; | |
2831 | else { | |
df8e7f76 N |
2832 | /* spinlock is needed as expand_progress may be |
2833 | * 64bit on a 32bit platform, and so it might be | |
2834 | * possible to see a half-updated value | |
2835 | * Ofcourse expand_progress could change after | |
2836 | * the lock is dropped, so once we get a reference | |
2837 | * to the stripe that we think it is, we will have | |
2838 | * to check again. | |
2839 | */ | |
7ecaa1e6 N |
2840 | spin_lock_irq(&conf->device_lock); |
2841 | disks = conf->raid_disks; | |
2842 | if (logical_sector >= conf->expand_progress) | |
2843 | disks = conf->previous_raid_disks; | |
b578d55f N |
2844 | else { |
2845 | if (logical_sector >= conf->expand_lo) { | |
2846 | spin_unlock_irq(&conf->device_lock); | |
2847 | schedule(); | |
2848 | goto retry; | |
2849 | } | |
2850 | } | |
7ecaa1e6 N |
2851 | spin_unlock_irq(&conf->device_lock); |
2852 | } | |
16a53ecc N |
2853 | data_disks = disks - conf->max_degraded; |
2854 | ||
2855 | new_sector = raid5_compute_sector(logical_sector, disks, data_disks, | |
7ecaa1e6 | 2856 | &dd_idx, &pd_idx, conf); |
1da177e4 LT |
2857 | PRINTK("raid5: make_request, sector %llu logical %llu\n", |
2858 | (unsigned long long)new_sector, | |
2859 | (unsigned long long)logical_sector); | |
2860 | ||
7ecaa1e6 | 2861 | sh = get_active_stripe(conf, new_sector, disks, pd_idx, (bi->bi_rw&RWA_MASK)); |
1da177e4 | 2862 | if (sh) { |
7ecaa1e6 N |
2863 | if (unlikely(conf->expand_progress != MaxSector)) { |
2864 | /* expansion might have moved on while waiting for a | |
df8e7f76 N |
2865 | * stripe, so we must do the range check again. |
2866 | * Expansion could still move past after this | |
2867 | * test, but as we are holding a reference to | |
2868 | * 'sh', we know that if that happens, | |
2869 | * STRIPE_EXPANDING will get set and the expansion | |
2870 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
2871 | */ |
2872 | int must_retry = 0; | |
2873 | spin_lock_irq(&conf->device_lock); | |
2874 | if (logical_sector < conf->expand_progress && | |
2875 | disks == conf->previous_raid_disks) | |
2876 | /* mismatch, need to try again */ | |
2877 | must_retry = 1; | |
2878 | spin_unlock_irq(&conf->device_lock); | |
2879 | if (must_retry) { | |
2880 | release_stripe(sh); | |
2881 | goto retry; | |
2882 | } | |
2883 | } | |
e464eafd N |
2884 | /* FIXME what if we get a false positive because these |
2885 | * are being updated. | |
2886 | */ | |
2887 | if (logical_sector >= mddev->suspend_lo && | |
2888 | logical_sector < mddev->suspend_hi) { | |
2889 | release_stripe(sh); | |
2890 | schedule(); | |
2891 | goto retry; | |
2892 | } | |
7ecaa1e6 N |
2893 | |
2894 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
2895 | !add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { | |
2896 | /* Stripe is busy expanding or | |
2897 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
2898 | * and wait a while |
2899 | */ | |
2900 | raid5_unplug_device(mddev->queue); | |
2901 | release_stripe(sh); | |
2902 | schedule(); | |
2903 | goto retry; | |
2904 | } | |
2905 | finish_wait(&conf->wait_for_overlap, &w); | |
16a53ecc | 2906 | handle_stripe(sh, NULL); |
1da177e4 | 2907 | release_stripe(sh); |
1da177e4 LT |
2908 | } else { |
2909 | /* cannot get stripe for read-ahead, just give-up */ | |
2910 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
2911 | finish_wait(&conf->wait_for_overlap, &w); | |
2912 | break; | |
2913 | } | |
2914 | ||
2915 | } | |
2916 | spin_lock_irq(&conf->device_lock); | |
f6344757 N |
2917 | remaining = --bi->bi_phys_segments; |
2918 | spin_unlock_irq(&conf->device_lock); | |
2919 | if (remaining == 0) { | |
1da177e4 LT |
2920 | int bytes = bi->bi_size; |
2921 | ||
16a53ecc | 2922 | if ( rw == WRITE ) |
1da177e4 LT |
2923 | md_write_end(mddev); |
2924 | bi->bi_size = 0; | |
c2b00852 N |
2925 | bi->bi_end_io(bi, bytes, |
2926 | test_bit(BIO_UPTODATE, &bi->bi_flags) | |
2927 | ? 0 : -EIO); | |
1da177e4 | 2928 | } |
1da177e4 LT |
2929 | return 0; |
2930 | } | |
2931 | ||
52c03291 | 2932 | static sector_t reshape_request(mddev_t *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 2933 | { |
52c03291 N |
2934 | /* reshaping is quite different to recovery/resync so it is |
2935 | * handled quite separately ... here. | |
2936 | * | |
2937 | * On each call to sync_request, we gather one chunk worth of | |
2938 | * destination stripes and flag them as expanding. | |
2939 | * Then we find all the source stripes and request reads. | |
2940 | * As the reads complete, handle_stripe will copy the data | |
2941 | * into the destination stripe and release that stripe. | |
2942 | */ | |
1da177e4 LT |
2943 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; |
2944 | struct stripe_head *sh; | |
ccfcc3c1 N |
2945 | int pd_idx; |
2946 | sector_t first_sector, last_sector; | |
f416885e N |
2947 | int raid_disks = conf->previous_raid_disks; |
2948 | int data_disks = raid_disks - conf->max_degraded; | |
2949 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
2950 | int i; |
2951 | int dd_idx; | |
2952 | sector_t writepos, safepos, gap; | |
2953 | ||
2954 | if (sector_nr == 0 && | |
2955 | conf->expand_progress != 0) { | |
2956 | /* restarting in the middle, skip the initial sectors */ | |
2957 | sector_nr = conf->expand_progress; | |
f416885e | 2958 | sector_div(sector_nr, new_data_disks); |
52c03291 N |
2959 | *skipped = 1; |
2960 | return sector_nr; | |
2961 | } | |
2962 | ||
2963 | /* we update the metadata when there is more than 3Meg | |
2964 | * in the block range (that is rather arbitrary, should | |
2965 | * probably be time based) or when the data about to be | |
2966 | * copied would over-write the source of the data at | |
2967 | * the front of the range. | |
2968 | * i.e. one new_stripe forward from expand_progress new_maps | |
2969 | * to after where expand_lo old_maps to | |
2970 | */ | |
2971 | writepos = conf->expand_progress + | |
f416885e N |
2972 | conf->chunk_size/512*(new_data_disks); |
2973 | sector_div(writepos, new_data_disks); | |
52c03291 | 2974 | safepos = conf->expand_lo; |
f416885e | 2975 | sector_div(safepos, data_disks); |
52c03291 N |
2976 | gap = conf->expand_progress - conf->expand_lo; |
2977 | ||
2978 | if (writepos >= safepos || | |
f416885e | 2979 | gap > (new_data_disks)*3000*2 /*3Meg*/) { |
52c03291 N |
2980 | /* Cannot proceed until we've updated the superblock... */ |
2981 | wait_event(conf->wait_for_overlap, | |
2982 | atomic_read(&conf->reshape_stripes)==0); | |
2983 | mddev->reshape_position = conf->expand_progress; | |
850b2b42 | 2984 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 2985 | md_wakeup_thread(mddev->thread); |
850b2b42 | 2986 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
52c03291 N |
2987 | kthread_should_stop()); |
2988 | spin_lock_irq(&conf->device_lock); | |
2989 | conf->expand_lo = mddev->reshape_position; | |
2990 | spin_unlock_irq(&conf->device_lock); | |
2991 | wake_up(&conf->wait_for_overlap); | |
2992 | } | |
2993 | ||
2994 | for (i=0; i < conf->chunk_size/512; i+= STRIPE_SECTORS) { | |
2995 | int j; | |
2996 | int skipped = 0; | |
2997 | pd_idx = stripe_to_pdidx(sector_nr+i, conf, conf->raid_disks); | |
2998 | sh = get_active_stripe(conf, sector_nr+i, | |
2999 | conf->raid_disks, pd_idx, 0); | |
3000 | set_bit(STRIPE_EXPANDING, &sh->state); | |
3001 | atomic_inc(&conf->reshape_stripes); | |
3002 | /* If any of this stripe is beyond the end of the old | |
3003 | * array, then we need to zero those blocks | |
3004 | */ | |
3005 | for (j=sh->disks; j--;) { | |
3006 | sector_t s; | |
3007 | if (j == sh->pd_idx) | |
3008 | continue; | |
f416885e N |
3009 | if (conf->level == 6 && |
3010 | j == raid6_next_disk(sh->pd_idx, sh->disks)) | |
3011 | continue; | |
52c03291 N |
3012 | s = compute_blocknr(sh, j); |
3013 | if (s < (mddev->array_size<<1)) { | |
3014 | skipped = 1; | |
3015 | continue; | |
3016 | } | |
3017 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
3018 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
3019 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
3020 | } | |
3021 | if (!skipped) { | |
3022 | set_bit(STRIPE_EXPAND_READY, &sh->state); | |
3023 | set_bit(STRIPE_HANDLE, &sh->state); | |
3024 | } | |
3025 | release_stripe(sh); | |
3026 | } | |
3027 | spin_lock_irq(&conf->device_lock); | |
6d3baf2e | 3028 | conf->expand_progress = (sector_nr + i) * new_data_disks; |
52c03291 N |
3029 | spin_unlock_irq(&conf->device_lock); |
3030 | /* Ok, those stripe are ready. We can start scheduling | |
3031 | * reads on the source stripes. | |
3032 | * The source stripes are determined by mapping the first and last | |
3033 | * block on the destination stripes. | |
3034 | */ | |
52c03291 | 3035 | first_sector = |
f416885e | 3036 | raid5_compute_sector(sector_nr*(new_data_disks), |
52c03291 N |
3037 | raid_disks, data_disks, |
3038 | &dd_idx, &pd_idx, conf); | |
3039 | last_sector = | |
3040 | raid5_compute_sector((sector_nr+conf->chunk_size/512) | |
f416885e | 3041 | *(new_data_disks) -1, |
52c03291 N |
3042 | raid_disks, data_disks, |
3043 | &dd_idx, &pd_idx, conf); | |
3044 | if (last_sector >= (mddev->size<<1)) | |
3045 | last_sector = (mddev->size<<1)-1; | |
3046 | while (first_sector <= last_sector) { | |
f416885e N |
3047 | pd_idx = stripe_to_pdidx(first_sector, conf, |
3048 | conf->previous_raid_disks); | |
52c03291 N |
3049 | sh = get_active_stripe(conf, first_sector, |
3050 | conf->previous_raid_disks, pd_idx, 0); | |
3051 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); | |
3052 | set_bit(STRIPE_HANDLE, &sh->state); | |
3053 | release_stripe(sh); | |
3054 | first_sector += STRIPE_SECTORS; | |
3055 | } | |
3056 | return conf->chunk_size>>9; | |
3057 | } | |
3058 | ||
3059 | /* FIXME go_faster isn't used */ | |
3060 | static inline sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster) | |
3061 | { | |
3062 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3063 | struct stripe_head *sh; | |
3064 | int pd_idx; | |
1da177e4 | 3065 | int raid_disks = conf->raid_disks; |
72626685 N |
3066 | sector_t max_sector = mddev->size << 1; |
3067 | int sync_blocks; | |
16a53ecc N |
3068 | int still_degraded = 0; |
3069 | int i; | |
1da177e4 | 3070 | |
72626685 | 3071 | if (sector_nr >= max_sector) { |
1da177e4 LT |
3072 | /* just being told to finish up .. nothing much to do */ |
3073 | unplug_slaves(mddev); | |
29269553 N |
3074 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3075 | end_reshape(conf); | |
3076 | return 0; | |
3077 | } | |
72626685 N |
3078 | |
3079 | if (mddev->curr_resync < max_sector) /* aborted */ | |
3080 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3081 | &sync_blocks, 1); | |
16a53ecc | 3082 | else /* completed sync */ |
72626685 N |
3083 | conf->fullsync = 0; |
3084 | bitmap_close_sync(mddev->bitmap); | |
3085 | ||
1da177e4 LT |
3086 | return 0; |
3087 | } | |
ccfcc3c1 | 3088 | |
52c03291 N |
3089 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
3090 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 3091 | |
16a53ecc | 3092 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
3093 | * to resync, then assert that we are finished, because there is |
3094 | * nothing we can do. | |
3095 | */ | |
3285edf1 | 3096 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 3097 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
57afd89f N |
3098 | sector_t rv = (mddev->size << 1) - sector_nr; |
3099 | *skipped = 1; | |
1da177e4 LT |
3100 | return rv; |
3101 | } | |
72626685 | 3102 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
3855ad9f | 3103 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
72626685 N |
3104 | !conf->fullsync && sync_blocks >= STRIPE_SECTORS) { |
3105 | /* we can skip this block, and probably more */ | |
3106 | sync_blocks /= STRIPE_SECTORS; | |
3107 | *skipped = 1; | |
3108 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
3109 | } | |
1da177e4 | 3110 | |
ccfcc3c1 | 3111 | pd_idx = stripe_to_pdidx(sector_nr, conf, raid_disks); |
7ecaa1e6 | 3112 | sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 1); |
1da177e4 | 3113 | if (sh == NULL) { |
7ecaa1e6 | 3114 | sh = get_active_stripe(conf, sector_nr, raid_disks, pd_idx, 0); |
1da177e4 | 3115 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 3116 | * is trying to get access |
1da177e4 | 3117 | */ |
66c006a5 | 3118 | schedule_timeout_uninterruptible(1); |
1da177e4 | 3119 | } |
16a53ecc N |
3120 | /* Need to check if array will still be degraded after recovery/resync |
3121 | * We don't need to check the 'failed' flag as when that gets set, | |
3122 | * recovery aborts. | |
3123 | */ | |
3124 | for (i=0; i<mddev->raid_disks; i++) | |
3125 | if (conf->disks[i].rdev == NULL) | |
3126 | still_degraded = 1; | |
3127 | ||
3128 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
3129 | ||
3130 | spin_lock(&sh->lock); | |
1da177e4 LT |
3131 | set_bit(STRIPE_SYNCING, &sh->state); |
3132 | clear_bit(STRIPE_INSYNC, &sh->state); | |
3133 | spin_unlock(&sh->lock); | |
3134 | ||
16a53ecc | 3135 | handle_stripe(sh, NULL); |
1da177e4 LT |
3136 | release_stripe(sh); |
3137 | ||
3138 | return STRIPE_SECTORS; | |
3139 | } | |
3140 | ||
46031f9a RBJ |
3141 | static int retry_aligned_read(raid5_conf_t *conf, struct bio *raid_bio) |
3142 | { | |
3143 | /* We may not be able to submit a whole bio at once as there | |
3144 | * may not be enough stripe_heads available. | |
3145 | * We cannot pre-allocate enough stripe_heads as we may need | |
3146 | * more than exist in the cache (if we allow ever large chunks). | |
3147 | * So we do one stripe head at a time and record in | |
3148 | * ->bi_hw_segments how many have been done. | |
3149 | * | |
3150 | * We *know* that this entire raid_bio is in one chunk, so | |
3151 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
3152 | */ | |
3153 | struct stripe_head *sh; | |
3154 | int dd_idx, pd_idx; | |
3155 | sector_t sector, logical_sector, last_sector; | |
3156 | int scnt = 0; | |
3157 | int remaining; | |
3158 | int handled = 0; | |
3159 | ||
3160 | logical_sector = raid_bio->bi_sector & ~((sector_t)STRIPE_SECTORS-1); | |
3161 | sector = raid5_compute_sector( logical_sector, | |
3162 | conf->raid_disks, | |
3163 | conf->raid_disks - conf->max_degraded, | |
3164 | &dd_idx, | |
3165 | &pd_idx, | |
3166 | conf); | |
3167 | last_sector = raid_bio->bi_sector + (raid_bio->bi_size>>9); | |
3168 | ||
3169 | for (; logical_sector < last_sector; | |
387bb173 NB |
3170 | logical_sector += STRIPE_SECTORS, |
3171 | sector += STRIPE_SECTORS, | |
3172 | scnt++) { | |
46031f9a RBJ |
3173 | |
3174 | if (scnt < raid_bio->bi_hw_segments) | |
3175 | /* already done this stripe */ | |
3176 | continue; | |
3177 | ||
3178 | sh = get_active_stripe(conf, sector, conf->raid_disks, pd_idx, 1); | |
3179 | ||
3180 | if (!sh) { | |
3181 | /* failed to get a stripe - must wait */ | |
3182 | raid_bio->bi_hw_segments = scnt; | |
3183 | conf->retry_read_aligned = raid_bio; | |
3184 | return handled; | |
3185 | } | |
3186 | ||
3187 | set_bit(R5_ReadError, &sh->dev[dd_idx].flags); | |
387bb173 NB |
3188 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0)) { |
3189 | release_stripe(sh); | |
3190 | raid_bio->bi_hw_segments = scnt; | |
3191 | conf->retry_read_aligned = raid_bio; | |
3192 | return handled; | |
3193 | } | |
3194 | ||
46031f9a RBJ |
3195 | handle_stripe(sh, NULL); |
3196 | release_stripe(sh); | |
3197 | handled++; | |
3198 | } | |
3199 | spin_lock_irq(&conf->device_lock); | |
3200 | remaining = --raid_bio->bi_phys_segments; | |
3201 | spin_unlock_irq(&conf->device_lock); | |
3202 | if (remaining == 0) { | |
3203 | int bytes = raid_bio->bi_size; | |
3204 | ||
3205 | raid_bio->bi_size = 0; | |
c2b00852 N |
3206 | raid_bio->bi_end_io(raid_bio, bytes, |
3207 | test_bit(BIO_UPTODATE, &raid_bio->bi_flags) | |
3208 | ? 0 : -EIO); | |
46031f9a RBJ |
3209 | } |
3210 | if (atomic_dec_and_test(&conf->active_aligned_reads)) | |
3211 | wake_up(&conf->wait_for_stripe); | |
3212 | return handled; | |
3213 | } | |
3214 | ||
3215 | ||
3216 | ||
1da177e4 LT |
3217 | /* |
3218 | * This is our raid5 kernel thread. | |
3219 | * | |
3220 | * We scan the hash table for stripes which can be handled now. | |
3221 | * During the scan, completed stripes are saved for us by the interrupt | |
3222 | * handler, so that they will not have to wait for our next wakeup. | |
3223 | */ | |
3224 | static void raid5d (mddev_t *mddev) | |
3225 | { | |
3226 | struct stripe_head *sh; | |
3227 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3228 | int handled; | |
3229 | ||
3230 | PRINTK("+++ raid5d active\n"); | |
3231 | ||
3232 | md_check_recovery(mddev); | |
1da177e4 LT |
3233 | |
3234 | handled = 0; | |
3235 | spin_lock_irq(&conf->device_lock); | |
3236 | while (1) { | |
3237 | struct list_head *first; | |
46031f9a | 3238 | struct bio *bio; |
1da177e4 | 3239 | |
ae3c20cc | 3240 | if (conf->seq_flush != conf->seq_write) { |
72626685 | 3241 | int seq = conf->seq_flush; |
700e432d | 3242 | spin_unlock_irq(&conf->device_lock); |
72626685 | 3243 | bitmap_unplug(mddev->bitmap); |
700e432d | 3244 | spin_lock_irq(&conf->device_lock); |
72626685 N |
3245 | conf->seq_write = seq; |
3246 | activate_bit_delay(conf); | |
3247 | } | |
3248 | ||
1da177e4 LT |
3249 | if (list_empty(&conf->handle_list) && |
3250 | atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD && | |
3251 | !blk_queue_plugged(mddev->queue) && | |
3252 | !list_empty(&conf->delayed_list)) | |
3253 | raid5_activate_delayed(conf); | |
3254 | ||
46031f9a RBJ |
3255 | while ((bio = remove_bio_from_retry(conf))) { |
3256 | int ok; | |
3257 | spin_unlock_irq(&conf->device_lock); | |
3258 | ok = retry_aligned_read(conf, bio); | |
3259 | spin_lock_irq(&conf->device_lock); | |
3260 | if (!ok) | |
3261 | break; | |
3262 | handled++; | |
3263 | } | |
3264 | ||
1da177e4 LT |
3265 | if (list_empty(&conf->handle_list)) |
3266 | break; | |
3267 | ||
3268 | first = conf->handle_list.next; | |
3269 | sh = list_entry(first, struct stripe_head, lru); | |
3270 | ||
3271 | list_del_init(first); | |
3272 | atomic_inc(&sh->count); | |
78bafebd | 3273 | BUG_ON(atomic_read(&sh->count)!= 1); |
1da177e4 LT |
3274 | spin_unlock_irq(&conf->device_lock); |
3275 | ||
3276 | handled++; | |
16a53ecc | 3277 | handle_stripe(sh, conf->spare_page); |
1da177e4 LT |
3278 | release_stripe(sh); |
3279 | ||
3280 | spin_lock_irq(&conf->device_lock); | |
3281 | } | |
3282 | PRINTK("%d stripes handled\n", handled); | |
3283 | ||
3284 | spin_unlock_irq(&conf->device_lock); | |
3285 | ||
3286 | unplug_slaves(mddev); | |
3287 | ||
3288 | PRINTK("--- raid5d inactive\n"); | |
3289 | } | |
3290 | ||
3f294f4f | 3291 | static ssize_t |
007583c9 | 3292 | raid5_show_stripe_cache_size(mddev_t *mddev, char *page) |
3f294f4f | 3293 | { |
007583c9 | 3294 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
3295 | if (conf) |
3296 | return sprintf(page, "%d\n", conf->max_nr_stripes); | |
3297 | else | |
3298 | return 0; | |
3f294f4f N |
3299 | } |
3300 | ||
3301 | static ssize_t | |
007583c9 | 3302 | raid5_store_stripe_cache_size(mddev_t *mddev, const char *page, size_t len) |
3f294f4f | 3303 | { |
007583c9 | 3304 | raid5_conf_t *conf = mddev_to_conf(mddev); |
3f294f4f N |
3305 | char *end; |
3306 | int new; | |
3307 | if (len >= PAGE_SIZE) | |
3308 | return -EINVAL; | |
96de1e66 N |
3309 | if (!conf) |
3310 | return -ENODEV; | |
3f294f4f N |
3311 | |
3312 | new = simple_strtoul(page, &end, 10); | |
3313 | if (!*page || (*end && *end != '\n') ) | |
3314 | return -EINVAL; | |
3315 | if (new <= 16 || new > 32768) | |
3316 | return -EINVAL; | |
3317 | while (new < conf->max_nr_stripes) { | |
3318 | if (drop_one_stripe(conf)) | |
3319 | conf->max_nr_stripes--; | |
3320 | else | |
3321 | break; | |
3322 | } | |
2a2275d6 | 3323 | md_allow_write(mddev); |
3f294f4f N |
3324 | while (new > conf->max_nr_stripes) { |
3325 | if (grow_one_stripe(conf)) | |
3326 | conf->max_nr_stripes++; | |
3327 | else break; | |
3328 | } | |
3329 | return len; | |
3330 | } | |
007583c9 | 3331 | |
96de1e66 N |
3332 | static struct md_sysfs_entry |
3333 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
3334 | raid5_show_stripe_cache_size, | |
3335 | raid5_store_stripe_cache_size); | |
3f294f4f N |
3336 | |
3337 | static ssize_t | |
96de1e66 | 3338 | stripe_cache_active_show(mddev_t *mddev, char *page) |
3f294f4f | 3339 | { |
007583c9 | 3340 | raid5_conf_t *conf = mddev_to_conf(mddev); |
96de1e66 N |
3341 | if (conf) |
3342 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
3343 | else | |
3344 | return 0; | |
3f294f4f N |
3345 | } |
3346 | ||
96de1e66 N |
3347 | static struct md_sysfs_entry |
3348 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 3349 | |
007583c9 | 3350 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
3351 | &raid5_stripecache_size.attr, |
3352 | &raid5_stripecache_active.attr, | |
3353 | NULL, | |
3354 | }; | |
007583c9 N |
3355 | static struct attribute_group raid5_attrs_group = { |
3356 | .name = NULL, | |
3357 | .attrs = raid5_attrs, | |
3f294f4f N |
3358 | }; |
3359 | ||
72626685 | 3360 | static int run(mddev_t *mddev) |
1da177e4 LT |
3361 | { |
3362 | raid5_conf_t *conf; | |
3363 | int raid_disk, memory; | |
3364 | mdk_rdev_t *rdev; | |
3365 | struct disk_info *disk; | |
3366 | struct list_head *tmp; | |
02c2de8c | 3367 | int working_disks = 0; |
1da177e4 | 3368 | |
16a53ecc N |
3369 | if (mddev->level != 5 && mddev->level != 4 && mddev->level != 6) { |
3370 | printk(KERN_ERR "raid5: %s: raid level not set to 4/5/6 (%d)\n", | |
14f8d26b | 3371 | mdname(mddev), mddev->level); |
1da177e4 LT |
3372 | return -EIO; |
3373 | } | |
3374 | ||
f6705578 N |
3375 | if (mddev->reshape_position != MaxSector) { |
3376 | /* Check that we can continue the reshape. | |
3377 | * Currently only disks can change, it must | |
3378 | * increase, and we must be past the point where | |
3379 | * a stripe over-writes itself | |
3380 | */ | |
3381 | sector_t here_new, here_old; | |
3382 | int old_disks; | |
f416885e | 3383 | int max_degraded = (mddev->level == 5 ? 1 : 2); |
f6705578 N |
3384 | |
3385 | if (mddev->new_level != mddev->level || | |
3386 | mddev->new_layout != mddev->layout || | |
3387 | mddev->new_chunk != mddev->chunk_size) { | |
f416885e N |
3388 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
3389 | "required - aborting.\n", | |
f6705578 N |
3390 | mdname(mddev)); |
3391 | return -EINVAL; | |
3392 | } | |
3393 | if (mddev->delta_disks <= 0) { | |
f416885e N |
3394 | printk(KERN_ERR "raid5: %s: unsupported reshape " |
3395 | "(reduce disks) required - aborting.\n", | |
f6705578 N |
3396 | mdname(mddev)); |
3397 | return -EINVAL; | |
3398 | } | |
3399 | old_disks = mddev->raid_disks - mddev->delta_disks; | |
3400 | /* reshape_position must be on a new-stripe boundary, and one | |
f416885e N |
3401 | * further up in new geometry must map after here in old |
3402 | * geometry. | |
f6705578 N |
3403 | */ |
3404 | here_new = mddev->reshape_position; | |
f416885e N |
3405 | if (sector_div(here_new, (mddev->chunk_size>>9)* |
3406 | (mddev->raid_disks - max_degraded))) { | |
3407 | printk(KERN_ERR "raid5: reshape_position not " | |
3408 | "on a stripe boundary\n"); | |
f6705578 N |
3409 | return -EINVAL; |
3410 | } | |
3411 | /* here_new is the stripe we will write to */ | |
3412 | here_old = mddev->reshape_position; | |
f416885e N |
3413 | sector_div(here_old, (mddev->chunk_size>>9)* |
3414 | (old_disks-max_degraded)); | |
3415 | /* here_old is the first stripe that we might need to read | |
3416 | * from */ | |
f6705578 N |
3417 | if (here_new >= here_old) { |
3418 | /* Reading from the same stripe as writing to - bad */ | |
f416885e N |
3419 | printk(KERN_ERR "raid5: reshape_position too early for " |
3420 | "auto-recovery - aborting.\n"); | |
f6705578 N |
3421 | return -EINVAL; |
3422 | } | |
3423 | printk(KERN_INFO "raid5: reshape will continue\n"); | |
3424 | /* OK, we should be able to continue; */ | |
3425 | } | |
3426 | ||
3427 | ||
b55e6bfc | 3428 | mddev->private = kzalloc(sizeof (raid5_conf_t), GFP_KERNEL); |
1da177e4 LT |
3429 | if ((conf = mddev->private) == NULL) |
3430 | goto abort; | |
f6705578 N |
3431 | if (mddev->reshape_position == MaxSector) { |
3432 | conf->previous_raid_disks = conf->raid_disks = mddev->raid_disks; | |
3433 | } else { | |
3434 | conf->raid_disks = mddev->raid_disks; | |
3435 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; | |
3436 | } | |
3437 | ||
3438 | conf->disks = kzalloc(conf->raid_disks * sizeof(struct disk_info), | |
b55e6bfc N |
3439 | GFP_KERNEL); |
3440 | if (!conf->disks) | |
3441 | goto abort; | |
9ffae0cf | 3442 | |
1da177e4 LT |
3443 | conf->mddev = mddev; |
3444 | ||
fccddba0 | 3445 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 3446 | goto abort; |
1da177e4 | 3447 | |
16a53ecc N |
3448 | if (mddev->level == 6) { |
3449 | conf->spare_page = alloc_page(GFP_KERNEL); | |
3450 | if (!conf->spare_page) | |
3451 | goto abort; | |
3452 | } | |
1da177e4 LT |
3453 | spin_lock_init(&conf->device_lock); |
3454 | init_waitqueue_head(&conf->wait_for_stripe); | |
3455 | init_waitqueue_head(&conf->wait_for_overlap); | |
3456 | INIT_LIST_HEAD(&conf->handle_list); | |
3457 | INIT_LIST_HEAD(&conf->delayed_list); | |
72626685 | 3458 | INIT_LIST_HEAD(&conf->bitmap_list); |
1da177e4 LT |
3459 | INIT_LIST_HEAD(&conf->inactive_list); |
3460 | atomic_set(&conf->active_stripes, 0); | |
3461 | atomic_set(&conf->preread_active_stripes, 0); | |
46031f9a | 3462 | atomic_set(&conf->active_aligned_reads, 0); |
1da177e4 | 3463 | |
1da177e4 LT |
3464 | PRINTK("raid5: run(%s) called.\n", mdname(mddev)); |
3465 | ||
3466 | ITERATE_RDEV(mddev,rdev,tmp) { | |
3467 | raid_disk = rdev->raid_disk; | |
f6705578 | 3468 | if (raid_disk >= conf->raid_disks |
1da177e4 LT |
3469 | || raid_disk < 0) |
3470 | continue; | |
3471 | disk = conf->disks + raid_disk; | |
3472 | ||
3473 | disk->rdev = rdev; | |
3474 | ||
b2d444d7 | 3475 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 LT |
3476 | char b[BDEVNAME_SIZE]; |
3477 | printk(KERN_INFO "raid5: device %s operational as raid" | |
3478 | " disk %d\n", bdevname(rdev->bdev,b), | |
3479 | raid_disk); | |
02c2de8c | 3480 | working_disks++; |
1da177e4 LT |
3481 | } |
3482 | } | |
3483 | ||
1da177e4 | 3484 | /* |
16a53ecc | 3485 | * 0 for a fully functional array, 1 or 2 for a degraded array. |
1da177e4 | 3486 | */ |
02c2de8c | 3487 | mddev->degraded = conf->raid_disks - working_disks; |
1da177e4 LT |
3488 | conf->mddev = mddev; |
3489 | conf->chunk_size = mddev->chunk_size; | |
3490 | conf->level = mddev->level; | |
16a53ecc N |
3491 | if (conf->level == 6) |
3492 | conf->max_degraded = 2; | |
3493 | else | |
3494 | conf->max_degraded = 1; | |
1da177e4 LT |
3495 | conf->algorithm = mddev->layout; |
3496 | conf->max_nr_stripes = NR_STRIPES; | |
f6705578 | 3497 | conf->expand_progress = mddev->reshape_position; |
1da177e4 LT |
3498 | |
3499 | /* device size must be a multiple of chunk size */ | |
3500 | mddev->size &= ~(mddev->chunk_size/1024 -1); | |
b1581566 | 3501 | mddev->resync_max_sectors = mddev->size << 1; |
1da177e4 | 3502 | |
16a53ecc N |
3503 | if (conf->level == 6 && conf->raid_disks < 4) { |
3504 | printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", | |
3505 | mdname(mddev), conf->raid_disks); | |
3506 | goto abort; | |
3507 | } | |
1da177e4 LT |
3508 | if (!conf->chunk_size || conf->chunk_size % 4) { |
3509 | printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", | |
3510 | conf->chunk_size, mdname(mddev)); | |
3511 | goto abort; | |
3512 | } | |
3513 | if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { | |
3514 | printk(KERN_ERR | |
3515 | "raid5: unsupported parity algorithm %d for %s\n", | |
3516 | conf->algorithm, mdname(mddev)); | |
3517 | goto abort; | |
3518 | } | |
16a53ecc | 3519 | if (mddev->degraded > conf->max_degraded) { |
1da177e4 LT |
3520 | printk(KERN_ERR "raid5: not enough operational devices for %s" |
3521 | " (%d/%d failed)\n", | |
02c2de8c | 3522 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
3523 | goto abort; |
3524 | } | |
3525 | ||
16a53ecc | 3526 | if (mddev->degraded > 0 && |
1da177e4 | 3527 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
3528 | if (mddev->ok_start_degraded) |
3529 | printk(KERN_WARNING | |
3530 | "raid5: starting dirty degraded array: %s" | |
3531 | "- data corruption possible.\n", | |
3532 | mdname(mddev)); | |
3533 | else { | |
3534 | printk(KERN_ERR | |
3535 | "raid5: cannot start dirty degraded array for %s\n", | |
3536 | mdname(mddev)); | |
3537 | goto abort; | |
3538 | } | |
1da177e4 LT |
3539 | } |
3540 | ||
3541 | { | |
3542 | mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); | |
3543 | if (!mddev->thread) { | |
3544 | printk(KERN_ERR | |
3545 | "raid5: couldn't allocate thread for %s\n", | |
3546 | mdname(mddev)); | |
3547 | goto abort; | |
3548 | } | |
3549 | } | |
5036805b | 3550 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
1da177e4 LT |
3551 | conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
3552 | if (grow_stripes(conf, conf->max_nr_stripes)) { | |
3553 | printk(KERN_ERR | |
3554 | "raid5: couldn't allocate %dkB for buffers\n", memory); | |
3555 | shrink_stripes(conf); | |
3556 | md_unregister_thread(mddev->thread); | |
3557 | goto abort; | |
3558 | } else | |
3559 | printk(KERN_INFO "raid5: allocated %dkB for %s\n", | |
3560 | memory, mdname(mddev)); | |
3561 | ||
3562 | if (mddev->degraded == 0) | |
3563 | printk("raid5: raid level %d set %s active with %d out of %d" | |
3564 | " devices, algorithm %d\n", conf->level, mdname(mddev), | |
3565 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
3566 | conf->algorithm); | |
3567 | else | |
3568 | printk(KERN_ALERT "raid5: raid level %d set %s active with %d" | |
3569 | " out of %d devices, algorithm %d\n", conf->level, | |
3570 | mdname(mddev), mddev->raid_disks - mddev->degraded, | |
3571 | mddev->raid_disks, conf->algorithm); | |
3572 | ||
3573 | print_raid5_conf(conf); | |
3574 | ||
f6705578 N |
3575 | if (conf->expand_progress != MaxSector) { |
3576 | printk("...ok start reshape thread\n"); | |
b578d55f | 3577 | conf->expand_lo = conf->expand_progress; |
f6705578 N |
3578 | atomic_set(&conf->reshape_stripes, 0); |
3579 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
3580 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3581 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3582 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3583 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3584 | "%s_reshape"); | |
f6705578 N |
3585 | } |
3586 | ||
1da177e4 | 3587 | /* read-ahead size must cover two whole stripes, which is |
16a53ecc | 3588 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices |
1da177e4 LT |
3589 | */ |
3590 | { | |
16a53ecc N |
3591 | int data_disks = conf->previous_raid_disks - conf->max_degraded; |
3592 | int stripe = data_disks * | |
8932c2e0 | 3593 | (mddev->chunk_size / PAGE_SIZE); |
1da177e4 LT |
3594 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
3595 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
3596 | } | |
3597 | ||
3598 | /* Ok, everything is just fine now */ | |
5e55e2f5 N |
3599 | if (sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) |
3600 | printk(KERN_WARNING | |
3601 | "raid5: failed to create sysfs attributes for %s\n", | |
3602 | mdname(mddev)); | |
7a5febe9 N |
3603 | |
3604 | mddev->queue->unplug_fn = raid5_unplug_device; | |
3605 | mddev->queue->issue_flush_fn = raid5_issue_flush; | |
f022b2fd | 3606 | mddev->queue->backing_dev_info.congested_data = mddev; |
041ae52e | 3607 | mddev->queue->backing_dev_info.congested_fn = raid5_congested; |
f022b2fd | 3608 | |
16a53ecc N |
3609 | mddev->array_size = mddev->size * (conf->previous_raid_disks - |
3610 | conf->max_degraded); | |
7a5febe9 | 3611 | |
23032a0e RBJ |
3612 | blk_queue_merge_bvec(mddev->queue, raid5_mergeable_bvec); |
3613 | ||
1da177e4 LT |
3614 | return 0; |
3615 | abort: | |
3616 | if (conf) { | |
3617 | print_raid5_conf(conf); | |
16a53ecc | 3618 | safe_put_page(conf->spare_page); |
b55e6bfc | 3619 | kfree(conf->disks); |
fccddba0 | 3620 | kfree(conf->stripe_hashtbl); |
1da177e4 LT |
3621 | kfree(conf); |
3622 | } | |
3623 | mddev->private = NULL; | |
3624 | printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); | |
3625 | return -EIO; | |
3626 | } | |
3627 | ||
3628 | ||
3629 | ||
3f294f4f | 3630 | static int stop(mddev_t *mddev) |
1da177e4 LT |
3631 | { |
3632 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3633 | ||
3634 | md_unregister_thread(mddev->thread); | |
3635 | mddev->thread = NULL; | |
3636 | shrink_stripes(conf); | |
fccddba0 | 3637 | kfree(conf->stripe_hashtbl); |
041ae52e | 3638 | mddev->queue->backing_dev_info.congested_fn = NULL; |
1da177e4 | 3639 | blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ |
007583c9 | 3640 | sysfs_remove_group(&mddev->kobj, &raid5_attrs_group); |
b55e6bfc | 3641 | kfree(conf->disks); |
96de1e66 | 3642 | kfree(conf); |
1da177e4 LT |
3643 | mddev->private = NULL; |
3644 | return 0; | |
3645 | } | |
3646 | ||
3647 | #if RAID5_DEBUG | |
16a53ecc | 3648 | static void print_sh (struct seq_file *seq, struct stripe_head *sh) |
1da177e4 LT |
3649 | { |
3650 | int i; | |
3651 | ||
16a53ecc N |
3652 | seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", |
3653 | (unsigned long long)sh->sector, sh->pd_idx, sh->state); | |
3654 | seq_printf(seq, "sh %llu, count %d.\n", | |
3655 | (unsigned long long)sh->sector, atomic_read(&sh->count)); | |
3656 | seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); | |
7ecaa1e6 | 3657 | for (i = 0; i < sh->disks; i++) { |
16a53ecc N |
3658 | seq_printf(seq, "(cache%d: %p %ld) ", |
3659 | i, sh->dev[i].page, sh->dev[i].flags); | |
1da177e4 | 3660 | } |
16a53ecc | 3661 | seq_printf(seq, "\n"); |
1da177e4 LT |
3662 | } |
3663 | ||
16a53ecc | 3664 | static void printall (struct seq_file *seq, raid5_conf_t *conf) |
1da177e4 LT |
3665 | { |
3666 | struct stripe_head *sh; | |
fccddba0 | 3667 | struct hlist_node *hn; |
1da177e4 LT |
3668 | int i; |
3669 | ||
3670 | spin_lock_irq(&conf->device_lock); | |
3671 | for (i = 0; i < NR_HASH; i++) { | |
fccddba0 | 3672 | hlist_for_each_entry(sh, hn, &conf->stripe_hashtbl[i], hash) { |
1da177e4 LT |
3673 | if (sh->raid_conf != conf) |
3674 | continue; | |
16a53ecc | 3675 | print_sh(seq, sh); |
1da177e4 LT |
3676 | } |
3677 | } | |
3678 | spin_unlock_irq(&conf->device_lock); | |
3679 | } | |
3680 | #endif | |
3681 | ||
3682 | static void status (struct seq_file *seq, mddev_t *mddev) | |
3683 | { | |
3684 | raid5_conf_t *conf = (raid5_conf_t *) mddev->private; | |
3685 | int i; | |
3686 | ||
3687 | seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); | |
02c2de8c | 3688 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
3689 | for (i = 0; i < conf->raid_disks; i++) |
3690 | seq_printf (seq, "%s", | |
3691 | conf->disks[i].rdev && | |
b2d444d7 | 3692 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 LT |
3693 | seq_printf (seq, "]"); |
3694 | #if RAID5_DEBUG | |
16a53ecc N |
3695 | seq_printf (seq, "\n"); |
3696 | printall(seq, conf); | |
1da177e4 LT |
3697 | #endif |
3698 | } | |
3699 | ||
3700 | static void print_raid5_conf (raid5_conf_t *conf) | |
3701 | { | |
3702 | int i; | |
3703 | struct disk_info *tmp; | |
3704 | ||
3705 | printk("RAID5 conf printout:\n"); | |
3706 | if (!conf) { | |
3707 | printk("(conf==NULL)\n"); | |
3708 | return; | |
3709 | } | |
02c2de8c N |
3710 | printk(" --- rd:%d wd:%d\n", conf->raid_disks, |
3711 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
3712 | |
3713 | for (i = 0; i < conf->raid_disks; i++) { | |
3714 | char b[BDEVNAME_SIZE]; | |
3715 | tmp = conf->disks + i; | |
3716 | if (tmp->rdev) | |
3717 | printk(" disk %d, o:%d, dev:%s\n", | |
b2d444d7 | 3718 | i, !test_bit(Faulty, &tmp->rdev->flags), |
1da177e4 LT |
3719 | bdevname(tmp->rdev->bdev,b)); |
3720 | } | |
3721 | } | |
3722 | ||
3723 | static int raid5_spare_active(mddev_t *mddev) | |
3724 | { | |
3725 | int i; | |
3726 | raid5_conf_t *conf = mddev->private; | |
3727 | struct disk_info *tmp; | |
3728 | ||
3729 | for (i = 0; i < conf->raid_disks; i++) { | |
3730 | tmp = conf->disks + i; | |
3731 | if (tmp->rdev | |
b2d444d7 | 3732 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa N |
3733 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
3734 | unsigned long flags; | |
3735 | spin_lock_irqsave(&conf->device_lock, flags); | |
1da177e4 | 3736 | mddev->degraded--; |
c04be0aa | 3737 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 LT |
3738 | } |
3739 | } | |
3740 | print_raid5_conf(conf); | |
3741 | return 0; | |
3742 | } | |
3743 | ||
3744 | static int raid5_remove_disk(mddev_t *mddev, int number) | |
3745 | { | |
3746 | raid5_conf_t *conf = mddev->private; | |
3747 | int err = 0; | |
3748 | mdk_rdev_t *rdev; | |
3749 | struct disk_info *p = conf->disks + number; | |
3750 | ||
3751 | print_raid5_conf(conf); | |
3752 | rdev = p->rdev; | |
3753 | if (rdev) { | |
b2d444d7 | 3754 | if (test_bit(In_sync, &rdev->flags) || |
1da177e4 LT |
3755 | atomic_read(&rdev->nr_pending)) { |
3756 | err = -EBUSY; | |
3757 | goto abort; | |
3758 | } | |
3759 | p->rdev = NULL; | |
fbd568a3 | 3760 | synchronize_rcu(); |
1da177e4 LT |
3761 | if (atomic_read(&rdev->nr_pending)) { |
3762 | /* lost the race, try later */ | |
3763 | err = -EBUSY; | |
3764 | p->rdev = rdev; | |
3765 | } | |
3766 | } | |
3767 | abort: | |
3768 | ||
3769 | print_raid5_conf(conf); | |
3770 | return err; | |
3771 | } | |
3772 | ||
3773 | static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) | |
3774 | { | |
3775 | raid5_conf_t *conf = mddev->private; | |
3776 | int found = 0; | |
3777 | int disk; | |
3778 | struct disk_info *p; | |
3779 | ||
16a53ecc | 3780 | if (mddev->degraded > conf->max_degraded) |
1da177e4 LT |
3781 | /* no point adding a device */ |
3782 | return 0; | |
3783 | ||
3784 | /* | |
16a53ecc N |
3785 | * find the disk ... but prefer rdev->saved_raid_disk |
3786 | * if possible. | |
1da177e4 | 3787 | */ |
16a53ecc N |
3788 | if (rdev->saved_raid_disk >= 0 && |
3789 | conf->disks[rdev->saved_raid_disk].rdev == NULL) | |
3790 | disk = rdev->saved_raid_disk; | |
3791 | else | |
3792 | disk = 0; | |
3793 | for ( ; disk < conf->raid_disks; disk++) | |
1da177e4 | 3794 | if ((p=conf->disks + disk)->rdev == NULL) { |
b2d444d7 | 3795 | clear_bit(In_sync, &rdev->flags); |
1da177e4 LT |
3796 | rdev->raid_disk = disk; |
3797 | found = 1; | |
72626685 N |
3798 | if (rdev->saved_raid_disk != disk) |
3799 | conf->fullsync = 1; | |
d6065f7b | 3800 | rcu_assign_pointer(p->rdev, rdev); |
1da177e4 LT |
3801 | break; |
3802 | } | |
3803 | print_raid5_conf(conf); | |
3804 | return found; | |
3805 | } | |
3806 | ||
3807 | static int raid5_resize(mddev_t *mddev, sector_t sectors) | |
3808 | { | |
3809 | /* no resync is happening, and there is enough space | |
3810 | * on all devices, so we can resize. | |
3811 | * We need to make sure resync covers any new space. | |
3812 | * If the array is shrinking we should possibly wait until | |
3813 | * any io in the removed space completes, but it hardly seems | |
3814 | * worth it. | |
3815 | */ | |
16a53ecc N |
3816 | raid5_conf_t *conf = mddev_to_conf(mddev); |
3817 | ||
1da177e4 | 3818 | sectors &= ~((sector_t)mddev->chunk_size/512 - 1); |
16a53ecc | 3819 | mddev->array_size = (sectors * (mddev->raid_disks-conf->max_degraded))>>1; |
1da177e4 | 3820 | set_capacity(mddev->gendisk, mddev->array_size << 1); |
44ce6294 | 3821 | mddev->changed = 1; |
1da177e4 LT |
3822 | if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) { |
3823 | mddev->recovery_cp = mddev->size << 1; | |
3824 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3825 | } | |
3826 | mddev->size = sectors /2; | |
4b5c7ae8 | 3827 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
3828 | return 0; |
3829 | } | |
3830 | ||
29269553 | 3831 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f | 3832 | static int raid5_check_reshape(mddev_t *mddev) |
29269553 N |
3833 | { |
3834 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3835 | int err; | |
29269553 | 3836 | |
63c70c4f N |
3837 | if (mddev->delta_disks < 0 || |
3838 | mddev->new_level != mddev->level) | |
3839 | return -EINVAL; /* Cannot shrink array or change level yet */ | |
3840 | if (mddev->delta_disks == 0) | |
29269553 N |
3841 | return 0; /* nothing to do */ |
3842 | ||
3843 | /* Can only proceed if there are plenty of stripe_heads. | |
3844 | * We need a minimum of one full stripe,, and for sensible progress | |
3845 | * it is best to have about 4 times that. | |
3846 | * If we require 4 times, then the default 256 4K stripe_heads will | |
3847 | * allow for chunk sizes up to 256K, which is probably OK. | |
3848 | * If the chunk size is greater, user-space should request more | |
3849 | * stripe_heads first. | |
3850 | */ | |
63c70c4f N |
3851 | if ((mddev->chunk_size / STRIPE_SIZE) * 4 > conf->max_nr_stripes || |
3852 | (mddev->new_chunk / STRIPE_SIZE) * 4 > conf->max_nr_stripes) { | |
29269553 N |
3853 | printk(KERN_WARNING "raid5: reshape: not enough stripes. Needed %lu\n", |
3854 | (mddev->chunk_size / STRIPE_SIZE)*4); | |
3855 | return -ENOSPC; | |
3856 | } | |
3857 | ||
63c70c4f N |
3858 | err = resize_stripes(conf, conf->raid_disks + mddev->delta_disks); |
3859 | if (err) | |
3860 | return err; | |
3861 | ||
b4c4c7b8 N |
3862 | if (mddev->degraded > conf->max_degraded) |
3863 | return -EINVAL; | |
63c70c4f N |
3864 | /* looks like we might be able to manage this */ |
3865 | return 0; | |
3866 | } | |
3867 | ||
3868 | static int raid5_start_reshape(mddev_t *mddev) | |
3869 | { | |
3870 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3871 | mdk_rdev_t *rdev; | |
3872 | struct list_head *rtmp; | |
3873 | int spares = 0; | |
3874 | int added_devices = 0; | |
c04be0aa | 3875 | unsigned long flags; |
63c70c4f | 3876 | |
f416885e | 3877 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
3878 | return -EBUSY; |
3879 | ||
29269553 N |
3880 | ITERATE_RDEV(mddev, rdev, rtmp) |
3881 | if (rdev->raid_disk < 0 && | |
3882 | !test_bit(Faulty, &rdev->flags)) | |
3883 | spares++; | |
63c70c4f | 3884 | |
f416885e | 3885 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
3886 | /* Not enough devices even to make a degraded array |
3887 | * of that size | |
3888 | */ | |
3889 | return -EINVAL; | |
3890 | ||
f6705578 | 3891 | atomic_set(&conf->reshape_stripes, 0); |
29269553 N |
3892 | spin_lock_irq(&conf->device_lock); |
3893 | conf->previous_raid_disks = conf->raid_disks; | |
63c70c4f | 3894 | conf->raid_disks += mddev->delta_disks; |
29269553 | 3895 | conf->expand_progress = 0; |
b578d55f | 3896 | conf->expand_lo = 0; |
29269553 N |
3897 | spin_unlock_irq(&conf->device_lock); |
3898 | ||
3899 | /* Add some new drives, as many as will fit. | |
3900 | * We know there are enough to make the newly sized array work. | |
3901 | */ | |
3902 | ITERATE_RDEV(mddev, rdev, rtmp) | |
3903 | if (rdev->raid_disk < 0 && | |
3904 | !test_bit(Faulty, &rdev->flags)) { | |
3905 | if (raid5_add_disk(mddev, rdev)) { | |
3906 | char nm[20]; | |
3907 | set_bit(In_sync, &rdev->flags); | |
29269553 | 3908 | added_devices++; |
5fd6c1dc | 3909 | rdev->recovery_offset = 0; |
29269553 | 3910 | sprintf(nm, "rd%d", rdev->raid_disk); |
5e55e2f5 N |
3911 | if (sysfs_create_link(&mddev->kobj, |
3912 | &rdev->kobj, nm)) | |
3913 | printk(KERN_WARNING | |
3914 | "raid5: failed to create " | |
3915 | " link %s for %s\n", | |
3916 | nm, mdname(mddev)); | |
29269553 N |
3917 | } else |
3918 | break; | |
3919 | } | |
3920 | ||
c04be0aa | 3921 | spin_lock_irqsave(&conf->device_lock, flags); |
63c70c4f | 3922 | mddev->degraded = (conf->raid_disks - conf->previous_raid_disks) - added_devices; |
c04be0aa | 3923 | spin_unlock_irqrestore(&conf->device_lock, flags); |
63c70c4f | 3924 | mddev->raid_disks = conf->raid_disks; |
f6705578 | 3925 | mddev->reshape_position = 0; |
850b2b42 | 3926 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 3927 | |
29269553 N |
3928 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3929 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3930 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3931 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3932 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3933 | "%s_reshape"); | |
3934 | if (!mddev->sync_thread) { | |
3935 | mddev->recovery = 0; | |
3936 | spin_lock_irq(&conf->device_lock); | |
3937 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; | |
3938 | conf->expand_progress = MaxSector; | |
3939 | spin_unlock_irq(&conf->device_lock); | |
3940 | return -EAGAIN; | |
3941 | } | |
3942 | md_wakeup_thread(mddev->sync_thread); | |
3943 | md_new_event(mddev); | |
3944 | return 0; | |
3945 | } | |
3946 | #endif | |
3947 | ||
3948 | static void end_reshape(raid5_conf_t *conf) | |
3949 | { | |
3950 | struct block_device *bdev; | |
3951 | ||
f6705578 | 3952 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f416885e N |
3953 | conf->mddev->array_size = conf->mddev->size * |
3954 | (conf->raid_disks - conf->max_degraded); | |
f6705578 | 3955 | set_capacity(conf->mddev->gendisk, conf->mddev->array_size << 1); |
44ce6294 | 3956 | conf->mddev->changed = 1; |
f6705578 N |
3957 | |
3958 | bdev = bdget_disk(conf->mddev->gendisk, 0); | |
3959 | if (bdev) { | |
3960 | mutex_lock(&bdev->bd_inode->i_mutex); | |
0692c6b1 | 3961 | i_size_write(bdev->bd_inode, (loff_t)conf->mddev->array_size << 10); |
f6705578 N |
3962 | mutex_unlock(&bdev->bd_inode->i_mutex); |
3963 | bdput(bdev); | |
3964 | } | |
3965 | spin_lock_irq(&conf->device_lock); | |
3966 | conf->expand_progress = MaxSector; | |
3967 | spin_unlock_irq(&conf->device_lock); | |
3968 | conf->mddev->reshape_position = MaxSector; | |
16a53ecc N |
3969 | |
3970 | /* read-ahead size must cover two whole stripes, which is | |
3971 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
3972 | */ | |
3973 | { | |
3974 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
3975 | int stripe = data_disks * | |
3976 | (conf->mddev->chunk_size / PAGE_SIZE); | |
3977 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
3978 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
3979 | } | |
29269553 | 3980 | } |
29269553 N |
3981 | } |
3982 | ||
72626685 N |
3983 | static void raid5_quiesce(mddev_t *mddev, int state) |
3984 | { | |
3985 | raid5_conf_t *conf = mddev_to_conf(mddev); | |
3986 | ||
3987 | switch(state) { | |
e464eafd N |
3988 | case 2: /* resume for a suspend */ |
3989 | wake_up(&conf->wait_for_overlap); | |
3990 | break; | |
3991 | ||
72626685 N |
3992 | case 1: /* stop all writes */ |
3993 | spin_lock_irq(&conf->device_lock); | |
3994 | conf->quiesce = 1; | |
3995 | wait_event_lock_irq(conf->wait_for_stripe, | |
46031f9a RBJ |
3996 | atomic_read(&conf->active_stripes) == 0 && |
3997 | atomic_read(&conf->active_aligned_reads) == 0, | |
72626685 N |
3998 | conf->device_lock, /* nothing */); |
3999 | spin_unlock_irq(&conf->device_lock); | |
4000 | break; | |
4001 | ||
4002 | case 0: /* re-enable writes */ | |
4003 | spin_lock_irq(&conf->device_lock); | |
4004 | conf->quiesce = 0; | |
4005 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 4006 | wake_up(&conf->wait_for_overlap); |
72626685 N |
4007 | spin_unlock_irq(&conf->device_lock); |
4008 | break; | |
4009 | } | |
72626685 | 4010 | } |
b15c2e57 | 4011 | |
16a53ecc N |
4012 | static struct mdk_personality raid6_personality = |
4013 | { | |
4014 | .name = "raid6", | |
4015 | .level = 6, | |
4016 | .owner = THIS_MODULE, | |
4017 | .make_request = make_request, | |
4018 | .run = run, | |
4019 | .stop = stop, | |
4020 | .status = status, | |
4021 | .error_handler = error, | |
4022 | .hot_add_disk = raid5_add_disk, | |
4023 | .hot_remove_disk= raid5_remove_disk, | |
4024 | .spare_active = raid5_spare_active, | |
4025 | .sync_request = sync_request, | |
4026 | .resize = raid5_resize, | |
f416885e N |
4027 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4028 | .check_reshape = raid5_check_reshape, | |
4029 | .start_reshape = raid5_start_reshape, | |
4030 | #endif | |
16a53ecc N |
4031 | .quiesce = raid5_quiesce, |
4032 | }; | |
2604b703 | 4033 | static struct mdk_personality raid5_personality = |
1da177e4 LT |
4034 | { |
4035 | .name = "raid5", | |
2604b703 | 4036 | .level = 5, |
1da177e4 LT |
4037 | .owner = THIS_MODULE, |
4038 | .make_request = make_request, | |
4039 | .run = run, | |
4040 | .stop = stop, | |
4041 | .status = status, | |
4042 | .error_handler = error, | |
4043 | .hot_add_disk = raid5_add_disk, | |
4044 | .hot_remove_disk= raid5_remove_disk, | |
4045 | .spare_active = raid5_spare_active, | |
4046 | .sync_request = sync_request, | |
4047 | .resize = raid5_resize, | |
29269553 | 4048 | #ifdef CONFIG_MD_RAID5_RESHAPE |
63c70c4f N |
4049 | .check_reshape = raid5_check_reshape, |
4050 | .start_reshape = raid5_start_reshape, | |
29269553 | 4051 | #endif |
72626685 | 4052 | .quiesce = raid5_quiesce, |
1da177e4 LT |
4053 | }; |
4054 | ||
2604b703 | 4055 | static struct mdk_personality raid4_personality = |
1da177e4 | 4056 | { |
2604b703 N |
4057 | .name = "raid4", |
4058 | .level = 4, | |
4059 | .owner = THIS_MODULE, | |
4060 | .make_request = make_request, | |
4061 | .run = run, | |
4062 | .stop = stop, | |
4063 | .status = status, | |
4064 | .error_handler = error, | |
4065 | .hot_add_disk = raid5_add_disk, | |
4066 | .hot_remove_disk= raid5_remove_disk, | |
4067 | .spare_active = raid5_spare_active, | |
4068 | .sync_request = sync_request, | |
4069 | .resize = raid5_resize, | |
3d37890b N |
4070 | #ifdef CONFIG_MD_RAID5_RESHAPE |
4071 | .check_reshape = raid5_check_reshape, | |
4072 | .start_reshape = raid5_start_reshape, | |
4073 | #endif | |
2604b703 N |
4074 | .quiesce = raid5_quiesce, |
4075 | }; | |
4076 | ||
4077 | static int __init raid5_init(void) | |
4078 | { | |
16a53ecc N |
4079 | int e; |
4080 | ||
4081 | e = raid6_select_algo(); | |
4082 | if ( e ) | |
4083 | return e; | |
4084 | register_md_personality(&raid6_personality); | |
2604b703 N |
4085 | register_md_personality(&raid5_personality); |
4086 | register_md_personality(&raid4_personality); | |
4087 | return 0; | |
1da177e4 LT |
4088 | } |
4089 | ||
2604b703 | 4090 | static void raid5_exit(void) |
1da177e4 | 4091 | { |
16a53ecc | 4092 | unregister_md_personality(&raid6_personality); |
2604b703 N |
4093 | unregister_md_personality(&raid5_personality); |
4094 | unregister_md_personality(&raid4_personality); | |
1da177e4 LT |
4095 | } |
4096 | ||
4097 | module_init(raid5_init); | |
4098 | module_exit(raid5_exit); | |
4099 | MODULE_LICENSE("GPL"); | |
4100 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ | |
d9d166c2 N |
4101 | MODULE_ALIAS("md-raid5"); |
4102 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
4103 | MODULE_ALIAS("md-level-5"); |
4104 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
4105 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
4106 | MODULE_ALIAS("md-raid6"); | |
4107 | MODULE_ALIAS("md-level-6"); | |
4108 | ||
4109 | /* This used to be two separate modules, they were: */ | |
4110 | MODULE_ALIAS("raid5"); | |
4111 | MODULE_ALIAS("raid6"); |