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. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
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 | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
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 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
851c30c9 | 56 | #include <linux/nodemask.h> |
46d5b785 | 57 | #include <linux/flex_array.h> |
a9add5d9 N |
58 | #include <trace/events/block.h> |
59 | ||
43b2e5d8 | 60 | #include "md.h" |
bff61975 | 61 | #include "raid5.h" |
54071b38 | 62 | #include "raid0.h" |
ef740c37 | 63 | #include "bitmap.h" |
72626685 | 64 | |
851c30c9 SL |
65 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
66 | #define ANY_GROUP NUMA_NO_NODE | |
67 | ||
8e0e99ba N |
68 | static bool devices_handle_discard_safely = false; |
69 | module_param(devices_handle_discard_safely, bool, 0644); | |
70 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
71 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 72 | static struct workqueue_struct *raid5_wq; |
1da177e4 LT |
73 | /* |
74 | * Stripe cache | |
75 | */ | |
76 | ||
77 | #define NR_STRIPES 256 | |
78 | #define STRIPE_SIZE PAGE_SIZE | |
79 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
80 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
81 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 82 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 83 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 | 84 | #define HASH_MASK (NR_HASH - 1) |
bfc90cb0 | 85 | #define MAX_STRIPE_BATCH 8 |
1da177e4 | 86 | |
d1688a6d | 87 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
88 | { |
89 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
90 | return &conf->stripe_hashtbl[hash]; | |
91 | } | |
1da177e4 | 92 | |
566c09c5 SL |
93 | static inline int stripe_hash_locks_hash(sector_t sect) |
94 | { | |
95 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
96 | } | |
97 | ||
98 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
99 | { | |
100 | spin_lock_irq(conf->hash_locks + hash); | |
101 | spin_lock(&conf->device_lock); | |
102 | } | |
103 | ||
104 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
105 | { | |
106 | spin_unlock(&conf->device_lock); | |
107 | spin_unlock_irq(conf->hash_locks + hash); | |
108 | } | |
109 | ||
110 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
111 | { | |
112 | int i; | |
113 | local_irq_disable(); | |
114 | spin_lock(conf->hash_locks); | |
115 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
116 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
117 | spin_lock(&conf->device_lock); | |
118 | } | |
119 | ||
120 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
121 | { | |
122 | int i; | |
123 | spin_unlock(&conf->device_lock); | |
124 | for (i = NR_STRIPE_HASH_LOCKS; i; i--) | |
125 | spin_unlock(conf->hash_locks + i - 1); | |
126 | local_irq_enable(); | |
127 | } | |
128 | ||
1da177e4 LT |
129 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector |
130 | * order without overlap. There may be several bio's per stripe+device, and | |
131 | * a bio could span several devices. | |
132 | * When walking this list for a particular stripe+device, we must never proceed | |
133 | * beyond a bio that extends past this device, as the next bio might no longer | |
134 | * be valid. | |
db298e19 | 135 | * This function is used to determine the 'next' bio in the list, given the sector |
1da177e4 LT |
136 | * of the current stripe+device |
137 | */ | |
db298e19 N |
138 | static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector) |
139 | { | |
aa8b57aa | 140 | int sectors = bio_sectors(bio); |
4f024f37 | 141 | if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS) |
db298e19 N |
142 | return bio->bi_next; |
143 | else | |
144 | return NULL; | |
145 | } | |
1da177e4 | 146 | |
960e739d | 147 | /* |
5b99c2ff JA |
148 | * We maintain a biased count of active stripes in the bottom 16 bits of |
149 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d | 150 | */ |
e7836bd6 | 151 | static inline int raid5_bi_processed_stripes(struct bio *bio) |
960e739d | 152 | { |
e7836bd6 SL |
153 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
154 | return (atomic_read(segments) >> 16) & 0xffff; | |
960e739d JA |
155 | } |
156 | ||
e7836bd6 | 157 | static inline int raid5_dec_bi_active_stripes(struct bio *bio) |
960e739d | 158 | { |
e7836bd6 SL |
159 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
160 | return atomic_sub_return(1, segments) & 0xffff; | |
960e739d JA |
161 | } |
162 | ||
e7836bd6 | 163 | static inline void raid5_inc_bi_active_stripes(struct bio *bio) |
960e739d | 164 | { |
e7836bd6 SL |
165 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
166 | atomic_inc(segments); | |
960e739d JA |
167 | } |
168 | ||
e7836bd6 SL |
169 | static inline void raid5_set_bi_processed_stripes(struct bio *bio, |
170 | unsigned int cnt) | |
960e739d | 171 | { |
e7836bd6 SL |
172 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
173 | int old, new; | |
960e739d | 174 | |
e7836bd6 SL |
175 | do { |
176 | old = atomic_read(segments); | |
177 | new = (old & 0xffff) | (cnt << 16); | |
178 | } while (atomic_cmpxchg(segments, old, new) != old); | |
960e739d JA |
179 | } |
180 | ||
e7836bd6 | 181 | static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt) |
960e739d | 182 | { |
e7836bd6 SL |
183 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
184 | atomic_set(segments, cnt); | |
960e739d JA |
185 | } |
186 | ||
d0dabf7e N |
187 | /* Find first data disk in a raid6 stripe */ |
188 | static inline int raid6_d0(struct stripe_head *sh) | |
189 | { | |
67cc2b81 N |
190 | if (sh->ddf_layout) |
191 | /* ddf always start from first device */ | |
192 | return 0; | |
193 | /* md starts just after Q block */ | |
d0dabf7e N |
194 | if (sh->qd_idx == sh->disks - 1) |
195 | return 0; | |
196 | else | |
197 | return sh->qd_idx + 1; | |
198 | } | |
16a53ecc N |
199 | static inline int raid6_next_disk(int disk, int raid_disks) |
200 | { | |
201 | disk++; | |
202 | return (disk < raid_disks) ? disk : 0; | |
203 | } | |
a4456856 | 204 | |
d0dabf7e N |
205 | /* When walking through the disks in a raid5, starting at raid6_d0, |
206 | * We need to map each disk to a 'slot', where the data disks are slot | |
207 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
208 | * is raid_disks-1. This help does that mapping. | |
209 | */ | |
67cc2b81 N |
210 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
211 | int *count, int syndrome_disks) | |
d0dabf7e | 212 | { |
6629542e | 213 | int slot = *count; |
67cc2b81 | 214 | |
e4424fee | 215 | if (sh->ddf_layout) |
6629542e | 216 | (*count)++; |
d0dabf7e | 217 | if (idx == sh->pd_idx) |
67cc2b81 | 218 | return syndrome_disks; |
d0dabf7e | 219 | if (idx == sh->qd_idx) |
67cc2b81 | 220 | return syndrome_disks + 1; |
e4424fee | 221 | if (!sh->ddf_layout) |
6629542e | 222 | (*count)++; |
d0dabf7e N |
223 | return slot; |
224 | } | |
225 | ||
a4456856 DW |
226 | static void return_io(struct bio *return_bi) |
227 | { | |
228 | struct bio *bi = return_bi; | |
229 | while (bi) { | |
a4456856 DW |
230 | |
231 | return_bi = bi->bi_next; | |
232 | bi->bi_next = NULL; | |
4f024f37 | 233 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
234 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
235 | bi, 0); | |
0e13fe23 | 236 | bio_endio(bi, 0); |
a4456856 DW |
237 | bi = return_bi; |
238 | } | |
239 | } | |
240 | ||
d1688a6d | 241 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 242 | |
600aa109 DW |
243 | static int stripe_operations_active(struct stripe_head *sh) |
244 | { | |
245 | return sh->check_state || sh->reconstruct_state || | |
246 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
247 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
248 | } | |
249 | ||
851c30c9 SL |
250 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
251 | { | |
252 | struct r5conf *conf = sh->raid_conf; | |
253 | struct r5worker_group *group; | |
bfc90cb0 | 254 | int thread_cnt; |
851c30c9 SL |
255 | int i, cpu = sh->cpu; |
256 | ||
257 | if (!cpu_online(cpu)) { | |
258 | cpu = cpumask_any(cpu_online_mask); | |
259 | sh->cpu = cpu; | |
260 | } | |
261 | ||
262 | if (list_empty(&sh->lru)) { | |
263 | struct r5worker_group *group; | |
264 | group = conf->worker_groups + cpu_to_group(cpu); | |
265 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
266 | group->stripes_cnt++; |
267 | sh->group = group; | |
851c30c9 SL |
268 | } |
269 | ||
270 | if (conf->worker_cnt_per_group == 0) { | |
271 | md_wakeup_thread(conf->mddev->thread); | |
272 | return; | |
273 | } | |
274 | ||
275 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
276 | ||
bfc90cb0 SL |
277 | group->workers[0].working = true; |
278 | /* at least one worker should run to avoid race */ | |
279 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
280 | ||
281 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
282 | /* wakeup more workers */ | |
283 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
284 | if (group->workers[i].working == false) { | |
285 | group->workers[i].working = true; | |
286 | queue_work_on(sh->cpu, raid5_wq, | |
287 | &group->workers[i].work); | |
288 | thread_cnt--; | |
289 | } | |
290 | } | |
851c30c9 SL |
291 | } |
292 | ||
566c09c5 SL |
293 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
294 | struct list_head *temp_inactive_list) | |
1da177e4 | 295 | { |
4eb788df SL |
296 | BUG_ON(!list_empty(&sh->lru)); |
297 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
298 | if (test_bit(STRIPE_HANDLE, &sh->state)) { | |
299 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 300 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 301 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 302 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
303 | sh->bm_seq - conf->seq_write > 0) |
304 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
305 | else { | |
306 | clear_bit(STRIPE_DELAYED, &sh->state); | |
307 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 SL |
308 | if (conf->worker_cnt_per_group == 0) { |
309 | list_add_tail(&sh->lru, &conf->handle_list); | |
310 | } else { | |
311 | raid5_wakeup_stripe_thread(sh); | |
312 | return; | |
313 | } | |
4eb788df SL |
314 | } |
315 | md_wakeup_thread(conf->mddev->thread); | |
316 | } else { | |
317 | BUG_ON(stripe_operations_active(sh)); | |
318 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
319 | if (atomic_dec_return(&conf->preread_active_stripes) | |
320 | < IO_THRESHOLD) | |
321 | md_wakeup_thread(conf->mddev->thread); | |
322 | atomic_dec(&conf->active_stripes); | |
566c09c5 SL |
323 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) |
324 | list_add_tail(&sh->lru, temp_inactive_list); | |
1da177e4 LT |
325 | } |
326 | } | |
d0dabf7e | 327 | |
566c09c5 SL |
328 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
329 | struct list_head *temp_inactive_list) | |
4eb788df SL |
330 | { |
331 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
332 | do_release_stripe(conf, sh, temp_inactive_list); |
333 | } | |
334 | ||
335 | /* | |
336 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
337 | * | |
338 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
339 | * given time. Adding stripes only takes device lock, while deleting stripes | |
340 | * only takes hash lock. | |
341 | */ | |
342 | static void release_inactive_stripe_list(struct r5conf *conf, | |
343 | struct list_head *temp_inactive_list, | |
344 | int hash) | |
345 | { | |
346 | int size; | |
347 | bool do_wakeup = false; | |
348 | unsigned long flags; | |
349 | ||
350 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
351 | size = NR_STRIPE_HASH_LOCKS; | |
352 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
353 | } else | |
354 | size = 1; | |
355 | while (size) { | |
356 | struct list_head *list = &temp_inactive_list[size - 1]; | |
357 | ||
358 | /* | |
359 | * We don't hold any lock here yet, get_active_stripe() might | |
360 | * remove stripes from the list | |
361 | */ | |
362 | if (!list_empty_careful(list)) { | |
363 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
364 | if (list_empty(conf->inactive_list + hash) && |
365 | !list_empty(list)) | |
366 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 SL |
367 | list_splice_tail_init(list, conf->inactive_list + hash); |
368 | do_wakeup = true; | |
369 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); | |
370 | } | |
371 | size--; | |
372 | hash--; | |
373 | } | |
374 | ||
375 | if (do_wakeup) { | |
376 | wake_up(&conf->wait_for_stripe); | |
377 | if (conf->retry_read_aligned) | |
378 | md_wakeup_thread(conf->mddev->thread); | |
379 | } | |
4eb788df SL |
380 | } |
381 | ||
773ca82f | 382 | /* should hold conf->device_lock already */ |
566c09c5 SL |
383 | static int release_stripe_list(struct r5conf *conf, |
384 | struct list_head *temp_inactive_list) | |
773ca82f SL |
385 | { |
386 | struct stripe_head *sh; | |
387 | int count = 0; | |
388 | struct llist_node *head; | |
389 | ||
390 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 391 | head = llist_reverse_order(head); |
773ca82f | 392 | while (head) { |
566c09c5 SL |
393 | int hash; |
394 | ||
773ca82f SL |
395 | sh = llist_entry(head, struct stripe_head, release_list); |
396 | head = llist_next(head); | |
397 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ | |
398 | smp_mb(); | |
399 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
400 | /* | |
401 | * Don't worry the bit is set here, because if the bit is set | |
402 | * again, the count is always > 1. This is true for | |
403 | * STRIPE_ON_UNPLUG_LIST bit too. | |
404 | */ | |
566c09c5 SL |
405 | hash = sh->hash_lock_index; |
406 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
407 | count++; |
408 | } | |
409 | ||
410 | return count; | |
411 | } | |
412 | ||
1da177e4 LT |
413 | static void release_stripe(struct stripe_head *sh) |
414 | { | |
d1688a6d | 415 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 416 | unsigned long flags; |
566c09c5 SL |
417 | struct list_head list; |
418 | int hash; | |
773ca82f | 419 | bool wakeup; |
16a53ecc | 420 | |
cf170f3f ES |
421 | /* Avoid release_list until the last reference. |
422 | */ | |
423 | if (atomic_add_unless(&sh->count, -1, 1)) | |
424 | return; | |
425 | ||
ad4068de | 426 | if (unlikely(!conf->mddev->thread) || |
427 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
428 | goto slow_path; |
429 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
430 | if (wakeup) | |
431 | md_wakeup_thread(conf->mddev->thread); | |
432 | return; | |
433 | slow_path: | |
4eb788df | 434 | local_irq_save(flags); |
773ca82f | 435 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 436 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
437 | INIT_LIST_HEAD(&list); |
438 | hash = sh->hash_lock_index; | |
439 | do_release_stripe(conf, sh, &list); | |
4eb788df | 440 | spin_unlock(&conf->device_lock); |
566c09c5 | 441 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
442 | } |
443 | local_irq_restore(flags); | |
1da177e4 LT |
444 | } |
445 | ||
fccddba0 | 446 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 447 | { |
45b4233c DW |
448 | pr_debug("remove_hash(), stripe %llu\n", |
449 | (unsigned long long)sh->sector); | |
1da177e4 | 450 | |
fccddba0 | 451 | hlist_del_init(&sh->hash); |
1da177e4 LT |
452 | } |
453 | ||
d1688a6d | 454 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 455 | { |
fccddba0 | 456 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 457 | |
45b4233c DW |
458 | pr_debug("insert_hash(), stripe %llu\n", |
459 | (unsigned long long)sh->sector); | |
1da177e4 | 460 | |
fccddba0 | 461 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
462 | } |
463 | ||
1da177e4 | 464 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 465 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
466 | { |
467 | struct stripe_head *sh = NULL; | |
468 | struct list_head *first; | |
469 | ||
566c09c5 | 470 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 471 | goto out; |
566c09c5 | 472 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
473 | sh = list_entry(first, struct stripe_head, lru); |
474 | list_del_init(first); | |
475 | remove_hash(sh); | |
476 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 477 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
478 | if (list_empty(conf->inactive_list + hash)) |
479 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
480 | out: |
481 | return sh; | |
482 | } | |
483 | ||
e4e11e38 | 484 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
485 | { |
486 | struct page *p; | |
487 | int i; | |
e4e11e38 | 488 | int num = sh->raid_conf->pool_size; |
1da177e4 | 489 | |
e4e11e38 | 490 | for (i = 0; i < num ; i++) { |
d592a996 | 491 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
492 | p = sh->dev[i].page; |
493 | if (!p) | |
494 | continue; | |
495 | sh->dev[i].page = NULL; | |
2d1f3b5d | 496 | put_page(p); |
1da177e4 LT |
497 | } |
498 | } | |
499 | ||
e4e11e38 | 500 | static int grow_buffers(struct stripe_head *sh) |
1da177e4 LT |
501 | { |
502 | int i; | |
e4e11e38 | 503 | int num = sh->raid_conf->pool_size; |
1da177e4 | 504 | |
e4e11e38 | 505 | for (i = 0; i < num; i++) { |
1da177e4 LT |
506 | struct page *page; |
507 | ||
508 | if (!(page = alloc_page(GFP_KERNEL))) { | |
509 | return 1; | |
510 | } | |
511 | sh->dev[i].page = page; | |
d592a996 | 512 | sh->dev[i].orig_page = page; |
1da177e4 LT |
513 | } |
514 | return 0; | |
515 | } | |
516 | ||
784052ec | 517 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 518 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 519 | struct stripe_head *sh); |
1da177e4 | 520 | |
b5663ba4 | 521 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 522 | { |
d1688a6d | 523 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 524 | int i, seq; |
1da177e4 | 525 | |
78bafebd ES |
526 | BUG_ON(atomic_read(&sh->count) != 0); |
527 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 528 | BUG_ON(stripe_operations_active(sh)); |
d84e0f10 | 529 | |
45b4233c | 530 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 531 | (unsigned long long)sector); |
566c09c5 SL |
532 | retry: |
533 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 534 | sh->generation = conf->generation - previous; |
b5663ba4 | 535 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 536 | sh->sector = sector; |
911d4ee8 | 537 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
538 | sh->state = 0; |
539 | ||
7ecaa1e6 | 540 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
541 | struct r5dev *dev = &sh->dev[i]; |
542 | ||
d84e0f10 | 543 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 544 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 545 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 546 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 547 | dev->read, dev->towrite, dev->written, |
1da177e4 | 548 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 549 | WARN_ON(1); |
1da177e4 LT |
550 | } |
551 | dev->flags = 0; | |
784052ec | 552 | raid5_build_block(sh, i, previous); |
1da177e4 | 553 | } |
566c09c5 SL |
554 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
555 | goto retry; | |
1da177e4 | 556 | insert_hash(conf, sh); |
851c30c9 | 557 | sh->cpu = smp_processor_id(); |
da41ba65 | 558 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
559 | } |
560 | ||
d1688a6d | 561 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 562 | short generation) |
1da177e4 LT |
563 | { |
564 | struct stripe_head *sh; | |
565 | ||
45b4233c | 566 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 567 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 568 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 569 | return sh; |
45b4233c | 570 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
571 | return NULL; |
572 | } | |
573 | ||
674806d6 N |
574 | /* |
575 | * Need to check if array has failed when deciding whether to: | |
576 | * - start an array | |
577 | * - remove non-faulty devices | |
578 | * - add a spare | |
579 | * - allow a reshape | |
580 | * This determination is simple when no reshape is happening. | |
581 | * However if there is a reshape, we need to carefully check | |
582 | * both the before and after sections. | |
583 | * This is because some failed devices may only affect one | |
584 | * of the two sections, and some non-in_sync devices may | |
585 | * be insync in the section most affected by failed devices. | |
586 | */ | |
908f4fbd | 587 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 588 | { |
908f4fbd | 589 | int degraded, degraded2; |
674806d6 | 590 | int i; |
674806d6 N |
591 | |
592 | rcu_read_lock(); | |
593 | degraded = 0; | |
594 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 595 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
596 | if (rdev && test_bit(Faulty, &rdev->flags)) |
597 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
598 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
599 | degraded++; | |
600 | else if (test_bit(In_sync, &rdev->flags)) | |
601 | ; | |
602 | else | |
603 | /* not in-sync or faulty. | |
604 | * If the reshape increases the number of devices, | |
605 | * this is being recovered by the reshape, so | |
606 | * this 'previous' section is not in_sync. | |
607 | * If the number of devices is being reduced however, | |
608 | * the device can only be part of the array if | |
609 | * we are reverting a reshape, so this section will | |
610 | * be in-sync. | |
611 | */ | |
612 | if (conf->raid_disks >= conf->previous_raid_disks) | |
613 | degraded++; | |
614 | } | |
615 | rcu_read_unlock(); | |
908f4fbd N |
616 | if (conf->raid_disks == conf->previous_raid_disks) |
617 | return degraded; | |
674806d6 | 618 | rcu_read_lock(); |
908f4fbd | 619 | degraded2 = 0; |
674806d6 | 620 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 621 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
622 | if (rdev && test_bit(Faulty, &rdev->flags)) |
623 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 624 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 625 | degraded2++; |
674806d6 N |
626 | else if (test_bit(In_sync, &rdev->flags)) |
627 | ; | |
628 | else | |
629 | /* not in-sync or faulty. | |
630 | * If reshape increases the number of devices, this | |
631 | * section has already been recovered, else it | |
632 | * almost certainly hasn't. | |
633 | */ | |
634 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 635 | degraded2++; |
674806d6 N |
636 | } |
637 | rcu_read_unlock(); | |
908f4fbd N |
638 | if (degraded2 > degraded) |
639 | return degraded2; | |
640 | return degraded; | |
641 | } | |
642 | ||
643 | static int has_failed(struct r5conf *conf) | |
644 | { | |
645 | int degraded; | |
646 | ||
647 | if (conf->mddev->reshape_position == MaxSector) | |
648 | return conf->mddev->degraded > conf->max_degraded; | |
649 | ||
650 | degraded = calc_degraded(conf); | |
674806d6 N |
651 | if (degraded > conf->max_degraded) |
652 | return 1; | |
653 | return 0; | |
654 | } | |
655 | ||
b5663ba4 | 656 | static struct stripe_head * |
d1688a6d | 657 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 658 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
659 | { |
660 | struct stripe_head *sh; | |
566c09c5 | 661 | int hash = stripe_hash_locks_hash(sector); |
1da177e4 | 662 | |
45b4233c | 663 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 664 | |
566c09c5 | 665 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
666 | |
667 | do { | |
72626685 | 668 | wait_event_lock_irq(conf->wait_for_stripe, |
a8c906ca | 669 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 670 | *(conf->hash_locks + hash)); |
86b42c71 | 671 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 LT |
672 | if (!sh) { |
673 | if (!conf->inactive_blocked) | |
566c09c5 | 674 | sh = get_free_stripe(conf, hash); |
1da177e4 LT |
675 | if (noblock && sh == NULL) |
676 | break; | |
677 | if (!sh) { | |
678 | conf->inactive_blocked = 1; | |
566c09c5 SL |
679 | wait_event_lock_irq( |
680 | conf->wait_for_stripe, | |
681 | !list_empty(conf->inactive_list + hash) && | |
682 | (atomic_read(&conf->active_stripes) | |
683 | < (conf->max_nr_stripes * 3 / 4) | |
684 | || !conf->inactive_blocked), | |
685 | *(conf->hash_locks + hash)); | |
1da177e4 | 686 | conf->inactive_blocked = 0; |
7da9d450 | 687 | } else { |
b5663ba4 | 688 | init_stripe(sh, sector, previous); |
7da9d450 N |
689 | atomic_inc(&sh->count); |
690 | } | |
e240c183 | 691 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 692 | spin_lock(&conf->device_lock); |
e240c183 | 693 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
694 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
695 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
696 | BUG_ON(list_empty(&sh->lru) && |
697 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
16a53ecc | 698 | list_del_init(&sh->lru); |
bfc90cb0 SL |
699 | if (sh->group) { |
700 | sh->group->stripes_cnt--; | |
701 | sh->group = NULL; | |
702 | } | |
1da177e4 | 703 | } |
7da9d450 | 704 | atomic_inc(&sh->count); |
6d183de4 | 705 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
706 | } |
707 | } while (sh == NULL); | |
708 | ||
566c09c5 | 709 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
710 | return sh; |
711 | } | |
712 | ||
05616be5 N |
713 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
714 | * in this stripe_head. | |
715 | */ | |
716 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
717 | { | |
718 | sector_t progress = conf->reshape_progress; | |
719 | /* Need a memory barrier to make sure we see the value | |
720 | * of conf->generation, or ->data_offset that was set before | |
721 | * reshape_progress was updated. | |
722 | */ | |
723 | smp_rmb(); | |
724 | if (progress == MaxSector) | |
725 | return 0; | |
726 | if (sh->generation == conf->generation - 1) | |
727 | return 0; | |
728 | /* We are in a reshape, and this is a new-generation stripe, | |
729 | * so use new_data_offset. | |
730 | */ | |
731 | return 1; | |
732 | } | |
733 | ||
6712ecf8 N |
734 | static void |
735 | raid5_end_read_request(struct bio *bi, int error); | |
736 | static void | |
737 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 738 | |
c4e5ac0a | 739 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 740 | { |
d1688a6d | 741 | struct r5conf *conf = sh->raid_conf; |
91c00924 DW |
742 | int i, disks = sh->disks; |
743 | ||
744 | might_sleep(); | |
745 | ||
746 | for (i = disks; i--; ) { | |
747 | int rw; | |
9a3e1101 | 748 | int replace_only = 0; |
977df362 N |
749 | struct bio *bi, *rbi; |
750 | struct md_rdev *rdev, *rrdev = NULL; | |
e9c7469b TH |
751 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
752 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
753 | rw = WRITE_FUA; | |
754 | else | |
755 | rw = WRITE; | |
9e444768 | 756 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
620125f2 | 757 | rw |= REQ_DISCARD; |
e9c7469b | 758 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
91c00924 | 759 | rw = READ; |
9a3e1101 N |
760 | else if (test_and_clear_bit(R5_WantReplace, |
761 | &sh->dev[i].flags)) { | |
762 | rw = WRITE; | |
763 | replace_only = 1; | |
764 | } else | |
91c00924 | 765 | continue; |
bc0934f0 SL |
766 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
767 | rw |= REQ_SYNC; | |
91c00924 DW |
768 | |
769 | bi = &sh->dev[i].req; | |
977df362 | 770 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 771 | |
91c00924 | 772 | rcu_read_lock(); |
9a3e1101 | 773 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
774 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
775 | rdev = rcu_dereference(conf->disks[i].rdev); | |
776 | if (!rdev) { | |
777 | rdev = rrdev; | |
778 | rrdev = NULL; | |
779 | } | |
9a3e1101 N |
780 | if (rw & WRITE) { |
781 | if (replace_only) | |
782 | rdev = NULL; | |
dd054fce N |
783 | if (rdev == rrdev) |
784 | /* We raced and saw duplicates */ | |
785 | rrdev = NULL; | |
9a3e1101 | 786 | } else { |
dd054fce | 787 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags) && rrdev) |
9a3e1101 N |
788 | rdev = rrdev; |
789 | rrdev = NULL; | |
790 | } | |
977df362 | 791 | |
91c00924 DW |
792 | if (rdev && test_bit(Faulty, &rdev->flags)) |
793 | rdev = NULL; | |
794 | if (rdev) | |
795 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
796 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
797 | rrdev = NULL; | |
798 | if (rrdev) | |
799 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
800 | rcu_read_unlock(); |
801 | ||
73e92e51 | 802 | /* We have already checked bad blocks for reads. Now |
977df362 N |
803 | * need to check for writes. We never accept write errors |
804 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
805 | */ |
806 | while ((rw & WRITE) && rdev && | |
807 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
808 | sector_t first_bad; | |
809 | int bad_sectors; | |
810 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
811 | &first_bad, &bad_sectors); | |
812 | if (!bad) | |
813 | break; | |
814 | ||
815 | if (bad < 0) { | |
816 | set_bit(BlockedBadBlocks, &rdev->flags); | |
817 | if (!conf->mddev->external && | |
818 | conf->mddev->flags) { | |
819 | /* It is very unlikely, but we might | |
820 | * still need to write out the | |
821 | * bad block log - better give it | |
822 | * a chance*/ | |
823 | md_check_recovery(conf->mddev); | |
824 | } | |
1850753d | 825 | /* |
826 | * Because md_wait_for_blocked_rdev | |
827 | * will dec nr_pending, we must | |
828 | * increment it first. | |
829 | */ | |
830 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
831 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
832 | } else { | |
833 | /* Acknowledged bad block - skip the write */ | |
834 | rdev_dec_pending(rdev, conf->mddev); | |
835 | rdev = NULL; | |
836 | } | |
837 | } | |
838 | ||
91c00924 | 839 | if (rdev) { |
9a3e1101 N |
840 | if (s->syncing || s->expanding || s->expanded |
841 | || s->replacing) | |
91c00924 DW |
842 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
843 | ||
2b7497f0 DW |
844 | set_bit(STRIPE_IO_STARTED, &sh->state); |
845 | ||
2f6db2a7 | 846 | bio_reset(bi); |
91c00924 | 847 | bi->bi_bdev = rdev->bdev; |
2f6db2a7 KO |
848 | bi->bi_rw = rw; |
849 | bi->bi_end_io = (rw & WRITE) | |
850 | ? raid5_end_write_request | |
851 | : raid5_end_read_request; | |
852 | bi->bi_private = sh; | |
853 | ||
91c00924 | 854 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", |
e46b272b | 855 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
856 | bi->bi_rw, i); |
857 | atomic_inc(&sh->count); | |
05616be5 | 858 | if (use_new_offset(conf, sh)) |
4f024f37 | 859 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
860 | + rdev->new_data_offset); |
861 | else | |
4f024f37 | 862 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 863 | + rdev->data_offset); |
3f9e7c14 | 864 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
e59aa23f | 865 | bi->bi_rw |= REQ_NOMERGE; |
3f9e7c14 | 866 | |
d592a996 SL |
867 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
868 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
869 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 870 | bi->bi_vcnt = 1; |
91c00924 DW |
871 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
872 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 873 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
874 | /* |
875 | * If this is discard request, set bi_vcnt 0. We don't | |
876 | * want to confuse SCSI because SCSI will replace payload | |
877 | */ | |
878 | if (rw & REQ_DISCARD) | |
879 | bi->bi_vcnt = 0; | |
977df362 N |
880 | if (rrdev) |
881 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
882 | |
883 | if (conf->mddev->gendisk) | |
884 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
885 | bi, disk_devt(conf->mddev->gendisk), | |
886 | sh->dev[i].sector); | |
91c00924 | 887 | generic_make_request(bi); |
977df362 N |
888 | } |
889 | if (rrdev) { | |
9a3e1101 N |
890 | if (s->syncing || s->expanding || s->expanded |
891 | || s->replacing) | |
977df362 N |
892 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
893 | ||
894 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
895 | ||
2f6db2a7 | 896 | bio_reset(rbi); |
977df362 | 897 | rbi->bi_bdev = rrdev->bdev; |
2f6db2a7 KO |
898 | rbi->bi_rw = rw; |
899 | BUG_ON(!(rw & WRITE)); | |
900 | rbi->bi_end_io = raid5_end_write_request; | |
901 | rbi->bi_private = sh; | |
902 | ||
977df362 N |
903 | pr_debug("%s: for %llu schedule op %ld on " |
904 | "replacement disc %d\n", | |
905 | __func__, (unsigned long long)sh->sector, | |
906 | rbi->bi_rw, i); | |
907 | atomic_inc(&sh->count); | |
05616be5 | 908 | if (use_new_offset(conf, sh)) |
4f024f37 | 909 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
910 | + rrdev->new_data_offset); |
911 | else | |
4f024f37 | 912 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 913 | + rrdev->data_offset); |
d592a996 SL |
914 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
915 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
916 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 917 | rbi->bi_vcnt = 1; |
977df362 N |
918 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
919 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 920 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
921 | /* |
922 | * If this is discard request, set bi_vcnt 0. We don't | |
923 | * want to confuse SCSI because SCSI will replace payload | |
924 | */ | |
925 | if (rw & REQ_DISCARD) | |
926 | rbi->bi_vcnt = 0; | |
e3620a3a JB |
927 | if (conf->mddev->gendisk) |
928 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
929 | rbi, disk_devt(conf->mddev->gendisk), | |
930 | sh->dev[i].sector); | |
977df362 N |
931 | generic_make_request(rbi); |
932 | } | |
933 | if (!rdev && !rrdev) { | |
b062962e | 934 | if (rw & WRITE) |
91c00924 DW |
935 | set_bit(STRIPE_DEGRADED, &sh->state); |
936 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
937 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
938 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
939 | set_bit(STRIPE_HANDLE, &sh->state); | |
940 | } | |
941 | } | |
942 | } | |
943 | ||
944 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
945 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
946 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
947 | struct stripe_head *sh) | |
91c00924 | 948 | { |
7988613b KO |
949 | struct bio_vec bvl; |
950 | struct bvec_iter iter; | |
91c00924 | 951 | struct page *bio_page; |
91c00924 | 952 | int page_offset; |
a08abd8c | 953 | struct async_submit_ctl submit; |
0403e382 | 954 | enum async_tx_flags flags = 0; |
91c00924 | 955 | |
4f024f37 KO |
956 | if (bio->bi_iter.bi_sector >= sector) |
957 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 958 | else |
4f024f37 | 959 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 960 | |
0403e382 DW |
961 | if (frombio) |
962 | flags |= ASYNC_TX_FENCE; | |
963 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
964 | ||
7988613b KO |
965 | bio_for_each_segment(bvl, bio, iter) { |
966 | int len = bvl.bv_len; | |
91c00924 DW |
967 | int clen; |
968 | int b_offset = 0; | |
969 | ||
970 | if (page_offset < 0) { | |
971 | b_offset = -page_offset; | |
972 | page_offset += b_offset; | |
973 | len -= b_offset; | |
974 | } | |
975 | ||
976 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
977 | clen = STRIPE_SIZE - page_offset; | |
978 | else | |
979 | clen = len; | |
980 | ||
981 | if (clen > 0) { | |
7988613b KO |
982 | b_offset += bvl.bv_offset; |
983 | bio_page = bvl.bv_page; | |
d592a996 SL |
984 | if (frombio) { |
985 | if (sh->raid_conf->skip_copy && | |
986 | b_offset == 0 && page_offset == 0 && | |
987 | clen == STRIPE_SIZE) | |
988 | *page = bio_page; | |
989 | else | |
990 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 991 | b_offset, clen, &submit); |
d592a996 SL |
992 | } else |
993 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 994 | page_offset, clen, &submit); |
91c00924 | 995 | } |
a08abd8c DW |
996 | /* chain the operations */ |
997 | submit.depend_tx = tx; | |
998 | ||
91c00924 DW |
999 | if (clen < len) /* hit end of page */ |
1000 | break; | |
1001 | page_offset += len; | |
1002 | } | |
1003 | ||
1004 | return tx; | |
1005 | } | |
1006 | ||
1007 | static void ops_complete_biofill(void *stripe_head_ref) | |
1008 | { | |
1009 | struct stripe_head *sh = stripe_head_ref; | |
1010 | struct bio *return_bi = NULL; | |
e4d84909 | 1011 | int i; |
91c00924 | 1012 | |
e46b272b | 1013 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1014 | (unsigned long long)sh->sector); |
1015 | ||
1016 | /* clear completed biofills */ | |
1017 | for (i = sh->disks; i--; ) { | |
1018 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1019 | |
1020 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1021 | /* and check if we need to reply to a read request, |
1022 | * new R5_Wantfill requests are held off until | |
83de75cc | 1023 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1024 | */ |
1025 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1026 | struct bio *rbi, *rbi2; |
91c00924 | 1027 | |
91c00924 DW |
1028 | BUG_ON(!dev->read); |
1029 | rbi = dev->read; | |
1030 | dev->read = NULL; | |
4f024f37 | 1031 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1032 | dev->sector + STRIPE_SECTORS) { |
1033 | rbi2 = r5_next_bio(rbi, dev->sector); | |
e7836bd6 | 1034 | if (!raid5_dec_bi_active_stripes(rbi)) { |
91c00924 DW |
1035 | rbi->bi_next = return_bi; |
1036 | return_bi = rbi; | |
1037 | } | |
91c00924 DW |
1038 | rbi = rbi2; |
1039 | } | |
1040 | } | |
1041 | } | |
83de75cc | 1042 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 DW |
1043 | |
1044 | return_io(return_bi); | |
1045 | ||
e4d84909 | 1046 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
1047 | release_stripe(sh); |
1048 | } | |
1049 | ||
1050 | static void ops_run_biofill(struct stripe_head *sh) | |
1051 | { | |
1052 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1053 | struct async_submit_ctl submit; |
91c00924 DW |
1054 | int i; |
1055 | ||
e46b272b | 1056 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1057 | (unsigned long long)sh->sector); |
1058 | ||
1059 | for (i = sh->disks; i--; ) { | |
1060 | struct r5dev *dev = &sh->dev[i]; | |
1061 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1062 | struct bio *rbi; | |
b17459c0 | 1063 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1064 | dev->read = rbi = dev->toread; |
1065 | dev->toread = NULL; | |
b17459c0 | 1066 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1067 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1068 | dev->sector + STRIPE_SECTORS) { |
d592a996 SL |
1069 | tx = async_copy_data(0, rbi, &dev->page, |
1070 | dev->sector, tx, sh); | |
91c00924 DW |
1071 | rbi = r5_next_bio(rbi, dev->sector); |
1072 | } | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | atomic_inc(&sh->count); | |
a08abd8c DW |
1077 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1078 | async_trigger_callback(&submit); | |
91c00924 DW |
1079 | } |
1080 | ||
4e7d2c0a | 1081 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1082 | { |
4e7d2c0a | 1083 | struct r5dev *tgt; |
91c00924 | 1084 | |
4e7d2c0a DW |
1085 | if (target < 0) |
1086 | return; | |
91c00924 | 1087 | |
4e7d2c0a | 1088 | tgt = &sh->dev[target]; |
91c00924 DW |
1089 | set_bit(R5_UPTODATE, &tgt->flags); |
1090 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1091 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1092 | } |
1093 | ||
ac6b53b6 | 1094 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1095 | { |
1096 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1097 | |
e46b272b | 1098 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1099 | (unsigned long long)sh->sector); |
1100 | ||
ac6b53b6 | 1101 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1102 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1103 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1104 | |
ecc65c9b DW |
1105 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1106 | if (sh->check_state == check_state_compute_run) | |
1107 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
1108 | set_bit(STRIPE_HANDLE, &sh->state); |
1109 | release_stripe(sh); | |
1110 | } | |
1111 | ||
d6f38f31 DW |
1112 | /* return a pointer to the address conversion region of the scribble buffer */ |
1113 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1114 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1115 | { |
46d5b785 | 1116 | void *addr; |
1117 | ||
1118 | addr = flex_array_get(percpu->scribble, i); | |
1119 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1120 | } | |
1121 | ||
1122 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1123 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1124 | { | |
1125 | void *addr; | |
1126 | ||
1127 | addr = flex_array_get(percpu->scribble, i); | |
1128 | return addr; | |
d6f38f31 DW |
1129 | } |
1130 | ||
1131 | static struct dma_async_tx_descriptor * | |
1132 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1133 | { |
91c00924 | 1134 | int disks = sh->disks; |
46d5b785 | 1135 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1136 | int target = sh->ops.target; |
1137 | struct r5dev *tgt = &sh->dev[target]; | |
1138 | struct page *xor_dest = tgt->page; | |
1139 | int count = 0; | |
1140 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1141 | struct async_submit_ctl submit; |
91c00924 DW |
1142 | int i; |
1143 | ||
1144 | pr_debug("%s: stripe %llu block: %d\n", | |
e46b272b | 1145 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1146 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1147 | ||
1148 | for (i = disks; i--; ) | |
1149 | if (i != target) | |
1150 | xor_srcs[count++] = sh->dev[i].page; | |
1151 | ||
1152 | atomic_inc(&sh->count); | |
1153 | ||
0403e382 | 1154 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1155 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1156 | if (unlikely(count == 1)) |
a08abd8c | 1157 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1158 | else |
a08abd8c | 1159 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1160 | |
91c00924 DW |
1161 | return tx; |
1162 | } | |
1163 | ||
ac6b53b6 DW |
1164 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1165 | * @srcs - (struct page *) array of size sh->disks | |
1166 | * @sh - stripe_head to parse | |
1167 | * | |
1168 | * Populates srcs in proper layout order for the stripe and returns the | |
1169 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1170 | * destination buffer is recorded in srcs[count] and the Q destination | |
1171 | * is recorded in srcs[count+1]]. | |
1172 | */ | |
1173 | static int set_syndrome_sources(struct page **srcs, struct stripe_head *sh) | |
1174 | { | |
1175 | int disks = sh->disks; | |
1176 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1177 | int d0_idx = raid6_d0(sh); | |
1178 | int count; | |
1179 | int i; | |
1180 | ||
1181 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1182 | srcs[i] = NULL; |
ac6b53b6 DW |
1183 | |
1184 | count = 0; | |
1185 | i = d0_idx; | |
1186 | do { | |
1187 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1188 | ||
1189 | srcs[slot] = sh->dev[i].page; | |
1190 | i = raid6_next_disk(i, disks); | |
1191 | } while (i != d0_idx); | |
ac6b53b6 | 1192 | |
e4424fee | 1193 | return syndrome_disks; |
ac6b53b6 DW |
1194 | } |
1195 | ||
1196 | static struct dma_async_tx_descriptor * | |
1197 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1198 | { | |
1199 | int disks = sh->disks; | |
46d5b785 | 1200 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1201 | int target; |
1202 | int qd_idx = sh->qd_idx; | |
1203 | struct dma_async_tx_descriptor *tx; | |
1204 | struct async_submit_ctl submit; | |
1205 | struct r5dev *tgt; | |
1206 | struct page *dest; | |
1207 | int i; | |
1208 | int count; | |
1209 | ||
1210 | if (sh->ops.target < 0) | |
1211 | target = sh->ops.target2; | |
1212 | else if (sh->ops.target2 < 0) | |
1213 | target = sh->ops.target; | |
91c00924 | 1214 | else |
ac6b53b6 DW |
1215 | /* we should only have one valid target */ |
1216 | BUG(); | |
1217 | BUG_ON(target < 0); | |
1218 | pr_debug("%s: stripe %llu block: %d\n", | |
1219 | __func__, (unsigned long long)sh->sector, target); | |
1220 | ||
1221 | tgt = &sh->dev[target]; | |
1222 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1223 | dest = tgt->page; | |
1224 | ||
1225 | atomic_inc(&sh->count); | |
1226 | ||
1227 | if (target == qd_idx) { | |
1228 | count = set_syndrome_sources(blocks, sh); | |
1229 | blocks[count] = NULL; /* regenerating p is not necessary */ | |
1230 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1231 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1232 | ops_complete_compute, sh, | |
46d5b785 | 1233 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1234 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1235 | } else { | |
1236 | /* Compute any data- or p-drive using XOR */ | |
1237 | count = 0; | |
1238 | for (i = disks; i-- ; ) { | |
1239 | if (i == target || i == qd_idx) | |
1240 | continue; | |
1241 | blocks[count++] = sh->dev[i].page; | |
1242 | } | |
1243 | ||
0403e382 DW |
1244 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1245 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1246 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1247 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1248 | } | |
91c00924 | 1249 | |
91c00924 DW |
1250 | return tx; |
1251 | } | |
1252 | ||
ac6b53b6 DW |
1253 | static struct dma_async_tx_descriptor * |
1254 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1255 | { | |
1256 | int i, count, disks = sh->disks; | |
1257 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1258 | int d0_idx = raid6_d0(sh); | |
1259 | int faila = -1, failb = -1; | |
1260 | int target = sh->ops.target; | |
1261 | int target2 = sh->ops.target2; | |
1262 | struct r5dev *tgt = &sh->dev[target]; | |
1263 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1264 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1265 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1266 | struct async_submit_ctl submit; |
1267 | ||
1268 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", | |
1269 | __func__, (unsigned long long)sh->sector, target, target2); | |
1270 | BUG_ON(target < 0 || target2 < 0); | |
1271 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1272 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1273 | ||
6c910a78 | 1274 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1275 | * slot number conversion for 'faila' and 'failb' |
1276 | */ | |
1277 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1278 | blocks[i] = NULL; |
ac6b53b6 DW |
1279 | count = 0; |
1280 | i = d0_idx; | |
1281 | do { | |
1282 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1283 | ||
1284 | blocks[slot] = sh->dev[i].page; | |
1285 | ||
1286 | if (i == target) | |
1287 | faila = slot; | |
1288 | if (i == target2) | |
1289 | failb = slot; | |
1290 | i = raid6_next_disk(i, disks); | |
1291 | } while (i != d0_idx); | |
ac6b53b6 DW |
1292 | |
1293 | BUG_ON(faila == failb); | |
1294 | if (failb < faila) | |
1295 | swap(faila, failb); | |
1296 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1297 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1298 | ||
1299 | atomic_inc(&sh->count); | |
1300 | ||
1301 | if (failb == syndrome_disks+1) { | |
1302 | /* Q disk is one of the missing disks */ | |
1303 | if (faila == syndrome_disks) { | |
1304 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1305 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1306 | ops_complete_compute, sh, | |
46d5b785 | 1307 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1308 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1309 | STRIPE_SIZE, &submit); |
1310 | } else { | |
1311 | struct page *dest; | |
1312 | int data_target; | |
1313 | int qd_idx = sh->qd_idx; | |
1314 | ||
1315 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1316 | if (target == qd_idx) | |
1317 | data_target = target2; | |
1318 | else | |
1319 | data_target = target; | |
1320 | ||
1321 | count = 0; | |
1322 | for (i = disks; i-- ; ) { | |
1323 | if (i == data_target || i == qd_idx) | |
1324 | continue; | |
1325 | blocks[count++] = sh->dev[i].page; | |
1326 | } | |
1327 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1328 | init_async_submit(&submit, |
1329 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1330 | NULL, NULL, NULL, | |
46d5b785 | 1331 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1332 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1333 | &submit); | |
1334 | ||
1335 | count = set_syndrome_sources(blocks, sh); | |
0403e382 DW |
1336 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1337 | ops_complete_compute, sh, | |
46d5b785 | 1338 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1339 | return async_gen_syndrome(blocks, 0, count+2, |
1340 | STRIPE_SIZE, &submit); | |
1341 | } | |
ac6b53b6 | 1342 | } else { |
6c910a78 DW |
1343 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1344 | ops_complete_compute, sh, | |
46d5b785 | 1345 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1346 | if (failb == syndrome_disks) { |
1347 | /* We're missing D+P. */ | |
1348 | return async_raid6_datap_recov(syndrome_disks+2, | |
1349 | STRIPE_SIZE, faila, | |
1350 | blocks, &submit); | |
1351 | } else { | |
1352 | /* We're missing D+D. */ | |
1353 | return async_raid6_2data_recov(syndrome_disks+2, | |
1354 | STRIPE_SIZE, faila, failb, | |
1355 | blocks, &submit); | |
1356 | } | |
ac6b53b6 DW |
1357 | } |
1358 | } | |
1359 | ||
91c00924 DW |
1360 | static void ops_complete_prexor(void *stripe_head_ref) |
1361 | { | |
1362 | struct stripe_head *sh = stripe_head_ref; | |
1363 | ||
e46b272b | 1364 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1365 | (unsigned long long)sh->sector); |
91c00924 DW |
1366 | } |
1367 | ||
1368 | static struct dma_async_tx_descriptor * | |
d6f38f31 DW |
1369 | ops_run_prexor(struct stripe_head *sh, struct raid5_percpu *percpu, |
1370 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1371 | { |
91c00924 | 1372 | int disks = sh->disks; |
46d5b785 | 1373 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1374 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1375 | struct async_submit_ctl submit; |
91c00924 DW |
1376 | |
1377 | /* existing parity data subtracted */ | |
1378 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1379 | ||
e46b272b | 1380 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1381 | (unsigned long long)sh->sector); |
1382 | ||
1383 | for (i = disks; i--; ) { | |
1384 | struct r5dev *dev = &sh->dev[i]; | |
1385 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1386 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1387 | xor_srcs[count++] = dev->page; |
1388 | } | |
1389 | ||
0403e382 | 1390 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1391 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1392 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1393 | |
1394 | return tx; | |
1395 | } | |
1396 | ||
1397 | static struct dma_async_tx_descriptor * | |
d8ee0728 | 1398 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1399 | { |
1400 | int disks = sh->disks; | |
d8ee0728 | 1401 | int i; |
91c00924 | 1402 | |
e46b272b | 1403 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1404 | (unsigned long long)sh->sector); |
1405 | ||
1406 | for (i = disks; i--; ) { | |
1407 | struct r5dev *dev = &sh->dev[i]; | |
1408 | struct bio *chosen; | |
91c00924 | 1409 | |
d8ee0728 | 1410 | if (test_and_clear_bit(R5_Wantdrain, &dev->flags)) { |
91c00924 DW |
1411 | struct bio *wbi; |
1412 | ||
b17459c0 | 1413 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1414 | chosen = dev->towrite; |
1415 | dev->towrite = NULL; | |
1416 | BUG_ON(dev->written); | |
1417 | wbi = dev->written = chosen; | |
b17459c0 | 1418 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1419 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1420 | |
4f024f37 | 1421 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1422 | dev->sector + STRIPE_SECTORS) { |
e9c7469b TH |
1423 | if (wbi->bi_rw & REQ_FUA) |
1424 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1425 | if (wbi->bi_rw & REQ_SYNC) |
1426 | set_bit(R5_SyncIO, &dev->flags); | |
9e444768 | 1427 | if (wbi->bi_rw & REQ_DISCARD) |
620125f2 | 1428 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1429 | else { |
1430 | tx = async_copy_data(1, wbi, &dev->page, | |
1431 | dev->sector, tx, sh); | |
1432 | if (dev->page != dev->orig_page) { | |
1433 | set_bit(R5_SkipCopy, &dev->flags); | |
1434 | clear_bit(R5_UPTODATE, &dev->flags); | |
1435 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1436 | } | |
1437 | } | |
91c00924 DW |
1438 | wbi = r5_next_bio(wbi, dev->sector); |
1439 | } | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | return tx; | |
1444 | } | |
1445 | ||
ac6b53b6 | 1446 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1447 | { |
1448 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1449 | int disks = sh->disks; |
1450 | int pd_idx = sh->pd_idx; | |
1451 | int qd_idx = sh->qd_idx; | |
1452 | int i; | |
9e444768 | 1453 | bool fua = false, sync = false, discard = false; |
91c00924 | 1454 | |
e46b272b | 1455 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1456 | (unsigned long long)sh->sector); |
1457 | ||
bc0934f0 | 1458 | for (i = disks; i--; ) { |
e9c7469b | 1459 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1460 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1461 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1462 | } |
e9c7469b | 1463 | |
91c00924 DW |
1464 | for (i = disks; i--; ) { |
1465 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1466 | |
e9c7469b | 1467 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1468 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1469 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1470 | if (fua) |
1471 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1472 | if (sync) |
1473 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1474 | } |
91c00924 DW |
1475 | } |
1476 | ||
d8ee0728 DW |
1477 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1478 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1479 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1480 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1481 | else { | |
1482 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1483 | sh->reconstruct_state = reconstruct_state_result; | |
1484 | } | |
91c00924 DW |
1485 | |
1486 | set_bit(STRIPE_HANDLE, &sh->state); | |
1487 | release_stripe(sh); | |
1488 | } | |
1489 | ||
1490 | static void | |
ac6b53b6 DW |
1491 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1492 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1493 | { |
91c00924 | 1494 | int disks = sh->disks; |
46d5b785 | 1495 | struct page **xor_srcs = to_addr_page(percpu, 0); |
a08abd8c | 1496 | struct async_submit_ctl submit; |
91c00924 DW |
1497 | int count = 0, pd_idx = sh->pd_idx, i; |
1498 | struct page *xor_dest; | |
d8ee0728 | 1499 | int prexor = 0; |
91c00924 | 1500 | unsigned long flags; |
91c00924 | 1501 | |
e46b272b | 1502 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1503 | (unsigned long long)sh->sector); |
1504 | ||
620125f2 SL |
1505 | for (i = 0; i < sh->disks; i++) { |
1506 | if (pd_idx == i) | |
1507 | continue; | |
1508 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1509 | break; | |
1510 | } | |
1511 | if (i >= sh->disks) { | |
1512 | atomic_inc(&sh->count); | |
620125f2 SL |
1513 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1514 | ops_complete_reconstruct(sh); | |
1515 | return; | |
1516 | } | |
91c00924 DW |
1517 | /* check if prexor is active which means only process blocks |
1518 | * that are part of a read-modify-write (written) | |
1519 | */ | |
d8ee0728 DW |
1520 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
1521 | prexor = 1; | |
91c00924 DW |
1522 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1523 | for (i = disks; i--; ) { | |
1524 | struct r5dev *dev = &sh->dev[i]; | |
1525 | if (dev->written) | |
1526 | xor_srcs[count++] = dev->page; | |
1527 | } | |
1528 | } else { | |
1529 | xor_dest = sh->dev[pd_idx].page; | |
1530 | for (i = disks; i--; ) { | |
1531 | struct r5dev *dev = &sh->dev[i]; | |
1532 | if (i != pd_idx) | |
1533 | xor_srcs[count++] = dev->page; | |
1534 | } | |
1535 | } | |
1536 | ||
91c00924 DW |
1537 | /* 1/ if we prexor'd then the dest is reused as a source |
1538 | * 2/ if we did not prexor then we are redoing the parity | |
1539 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1540 | * for the synchronous xor case | |
1541 | */ | |
88ba2aa5 | 1542 | flags = ASYNC_TX_ACK | |
91c00924 DW |
1543 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); |
1544 | ||
1545 | atomic_inc(&sh->count); | |
1546 | ||
ac6b53b6 | 1547 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, sh, |
46d5b785 | 1548 | to_addr_conv(sh, percpu, 0)); |
a08abd8c DW |
1549 | if (unlikely(count == 1)) |
1550 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1551 | else | |
1552 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
91c00924 DW |
1553 | } |
1554 | ||
ac6b53b6 DW |
1555 | static void |
1556 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1557 | struct dma_async_tx_descriptor *tx) | |
1558 | { | |
1559 | struct async_submit_ctl submit; | |
46d5b785 | 1560 | struct page **blocks = to_addr_page(percpu, 0); |
620125f2 | 1561 | int count, i; |
ac6b53b6 DW |
1562 | |
1563 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1564 | ||
620125f2 SL |
1565 | for (i = 0; i < sh->disks; i++) { |
1566 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1567 | continue; | |
1568 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1569 | break; | |
1570 | } | |
1571 | if (i >= sh->disks) { | |
1572 | atomic_inc(&sh->count); | |
620125f2 SL |
1573 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1574 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1575 | ops_complete_reconstruct(sh); | |
1576 | return; | |
1577 | } | |
1578 | ||
ac6b53b6 DW |
1579 | count = set_syndrome_sources(blocks, sh); |
1580 | ||
1581 | atomic_inc(&sh->count); | |
1582 | ||
1583 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_reconstruct, | |
46d5b785 | 1584 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 | 1585 | async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
91c00924 DW |
1586 | } |
1587 | ||
1588 | static void ops_complete_check(void *stripe_head_ref) | |
1589 | { | |
1590 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1591 | |
e46b272b | 1592 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1593 | (unsigned long long)sh->sector); |
1594 | ||
ecc65c9b | 1595 | sh->check_state = check_state_check_result; |
91c00924 DW |
1596 | set_bit(STRIPE_HANDLE, &sh->state); |
1597 | release_stripe(sh); | |
1598 | } | |
1599 | ||
ac6b53b6 | 1600 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1601 | { |
91c00924 | 1602 | int disks = sh->disks; |
ac6b53b6 DW |
1603 | int pd_idx = sh->pd_idx; |
1604 | int qd_idx = sh->qd_idx; | |
1605 | struct page *xor_dest; | |
46d5b785 | 1606 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1607 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1608 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1609 | int count; |
1610 | int i; | |
91c00924 | 1611 | |
e46b272b | 1612 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1613 | (unsigned long long)sh->sector); |
1614 | ||
ac6b53b6 DW |
1615 | count = 0; |
1616 | xor_dest = sh->dev[pd_idx].page; | |
1617 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1618 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1619 | if (i == pd_idx || i == qd_idx) |
1620 | continue; | |
1621 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1622 | } |
1623 | ||
d6f38f31 | 1624 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1625 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1626 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1627 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1628 | |
91c00924 | 1629 | atomic_inc(&sh->count); |
a08abd8c DW |
1630 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1631 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1632 | } |
1633 | ||
ac6b53b6 DW |
1634 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1635 | { | |
46d5b785 | 1636 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1637 | struct async_submit_ctl submit; |
1638 | int count; | |
1639 | ||
1640 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1641 | (unsigned long long)sh->sector, checkp); | |
1642 | ||
1643 | count = set_syndrome_sources(srcs, sh); | |
1644 | if (!checkp) | |
1645 | srcs[count] = NULL; | |
91c00924 | 1646 | |
91c00924 | 1647 | atomic_inc(&sh->count); |
ac6b53b6 | 1648 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1649 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1650 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1651 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1652 | } |
1653 | ||
51acbcec | 1654 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1655 | { |
1656 | int overlap_clear = 0, i, disks = sh->disks; | |
1657 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1658 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1659 | int level = conf->level; |
d6f38f31 DW |
1660 | struct raid5_percpu *percpu; |
1661 | unsigned long cpu; | |
91c00924 | 1662 | |
d6f38f31 DW |
1663 | cpu = get_cpu(); |
1664 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1665 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1666 | ops_run_biofill(sh); |
1667 | overlap_clear++; | |
1668 | } | |
1669 | ||
7b3a871e | 1670 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1671 | if (level < 6) |
1672 | tx = ops_run_compute5(sh, percpu); | |
1673 | else { | |
1674 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1675 | tx = ops_run_compute6_1(sh, percpu); | |
1676 | else | |
1677 | tx = ops_run_compute6_2(sh, percpu); | |
1678 | } | |
1679 | /* terminate the chain if reconstruct is not set to be run */ | |
1680 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1681 | async_tx_ack(tx); |
1682 | } | |
91c00924 | 1683 | |
600aa109 | 1684 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) |
d6f38f31 | 1685 | tx = ops_run_prexor(sh, percpu, tx); |
91c00924 | 1686 | |
600aa109 | 1687 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1688 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1689 | overlap_clear++; |
1690 | } | |
1691 | ||
ac6b53b6 DW |
1692 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1693 | if (level < 6) | |
1694 | ops_run_reconstruct5(sh, percpu, tx); | |
1695 | else | |
1696 | ops_run_reconstruct6(sh, percpu, tx); | |
1697 | } | |
91c00924 | 1698 | |
ac6b53b6 DW |
1699 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1700 | if (sh->check_state == check_state_run) | |
1701 | ops_run_check_p(sh, percpu); | |
1702 | else if (sh->check_state == check_state_run_q) | |
1703 | ops_run_check_pq(sh, percpu, 0); | |
1704 | else if (sh->check_state == check_state_run_pq) | |
1705 | ops_run_check_pq(sh, percpu, 1); | |
1706 | else | |
1707 | BUG(); | |
1708 | } | |
91c00924 | 1709 | |
91c00924 DW |
1710 | if (overlap_clear) |
1711 | for (i = disks; i--; ) { | |
1712 | struct r5dev *dev = &sh->dev[i]; | |
1713 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1714 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1715 | } | |
d6f38f31 | 1716 | put_cpu(); |
91c00924 DW |
1717 | } |
1718 | ||
566c09c5 | 1719 | static int grow_one_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
1720 | { |
1721 | struct stripe_head *sh; | |
6ce32846 | 1722 | sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL); |
3f294f4f N |
1723 | if (!sh) |
1724 | return 0; | |
6ce32846 | 1725 | |
3f294f4f | 1726 | sh->raid_conf = conf; |
3f294f4f | 1727 | |
b17459c0 SL |
1728 | spin_lock_init(&sh->stripe_lock); |
1729 | ||
e4e11e38 N |
1730 | if (grow_buffers(sh)) { |
1731 | shrink_buffers(sh); | |
3f294f4f N |
1732 | kmem_cache_free(conf->slab_cache, sh); |
1733 | return 0; | |
1734 | } | |
566c09c5 | 1735 | sh->hash_lock_index = hash; |
3f294f4f N |
1736 | /* we just created an active stripe so... */ |
1737 | atomic_set(&sh->count, 1); | |
1738 | atomic_inc(&conf->active_stripes); | |
1739 | INIT_LIST_HEAD(&sh->lru); | |
1740 | release_stripe(sh); | |
1741 | return 1; | |
1742 | } | |
1743 | ||
d1688a6d | 1744 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 1745 | { |
e18b890b | 1746 | struct kmem_cache *sc; |
5e5e3e78 | 1747 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
566c09c5 | 1748 | int hash; |
1da177e4 | 1749 | |
f4be6b43 N |
1750 | if (conf->mddev->gendisk) |
1751 | sprintf(conf->cache_name[0], | |
1752 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
1753 | else | |
1754 | sprintf(conf->cache_name[0], | |
1755 | "raid%d-%p", conf->level, conf->mddev); | |
1756 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
1757 | ||
ad01c9e3 N |
1758 | conf->active_name = 0; |
1759 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 1760 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 1761 | 0, 0, NULL); |
1da177e4 LT |
1762 | if (!sc) |
1763 | return 1; | |
1764 | conf->slab_cache = sc; | |
ad01c9e3 | 1765 | conf->pool_size = devs; |
566c09c5 SL |
1766 | hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; |
1767 | while (num--) { | |
1768 | if (!grow_one_stripe(conf, hash)) | |
1da177e4 | 1769 | return 1; |
566c09c5 SL |
1770 | conf->max_nr_stripes++; |
1771 | hash = (hash + 1) % NR_STRIPE_HASH_LOCKS; | |
1772 | } | |
1da177e4 LT |
1773 | return 0; |
1774 | } | |
29269553 | 1775 | |
d6f38f31 DW |
1776 | /** |
1777 | * scribble_len - return the required size of the scribble region | |
1778 | * @num - total number of disks in the array | |
1779 | * | |
1780 | * The size must be enough to contain: | |
1781 | * 1/ a struct page pointer for each device in the array +2 | |
1782 | * 2/ room to convert each entry in (1) to its corresponding dma | |
1783 | * (dma_map_page()) or page (page_address()) address. | |
1784 | * | |
1785 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
1786 | * calculate over all devices (not just the data blocks), using zeros in place | |
1787 | * of the P and Q blocks. | |
1788 | */ | |
46d5b785 | 1789 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 1790 | { |
46d5b785 | 1791 | struct flex_array *ret; |
d6f38f31 DW |
1792 | size_t len; |
1793 | ||
1794 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 1795 | ret = flex_array_alloc(len, cnt, flags); |
1796 | if (!ret) | |
1797 | return NULL; | |
1798 | /* always prealloc all elements, so no locking is required */ | |
1799 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
1800 | flex_array_free(ret); | |
1801 | return NULL; | |
1802 | } | |
1803 | return ret; | |
d6f38f31 DW |
1804 | } |
1805 | ||
d1688a6d | 1806 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
1807 | { |
1808 | /* Make all the stripes able to hold 'newsize' devices. | |
1809 | * New slots in each stripe get 'page' set to a new page. | |
1810 | * | |
1811 | * This happens in stages: | |
1812 | * 1/ create a new kmem_cache and allocate the required number of | |
1813 | * stripe_heads. | |
83f0d77a | 1814 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
1815 | * to the new stripe_heads. This will have the side effect of |
1816 | * freezing the array as once all stripe_heads have been collected, | |
1817 | * no IO will be possible. Old stripe heads are freed once their | |
1818 | * pages have been transferred over, and the old kmem_cache is | |
1819 | * freed when all stripes are done. | |
1820 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
1821 | * we simple return a failre status - no need to clean anything up. | |
1822 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
1823 | * If this fails, we don't bother trying the shrink the | |
1824 | * stripe_heads down again, we just leave them as they are. | |
1825 | * As each stripe_head is processed the new one is released into | |
1826 | * active service. | |
1827 | * | |
1828 | * Once step2 is started, we cannot afford to wait for a write, | |
1829 | * so we use GFP_NOIO allocations. | |
1830 | */ | |
1831 | struct stripe_head *osh, *nsh; | |
1832 | LIST_HEAD(newstripes); | |
1833 | struct disk_info *ndisks; | |
d6f38f31 | 1834 | unsigned long cpu; |
b5470dc5 | 1835 | int err; |
e18b890b | 1836 | struct kmem_cache *sc; |
ad01c9e3 | 1837 | int i; |
566c09c5 | 1838 | int hash, cnt; |
ad01c9e3 N |
1839 | |
1840 | if (newsize <= conf->pool_size) | |
1841 | return 0; /* never bother to shrink */ | |
1842 | ||
b5470dc5 DW |
1843 | err = md_allow_write(conf->mddev); |
1844 | if (err) | |
1845 | return err; | |
2a2275d6 | 1846 | |
ad01c9e3 N |
1847 | /* Step 1 */ |
1848 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
1849 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 1850 | 0, 0, NULL); |
ad01c9e3 N |
1851 | if (!sc) |
1852 | return -ENOMEM; | |
1853 | ||
1854 | for (i = conf->max_nr_stripes; i; i--) { | |
6ce32846 | 1855 | nsh = kmem_cache_zalloc(sc, GFP_KERNEL); |
ad01c9e3 N |
1856 | if (!nsh) |
1857 | break; | |
1858 | ||
ad01c9e3 | 1859 | nsh->raid_conf = conf; |
cb13ff69 | 1860 | spin_lock_init(&nsh->stripe_lock); |
ad01c9e3 N |
1861 | |
1862 | list_add(&nsh->lru, &newstripes); | |
1863 | } | |
1864 | if (i) { | |
1865 | /* didn't get enough, give up */ | |
1866 | while (!list_empty(&newstripes)) { | |
1867 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1868 | list_del(&nsh->lru); | |
1869 | kmem_cache_free(sc, nsh); | |
1870 | } | |
1871 | kmem_cache_destroy(sc); | |
1872 | return -ENOMEM; | |
1873 | } | |
1874 | /* Step 2 - Must use GFP_NOIO now. | |
1875 | * OK, we have enough stripes, start collecting inactive | |
1876 | * stripes and copying them over | |
1877 | */ | |
566c09c5 SL |
1878 | hash = 0; |
1879 | cnt = 0; | |
ad01c9e3 | 1880 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 SL |
1881 | lock_device_hash_lock(conf, hash); |
1882 | wait_event_cmd(conf->wait_for_stripe, | |
1883 | !list_empty(conf->inactive_list + hash), | |
1884 | unlock_device_hash_lock(conf, hash), | |
1885 | lock_device_hash_lock(conf, hash)); | |
1886 | osh = get_free_stripe(conf, hash); | |
1887 | unlock_device_hash_lock(conf, hash); | |
ad01c9e3 | 1888 | atomic_set(&nsh->count, 1); |
d592a996 | 1889 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 1890 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
1891 | nsh->dev[i].orig_page = osh->dev[i].page; |
1892 | } | |
ad01c9e3 N |
1893 | for( ; i<newsize; i++) |
1894 | nsh->dev[i].page = NULL; | |
566c09c5 | 1895 | nsh->hash_lock_index = hash; |
ad01c9e3 | 1896 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
1897 | cnt++; |
1898 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
1899 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
1900 | hash++; | |
1901 | cnt = 0; | |
1902 | } | |
ad01c9e3 N |
1903 | } |
1904 | kmem_cache_destroy(conf->slab_cache); | |
1905 | ||
1906 | /* Step 3. | |
1907 | * At this point, we are holding all the stripes so the array | |
1908 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 1909 | * conf->disks and the scribble region |
ad01c9e3 N |
1910 | */ |
1911 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
1912 | if (ndisks) { | |
1913 | for (i=0; i<conf->raid_disks; i++) | |
1914 | ndisks[i] = conf->disks[i]; | |
1915 | kfree(conf->disks); | |
1916 | conf->disks = ndisks; | |
1917 | } else | |
1918 | err = -ENOMEM; | |
1919 | ||
d6f38f31 | 1920 | get_online_cpus(); |
d6f38f31 DW |
1921 | for_each_present_cpu(cpu) { |
1922 | struct raid5_percpu *percpu; | |
46d5b785 | 1923 | struct flex_array *scribble; |
d6f38f31 DW |
1924 | |
1925 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
46d5b785 | 1926 | scribble = scribble_alloc(newsize, conf->chunk_sectors / |
1927 | STRIPE_SECTORS, GFP_NOIO); | |
d6f38f31 DW |
1928 | |
1929 | if (scribble) { | |
46d5b785 | 1930 | flex_array_free(percpu->scribble); |
d6f38f31 DW |
1931 | percpu->scribble = scribble; |
1932 | } else { | |
1933 | err = -ENOMEM; | |
1934 | break; | |
1935 | } | |
1936 | } | |
1937 | put_online_cpus(); | |
1938 | ||
ad01c9e3 N |
1939 | /* Step 4, return new stripes to service */ |
1940 | while(!list_empty(&newstripes)) { | |
1941 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
1942 | list_del_init(&nsh->lru); | |
d6f38f31 | 1943 | |
ad01c9e3 N |
1944 | for (i=conf->raid_disks; i < newsize; i++) |
1945 | if (nsh->dev[i].page == NULL) { | |
1946 | struct page *p = alloc_page(GFP_NOIO); | |
1947 | nsh->dev[i].page = p; | |
d592a996 | 1948 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
1949 | if (!p) |
1950 | err = -ENOMEM; | |
1951 | } | |
1952 | release_stripe(nsh); | |
1953 | } | |
1954 | /* critical section pass, GFP_NOIO no longer needed */ | |
1955 | ||
1956 | conf->slab_cache = sc; | |
1957 | conf->active_name = 1-conf->active_name; | |
1958 | conf->pool_size = newsize; | |
1959 | return err; | |
1960 | } | |
1da177e4 | 1961 | |
566c09c5 | 1962 | static int drop_one_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
1963 | { |
1964 | struct stripe_head *sh; | |
1965 | ||
566c09c5 SL |
1966 | spin_lock_irq(conf->hash_locks + hash); |
1967 | sh = get_free_stripe(conf, hash); | |
1968 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
1969 | if (!sh) |
1970 | return 0; | |
78bafebd | 1971 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 1972 | shrink_buffers(sh); |
3f294f4f N |
1973 | kmem_cache_free(conf->slab_cache, sh); |
1974 | atomic_dec(&conf->active_stripes); | |
1975 | return 1; | |
1976 | } | |
1977 | ||
d1688a6d | 1978 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 1979 | { |
566c09c5 SL |
1980 | int hash; |
1981 | for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++) | |
1982 | while (drop_one_stripe(conf, hash)) | |
1983 | ; | |
3f294f4f | 1984 | |
29fc7e3e N |
1985 | if (conf->slab_cache) |
1986 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
1987 | conf->slab_cache = NULL; |
1988 | } | |
1989 | ||
6712ecf8 | 1990 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 1991 | { |
99c0fb5f | 1992 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 1993 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 1994 | int disks = sh->disks, i; |
1da177e4 | 1995 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 1996 | char b[BDEVNAME_SIZE]; |
dd054fce | 1997 | struct md_rdev *rdev = NULL; |
05616be5 | 1998 | sector_t s; |
1da177e4 LT |
1999 | |
2000 | for (i=0 ; i<disks; i++) | |
2001 | if (bi == &sh->dev[i].req) | |
2002 | break; | |
2003 | ||
45b4233c DW |
2004 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
2005 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
2006 | uptodate); |
2007 | if (i == disks) { | |
2008 | BUG(); | |
6712ecf8 | 2009 | return; |
1da177e4 | 2010 | } |
14a75d3e | 2011 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2012 | /* If replacement finished while this request was outstanding, |
2013 | * 'replacement' might be NULL already. | |
2014 | * In that case it moved down to 'rdev'. | |
2015 | * rdev is not removed until all requests are finished. | |
2016 | */ | |
14a75d3e | 2017 | rdev = conf->disks[i].replacement; |
dd054fce | 2018 | if (!rdev) |
14a75d3e | 2019 | rdev = conf->disks[i].rdev; |
1da177e4 | 2020 | |
05616be5 N |
2021 | if (use_new_offset(conf, sh)) |
2022 | s = sh->sector + rdev->new_data_offset; | |
2023 | else | |
2024 | s = sh->sector + rdev->data_offset; | |
1da177e4 | 2025 | if (uptodate) { |
1da177e4 | 2026 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2027 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2028 | /* Note that this cannot happen on a |
2029 | * replacement device. We just fail those on | |
2030 | * any error | |
2031 | */ | |
8bda470e CD |
2032 | printk_ratelimited( |
2033 | KERN_INFO | |
2034 | "md/raid:%s: read error corrected" | |
2035 | " (%lu sectors at %llu on %s)\n", | |
2036 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 2037 | (unsigned long long)s, |
8bda470e | 2038 | bdevname(rdev->bdev, b)); |
ddd5115f | 2039 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2040 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2041 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2042 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2043 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2044 | ||
14a75d3e N |
2045 | if (atomic_read(&rdev->read_errors)) |
2046 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2047 | } else { |
14a75d3e | 2048 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2049 | int retry = 0; |
2e8ac303 | 2050 | int set_bad = 0; |
d6950432 | 2051 | |
1da177e4 | 2052 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2053 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
2054 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
2055 | printk_ratelimited( | |
2056 | KERN_WARNING | |
2057 | "md/raid:%s: read error on replacement device " | |
2058 | "(sector %llu on %s).\n", | |
2059 | mdname(conf->mddev), | |
05616be5 | 2060 | (unsigned long long)s, |
14a75d3e | 2061 | bdn); |
2e8ac303 | 2062 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2063 | set_bad = 1; | |
8bda470e CD |
2064 | printk_ratelimited( |
2065 | KERN_WARNING | |
2066 | "md/raid:%s: read error not correctable " | |
2067 | "(sector %llu on %s).\n", | |
2068 | mdname(conf->mddev), | |
05616be5 | 2069 | (unsigned long long)s, |
8bda470e | 2070 | bdn); |
2e8ac303 | 2071 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2072 | /* Oh, no!!! */ |
2e8ac303 | 2073 | set_bad = 1; |
8bda470e CD |
2074 | printk_ratelimited( |
2075 | KERN_WARNING | |
2076 | "md/raid:%s: read error NOT corrected!! " | |
2077 | "(sector %llu on %s).\n", | |
2078 | mdname(conf->mddev), | |
05616be5 | 2079 | (unsigned long long)s, |
8bda470e | 2080 | bdn); |
2e8ac303 | 2081 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2082 | > conf->max_nr_stripes) |
14f8d26b | 2083 | printk(KERN_WARNING |
0c55e022 | 2084 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2085 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2086 | else |
2087 | retry = 1; | |
edfa1f65 BY |
2088 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2089 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2090 | retry = 1; | |
ba22dcbf | 2091 | if (retry) |
3f9e7c14 | 2092 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2093 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2094 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2095 | } else | |
2096 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2097 | else { |
4e5314b5 N |
2098 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2099 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2100 | if (!(set_bad |
2101 | && test_bit(In_sync, &rdev->flags) | |
2102 | && rdev_set_badblocks( | |
2103 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2104 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2105 | } |
1da177e4 | 2106 | } |
14a75d3e | 2107 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
2108 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2109 | set_bit(STRIPE_HANDLE, &sh->state); | |
2110 | release_stripe(sh); | |
1da177e4 LT |
2111 | } |
2112 | ||
d710e138 | 2113 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 2114 | { |
99c0fb5f | 2115 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2116 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2117 | int disks = sh->disks, i; |
977df362 | 2118 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 2119 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
2120 | sector_t first_bad; |
2121 | int bad_sectors; | |
977df362 | 2122 | int replacement = 0; |
1da177e4 | 2123 | |
977df362 N |
2124 | for (i = 0 ; i < disks; i++) { |
2125 | if (bi == &sh->dev[i].req) { | |
2126 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2127 | break; |
977df362 N |
2128 | } |
2129 | if (bi == &sh->dev[i].rreq) { | |
2130 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2131 | if (rdev) |
2132 | replacement = 1; | |
2133 | else | |
2134 | /* rdev was removed and 'replacement' | |
2135 | * replaced it. rdev is not removed | |
2136 | * until all requests are finished. | |
2137 | */ | |
2138 | rdev = conf->disks[i].rdev; | |
977df362 N |
2139 | break; |
2140 | } | |
2141 | } | |
45b4233c | 2142 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
2143 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
2144 | uptodate); | |
2145 | if (i == disks) { | |
2146 | BUG(); | |
6712ecf8 | 2147 | return; |
1da177e4 LT |
2148 | } |
2149 | ||
977df362 N |
2150 | if (replacement) { |
2151 | if (!uptodate) | |
2152 | md_error(conf->mddev, rdev); | |
2153 | else if (is_badblock(rdev, sh->sector, | |
2154 | STRIPE_SECTORS, | |
2155 | &first_bad, &bad_sectors)) | |
2156 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2157 | } else { | |
2158 | if (!uptodate) { | |
9f97e4b1 | 2159 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2160 | set_bit(WriteErrorSeen, &rdev->flags); |
2161 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2162 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2163 | set_bit(MD_RECOVERY_NEEDED, | |
2164 | &rdev->mddev->recovery); | |
977df362 N |
2165 | } else if (is_badblock(rdev, sh->sector, |
2166 | STRIPE_SECTORS, | |
c0b32972 | 2167 | &first_bad, &bad_sectors)) { |
977df362 | 2168 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2169 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2170 | /* That was a successful write so make | |
2171 | * sure it looks like we already did | |
2172 | * a re-write. | |
2173 | */ | |
2174 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2175 | } | |
977df362 N |
2176 | } |
2177 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2178 | |
977df362 N |
2179 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2180 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2181 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 2182 | release_stripe(sh); |
1da177e4 LT |
2183 | } |
2184 | ||
784052ec | 2185 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
d592a996 | 2186 | |
784052ec | 2187 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2188 | { |
2189 | struct r5dev *dev = &sh->dev[i]; | |
2190 | ||
2191 | bio_init(&dev->req); | |
2192 | dev->req.bi_io_vec = &dev->vec; | |
d592a996 | 2193 | dev->req.bi_max_vecs = 1; |
1da177e4 LT |
2194 | dev->req.bi_private = sh; |
2195 | ||
977df362 N |
2196 | bio_init(&dev->rreq); |
2197 | dev->rreq.bi_io_vec = &dev->rvec; | |
d592a996 | 2198 | dev->rreq.bi_max_vecs = 1; |
977df362 | 2199 | dev->rreq.bi_private = sh; |
977df362 | 2200 | |
1da177e4 | 2201 | dev->flags = 0; |
784052ec | 2202 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
2203 | } |
2204 | ||
fd01b88c | 2205 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2206 | { |
2207 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2208 | struct r5conf *conf = mddev->private; |
908f4fbd | 2209 | unsigned long flags; |
0c55e022 | 2210 | pr_debug("raid456: error called\n"); |
1da177e4 | 2211 | |
908f4fbd N |
2212 | spin_lock_irqsave(&conf->device_lock, flags); |
2213 | clear_bit(In_sync, &rdev->flags); | |
2214 | mddev->degraded = calc_degraded(conf); | |
2215 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2216 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2217 | ||
de393cde | 2218 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
2219 | set_bit(Faulty, &rdev->flags); |
2220 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2221 | printk(KERN_ALERT | |
2222 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
2223 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2224 | mdname(mddev), | |
2225 | bdevname(rdev->bdev, b), | |
2226 | mdname(mddev), | |
2227 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 2228 | } |
1da177e4 LT |
2229 | |
2230 | /* | |
2231 | * Input: a 'big' sector number, | |
2232 | * Output: index of the data and parity disk, and the sector # in them. | |
2233 | */ | |
d1688a6d | 2234 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
2235 | int previous, int *dd_idx, |
2236 | struct stripe_head *sh) | |
1da177e4 | 2237 | { |
6e3b96ed | 2238 | sector_t stripe, stripe2; |
35f2a591 | 2239 | sector_t chunk_number; |
1da177e4 | 2240 | unsigned int chunk_offset; |
911d4ee8 | 2241 | int pd_idx, qd_idx; |
67cc2b81 | 2242 | int ddf_layout = 0; |
1da177e4 | 2243 | sector_t new_sector; |
e183eaed N |
2244 | int algorithm = previous ? conf->prev_algo |
2245 | : conf->algorithm; | |
09c9e5fa AN |
2246 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2247 | : conf->chunk_sectors; | |
112bf897 N |
2248 | int raid_disks = previous ? conf->previous_raid_disks |
2249 | : conf->raid_disks; | |
2250 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2251 | |
2252 | /* First compute the information on this sector */ | |
2253 | ||
2254 | /* | |
2255 | * Compute the chunk number and the sector offset inside the chunk | |
2256 | */ | |
2257 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2258 | chunk_number = r_sector; | |
1da177e4 LT |
2259 | |
2260 | /* | |
2261 | * Compute the stripe number | |
2262 | */ | |
35f2a591 N |
2263 | stripe = chunk_number; |
2264 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2265 | stripe2 = stripe; |
1da177e4 LT |
2266 | /* |
2267 | * Select the parity disk based on the user selected algorithm. | |
2268 | */ | |
84789554 | 2269 | pd_idx = qd_idx = -1; |
16a53ecc N |
2270 | switch(conf->level) { |
2271 | case 4: | |
911d4ee8 | 2272 | pd_idx = data_disks; |
16a53ecc N |
2273 | break; |
2274 | case 5: | |
e183eaed | 2275 | switch (algorithm) { |
1da177e4 | 2276 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2277 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2278 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2279 | (*dd_idx)++; |
2280 | break; | |
2281 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2282 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2283 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2284 | (*dd_idx)++; |
2285 | break; | |
2286 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2287 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2288 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2289 | break; |
2290 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2291 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2292 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2293 | break; |
99c0fb5f N |
2294 | case ALGORITHM_PARITY_0: |
2295 | pd_idx = 0; | |
2296 | (*dd_idx)++; | |
2297 | break; | |
2298 | case ALGORITHM_PARITY_N: | |
2299 | pd_idx = data_disks; | |
2300 | break; | |
1da177e4 | 2301 | default: |
99c0fb5f | 2302 | BUG(); |
16a53ecc N |
2303 | } |
2304 | break; | |
2305 | case 6: | |
2306 | ||
e183eaed | 2307 | switch (algorithm) { |
16a53ecc | 2308 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2309 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2310 | qd_idx = pd_idx + 1; |
2311 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2312 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2313 | qd_idx = 0; |
2314 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2315 | (*dd_idx) += 2; /* D D P Q D */ |
2316 | break; | |
2317 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2318 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2319 | qd_idx = pd_idx + 1; |
2320 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2321 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2322 | qd_idx = 0; |
2323 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2324 | (*dd_idx) += 2; /* D D P Q D */ |
2325 | break; | |
2326 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2327 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2328 | qd_idx = (pd_idx + 1) % raid_disks; |
2329 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2330 | break; |
2331 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2332 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2333 | qd_idx = (pd_idx + 1) % raid_disks; |
2334 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2335 | break; |
99c0fb5f N |
2336 | |
2337 | case ALGORITHM_PARITY_0: | |
2338 | pd_idx = 0; | |
2339 | qd_idx = 1; | |
2340 | (*dd_idx) += 2; | |
2341 | break; | |
2342 | case ALGORITHM_PARITY_N: | |
2343 | pd_idx = data_disks; | |
2344 | qd_idx = data_disks + 1; | |
2345 | break; | |
2346 | ||
2347 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2348 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2349 | * of blocks for computing Q is different. | |
2350 | */ | |
6e3b96ed | 2351 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2352 | qd_idx = pd_idx + 1; |
2353 | if (pd_idx == raid_disks-1) { | |
2354 | (*dd_idx)++; /* Q D D D P */ | |
2355 | qd_idx = 0; | |
2356 | } else if (*dd_idx >= pd_idx) | |
2357 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2358 | ddf_layout = 1; |
99c0fb5f N |
2359 | break; |
2360 | ||
2361 | case ALGORITHM_ROTATING_N_RESTART: | |
2362 | /* Same a left_asymmetric, by first stripe is | |
2363 | * D D D P Q rather than | |
2364 | * Q D D D P | |
2365 | */ | |
6e3b96ed N |
2366 | stripe2 += 1; |
2367 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2368 | qd_idx = pd_idx + 1; |
2369 | if (pd_idx == raid_disks-1) { | |
2370 | (*dd_idx)++; /* Q D D D P */ | |
2371 | qd_idx = 0; | |
2372 | } else if (*dd_idx >= pd_idx) | |
2373 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2374 | ddf_layout = 1; |
99c0fb5f N |
2375 | break; |
2376 | ||
2377 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2378 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2379 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2380 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2381 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2382 | ddf_layout = 1; |
99c0fb5f N |
2383 | break; |
2384 | ||
2385 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2386 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2387 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2388 | if (*dd_idx >= pd_idx) |
2389 | (*dd_idx)++; | |
2390 | qd_idx = raid_disks - 1; | |
2391 | break; | |
2392 | ||
2393 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2394 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2395 | if (*dd_idx >= pd_idx) |
2396 | (*dd_idx)++; | |
2397 | qd_idx = raid_disks - 1; | |
2398 | break; | |
2399 | ||
2400 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2401 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2402 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2403 | qd_idx = raid_disks - 1; | |
2404 | break; | |
2405 | ||
2406 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2407 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2408 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2409 | qd_idx = raid_disks - 1; | |
2410 | break; | |
2411 | ||
2412 | case ALGORITHM_PARITY_0_6: | |
2413 | pd_idx = 0; | |
2414 | (*dd_idx)++; | |
2415 | qd_idx = raid_disks - 1; | |
2416 | break; | |
2417 | ||
16a53ecc | 2418 | default: |
99c0fb5f | 2419 | BUG(); |
16a53ecc N |
2420 | } |
2421 | break; | |
1da177e4 LT |
2422 | } |
2423 | ||
911d4ee8 N |
2424 | if (sh) { |
2425 | sh->pd_idx = pd_idx; | |
2426 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2427 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2428 | } |
1da177e4 LT |
2429 | /* |
2430 | * Finally, compute the new sector number | |
2431 | */ | |
2432 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2433 | return new_sector; | |
2434 | } | |
2435 | ||
784052ec | 2436 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2437 | { |
d1688a6d | 2438 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2439 | int raid_disks = sh->disks; |
2440 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2441 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2442 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2443 | : conf->chunk_sectors; | |
e183eaed N |
2444 | int algorithm = previous ? conf->prev_algo |
2445 | : conf->algorithm; | |
1da177e4 LT |
2446 | sector_t stripe; |
2447 | int chunk_offset; | |
35f2a591 N |
2448 | sector_t chunk_number; |
2449 | int dummy1, dd_idx = i; | |
1da177e4 | 2450 | sector_t r_sector; |
911d4ee8 | 2451 | struct stripe_head sh2; |
1da177e4 LT |
2452 | |
2453 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2454 | stripe = new_sector; | |
1da177e4 | 2455 | |
16a53ecc N |
2456 | if (i == sh->pd_idx) |
2457 | return 0; | |
2458 | switch(conf->level) { | |
2459 | case 4: break; | |
2460 | case 5: | |
e183eaed | 2461 | switch (algorithm) { |
1da177e4 LT |
2462 | case ALGORITHM_LEFT_ASYMMETRIC: |
2463 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2464 | if (i > sh->pd_idx) | |
2465 | i--; | |
2466 | break; | |
2467 | case ALGORITHM_LEFT_SYMMETRIC: | |
2468 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2469 | if (i < sh->pd_idx) | |
2470 | i += raid_disks; | |
2471 | i -= (sh->pd_idx + 1); | |
2472 | break; | |
99c0fb5f N |
2473 | case ALGORITHM_PARITY_0: |
2474 | i -= 1; | |
2475 | break; | |
2476 | case ALGORITHM_PARITY_N: | |
2477 | break; | |
1da177e4 | 2478 | default: |
99c0fb5f | 2479 | BUG(); |
16a53ecc N |
2480 | } |
2481 | break; | |
2482 | case 6: | |
d0dabf7e | 2483 | if (i == sh->qd_idx) |
16a53ecc | 2484 | return 0; /* It is the Q disk */ |
e183eaed | 2485 | switch (algorithm) { |
16a53ecc N |
2486 | case ALGORITHM_LEFT_ASYMMETRIC: |
2487 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2488 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2489 | case ALGORITHM_ROTATING_N_RESTART: | |
2490 | if (sh->pd_idx == raid_disks-1) | |
2491 | i--; /* Q D D D P */ | |
16a53ecc N |
2492 | else if (i > sh->pd_idx) |
2493 | i -= 2; /* D D P Q D */ | |
2494 | break; | |
2495 | case ALGORITHM_LEFT_SYMMETRIC: | |
2496 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2497 | if (sh->pd_idx == raid_disks-1) | |
2498 | i--; /* Q D D D P */ | |
2499 | else { | |
2500 | /* D D P Q D */ | |
2501 | if (i < sh->pd_idx) | |
2502 | i += raid_disks; | |
2503 | i -= (sh->pd_idx + 2); | |
2504 | } | |
2505 | break; | |
99c0fb5f N |
2506 | case ALGORITHM_PARITY_0: |
2507 | i -= 2; | |
2508 | break; | |
2509 | case ALGORITHM_PARITY_N: | |
2510 | break; | |
2511 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2512 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2513 | if (sh->pd_idx == 0) |
2514 | i--; /* P D D D Q */ | |
e4424fee N |
2515 | else { |
2516 | /* D D Q P D */ | |
2517 | if (i < sh->pd_idx) | |
2518 | i += raid_disks; | |
2519 | i -= (sh->pd_idx + 1); | |
2520 | } | |
99c0fb5f N |
2521 | break; |
2522 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2523 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2524 | if (i > sh->pd_idx) | |
2525 | i--; | |
2526 | break; | |
2527 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2528 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2529 | if (i < sh->pd_idx) | |
2530 | i += data_disks + 1; | |
2531 | i -= (sh->pd_idx + 1); | |
2532 | break; | |
2533 | case ALGORITHM_PARITY_0_6: | |
2534 | i -= 1; | |
2535 | break; | |
16a53ecc | 2536 | default: |
99c0fb5f | 2537 | BUG(); |
16a53ecc N |
2538 | } |
2539 | break; | |
1da177e4 LT |
2540 | } |
2541 | ||
2542 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2543 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2544 | |
112bf897 | 2545 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2546 | previous, &dummy1, &sh2); |
911d4ee8 N |
2547 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2548 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2549 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2550 | mdname(conf->mddev)); | |
1da177e4 LT |
2551 | return 0; |
2552 | } | |
2553 | return r_sector; | |
2554 | } | |
2555 | ||
600aa109 | 2556 | static void |
c0f7bddb | 2557 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2558 | int rcw, int expand) |
e33129d8 DW |
2559 | { |
2560 | int i, pd_idx = sh->pd_idx, disks = sh->disks; | |
d1688a6d | 2561 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2562 | int level = conf->level; |
e33129d8 DW |
2563 | |
2564 | if (rcw) { | |
e33129d8 DW |
2565 | |
2566 | for (i = disks; i--; ) { | |
2567 | struct r5dev *dev = &sh->dev[i]; | |
2568 | ||
2569 | if (dev->towrite) { | |
2570 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2571 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2572 | if (!expand) |
2573 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2574 | s->locked++; |
e33129d8 DW |
2575 | } |
2576 | } | |
ce7d363a N |
2577 | /* if we are not expanding this is a proper write request, and |
2578 | * there will be bios with new data to be drained into the | |
2579 | * stripe cache | |
2580 | */ | |
2581 | if (!expand) { | |
2582 | if (!s->locked) | |
2583 | /* False alarm, nothing to do */ | |
2584 | return; | |
2585 | sh->reconstruct_state = reconstruct_state_drain_run; | |
2586 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2587 | } else | |
2588 | sh->reconstruct_state = reconstruct_state_run; | |
2589 | ||
2590 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
2591 | ||
c0f7bddb | 2592 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2593 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2594 | atomic_inc(&conf->pending_full_writes); |
e33129d8 | 2595 | } else { |
c0f7bddb | 2596 | BUG_ON(level == 6); |
e33129d8 DW |
2597 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || |
2598 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
2599 | ||
e33129d8 DW |
2600 | for (i = disks; i--; ) { |
2601 | struct r5dev *dev = &sh->dev[i]; | |
2602 | if (i == pd_idx) | |
2603 | continue; | |
2604 | ||
e33129d8 DW |
2605 | if (dev->towrite && |
2606 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2607 | test_bit(R5_Wantcompute, &dev->flags))) { |
2608 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2609 | set_bit(R5_LOCKED, &dev->flags); |
2610 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2611 | s->locked++; |
e33129d8 DW |
2612 | } |
2613 | } | |
ce7d363a N |
2614 | if (!s->locked) |
2615 | /* False alarm - nothing to do */ | |
2616 | return; | |
2617 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
2618 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
2619 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2620 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
2621 | } |
2622 | ||
c0f7bddb | 2623 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2624 | * are in flight |
2625 | */ | |
2626 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2627 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2628 | s->locked++; |
e33129d8 | 2629 | |
c0f7bddb YT |
2630 | if (level == 6) { |
2631 | int qd_idx = sh->qd_idx; | |
2632 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2633 | ||
2634 | set_bit(R5_LOCKED, &dev->flags); | |
2635 | clear_bit(R5_UPTODATE, &dev->flags); | |
2636 | s->locked++; | |
2637 | } | |
2638 | ||
600aa109 | 2639 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2640 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2641 | s->locked, s->ops_request); |
e33129d8 | 2642 | } |
16a53ecc | 2643 | |
1da177e4 LT |
2644 | /* |
2645 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2646 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2647 | * The bi_next chain must be in order. |
2648 | */ | |
da41ba65 | 2649 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
2650 | int forwrite, int previous) | |
1da177e4 LT |
2651 | { |
2652 | struct bio **bip; | |
d1688a6d | 2653 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2654 | int firstwrite=0; |
1da177e4 | 2655 | |
cbe47ec5 | 2656 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 2657 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
2658 | (unsigned long long)sh->sector); |
2659 | ||
b17459c0 SL |
2660 | /* |
2661 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2662 | * reference count to avoid race. The reference count should already be | |
2663 | * increased before this function is called (for example, in | |
2664 | * make_request()), so other bio sharing this stripe will not free the | |
2665 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2666 | * protect it. | |
2667 | */ | |
2668 | spin_lock_irq(&sh->stripe_lock); | |
72626685 | 2669 | if (forwrite) { |
1da177e4 | 2670 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2671 | if (*bip == NULL) |
72626685 N |
2672 | firstwrite = 1; |
2673 | } else | |
1da177e4 | 2674 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
2675 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
2676 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
2677 | goto overlap; |
2678 | bip = & (*bip)->bi_next; | |
2679 | } | |
4f024f37 | 2680 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
2681 | goto overlap; |
2682 | ||
da41ba65 | 2683 | if (!forwrite || previous) |
2684 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
2685 | ||
78bafebd | 2686 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2687 | if (*bip) |
2688 | bi->bi_next = *bip; | |
2689 | *bip = bi; | |
e7836bd6 | 2690 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2691 | |
1da177e4 LT |
2692 | if (forwrite) { |
2693 | /* check if page is covered */ | |
2694 | sector_t sector = sh->dev[dd_idx].sector; | |
2695 | for (bi=sh->dev[dd_idx].towrite; | |
2696 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 2697 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 2698 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
2699 | if (bio_end_sector(bi) >= sector) |
2700 | sector = bio_end_sector(bi); | |
1da177e4 LT |
2701 | } |
2702 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
2703 | set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); | |
2704 | } | |
cbe47ec5 N |
2705 | |
2706 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 2707 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 | 2708 | (unsigned long long)sh->sector, dd_idx); |
b97390ae | 2709 | spin_unlock_irq(&sh->stripe_lock); |
cbe47ec5 N |
2710 | |
2711 | if (conf->mddev->bitmap && firstwrite) { | |
2712 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, | |
2713 | STRIPE_SECTORS, 0); | |
2714 | sh->bm_seq = conf->seq_flush+1; | |
2715 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
2716 | } | |
1da177e4 LT |
2717 | return 1; |
2718 | ||
2719 | overlap: | |
2720 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 2721 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
2722 | return 0; |
2723 | } | |
2724 | ||
d1688a6d | 2725 | static void end_reshape(struct r5conf *conf); |
29269553 | 2726 | |
d1688a6d | 2727 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 2728 | struct stripe_head *sh) |
ccfcc3c1 | 2729 | { |
784052ec | 2730 | int sectors_per_chunk = |
09c9e5fa | 2731 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 2732 | int dd_idx; |
2d2063ce | 2733 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 2734 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 2735 | |
112bf897 N |
2736 | raid5_compute_sector(conf, |
2737 | stripe * (disks - conf->max_degraded) | |
b875e531 | 2738 | *sectors_per_chunk + chunk_offset, |
112bf897 | 2739 | previous, |
911d4ee8 | 2740 | &dd_idx, sh); |
ccfcc3c1 N |
2741 | } |
2742 | ||
a4456856 | 2743 | static void |
d1688a6d | 2744 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
2745 | struct stripe_head_state *s, int disks, |
2746 | struct bio **return_bi) | |
2747 | { | |
2748 | int i; | |
2749 | for (i = disks; i--; ) { | |
2750 | struct bio *bi; | |
2751 | int bitmap_end = 0; | |
2752 | ||
2753 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 2754 | struct md_rdev *rdev; |
a4456856 DW |
2755 | rcu_read_lock(); |
2756 | rdev = rcu_dereference(conf->disks[i].rdev); | |
2757 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
2758 | atomic_inc(&rdev->nr_pending); |
2759 | else | |
2760 | rdev = NULL; | |
a4456856 | 2761 | rcu_read_unlock(); |
7f0da59b N |
2762 | if (rdev) { |
2763 | if (!rdev_set_badblocks( | |
2764 | rdev, | |
2765 | sh->sector, | |
2766 | STRIPE_SECTORS, 0)) | |
2767 | md_error(conf->mddev, rdev); | |
2768 | rdev_dec_pending(rdev, conf->mddev); | |
2769 | } | |
a4456856 | 2770 | } |
b17459c0 | 2771 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
2772 | /* fail all writes first */ |
2773 | bi = sh->dev[i].towrite; | |
2774 | sh->dev[i].towrite = NULL; | |
b17459c0 | 2775 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 2776 | if (bi) |
a4456856 | 2777 | bitmap_end = 1; |
a4456856 DW |
2778 | |
2779 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
2780 | wake_up(&conf->wait_for_overlap); | |
2781 | ||
4f024f37 | 2782 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
2783 | sh->dev[i].sector + STRIPE_SECTORS) { |
2784 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
2785 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2786 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2787 | md_write_end(conf->mddev); |
2788 | bi->bi_next = *return_bi; | |
2789 | *return_bi = bi; | |
2790 | } | |
2791 | bi = nextbi; | |
2792 | } | |
7eaf7e8e SL |
2793 | if (bitmap_end) |
2794 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2795 | STRIPE_SECTORS, 0, 0); | |
2796 | bitmap_end = 0; | |
a4456856 DW |
2797 | /* and fail all 'written' */ |
2798 | bi = sh->dev[i].written; | |
2799 | sh->dev[i].written = NULL; | |
d592a996 SL |
2800 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
2801 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
2802 | sh->dev[i].page = sh->dev[i].orig_page; | |
2803 | } | |
2804 | ||
a4456856 | 2805 | if (bi) bitmap_end = 1; |
4f024f37 | 2806 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
2807 | sh->dev[i].sector + STRIPE_SECTORS) { |
2808 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
2809 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2810 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2811 | md_write_end(conf->mddev); |
2812 | bi->bi_next = *return_bi; | |
2813 | *return_bi = bi; | |
2814 | } | |
2815 | bi = bi2; | |
2816 | } | |
2817 | ||
b5e98d65 DW |
2818 | /* fail any reads if this device is non-operational and |
2819 | * the data has not reached the cache yet. | |
2820 | */ | |
2821 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
2822 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
2823 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 2824 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
2825 | bi = sh->dev[i].toread; |
2826 | sh->dev[i].toread = NULL; | |
143c4d05 | 2827 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
2828 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
2829 | wake_up(&conf->wait_for_overlap); | |
4f024f37 | 2830 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
2831 | sh->dev[i].sector + STRIPE_SECTORS) { |
2832 | struct bio *nextbi = | |
2833 | r5_next_bio(bi, sh->dev[i].sector); | |
2834 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 2835 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
2836 | bi->bi_next = *return_bi; |
2837 | *return_bi = bi; | |
2838 | } | |
2839 | bi = nextbi; | |
2840 | } | |
2841 | } | |
a4456856 DW |
2842 | if (bitmap_end) |
2843 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
2844 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
2845 | /* If we were in the middle of a write the parity block might |
2846 | * still be locked - so just clear all R5_LOCKED flags | |
2847 | */ | |
2848 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
2849 | } |
2850 | ||
8b3e6cdc DW |
2851 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
2852 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
2853 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
2854 | } |
2855 | ||
7f0da59b | 2856 | static void |
d1688a6d | 2857 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
2858 | struct stripe_head_state *s) |
2859 | { | |
2860 | int abort = 0; | |
2861 | int i; | |
2862 | ||
7f0da59b | 2863 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
2864 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
2865 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 2866 | s->syncing = 0; |
9a3e1101 | 2867 | s->replacing = 0; |
7f0da59b | 2868 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
2869 | * Don't even need to abort as that is handled elsewhere |
2870 | * if needed, and not always wanted e.g. if there is a known | |
2871 | * bad block here. | |
9a3e1101 | 2872 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
2873 | * non-sync devices, or abort the recovery |
2874 | */ | |
18b9837e N |
2875 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
2876 | /* During recovery devices cannot be removed, so | |
2877 | * locking and refcounting of rdevs is not needed | |
2878 | */ | |
2879 | for (i = 0; i < conf->raid_disks; i++) { | |
2880 | struct md_rdev *rdev = conf->disks[i].rdev; | |
2881 | if (rdev | |
2882 | && !test_bit(Faulty, &rdev->flags) | |
2883 | && !test_bit(In_sync, &rdev->flags) | |
2884 | && !rdev_set_badblocks(rdev, sh->sector, | |
2885 | STRIPE_SECTORS, 0)) | |
2886 | abort = 1; | |
2887 | rdev = conf->disks[i].replacement; | |
2888 | if (rdev | |
2889 | && !test_bit(Faulty, &rdev->flags) | |
2890 | && !test_bit(In_sync, &rdev->flags) | |
2891 | && !rdev_set_badblocks(rdev, sh->sector, | |
2892 | STRIPE_SECTORS, 0)) | |
2893 | abort = 1; | |
2894 | } | |
2895 | if (abort) | |
2896 | conf->recovery_disabled = | |
2897 | conf->mddev->recovery_disabled; | |
7f0da59b | 2898 | } |
18b9837e | 2899 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
2900 | } |
2901 | ||
9a3e1101 N |
2902 | static int want_replace(struct stripe_head *sh, int disk_idx) |
2903 | { | |
2904 | struct md_rdev *rdev; | |
2905 | int rv = 0; | |
2906 | /* Doing recovery so rcu locking not required */ | |
2907 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
2908 | if (rdev | |
2909 | && !test_bit(Faulty, &rdev->flags) | |
2910 | && !test_bit(In_sync, &rdev->flags) | |
2911 | && (rdev->recovery_offset <= sh->sector | |
2912 | || rdev->mddev->recovery_cp <= sh->sector)) | |
2913 | rv = 1; | |
2914 | ||
2915 | return rv; | |
2916 | } | |
2917 | ||
93b3dbce | 2918 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
2919 | * to be read or computed to satisfy a request. |
2920 | * | |
2921 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 2922 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 2923 | */ |
2c58f06e N |
2924 | |
2925 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, | |
2926 | int disk_idx, int disks) | |
a4456856 | 2927 | { |
5599becc | 2928 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
2929 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
2930 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 2931 | int i; |
5599becc | 2932 | |
a79cfe12 N |
2933 | |
2934 | if (test_bit(R5_LOCKED, &dev->flags) || | |
2935 | test_bit(R5_UPTODATE, &dev->flags)) | |
2936 | /* No point reading this as we already have it or have | |
2937 | * decided to get it. | |
2938 | */ | |
2939 | return 0; | |
2940 | ||
2941 | if (dev->toread || | |
2942 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
2943 | /* We need this block to directly satisfy a request */ | |
2944 | return 1; | |
2945 | ||
2946 | if (s->syncing || s->expanding || | |
2947 | (s->replacing && want_replace(sh, disk_idx))) | |
2948 | /* When syncing, or expanding we read everything. | |
2949 | * When replacing, we need the replaced block. | |
2950 | */ | |
2951 | return 1; | |
2952 | ||
2953 | if ((s->failed >= 1 && fdev[0]->toread) || | |
2954 | (s->failed >= 2 && fdev[1]->toread)) | |
2955 | /* If we want to read from a failed device, then | |
2956 | * we need to actually read every other device. | |
2957 | */ | |
2958 | return 1; | |
2959 | ||
a9d56950 N |
2960 | /* Sometimes neither read-modify-write nor reconstruct-write |
2961 | * cycles can work. In those cases we read every block we | |
2962 | * can. Then the parity-update is certain to have enough to | |
2963 | * work with. | |
2964 | * This can only be a problem when we need to write something, | |
2965 | * and some device has failed. If either of those tests | |
2966 | * fail we need look no further. | |
2967 | */ | |
2968 | if (!s->failed || !s->to_write) | |
2969 | return 0; | |
2970 | ||
2971 | if (test_bit(R5_Insync, &dev->flags) && | |
2972 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
2973 | /* Pre-reads at not permitted until after short delay | |
2974 | * to gather multiple requests. However if this | |
2975 | * device is no Insync, the block could only be be computed | |
2976 | * and there is no need to delay that. | |
2977 | */ | |
2978 | return 0; | |
ea664c82 N |
2979 | |
2980 | for (i = 0; i < s->failed; i++) { | |
2981 | if (fdev[i]->towrite && | |
2982 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
2983 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
2984 | /* If we have a partial write to a failed | |
2985 | * device, then we will need to reconstruct | |
2986 | * the content of that device, so all other | |
2987 | * devices must be read. | |
2988 | */ | |
2989 | return 1; | |
2990 | } | |
2991 | ||
2992 | /* If we are forced to do a reconstruct-write, either because | |
2993 | * the current RAID6 implementation only supports that, or | |
2994 | * or because parity cannot be trusted and we are currently | |
2995 | * recovering it, there is extra need to be careful. | |
2996 | * If one of the devices that we would need to read, because | |
2997 | * it is not being overwritten (and maybe not written at all) | |
2998 | * is missing/faulty, then we need to read everything we can. | |
2999 | */ | |
3000 | if (sh->raid_conf->level != 6 && | |
3001 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3002 | /* reconstruct-write isn't being forced */ | |
3003 | return 0; | |
3004 | for (i = 0; i < s->failed; i++) { | |
3005 | if (!test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3006 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3007 | return 1; | |
3008 | } | |
3009 | ||
2c58f06e N |
3010 | return 0; |
3011 | } | |
3012 | ||
3013 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3014 | int disk_idx, int disks) | |
3015 | { | |
3016 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3017 | ||
3018 | /* is the data in this block needed, and can we get it? */ | |
3019 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3020 | /* we would like to get this block, possibly by computing it, |
3021 | * otherwise read it if the backing disk is insync | |
3022 | */ | |
3023 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3024 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
3025 | if ((s->uptodate == disks - 1) && | |
f2b3b44d N |
3026 | (s->failed && (disk_idx == s->failed_num[0] || |
3027 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3028 | /* have disk failed, and we're requested to fetch it; |
3029 | * do compute it | |
a4456856 | 3030 | */ |
5599becc YT |
3031 | pr_debug("Computing stripe %llu block %d\n", |
3032 | (unsigned long long)sh->sector, disk_idx); | |
3033 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3034 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3035 | set_bit(R5_Wantcompute, &dev->flags); | |
3036 | sh->ops.target = disk_idx; | |
3037 | sh->ops.target2 = -1; /* no 2nd target */ | |
3038 | s->req_compute = 1; | |
93b3dbce N |
3039 | /* Careful: from this point on 'uptodate' is in the eye |
3040 | * of raid_run_ops which services 'compute' operations | |
3041 | * before writes. R5_Wantcompute flags a block that will | |
3042 | * be R5_UPTODATE by the time it is needed for a | |
3043 | * subsequent operation. | |
3044 | */ | |
5599becc YT |
3045 | s->uptodate++; |
3046 | return 1; | |
3047 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3048 | /* Computing 2-failure is *very* expensive; only | |
3049 | * do it if failed >= 2 | |
3050 | */ | |
3051 | int other; | |
3052 | for (other = disks; other--; ) { | |
3053 | if (other == disk_idx) | |
3054 | continue; | |
3055 | if (!test_bit(R5_UPTODATE, | |
3056 | &sh->dev[other].flags)) | |
3057 | break; | |
a4456856 | 3058 | } |
5599becc YT |
3059 | BUG_ON(other < 0); |
3060 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3061 | (unsigned long long)sh->sector, | |
3062 | disk_idx, other); | |
3063 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3064 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3065 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3066 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3067 | sh->ops.target = disk_idx; | |
3068 | sh->ops.target2 = other; | |
3069 | s->uptodate += 2; | |
3070 | s->req_compute = 1; | |
3071 | return 1; | |
3072 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3073 | set_bit(R5_LOCKED, &dev->flags); | |
3074 | set_bit(R5_Wantread, &dev->flags); | |
3075 | s->locked++; | |
3076 | pr_debug("Reading block %d (sync=%d)\n", | |
3077 | disk_idx, s->syncing); | |
a4456856 DW |
3078 | } |
3079 | } | |
5599becc YT |
3080 | |
3081 | return 0; | |
3082 | } | |
3083 | ||
3084 | /** | |
93b3dbce | 3085 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3086 | */ |
93b3dbce N |
3087 | static void handle_stripe_fill(struct stripe_head *sh, |
3088 | struct stripe_head_state *s, | |
3089 | int disks) | |
5599becc YT |
3090 | { |
3091 | int i; | |
3092 | ||
3093 | /* look for blocks to read/compute, skip this if a compute | |
3094 | * is already in flight, or if the stripe contents are in the | |
3095 | * midst of changing due to a write | |
3096 | */ | |
3097 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
3098 | !sh->reconstruct_state) | |
3099 | for (i = disks; i--; ) | |
93b3dbce | 3100 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3101 | break; |
a4456856 DW |
3102 | set_bit(STRIPE_HANDLE, &sh->state); |
3103 | } | |
3104 | ||
1fe797e6 | 3105 | /* handle_stripe_clean_event |
a4456856 DW |
3106 | * any written block on an uptodate or failed drive can be returned. |
3107 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3108 | * never LOCKED, so we don't need to test 'failed' directly. | |
3109 | */ | |
d1688a6d | 3110 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
3111 | struct stripe_head *sh, int disks, struct bio **return_bi) |
3112 | { | |
3113 | int i; | |
3114 | struct r5dev *dev; | |
f8dfcffd | 3115 | int discard_pending = 0; |
a4456856 DW |
3116 | |
3117 | for (i = disks; i--; ) | |
3118 | if (sh->dev[i].written) { | |
3119 | dev = &sh->dev[i]; | |
3120 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3121 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3122 | test_bit(R5_Discard, &dev->flags) || |
3123 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3124 | /* We can return any write requests */ |
3125 | struct bio *wbi, *wbi2; | |
45b4233c | 3126 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3127 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3128 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3129 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3130 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
3131 | dev->page = dev->orig_page; | |
3132 | } | |
a4456856 DW |
3133 | wbi = dev->written; |
3134 | dev->written = NULL; | |
4f024f37 | 3135 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3136 | dev->sector + STRIPE_SECTORS) { |
3137 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3138 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 DW |
3139 | md_write_end(conf->mddev); |
3140 | wbi->bi_next = *return_bi; | |
3141 | *return_bi = wbi; | |
3142 | } | |
3143 | wbi = wbi2; | |
3144 | } | |
7eaf7e8e SL |
3145 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3146 | STRIPE_SECTORS, | |
a4456856 | 3147 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3148 | 0); |
f8dfcffd N |
3149 | } else if (test_bit(R5_Discard, &dev->flags)) |
3150 | discard_pending = 1; | |
d592a996 SL |
3151 | WARN_ON(test_bit(R5_SkipCopy, &dev->flags)); |
3152 | WARN_ON(dev->page != dev->orig_page); | |
f8dfcffd N |
3153 | } |
3154 | if (!discard_pending && | |
3155 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
3156 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); | |
3157 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3158 | if (sh->qd_idx >= 0) { | |
3159 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3160 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3161 | } | |
3162 | /* now that discard is done we can proceed with any sync */ | |
3163 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3164 | /* |
3165 | * SCSI discard will change some bio fields and the stripe has | |
3166 | * no updated data, so remove it from hash list and the stripe | |
3167 | * will be reinitialized | |
3168 | */ | |
3169 | spin_lock_irq(&conf->device_lock); | |
3170 | remove_hash(sh); | |
3171 | spin_unlock_irq(&conf->device_lock); | |
f8dfcffd N |
3172 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3173 | set_bit(STRIPE_HANDLE, &sh->state); | |
3174 | ||
3175 | } | |
8b3e6cdc DW |
3176 | |
3177 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3178 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3179 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3180 | } |
3181 | ||
d1688a6d | 3182 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
3183 | struct stripe_head *sh, |
3184 | struct stripe_head_state *s, | |
3185 | int disks) | |
a4456856 DW |
3186 | { |
3187 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3188 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3189 | ||
3190 | /* RAID6 requires 'rcw' in current implementation. | |
3191 | * Otherwise, check whether resync is now happening or should start. | |
3192 | * If yes, then the array is dirty (after unclean shutdown or | |
3193 | * initial creation), so parity in some stripes might be inconsistent. | |
3194 | * In this case, we need to always do reconstruct-write, to ensure | |
3195 | * that in case of drive failure or read-error correction, we | |
3196 | * generate correct data from the parity. | |
3197 | */ | |
3198 | if (conf->max_degraded == 2 || | |
26ac1073 N |
3199 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3200 | s->failed == 0)) { | |
a7854487 | 3201 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3202 | * look like rcw is cheaper |
3203 | */ | |
3204 | rcw = 1; rmw = 2; | |
a7854487 AL |
3205 | pr_debug("force RCW max_degraded=%u, recovery_cp=%llu sh->sector=%llu\n", |
3206 | conf->max_degraded, (unsigned long long)recovery_cp, | |
3207 | (unsigned long long)sh->sector); | |
c8ac1803 | 3208 | } else for (i = disks; i--; ) { |
a4456856 DW |
3209 | /* would I have to read this buffer for read_modify_write */ |
3210 | struct r5dev *dev = &sh->dev[i]; | |
3211 | if ((dev->towrite || i == sh->pd_idx) && | |
3212 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
3213 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3214 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
3215 | if (test_bit(R5_Insync, &dev->flags)) |
3216 | rmw++; | |
3217 | else | |
3218 | rmw += 2*disks; /* cannot read it */ | |
3219 | } | |
3220 | /* Would I have to read this buffer for reconstruct_write */ | |
3221 | if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && | |
3222 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
3223 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3224 | test_bit(R5_Wantcompute, &dev->flags))) { | |
67f45548 N |
3225 | if (test_bit(R5_Insync, &dev->flags)) |
3226 | rcw++; | |
a4456856 DW |
3227 | else |
3228 | rcw += 2*disks; | |
3229 | } | |
3230 | } | |
45b4233c | 3231 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
3232 | (unsigned long long)sh->sector, rmw, rcw); |
3233 | set_bit(STRIPE_HANDLE, &sh->state); | |
a9add5d9 | 3234 | if (rmw < rcw && rmw > 0) { |
a4456856 | 3235 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3236 | if (conf->mddev->queue) |
3237 | blk_add_trace_msg(conf->mddev->queue, | |
3238 | "raid5 rmw %llu %d", | |
3239 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3240 | for (i = disks; i--; ) { |
3241 | struct r5dev *dev = &sh->dev[i]; | |
3242 | if ((dev->towrite || i == sh->pd_idx) && | |
3243 | !test_bit(R5_LOCKED, &dev->flags) && | |
f38e1219 DW |
3244 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3245 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 | 3246 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3247 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3248 | &sh->state)) { | |
3249 | pr_debug("Read_old block %d for r-m-w\n", | |
3250 | i); | |
a4456856 DW |
3251 | set_bit(R5_LOCKED, &dev->flags); |
3252 | set_bit(R5_Wantread, &dev->flags); | |
3253 | s->locked++; | |
3254 | } else { | |
3255 | set_bit(STRIPE_DELAYED, &sh->state); | |
3256 | set_bit(STRIPE_HANDLE, &sh->state); | |
3257 | } | |
3258 | } | |
3259 | } | |
a9add5d9 | 3260 | } |
c8ac1803 | 3261 | if (rcw <= rmw && rcw > 0) { |
a4456856 | 3262 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3263 | int qread =0; |
c8ac1803 | 3264 | rcw = 0; |
a4456856 DW |
3265 | for (i = disks; i--; ) { |
3266 | struct r5dev *dev = &sh->dev[i]; | |
3267 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3268 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3269 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3270 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3271 | test_bit(R5_Wantcompute, &dev->flags))) { |
3272 | rcw++; | |
67f45548 N |
3273 | if (test_bit(R5_Insync, &dev->flags) && |
3274 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3275 | &sh->state)) { | |
45b4233c | 3276 | pr_debug("Read_old block " |
a4456856 DW |
3277 | "%d for Reconstruct\n", i); |
3278 | set_bit(R5_LOCKED, &dev->flags); | |
3279 | set_bit(R5_Wantread, &dev->flags); | |
3280 | s->locked++; | |
a9add5d9 | 3281 | qread++; |
a4456856 DW |
3282 | } else { |
3283 | set_bit(STRIPE_DELAYED, &sh->state); | |
3284 | set_bit(STRIPE_HANDLE, &sh->state); | |
3285 | } | |
3286 | } | |
3287 | } | |
e3620a3a | 3288 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3289 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3290 | (unsigned long long)sh->sector, | |
3291 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3292 | } |
b1b02fe9 N |
3293 | |
3294 | if (rcw > disks && rmw > disks && | |
3295 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3296 | set_bit(STRIPE_DELAYED, &sh->state); | |
3297 | ||
a4456856 DW |
3298 | /* now if nothing is locked, and if we have enough data, |
3299 | * we can start a write request | |
3300 | */ | |
f38e1219 DW |
3301 | /* since handle_stripe can be called at any time we need to handle the |
3302 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3303 | * subsequent call wants to start a write request. raid_run_ops only |
3304 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3305 | * simultaneously. If this is not the case then new writes need to be |
3306 | * held off until the compute completes. | |
3307 | */ | |
976ea8d4 DW |
3308 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3309 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
3310 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 3311 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
3312 | } |
3313 | ||
d1688a6d | 3314 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3315 | struct stripe_head_state *s, int disks) |
3316 | { | |
ecc65c9b | 3317 | struct r5dev *dev = NULL; |
bd2ab670 | 3318 | |
a4456856 | 3319 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3320 | |
ecc65c9b DW |
3321 | switch (sh->check_state) { |
3322 | case check_state_idle: | |
3323 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3324 | if (s->failed == 0) { |
bd2ab670 | 3325 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3326 | sh->check_state = check_state_run; |
3327 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3328 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3329 | s->uptodate--; |
ecc65c9b | 3330 | break; |
bd2ab670 | 3331 | } |
f2b3b44d | 3332 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3333 | /* fall through */ |
3334 | case check_state_compute_result: | |
3335 | sh->check_state = check_state_idle; | |
3336 | if (!dev) | |
3337 | dev = &sh->dev[sh->pd_idx]; | |
3338 | ||
3339 | /* check that a write has not made the stripe insync */ | |
3340 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3341 | break; | |
c8894419 | 3342 | |
a4456856 | 3343 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3344 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3345 | BUG_ON(s->uptodate != disks); | |
3346 | ||
3347 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3348 | s->locked++; |
a4456856 | 3349 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3350 | |
a4456856 | 3351 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3352 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3353 | break; |
3354 | case check_state_run: | |
3355 | break; /* we will be called again upon completion */ | |
3356 | case check_state_check_result: | |
3357 | sh->check_state = check_state_idle; | |
3358 | ||
3359 | /* if a failure occurred during the check operation, leave | |
3360 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3361 | */ | |
3362 | if (s->failed) | |
3363 | break; | |
3364 | ||
3365 | /* handle a successful check operation, if parity is correct | |
3366 | * we are done. Otherwise update the mismatch count and repair | |
3367 | * parity if !MD_RECOVERY_CHECK | |
3368 | */ | |
ad283ea4 | 3369 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3370 | /* parity is correct (on disc, |
3371 | * not in buffer any more) | |
3372 | */ | |
3373 | set_bit(STRIPE_INSYNC, &sh->state); | |
3374 | else { | |
7f7583d4 | 3375 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3376 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3377 | /* don't try to repair!! */ | |
3378 | set_bit(STRIPE_INSYNC, &sh->state); | |
3379 | else { | |
3380 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3381 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3382 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3383 | set_bit(R5_Wantcompute, | |
3384 | &sh->dev[sh->pd_idx].flags); | |
3385 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3386 | sh->ops.target2 = -1; |
ecc65c9b DW |
3387 | s->uptodate++; |
3388 | } | |
3389 | } | |
3390 | break; | |
3391 | case check_state_compute_run: | |
3392 | break; | |
3393 | default: | |
3394 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3395 | __func__, sh->check_state, | |
3396 | (unsigned long long) sh->sector); | |
3397 | BUG(); | |
a4456856 DW |
3398 | } |
3399 | } | |
3400 | ||
d1688a6d | 3401 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3402 | struct stripe_head_state *s, |
f2b3b44d | 3403 | int disks) |
a4456856 | 3404 | { |
a4456856 | 3405 | int pd_idx = sh->pd_idx; |
34e04e87 | 3406 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3407 | struct r5dev *dev; |
a4456856 DW |
3408 | |
3409 | set_bit(STRIPE_HANDLE, &sh->state); | |
3410 | ||
3411 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3412 | |
a4456856 DW |
3413 | /* Want to check and possibly repair P and Q. |
3414 | * However there could be one 'failed' device, in which | |
3415 | * case we can only check one of them, possibly using the | |
3416 | * other to generate missing data | |
3417 | */ | |
3418 | ||
d82dfee0 DW |
3419 | switch (sh->check_state) { |
3420 | case check_state_idle: | |
3421 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 3422 | if (s->failed == s->q_failed) { |
d82dfee0 | 3423 | /* The only possible failed device holds Q, so it |
a4456856 DW |
3424 | * makes sense to check P (If anything else were failed, |
3425 | * we would have used P to recreate it). | |
3426 | */ | |
d82dfee0 | 3427 | sh->check_state = check_state_run; |
a4456856 | 3428 | } |
f2b3b44d | 3429 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 3430 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
3431 | * anything, so it makes sense to check it |
3432 | */ | |
d82dfee0 DW |
3433 | if (sh->check_state == check_state_run) |
3434 | sh->check_state = check_state_run_pq; | |
3435 | else | |
3436 | sh->check_state = check_state_run_q; | |
a4456856 | 3437 | } |
a4456856 | 3438 | |
d82dfee0 DW |
3439 | /* discard potentially stale zero_sum_result */ |
3440 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3441 | |
d82dfee0 DW |
3442 | if (sh->check_state == check_state_run) { |
3443 | /* async_xor_zero_sum destroys the contents of P */ | |
3444 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3445 | s->uptodate--; | |
a4456856 | 3446 | } |
d82dfee0 DW |
3447 | if (sh->check_state >= check_state_run && |
3448 | sh->check_state <= check_state_run_pq) { | |
3449 | /* async_syndrome_zero_sum preserves P and Q, so | |
3450 | * no need to mark them !uptodate here | |
3451 | */ | |
3452 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3453 | break; | |
a4456856 DW |
3454 | } |
3455 | ||
d82dfee0 DW |
3456 | /* we have 2-disk failure */ |
3457 | BUG_ON(s->failed != 2); | |
3458 | /* fall through */ | |
3459 | case check_state_compute_result: | |
3460 | sh->check_state = check_state_idle; | |
a4456856 | 3461 | |
d82dfee0 DW |
3462 | /* check that a write has not made the stripe insync */ |
3463 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3464 | break; | |
a4456856 DW |
3465 | |
3466 | /* now write out any block on a failed drive, | |
d82dfee0 | 3467 | * or P or Q if they were recomputed |
a4456856 | 3468 | */ |
d82dfee0 | 3469 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3470 | if (s->failed == 2) { |
f2b3b44d | 3471 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3472 | s->locked++; |
3473 | set_bit(R5_LOCKED, &dev->flags); | |
3474 | set_bit(R5_Wantwrite, &dev->flags); | |
3475 | } | |
3476 | if (s->failed >= 1) { | |
f2b3b44d | 3477 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3478 | s->locked++; |
3479 | set_bit(R5_LOCKED, &dev->flags); | |
3480 | set_bit(R5_Wantwrite, &dev->flags); | |
3481 | } | |
d82dfee0 | 3482 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3483 | dev = &sh->dev[pd_idx]; |
3484 | s->locked++; | |
3485 | set_bit(R5_LOCKED, &dev->flags); | |
3486 | set_bit(R5_Wantwrite, &dev->flags); | |
3487 | } | |
d82dfee0 | 3488 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3489 | dev = &sh->dev[qd_idx]; |
3490 | s->locked++; | |
3491 | set_bit(R5_LOCKED, &dev->flags); | |
3492 | set_bit(R5_Wantwrite, &dev->flags); | |
3493 | } | |
3494 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3495 | ||
3496 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3497 | break; |
3498 | case check_state_run: | |
3499 | case check_state_run_q: | |
3500 | case check_state_run_pq: | |
3501 | break; /* we will be called again upon completion */ | |
3502 | case check_state_check_result: | |
3503 | sh->check_state = check_state_idle; | |
3504 | ||
3505 | /* handle a successful check operation, if parity is correct | |
3506 | * we are done. Otherwise update the mismatch count and repair | |
3507 | * parity if !MD_RECOVERY_CHECK | |
3508 | */ | |
3509 | if (sh->ops.zero_sum_result == 0) { | |
3510 | /* both parities are correct */ | |
3511 | if (!s->failed) | |
3512 | set_bit(STRIPE_INSYNC, &sh->state); | |
3513 | else { | |
3514 | /* in contrast to the raid5 case we can validate | |
3515 | * parity, but still have a failure to write | |
3516 | * back | |
3517 | */ | |
3518 | sh->check_state = check_state_compute_result; | |
3519 | /* Returning at this point means that we may go | |
3520 | * off and bring p and/or q uptodate again so | |
3521 | * we make sure to check zero_sum_result again | |
3522 | * to verify if p or q need writeback | |
3523 | */ | |
3524 | } | |
3525 | } else { | |
7f7583d4 | 3526 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
3527 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3528 | /* don't try to repair!! */ | |
3529 | set_bit(STRIPE_INSYNC, &sh->state); | |
3530 | else { | |
3531 | int *target = &sh->ops.target; | |
3532 | ||
3533 | sh->ops.target = -1; | |
3534 | sh->ops.target2 = -1; | |
3535 | sh->check_state = check_state_compute_run; | |
3536 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3537 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3538 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3539 | set_bit(R5_Wantcompute, | |
3540 | &sh->dev[pd_idx].flags); | |
3541 | *target = pd_idx; | |
3542 | target = &sh->ops.target2; | |
3543 | s->uptodate++; | |
3544 | } | |
3545 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3546 | set_bit(R5_Wantcompute, | |
3547 | &sh->dev[qd_idx].flags); | |
3548 | *target = qd_idx; | |
3549 | s->uptodate++; | |
3550 | } | |
3551 | } | |
3552 | } | |
3553 | break; | |
3554 | case check_state_compute_run: | |
3555 | break; | |
3556 | default: | |
3557 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3558 | __func__, sh->check_state, | |
3559 | (unsigned long long) sh->sector); | |
3560 | BUG(); | |
a4456856 DW |
3561 | } |
3562 | } | |
3563 | ||
d1688a6d | 3564 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3565 | { |
3566 | int i; | |
3567 | ||
3568 | /* We have read all the blocks in this stripe and now we need to | |
3569 | * copy some of them into a target stripe for expand. | |
3570 | */ | |
f0a50d37 | 3571 | struct dma_async_tx_descriptor *tx = NULL; |
a4456856 DW |
3572 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3573 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3574 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3575 | int dd_idx, j; |
a4456856 | 3576 | struct stripe_head *sh2; |
a08abd8c | 3577 | struct async_submit_ctl submit; |
a4456856 | 3578 | |
784052ec | 3579 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3580 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3581 | &dd_idx, NULL); | |
a8c906ca | 3582 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3583 | if (sh2 == NULL) |
3584 | /* so far only the early blocks of this stripe | |
3585 | * have been requested. When later blocks | |
3586 | * get requested, we will try again | |
3587 | */ | |
3588 | continue; | |
3589 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3590 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3591 | /* must have already done this block */ | |
3592 | release_stripe(sh2); | |
3593 | continue; | |
3594 | } | |
f0a50d37 DW |
3595 | |
3596 | /* place all the copies on one channel */ | |
a08abd8c | 3597 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3598 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3599 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3600 | &submit); |
f0a50d37 | 3601 | |
a4456856 DW |
3602 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3603 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3604 | for (j = 0; j < conf->raid_disks; j++) | |
3605 | if (j != sh2->pd_idx && | |
86c374ba | 3606 | j != sh2->qd_idx && |
a4456856 DW |
3607 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3608 | break; | |
3609 | if (j == conf->raid_disks) { | |
3610 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3611 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3612 | } | |
3613 | release_stripe(sh2); | |
f0a50d37 | 3614 | |
a4456856 | 3615 | } |
a2e08551 | 3616 | /* done submitting copies, wait for them to complete */ |
749586b7 | 3617 | async_tx_quiesce(&tx); |
a4456856 | 3618 | } |
1da177e4 LT |
3619 | |
3620 | /* | |
3621 | * handle_stripe - do things to a stripe. | |
3622 | * | |
9a3e1101 N |
3623 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3624 | * state of various bits to see what needs to be done. | |
1da177e4 | 3625 | * Possible results: |
9a3e1101 N |
3626 | * return some read requests which now have data |
3627 | * return some write requests which are safely on storage | |
1da177e4 LT |
3628 | * schedule a read on some buffers |
3629 | * schedule a write of some buffers | |
3630 | * return confirmation of parity correctness | |
3631 | * | |
1da177e4 | 3632 | */ |
a4456856 | 3633 | |
acfe726b | 3634 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3635 | { |
d1688a6d | 3636 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3637 | int disks = sh->disks; |
474af965 N |
3638 | struct r5dev *dev; |
3639 | int i; | |
9a3e1101 | 3640 | int do_recovery = 0; |
1da177e4 | 3641 | |
acfe726b N |
3642 | memset(s, 0, sizeof(*s)); |
3643 | ||
acfe726b N |
3644 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3645 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state); | |
3646 | s->failed_num[0] = -1; | |
3647 | s->failed_num[1] = -1; | |
1da177e4 | 3648 | |
acfe726b | 3649 | /* Now to look around and see what can be done */ |
1da177e4 | 3650 | rcu_read_lock(); |
16a53ecc | 3651 | for (i=disks; i--; ) { |
3cb03002 | 3652 | struct md_rdev *rdev; |
31c176ec N |
3653 | sector_t first_bad; |
3654 | int bad_sectors; | |
3655 | int is_bad = 0; | |
acfe726b | 3656 | |
16a53ecc | 3657 | dev = &sh->dev[i]; |
1da177e4 | 3658 | |
45b4233c | 3659 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
3660 | i, dev->flags, |
3661 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
3662 | /* maybe we can reply to a read |
3663 | * | |
3664 | * new wantfill requests are only permitted while | |
3665 | * ops_complete_biofill is guaranteed to be inactive | |
3666 | */ | |
3667 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
3668 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
3669 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 3670 | |
16a53ecc | 3671 | /* now count some things */ |
cc94015a N |
3672 | if (test_bit(R5_LOCKED, &dev->flags)) |
3673 | s->locked++; | |
3674 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
3675 | s->uptodate++; | |
2d6e4ecc | 3676 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
3677 | s->compute++; |
3678 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 3679 | } |
1da177e4 | 3680 | |
acfe726b | 3681 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 3682 | s->to_fill++; |
acfe726b | 3683 | else if (dev->toread) |
cc94015a | 3684 | s->to_read++; |
16a53ecc | 3685 | if (dev->towrite) { |
cc94015a | 3686 | s->to_write++; |
16a53ecc | 3687 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 3688 | s->non_overwrite++; |
16a53ecc | 3689 | } |
a4456856 | 3690 | if (dev->written) |
cc94015a | 3691 | s->written++; |
14a75d3e N |
3692 | /* Prefer to use the replacement for reads, but only |
3693 | * if it is recovered enough and has no bad blocks. | |
3694 | */ | |
3695 | rdev = rcu_dereference(conf->disks[i].replacement); | |
3696 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
3697 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
3698 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3699 | &first_bad, &bad_sectors)) | |
3700 | set_bit(R5_ReadRepl, &dev->flags); | |
3701 | else { | |
9a3e1101 N |
3702 | if (rdev) |
3703 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
3704 | rdev = rcu_dereference(conf->disks[i].rdev); |
3705 | clear_bit(R5_ReadRepl, &dev->flags); | |
3706 | } | |
9283d8c5 N |
3707 | if (rdev && test_bit(Faulty, &rdev->flags)) |
3708 | rdev = NULL; | |
31c176ec N |
3709 | if (rdev) { |
3710 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
3711 | &first_bad, &bad_sectors); | |
3712 | if (s->blocked_rdev == NULL | |
3713 | && (test_bit(Blocked, &rdev->flags) | |
3714 | || is_bad < 0)) { | |
3715 | if (is_bad < 0) | |
3716 | set_bit(BlockedBadBlocks, | |
3717 | &rdev->flags); | |
3718 | s->blocked_rdev = rdev; | |
3719 | atomic_inc(&rdev->nr_pending); | |
3720 | } | |
6bfe0b49 | 3721 | } |
415e72d0 N |
3722 | clear_bit(R5_Insync, &dev->flags); |
3723 | if (!rdev) | |
3724 | /* Not in-sync */; | |
31c176ec N |
3725 | else if (is_bad) { |
3726 | /* also not in-sync */ | |
18b9837e N |
3727 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
3728 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
3729 | /* treat as in-sync, but with a read error |
3730 | * which we can now try to correct | |
3731 | */ | |
3732 | set_bit(R5_Insync, &dev->flags); | |
3733 | set_bit(R5_ReadError, &dev->flags); | |
3734 | } | |
3735 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 3736 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 3737 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 3738 | /* in sync if before recovery_offset */ |
30d7a483 N |
3739 | set_bit(R5_Insync, &dev->flags); |
3740 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
3741 | test_bit(R5_Expanded, &dev->flags)) | |
3742 | /* If we've reshaped into here, we assume it is Insync. | |
3743 | * We will shortly update recovery_offset to make | |
3744 | * it official. | |
3745 | */ | |
3746 | set_bit(R5_Insync, &dev->flags); | |
3747 | ||
1cc03eb9 | 3748 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
3749 | /* This flag does not apply to '.replacement' |
3750 | * only to .rdev, so make sure to check that*/ | |
3751 | struct md_rdev *rdev2 = rcu_dereference( | |
3752 | conf->disks[i].rdev); | |
3753 | if (rdev2 == rdev) | |
3754 | clear_bit(R5_Insync, &dev->flags); | |
3755 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 3756 | s->handle_bad_blocks = 1; |
14a75d3e | 3757 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
3758 | } else |
3759 | clear_bit(R5_WriteError, &dev->flags); | |
3760 | } | |
1cc03eb9 | 3761 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
3762 | /* This flag does not apply to '.replacement' |
3763 | * only to .rdev, so make sure to check that*/ | |
3764 | struct md_rdev *rdev2 = rcu_dereference( | |
3765 | conf->disks[i].rdev); | |
3766 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 3767 | s->handle_bad_blocks = 1; |
14a75d3e | 3768 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
3769 | } else |
3770 | clear_bit(R5_MadeGood, &dev->flags); | |
3771 | } | |
977df362 N |
3772 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
3773 | struct md_rdev *rdev2 = rcu_dereference( | |
3774 | conf->disks[i].replacement); | |
3775 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
3776 | s->handle_bad_blocks = 1; | |
3777 | atomic_inc(&rdev2->nr_pending); | |
3778 | } else | |
3779 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
3780 | } | |
415e72d0 | 3781 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
3782 | /* The ReadError flag will just be confusing now */ |
3783 | clear_bit(R5_ReadError, &dev->flags); | |
3784 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 3785 | } |
415e72d0 N |
3786 | if (test_bit(R5_ReadError, &dev->flags)) |
3787 | clear_bit(R5_Insync, &dev->flags); | |
3788 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
3789 | if (s->failed < 2) |
3790 | s->failed_num[s->failed] = i; | |
3791 | s->failed++; | |
9a3e1101 N |
3792 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
3793 | do_recovery = 1; | |
415e72d0 | 3794 | } |
1da177e4 | 3795 | } |
9a3e1101 N |
3796 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
3797 | /* If there is a failed device being replaced, | |
3798 | * we must be recovering. | |
3799 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 3800 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
3801 | * else we can only be replacing |
3802 | * sync and recovery both need to read all devices, and so | |
3803 | * use the same flag. | |
3804 | */ | |
3805 | if (do_recovery || | |
c6d2e084 | 3806 | sh->sector >= conf->mddev->recovery_cp || |
3807 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
3808 | s->syncing = 1; |
3809 | else | |
3810 | s->replacing = 1; | |
3811 | } | |
1da177e4 | 3812 | rcu_read_unlock(); |
cc94015a N |
3813 | } |
3814 | ||
3815 | static void handle_stripe(struct stripe_head *sh) | |
3816 | { | |
3817 | struct stripe_head_state s; | |
d1688a6d | 3818 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 3819 | int i; |
84789554 N |
3820 | int prexor; |
3821 | int disks = sh->disks; | |
474af965 | 3822 | struct r5dev *pdev, *qdev; |
cc94015a N |
3823 | |
3824 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 3825 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
3826 | /* already being handled, ensure it gets handled |
3827 | * again when current action finishes */ | |
3828 | set_bit(STRIPE_HANDLE, &sh->state); | |
3829 | return; | |
3830 | } | |
3831 | ||
da41ba65 | 3832 | clear_bit(STRIPE_BATCH_READY, &sh->state); |
f8dfcffd N |
3833 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
3834 | spin_lock(&sh->stripe_lock); | |
3835 | /* Cannot process 'sync' concurrently with 'discard' */ | |
3836 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
3837 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
3838 | set_bit(STRIPE_SYNCING, &sh->state); | |
3839 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 3840 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
3841 | } |
3842 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
3843 | } |
3844 | clear_bit(STRIPE_DELAYED, &sh->state); | |
3845 | ||
3846 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
3847 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
3848 | (unsigned long long)sh->sector, sh->state, | |
3849 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
3850 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 3851 | |
acfe726b | 3852 | analyse_stripe(sh, &s); |
c5a31000 | 3853 | |
bc2607f3 N |
3854 | if (s.handle_bad_blocks) { |
3855 | set_bit(STRIPE_HANDLE, &sh->state); | |
3856 | goto finish; | |
3857 | } | |
3858 | ||
474af965 N |
3859 | if (unlikely(s.blocked_rdev)) { |
3860 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 3861 | s.replacing || s.to_write || s.written) { |
474af965 N |
3862 | set_bit(STRIPE_HANDLE, &sh->state); |
3863 | goto finish; | |
3864 | } | |
3865 | /* There is nothing for the blocked_rdev to block */ | |
3866 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
3867 | s.blocked_rdev = NULL; | |
3868 | } | |
3869 | ||
3870 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
3871 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
3872 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
3873 | } | |
3874 | ||
3875 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
3876 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
3877 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
3878 | s.failed_num[0], s.failed_num[1]); | |
3879 | /* check if the array has lost more than max_degraded devices and, | |
3880 | * if so, some requests might need to be failed. | |
3881 | */ | |
9a3f530f N |
3882 | if (s.failed > conf->max_degraded) { |
3883 | sh->check_state = 0; | |
3884 | sh->reconstruct_state = 0; | |
3885 | if (s.to_read+s.to_write+s.written) | |
3886 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 3887 | if (s.syncing + s.replacing) |
9a3f530f N |
3888 | handle_failed_sync(conf, sh, &s); |
3889 | } | |
474af965 | 3890 | |
84789554 N |
3891 | /* Now we check to see if any write operations have recently |
3892 | * completed | |
3893 | */ | |
3894 | prexor = 0; | |
3895 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
3896 | prexor = 1; | |
3897 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
3898 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
3899 | sh->reconstruct_state = reconstruct_state_idle; | |
3900 | ||
3901 | /* All the 'written' buffers and the parity block are ready to | |
3902 | * be written back to disk | |
3903 | */ | |
9e444768 SL |
3904 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
3905 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 3906 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
3907 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
3908 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
3909 | for (i = disks; i--; ) { |
3910 | struct r5dev *dev = &sh->dev[i]; | |
3911 | if (test_bit(R5_LOCKED, &dev->flags) && | |
3912 | (i == sh->pd_idx || i == sh->qd_idx || | |
3913 | dev->written)) { | |
3914 | pr_debug("Writing block %d\n", i); | |
3915 | set_bit(R5_Wantwrite, &dev->flags); | |
3916 | if (prexor) | |
3917 | continue; | |
9c4bdf69 N |
3918 | if (s.failed > 1) |
3919 | continue; | |
84789554 N |
3920 | if (!test_bit(R5_Insync, &dev->flags) || |
3921 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
3922 | s.failed == 0)) | |
3923 | set_bit(STRIPE_INSYNC, &sh->state); | |
3924 | } | |
3925 | } | |
3926 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3927 | s.dec_preread_active = 1; | |
3928 | } | |
3929 | ||
ef5b7c69 N |
3930 | /* |
3931 | * might be able to return some write requests if the parity blocks | |
3932 | * are safe, or on a failed drive | |
3933 | */ | |
3934 | pdev = &sh->dev[sh->pd_idx]; | |
3935 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
3936 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
3937 | qdev = &sh->dev[sh->qd_idx]; | |
3938 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
3939 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
3940 | || conf->level < 6; | |
3941 | ||
3942 | if (s.written && | |
3943 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
3944 | && !test_bit(R5_LOCKED, &pdev->flags) | |
3945 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
3946 | test_bit(R5_Discard, &pdev->flags))))) && | |
3947 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
3948 | && !test_bit(R5_LOCKED, &qdev->flags) | |
3949 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
3950 | test_bit(R5_Discard, &qdev->flags)))))) | |
3951 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
3952 | ||
3953 | /* Now we might consider reading some blocks, either to check/generate | |
3954 | * parity, or to satisfy requests | |
3955 | * or to load a block that is being partially written. | |
3956 | */ | |
3957 | if (s.to_read || s.non_overwrite | |
3958 | || (conf->level == 6 && s.to_write && s.failed) | |
3959 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
3960 | || s.replacing | |
3961 | || s.expanding) | |
3962 | handle_stripe_fill(sh, &s, disks); | |
3963 | ||
84789554 N |
3964 | /* Now to consider new write requests and what else, if anything |
3965 | * should be read. We do not handle new writes when: | |
3966 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
3967 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
3968 | * block. | |
3969 | */ | |
3970 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
3971 | handle_stripe_dirtying(conf, sh, &s, disks); | |
3972 | ||
3973 | /* maybe we need to check and possibly fix the parity for this stripe | |
3974 | * Any reads will already have been scheduled, so we just see if enough | |
3975 | * data is available. The parity check is held off while parity | |
3976 | * dependent operations are in flight. | |
3977 | */ | |
3978 | if (sh->check_state || | |
3979 | (s.syncing && s.locked == 0 && | |
3980 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
3981 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
3982 | if (conf->level == 6) | |
3983 | handle_parity_checks6(conf, sh, &s, disks); | |
3984 | else | |
3985 | handle_parity_checks5(conf, sh, &s, disks); | |
3986 | } | |
c5a31000 | 3987 | |
f94c0b66 N |
3988 | if ((s.replacing || s.syncing) && s.locked == 0 |
3989 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
3990 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
3991 | /* Write out to replacement devices where possible */ |
3992 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
3993 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
3994 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
3995 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
3996 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
3997 | s.locked++; | |
3998 | } | |
f94c0b66 N |
3999 | if (s.replacing) |
4000 | set_bit(STRIPE_INSYNC, &sh->state); | |
4001 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4002 | } |
4003 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4004 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4005 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4006 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4007 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4008 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4009 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4010 | } |
4011 | ||
4012 | /* If the failed drives are just a ReadError, then we might need | |
4013 | * to progress the repair/check process | |
4014 | */ | |
4015 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4016 | for (i = 0; i < s.failed; i++) { | |
4017 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4018 | if (test_bit(R5_ReadError, &dev->flags) | |
4019 | && !test_bit(R5_LOCKED, &dev->flags) | |
4020 | && test_bit(R5_UPTODATE, &dev->flags) | |
4021 | ) { | |
4022 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4023 | set_bit(R5_Wantwrite, &dev->flags); | |
4024 | set_bit(R5_ReWrite, &dev->flags); | |
4025 | set_bit(R5_LOCKED, &dev->flags); | |
4026 | s.locked++; | |
4027 | } else { | |
4028 | /* let's read it back */ | |
4029 | set_bit(R5_Wantread, &dev->flags); | |
4030 | set_bit(R5_LOCKED, &dev->flags); | |
4031 | s.locked++; | |
4032 | } | |
4033 | } | |
4034 | } | |
4035 | ||
3687c061 N |
4036 | /* Finish reconstruct operations initiated by the expansion process */ |
4037 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4038 | struct stripe_head *sh_src | |
4039 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
4040 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
4041 | /* sh cannot be written until sh_src has been read. | |
4042 | * so arrange for sh to be delayed a little | |
4043 | */ | |
4044 | set_bit(STRIPE_DELAYED, &sh->state); | |
4045 | set_bit(STRIPE_HANDLE, &sh->state); | |
4046 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4047 | &sh_src->state)) | |
4048 | atomic_inc(&conf->preread_active_stripes); | |
4049 | release_stripe(sh_src); | |
4050 | goto finish; | |
4051 | } | |
4052 | if (sh_src) | |
4053 | release_stripe(sh_src); | |
4054 | ||
4055 | sh->reconstruct_state = reconstruct_state_idle; | |
4056 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4057 | for (i = conf->raid_disks; i--; ) { | |
4058 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4059 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4060 | s.locked++; | |
4061 | } | |
4062 | } | |
f416885e | 4063 | |
3687c061 N |
4064 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4065 | !sh->reconstruct_state) { | |
4066 | /* Need to write out all blocks after computing parity */ | |
4067 | sh->disks = conf->raid_disks; | |
4068 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4069 | schedule_reconstruction(sh, &s, 1, 1); | |
4070 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4071 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4072 | atomic_dec(&conf->reshape_stripes); | |
4073 | wake_up(&conf->wait_for_overlap); | |
4074 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4075 | } | |
4076 | ||
4077 | if (s.expanding && s.locked == 0 && | |
4078 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4079 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4080 | |
3687c061 | 4081 | finish: |
6bfe0b49 | 4082 | /* wait for this device to become unblocked */ |
5f066c63 N |
4083 | if (unlikely(s.blocked_rdev)) { |
4084 | if (conf->mddev->external) | |
4085 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4086 | conf->mddev); | |
4087 | else | |
4088 | /* Internal metadata will immediately | |
4089 | * be written by raid5d, so we don't | |
4090 | * need to wait here. | |
4091 | */ | |
4092 | rdev_dec_pending(s.blocked_rdev, | |
4093 | conf->mddev); | |
4094 | } | |
6bfe0b49 | 4095 | |
bc2607f3 N |
4096 | if (s.handle_bad_blocks) |
4097 | for (i = disks; i--; ) { | |
3cb03002 | 4098 | struct md_rdev *rdev; |
bc2607f3 N |
4099 | struct r5dev *dev = &sh->dev[i]; |
4100 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4101 | /* We own a safe reference to the rdev */ | |
4102 | rdev = conf->disks[i].rdev; | |
4103 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4104 | STRIPE_SECTORS, 0)) | |
4105 | md_error(conf->mddev, rdev); | |
4106 | rdev_dec_pending(rdev, conf->mddev); | |
4107 | } | |
b84db560 N |
4108 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4109 | rdev = conf->disks[i].rdev; | |
4110 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4111 | STRIPE_SECTORS, 0); |
b84db560 N |
4112 | rdev_dec_pending(rdev, conf->mddev); |
4113 | } | |
977df362 N |
4114 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4115 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4116 | if (!rdev) |
4117 | /* rdev have been moved down */ | |
4118 | rdev = conf->disks[i].rdev; | |
977df362 | 4119 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4120 | STRIPE_SECTORS, 0); |
977df362 N |
4121 | rdev_dec_pending(rdev, conf->mddev); |
4122 | } | |
bc2607f3 N |
4123 | } |
4124 | ||
6c0069c0 YT |
4125 | if (s.ops_request) |
4126 | raid_run_ops(sh, s.ops_request); | |
4127 | ||
f0e43bcd | 4128 | ops_run_io(sh, &s); |
16a53ecc | 4129 | |
c5709ef6 | 4130 | if (s.dec_preread_active) { |
729a1866 | 4131 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4132 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4133 | * have actually been submitted. |
4134 | */ | |
4135 | atomic_dec(&conf->preread_active_stripes); | |
4136 | if (atomic_read(&conf->preread_active_stripes) < | |
4137 | IO_THRESHOLD) | |
4138 | md_wakeup_thread(conf->mddev->thread); | |
4139 | } | |
4140 | ||
c5709ef6 | 4141 | return_io(s.return_bi); |
16a53ecc | 4142 | |
257a4b42 | 4143 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4144 | } |
4145 | ||
d1688a6d | 4146 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4147 | { |
4148 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4149 | while (!list_empty(&conf->delayed_list)) { | |
4150 | struct list_head *l = conf->delayed_list.next; | |
4151 | struct stripe_head *sh; | |
4152 | sh = list_entry(l, struct stripe_head, lru); | |
4153 | list_del_init(l); | |
4154 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4155 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4156 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4157 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4158 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4159 | } |
482c0834 | 4160 | } |
16a53ecc N |
4161 | } |
4162 | ||
566c09c5 SL |
4163 | static void activate_bit_delay(struct r5conf *conf, |
4164 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4165 | { |
4166 | /* device_lock is held */ | |
4167 | struct list_head head; | |
4168 | list_add(&head, &conf->bitmap_list); | |
4169 | list_del_init(&conf->bitmap_list); | |
4170 | while (!list_empty(&head)) { | |
4171 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4172 | int hash; |
16a53ecc N |
4173 | list_del_init(&sh->lru); |
4174 | atomic_inc(&sh->count); | |
566c09c5 SL |
4175 | hash = sh->hash_lock_index; |
4176 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4177 | } |
4178 | } | |
4179 | ||
5c675f83 | 4180 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4181 | { |
d1688a6d | 4182 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4183 | |
4184 | /* No difference between reads and writes. Just check | |
4185 | * how busy the stripe_cache is | |
4186 | */ | |
3fa841d7 | 4187 | |
f022b2fd N |
4188 | if (conf->inactive_blocked) |
4189 | return 1; | |
4190 | if (conf->quiesce) | |
4191 | return 1; | |
4bda556a | 4192 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4193 | return 1; |
4194 | ||
4195 | return 0; | |
4196 | } | |
4197 | ||
23032a0e RBJ |
4198 | /* We want read requests to align with chunks where possible, |
4199 | * but write requests don't need to. | |
4200 | */ | |
64590f45 | 4201 | static int raid5_mergeable_bvec(struct mddev *mddev, |
cc371e66 AK |
4202 | struct bvec_merge_data *bvm, |
4203 | struct bio_vec *biovec) | |
23032a0e | 4204 | { |
cc371e66 | 4205 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 4206 | int max; |
9d8f0363 | 4207 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 4208 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 4209 | |
cc371e66 | 4210 | if ((bvm->bi_rw & 1) == WRITE) |
23032a0e RBJ |
4211 | return biovec->bv_len; /* always allow writes to be mergeable */ |
4212 | ||
664e7c41 AN |
4213 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
4214 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
4215 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
4216 | if (max < 0) max = 0; | |
4217 | if (max <= biovec->bv_len && bio_sectors == 0) | |
4218 | return biovec->bv_len; | |
4219 | else | |
4220 | return max; | |
4221 | } | |
4222 | ||
fd01b88c | 4223 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4224 | { |
4f024f37 | 4225 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
9d8f0363 | 4226 | unsigned int chunk_sectors = mddev->chunk_sectors; |
aa8b57aa | 4227 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4228 | |
664e7c41 AN |
4229 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
4230 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
4231 | return chunk_sectors >= |
4232 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4233 | } | |
4234 | ||
46031f9a RBJ |
4235 | /* |
4236 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4237 | * later sampled by raid5d. | |
4238 | */ | |
d1688a6d | 4239 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4240 | { |
4241 | unsigned long flags; | |
4242 | ||
4243 | spin_lock_irqsave(&conf->device_lock, flags); | |
4244 | ||
4245 | bi->bi_next = conf->retry_read_aligned_list; | |
4246 | conf->retry_read_aligned_list = bi; | |
4247 | ||
4248 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4249 | md_wakeup_thread(conf->mddev->thread); | |
4250 | } | |
4251 | ||
d1688a6d | 4252 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4253 | { |
4254 | struct bio *bi; | |
4255 | ||
4256 | bi = conf->retry_read_aligned; | |
4257 | if (bi) { | |
4258 | conf->retry_read_aligned = NULL; | |
4259 | return bi; | |
4260 | } | |
4261 | bi = conf->retry_read_aligned_list; | |
4262 | if(bi) { | |
387bb173 | 4263 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4264 | bi->bi_next = NULL; |
960e739d JA |
4265 | /* |
4266 | * this sets the active strip count to 1 and the processed | |
4267 | * strip count to zero (upper 8 bits) | |
4268 | */ | |
e7836bd6 | 4269 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
4270 | } |
4271 | ||
4272 | return bi; | |
4273 | } | |
4274 | ||
f679623f RBJ |
4275 | /* |
4276 | * The "raid5_align_endio" should check if the read succeeded and if it | |
4277 | * did, call bio_endio on the original bio (having bio_put the new bio | |
4278 | * first). | |
4279 | * If the read failed.. | |
4280 | */ | |
6712ecf8 | 4281 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
4282 | { |
4283 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 4284 | struct mddev *mddev; |
d1688a6d | 4285 | struct r5conf *conf; |
46031f9a | 4286 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 4287 | struct md_rdev *rdev; |
46031f9a | 4288 | |
f679623f | 4289 | bio_put(bi); |
46031f9a | 4290 | |
46031f9a RBJ |
4291 | rdev = (void*)raid_bi->bi_next; |
4292 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
4293 | mddev = rdev->mddev; |
4294 | conf = mddev->private; | |
46031f9a RBJ |
4295 | |
4296 | rdev_dec_pending(rdev, conf->mddev); | |
4297 | ||
4298 | if (!error && uptodate) { | |
0a82a8d1 LT |
4299 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
4300 | raid_bi, 0); | |
6712ecf8 | 4301 | bio_endio(raid_bi, 0); |
46031f9a RBJ |
4302 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
4303 | wake_up(&conf->wait_for_stripe); | |
6712ecf8 | 4304 | return; |
46031f9a RBJ |
4305 | } |
4306 | ||
45b4233c | 4307 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
4308 | |
4309 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
4310 | } |
4311 | ||
387bb173 NB |
4312 | static int bio_fits_rdev(struct bio *bi) |
4313 | { | |
165125e1 | 4314 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 4315 | |
aa8b57aa | 4316 | if (bio_sectors(bi) > queue_max_sectors(q)) |
387bb173 NB |
4317 | return 0; |
4318 | blk_recount_segments(q, bi); | |
8a78362c | 4319 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
4320 | return 0; |
4321 | ||
4322 | if (q->merge_bvec_fn) | |
4323 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
4324 | * just just give up | |
4325 | */ | |
4326 | return 0; | |
4327 | ||
4328 | return 1; | |
4329 | } | |
4330 | ||
fd01b88c | 4331 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 4332 | { |
d1688a6d | 4333 | struct r5conf *conf = mddev->private; |
8553fe7e | 4334 | int dd_idx; |
f679623f | 4335 | struct bio* align_bi; |
3cb03002 | 4336 | struct md_rdev *rdev; |
671488cc | 4337 | sector_t end_sector; |
f679623f RBJ |
4338 | |
4339 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 4340 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
4341 | return 0; |
4342 | } | |
4343 | /* | |
a167f663 | 4344 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 4345 | */ |
a167f663 | 4346 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
4347 | if (!align_bi) |
4348 | return 0; | |
4349 | /* | |
4350 | * set bi_end_io to a new function, and set bi_private to the | |
4351 | * original bio. | |
4352 | */ | |
4353 | align_bi->bi_end_io = raid5_align_endio; | |
4354 | align_bi->bi_private = raid_bio; | |
4355 | /* | |
4356 | * compute position | |
4357 | */ | |
4f024f37 KO |
4358 | align_bi->bi_iter.bi_sector = |
4359 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
4360 | 0, &dd_idx, NULL); | |
f679623f | 4361 | |
f73a1c7d | 4362 | end_sector = bio_end_sector(align_bi); |
f679623f | 4363 | rcu_read_lock(); |
671488cc N |
4364 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
4365 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
4366 | rdev->recovery_offset < end_sector) { | |
4367 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
4368 | if (rdev && | |
4369 | (test_bit(Faulty, &rdev->flags) || | |
4370 | !(test_bit(In_sync, &rdev->flags) || | |
4371 | rdev->recovery_offset >= end_sector))) | |
4372 | rdev = NULL; | |
4373 | } | |
4374 | if (rdev) { | |
31c176ec N |
4375 | sector_t first_bad; |
4376 | int bad_sectors; | |
4377 | ||
f679623f RBJ |
4378 | atomic_inc(&rdev->nr_pending); |
4379 | rcu_read_unlock(); | |
46031f9a RBJ |
4380 | raid_bio->bi_next = (void*)rdev; |
4381 | align_bi->bi_bdev = rdev->bdev; | |
3fd83717 | 4382 | __clear_bit(BIO_SEG_VALID, &align_bi->bi_flags); |
46031f9a | 4383 | |
31c176ec | 4384 | if (!bio_fits_rdev(align_bi) || |
4f024f37 KO |
4385 | is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4386 | bio_sectors(align_bi), | |
31c176ec N |
4387 | &first_bad, &bad_sectors)) { |
4388 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
4389 | bio_put(align_bi); |
4390 | rdev_dec_pending(rdev, mddev); | |
4391 | return 0; | |
4392 | } | |
4393 | ||
6c0544e2 | 4394 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 4395 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 4396 | |
46031f9a RBJ |
4397 | spin_lock_irq(&conf->device_lock); |
4398 | wait_event_lock_irq(conf->wait_for_stripe, | |
4399 | conf->quiesce == 0, | |
eed8c02e | 4400 | conf->device_lock); |
46031f9a RBJ |
4401 | atomic_inc(&conf->active_aligned_reads); |
4402 | spin_unlock_irq(&conf->device_lock); | |
4403 | ||
e3620a3a JB |
4404 | if (mddev->gendisk) |
4405 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
4406 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 4407 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
4408 | generic_make_request(align_bi); |
4409 | return 1; | |
4410 | } else { | |
4411 | rcu_read_unlock(); | |
46031f9a | 4412 | bio_put(align_bi); |
f679623f RBJ |
4413 | return 0; |
4414 | } | |
4415 | } | |
4416 | ||
8b3e6cdc DW |
4417 | /* __get_priority_stripe - get the next stripe to process |
4418 | * | |
4419 | * Full stripe writes are allowed to pass preread active stripes up until | |
4420 | * the bypass_threshold is exceeded. In general the bypass_count | |
4421 | * increments when the handle_list is handled before the hold_list; however, it | |
4422 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
4423 | * stripe with in flight i/o. The bypass_count will be reset when the | |
4424 | * head of the hold_list has changed, i.e. the head was promoted to the | |
4425 | * handle_list. | |
4426 | */ | |
851c30c9 | 4427 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 4428 | { |
851c30c9 SL |
4429 | struct stripe_head *sh = NULL, *tmp; |
4430 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 4431 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
4432 | |
4433 | if (conf->worker_cnt_per_group == 0) { | |
4434 | handle_list = &conf->handle_list; | |
4435 | } else if (group != ANY_GROUP) { | |
4436 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 4437 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
4438 | } else { |
4439 | int i; | |
4440 | for (i = 0; i < conf->group_cnt; i++) { | |
4441 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 4442 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
4443 | if (!list_empty(handle_list)) |
4444 | break; | |
4445 | } | |
4446 | } | |
8b3e6cdc DW |
4447 | |
4448 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
4449 | __func__, | |
851c30c9 | 4450 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
4451 | list_empty(&conf->hold_list) ? "empty" : "busy", |
4452 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
4453 | ||
851c30c9 SL |
4454 | if (!list_empty(handle_list)) { |
4455 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
4456 | |
4457 | if (list_empty(&conf->hold_list)) | |
4458 | conf->bypass_count = 0; | |
4459 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
4460 | if (conf->hold_list.next == conf->last_hold) | |
4461 | conf->bypass_count++; | |
4462 | else { | |
4463 | conf->last_hold = conf->hold_list.next; | |
4464 | conf->bypass_count -= conf->bypass_threshold; | |
4465 | if (conf->bypass_count < 0) | |
4466 | conf->bypass_count = 0; | |
4467 | } | |
4468 | } | |
4469 | } else if (!list_empty(&conf->hold_list) && | |
4470 | ((conf->bypass_threshold && | |
4471 | conf->bypass_count > conf->bypass_threshold) || | |
4472 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
4473 | |
4474 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
4475 | if (conf->worker_cnt_per_group == 0 || | |
4476 | group == ANY_GROUP || | |
4477 | !cpu_online(tmp->cpu) || | |
4478 | cpu_to_group(tmp->cpu) == group) { | |
4479 | sh = tmp; | |
4480 | break; | |
4481 | } | |
4482 | } | |
4483 | ||
4484 | if (sh) { | |
4485 | conf->bypass_count -= conf->bypass_threshold; | |
4486 | if (conf->bypass_count < 0) | |
4487 | conf->bypass_count = 0; | |
4488 | } | |
bfc90cb0 | 4489 | wg = NULL; |
851c30c9 SL |
4490 | } |
4491 | ||
4492 | if (!sh) | |
8b3e6cdc DW |
4493 | return NULL; |
4494 | ||
bfc90cb0 SL |
4495 | if (wg) { |
4496 | wg->stripes_cnt--; | |
4497 | sh->group = NULL; | |
4498 | } | |
8b3e6cdc | 4499 | list_del_init(&sh->lru); |
c7a6d35e | 4500 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
4501 | return sh; |
4502 | } | |
f679623f | 4503 | |
8811b596 SL |
4504 | struct raid5_plug_cb { |
4505 | struct blk_plug_cb cb; | |
4506 | struct list_head list; | |
566c09c5 | 4507 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
4508 | }; |
4509 | ||
4510 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4511 | { | |
4512 | struct raid5_plug_cb *cb = container_of( | |
4513 | blk_cb, struct raid5_plug_cb, cb); | |
4514 | struct stripe_head *sh; | |
4515 | struct mddev *mddev = cb->cb.data; | |
4516 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 4517 | int cnt = 0; |
566c09c5 | 4518 | int hash; |
8811b596 SL |
4519 | |
4520 | if (cb->list.next && !list_empty(&cb->list)) { | |
4521 | spin_lock_irq(&conf->device_lock); | |
4522 | while (!list_empty(&cb->list)) { | |
4523 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
4524 | list_del_init(&sh->lru); | |
4525 | /* | |
4526 | * avoid race release_stripe_plug() sees | |
4527 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
4528 | * is still in our list | |
4529 | */ | |
4e857c58 | 4530 | smp_mb__before_atomic(); |
8811b596 | 4531 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
4532 | /* |
4533 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
4534 | * case, the count is always > 1 here | |
4535 | */ | |
566c09c5 SL |
4536 | hash = sh->hash_lock_index; |
4537 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 4538 | cnt++; |
8811b596 SL |
4539 | } |
4540 | spin_unlock_irq(&conf->device_lock); | |
4541 | } | |
566c09c5 SL |
4542 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
4543 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
4544 | if (mddev->queue) |
4545 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
4546 | kfree(cb); |
4547 | } | |
4548 | ||
4549 | static void release_stripe_plug(struct mddev *mddev, | |
4550 | struct stripe_head *sh) | |
4551 | { | |
4552 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
4553 | raid5_unplug, mddev, | |
4554 | sizeof(struct raid5_plug_cb)); | |
4555 | struct raid5_plug_cb *cb; | |
4556 | ||
4557 | if (!blk_cb) { | |
4558 | release_stripe(sh); | |
4559 | return; | |
4560 | } | |
4561 | ||
4562 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
4563 | ||
566c09c5 SL |
4564 | if (cb->list.next == NULL) { |
4565 | int i; | |
8811b596 | 4566 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
4567 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
4568 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
4569 | } | |
8811b596 SL |
4570 | |
4571 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
4572 | list_add_tail(&sh->lru, &cb->list); | |
4573 | else | |
4574 | release_stripe(sh); | |
4575 | } | |
4576 | ||
620125f2 SL |
4577 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
4578 | { | |
4579 | struct r5conf *conf = mddev->private; | |
4580 | sector_t logical_sector, last_sector; | |
4581 | struct stripe_head *sh; | |
4582 | int remaining; | |
4583 | int stripe_sectors; | |
4584 | ||
4585 | if (mddev->reshape_position != MaxSector) | |
4586 | /* Skip discard while reshape is happening */ | |
4587 | return; | |
4588 | ||
4f024f37 KO |
4589 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
4590 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
4591 | |
4592 | bi->bi_next = NULL; | |
4593 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
4594 | ||
4595 | stripe_sectors = conf->chunk_sectors * | |
4596 | (conf->raid_disks - conf->max_degraded); | |
4597 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
4598 | stripe_sectors); | |
4599 | sector_div(last_sector, stripe_sectors); | |
4600 | ||
4601 | logical_sector *= conf->chunk_sectors; | |
4602 | last_sector *= conf->chunk_sectors; | |
4603 | ||
4604 | for (; logical_sector < last_sector; | |
4605 | logical_sector += STRIPE_SECTORS) { | |
4606 | DEFINE_WAIT(w); | |
4607 | int d; | |
4608 | again: | |
4609 | sh = get_active_stripe(conf, logical_sector, 0, 0, 0); | |
4610 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
4611 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
4612 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
4613 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
4614 | release_stripe(sh); | |
4615 | schedule(); | |
4616 | goto again; | |
4617 | } | |
4618 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
4619 | spin_lock_irq(&sh->stripe_lock); |
4620 | for (d = 0; d < conf->raid_disks; d++) { | |
4621 | if (d == sh->pd_idx || d == sh->qd_idx) | |
4622 | continue; | |
4623 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
4624 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
4625 | spin_unlock_irq(&sh->stripe_lock); | |
4626 | release_stripe(sh); | |
4627 | schedule(); | |
4628 | goto again; | |
4629 | } | |
4630 | } | |
f8dfcffd | 4631 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 SL |
4632 | finish_wait(&conf->wait_for_overlap, &w); |
4633 | for (d = 0; d < conf->raid_disks; d++) { | |
4634 | if (d == sh->pd_idx || d == sh->qd_idx) | |
4635 | continue; | |
4636 | sh->dev[d].towrite = bi; | |
4637 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
4638 | raid5_inc_bi_active_stripes(bi); | |
4639 | } | |
4640 | spin_unlock_irq(&sh->stripe_lock); | |
4641 | if (conf->mddev->bitmap) { | |
4642 | for (d = 0; | |
4643 | d < conf->raid_disks - conf->max_degraded; | |
4644 | d++) | |
4645 | bitmap_startwrite(mddev->bitmap, | |
4646 | sh->sector, | |
4647 | STRIPE_SECTORS, | |
4648 | 0); | |
4649 | sh->bm_seq = conf->seq_flush + 1; | |
4650 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
4651 | } | |
4652 | ||
4653 | set_bit(STRIPE_HANDLE, &sh->state); | |
4654 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4655 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4656 | atomic_inc(&conf->preread_active_stripes); | |
4657 | release_stripe_plug(mddev, sh); | |
4658 | } | |
4659 | ||
4660 | remaining = raid5_dec_bi_active_stripes(bi); | |
4661 | if (remaining == 0) { | |
4662 | md_write_end(mddev); | |
4663 | bio_endio(bi, 0); | |
4664 | } | |
4665 | } | |
4666 | ||
b4fdcb02 | 4667 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 4668 | { |
d1688a6d | 4669 | struct r5conf *conf = mddev->private; |
911d4ee8 | 4670 | int dd_idx; |
1da177e4 LT |
4671 | sector_t new_sector; |
4672 | sector_t logical_sector, last_sector; | |
4673 | struct stripe_head *sh; | |
a362357b | 4674 | const int rw = bio_data_dir(bi); |
49077326 | 4675 | int remaining; |
27c0f68f SL |
4676 | DEFINE_WAIT(w); |
4677 | bool do_prepare; | |
1da177e4 | 4678 | |
e9c7469b TH |
4679 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
4680 | md_flush_request(mddev, bi); | |
5a7bbad2 | 4681 | return; |
e5dcdd80 N |
4682 | } |
4683 | ||
3d310eb7 | 4684 | md_write_start(mddev, bi); |
06d91a5f | 4685 | |
802ba064 | 4686 | if (rw == READ && |
52488615 | 4687 | mddev->reshape_position == MaxSector && |
21a52c6d | 4688 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 4689 | return; |
52488615 | 4690 | |
620125f2 SL |
4691 | if (unlikely(bi->bi_rw & REQ_DISCARD)) { |
4692 | make_discard_request(mddev, bi); | |
4693 | return; | |
4694 | } | |
4695 | ||
4f024f37 | 4696 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 4697 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
4698 | bi->bi_next = NULL; |
4699 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 4700 | |
27c0f68f | 4701 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4702 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 4703 | int previous; |
c46501b2 | 4704 | int seq; |
b578d55f | 4705 | |
27c0f68f | 4706 | do_prepare = false; |
7ecaa1e6 | 4707 | retry: |
c46501b2 | 4708 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 4709 | previous = 0; |
27c0f68f SL |
4710 | if (do_prepare) |
4711 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
4712 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 4713 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 4714 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
4715 | * 64bit on a 32bit platform, and so it might be |
4716 | * possible to see a half-updated value | |
aeb878b0 | 4717 | * Of course reshape_progress could change after |
df8e7f76 N |
4718 | * the lock is dropped, so once we get a reference |
4719 | * to the stripe that we think it is, we will have | |
4720 | * to check again. | |
4721 | */ | |
7ecaa1e6 | 4722 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 4723 | if (mddev->reshape_backwards |
fef9c61f N |
4724 | ? logical_sector < conf->reshape_progress |
4725 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
4726 | previous = 1; |
4727 | } else { | |
2c810cdd | 4728 | if (mddev->reshape_backwards |
fef9c61f N |
4729 | ? logical_sector < conf->reshape_safe |
4730 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
4731 | spin_unlock_irq(&conf->device_lock); |
4732 | schedule(); | |
27c0f68f | 4733 | do_prepare = true; |
b578d55f N |
4734 | goto retry; |
4735 | } | |
4736 | } | |
7ecaa1e6 N |
4737 | spin_unlock_irq(&conf->device_lock); |
4738 | } | |
16a53ecc | 4739 | |
112bf897 N |
4740 | new_sector = raid5_compute_sector(conf, logical_sector, |
4741 | previous, | |
911d4ee8 | 4742 | &dd_idx, NULL); |
0c55e022 | 4743 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
c46501b2 | 4744 | (unsigned long long)new_sector, |
1da177e4 LT |
4745 | (unsigned long long)logical_sector); |
4746 | ||
b5663ba4 | 4747 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 4748 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 4749 | if (sh) { |
b0f9ec04 | 4750 | if (unlikely(previous)) { |
7ecaa1e6 | 4751 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
4752 | * stripe, so we must do the range check again. |
4753 | * Expansion could still move past after this | |
4754 | * test, but as we are holding a reference to | |
4755 | * 'sh', we know that if that happens, | |
4756 | * STRIPE_EXPANDING will get set and the expansion | |
4757 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
4758 | */ |
4759 | int must_retry = 0; | |
4760 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 4761 | if (mddev->reshape_backwards |
b0f9ec04 N |
4762 | ? logical_sector >= conf->reshape_progress |
4763 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
4764 | /* mismatch, need to try again */ |
4765 | must_retry = 1; | |
4766 | spin_unlock_irq(&conf->device_lock); | |
4767 | if (must_retry) { | |
4768 | release_stripe(sh); | |
7a3ab908 | 4769 | schedule(); |
27c0f68f | 4770 | do_prepare = true; |
7ecaa1e6 N |
4771 | goto retry; |
4772 | } | |
4773 | } | |
c46501b2 N |
4774 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
4775 | /* Might have got the wrong stripe_head | |
4776 | * by accident | |
4777 | */ | |
4778 | release_stripe(sh); | |
4779 | goto retry; | |
4780 | } | |
e62e58a5 | 4781 | |
ffd96e35 | 4782 | if (rw == WRITE && |
a5c308d4 | 4783 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
4784 | logical_sector < mddev->suspend_hi) { |
4785 | release_stripe(sh); | |
e62e58a5 N |
4786 | /* As the suspend_* range is controlled by |
4787 | * userspace, we want an interruptible | |
4788 | * wait. | |
4789 | */ | |
4790 | flush_signals(current); | |
4791 | prepare_to_wait(&conf->wait_for_overlap, | |
4792 | &w, TASK_INTERRUPTIBLE); | |
4793 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 4794 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 4795 | schedule(); |
27c0f68f SL |
4796 | do_prepare = true; |
4797 | } | |
e464eafd N |
4798 | goto retry; |
4799 | } | |
7ecaa1e6 N |
4800 | |
4801 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 4802 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
4803 | /* Stripe is busy expanding or |
4804 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
4805 | * and wait a while |
4806 | */ | |
482c0834 | 4807 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
4808 | release_stripe(sh); |
4809 | schedule(); | |
27c0f68f | 4810 | do_prepare = true; |
1da177e4 LT |
4811 | goto retry; |
4812 | } | |
6ed3003c N |
4813 | set_bit(STRIPE_HANDLE, &sh->state); |
4814 | clear_bit(STRIPE_DELAYED, &sh->state); | |
a852d7b8 | 4815 | if ((bi->bi_rw & REQ_SYNC) && |
729a1866 N |
4816 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4817 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 4818 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
4819 | } else { |
4820 | /* cannot get stripe for read-ahead, just give-up */ | |
4821 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1da177e4 LT |
4822 | break; |
4823 | } | |
1da177e4 | 4824 | } |
27c0f68f | 4825 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 4826 | |
e7836bd6 | 4827 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 4828 | if (remaining == 0) { |
1da177e4 | 4829 | |
16a53ecc | 4830 | if ( rw == WRITE ) |
1da177e4 | 4831 | md_write_end(mddev); |
6712ecf8 | 4832 | |
0a82a8d1 LT |
4833 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
4834 | bi, 0); | |
0e13fe23 | 4835 | bio_endio(bi, 0); |
1da177e4 | 4836 | } |
1da177e4 LT |
4837 | } |
4838 | ||
fd01b88c | 4839 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 4840 | |
fd01b88c | 4841 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 4842 | { |
52c03291 N |
4843 | /* reshaping is quite different to recovery/resync so it is |
4844 | * handled quite separately ... here. | |
4845 | * | |
4846 | * On each call to sync_request, we gather one chunk worth of | |
4847 | * destination stripes and flag them as expanding. | |
4848 | * Then we find all the source stripes and request reads. | |
4849 | * As the reads complete, handle_stripe will copy the data | |
4850 | * into the destination stripe and release that stripe. | |
4851 | */ | |
d1688a6d | 4852 | struct r5conf *conf = mddev->private; |
1da177e4 | 4853 | struct stripe_head *sh; |
ccfcc3c1 | 4854 | sector_t first_sector, last_sector; |
f416885e N |
4855 | int raid_disks = conf->previous_raid_disks; |
4856 | int data_disks = raid_disks - conf->max_degraded; | |
4857 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
4858 | int i; |
4859 | int dd_idx; | |
c8f517c4 | 4860 | sector_t writepos, readpos, safepos; |
ec32a2bd | 4861 | sector_t stripe_addr; |
7a661381 | 4862 | int reshape_sectors; |
ab69ae12 | 4863 | struct list_head stripes; |
52c03291 | 4864 | |
fef9c61f N |
4865 | if (sector_nr == 0) { |
4866 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 4867 | if (mddev->reshape_backwards && |
fef9c61f N |
4868 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
4869 | sector_nr = raid5_size(mddev, 0, 0) | |
4870 | - conf->reshape_progress; | |
2c810cdd | 4871 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
4872 | conf->reshape_progress > 0) |
4873 | sector_nr = conf->reshape_progress; | |
f416885e | 4874 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 4875 | if (sector_nr) { |
8dee7211 N |
4876 | mddev->curr_resync_completed = sector_nr; |
4877 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
4878 | *skipped = 1; |
4879 | return sector_nr; | |
4880 | } | |
52c03291 N |
4881 | } |
4882 | ||
7a661381 N |
4883 | /* We need to process a full chunk at a time. |
4884 | * If old and new chunk sizes differ, we need to process the | |
4885 | * largest of these | |
4886 | */ | |
664e7c41 AN |
4887 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
4888 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 4889 | else |
9d8f0363 | 4890 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 4891 | |
b5254dd5 N |
4892 | /* We update the metadata at least every 10 seconds, or when |
4893 | * the data about to be copied would over-write the source of | |
4894 | * the data at the front of the range. i.e. one new_stripe | |
4895 | * along from reshape_progress new_maps to after where | |
4896 | * reshape_safe old_maps to | |
52c03291 | 4897 | */ |
fef9c61f | 4898 | writepos = conf->reshape_progress; |
f416885e | 4899 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
4900 | readpos = conf->reshape_progress; |
4901 | sector_div(readpos, data_disks); | |
fef9c61f | 4902 | safepos = conf->reshape_safe; |
f416885e | 4903 | sector_div(safepos, data_disks); |
2c810cdd | 4904 | if (mddev->reshape_backwards) { |
ed37d83e | 4905 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 4906 | readpos += reshape_sectors; |
7a661381 | 4907 | safepos += reshape_sectors; |
fef9c61f | 4908 | } else { |
7a661381 | 4909 | writepos += reshape_sectors; |
ed37d83e N |
4910 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
4911 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 4912 | } |
52c03291 | 4913 | |
b5254dd5 N |
4914 | /* Having calculated the 'writepos' possibly use it |
4915 | * to set 'stripe_addr' which is where we will write to. | |
4916 | */ | |
4917 | if (mddev->reshape_backwards) { | |
4918 | BUG_ON(conf->reshape_progress == 0); | |
4919 | stripe_addr = writepos; | |
4920 | BUG_ON((mddev->dev_sectors & | |
4921 | ~((sector_t)reshape_sectors - 1)) | |
4922 | - reshape_sectors - stripe_addr | |
4923 | != sector_nr); | |
4924 | } else { | |
4925 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
4926 | stripe_addr = sector_nr; | |
4927 | } | |
4928 | ||
c8f517c4 N |
4929 | /* 'writepos' is the most advanced device address we might write. |
4930 | * 'readpos' is the least advanced device address we might read. | |
4931 | * 'safepos' is the least address recorded in the metadata as having | |
4932 | * been reshaped. | |
b5254dd5 N |
4933 | * If there is a min_offset_diff, these are adjusted either by |
4934 | * increasing the safepos/readpos if diff is negative, or | |
4935 | * increasing writepos if diff is positive. | |
4936 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
4937 | * ensure safety in the face of a crash - that must be done by userspace |
4938 | * making a backup of the data. So in that case there is no particular | |
4939 | * rush to update metadata. | |
4940 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
4941 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
4942 | * we can be safe in the event of a crash. | |
4943 | * So we insist on updating metadata if safepos is behind writepos and | |
4944 | * readpos is beyond writepos. | |
4945 | * In any case, update the metadata every 10 seconds. | |
4946 | * Maybe that number should be configurable, but I'm not sure it is | |
4947 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
4948 | */ | |
b5254dd5 N |
4949 | if (conf->min_offset_diff < 0) { |
4950 | safepos += -conf->min_offset_diff; | |
4951 | readpos += -conf->min_offset_diff; | |
4952 | } else | |
4953 | writepos += conf->min_offset_diff; | |
4954 | ||
2c810cdd | 4955 | if ((mddev->reshape_backwards |
c8f517c4 N |
4956 | ? (safepos > writepos && readpos < writepos) |
4957 | : (safepos < writepos && readpos > writepos)) || | |
4958 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
4959 | /* Cannot proceed until we've updated the superblock... */ |
4960 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
4961 | atomic_read(&conf->reshape_stripes)==0 |
4962 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
4963 | if (atomic_read(&conf->reshape_stripes) != 0) | |
4964 | return 0; | |
fef9c61f | 4965 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 4966 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 4967 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 4968 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 4969 | md_wakeup_thread(mddev->thread); |
850b2b42 | 4970 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
c91abf5a N |
4971 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4972 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4973 | return 0; | |
52c03291 | 4974 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 4975 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
4976 | spin_unlock_irq(&conf->device_lock); |
4977 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 4978 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
4979 | } |
4980 | ||
ab69ae12 | 4981 | INIT_LIST_HEAD(&stripes); |
7a661381 | 4982 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 4983 | int j; |
a9f326eb | 4984 | int skipped_disk = 0; |
a8c906ca | 4985 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
4986 | set_bit(STRIPE_EXPANDING, &sh->state); |
4987 | atomic_inc(&conf->reshape_stripes); | |
4988 | /* If any of this stripe is beyond the end of the old | |
4989 | * array, then we need to zero those blocks | |
4990 | */ | |
4991 | for (j=sh->disks; j--;) { | |
4992 | sector_t s; | |
4993 | if (j == sh->pd_idx) | |
4994 | continue; | |
f416885e | 4995 | if (conf->level == 6 && |
d0dabf7e | 4996 | j == sh->qd_idx) |
f416885e | 4997 | continue; |
784052ec | 4998 | s = compute_blocknr(sh, j, 0); |
b522adcd | 4999 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5000 | skipped_disk = 1; |
52c03291 N |
5001 | continue; |
5002 | } | |
5003 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5004 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5005 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5006 | } | |
a9f326eb | 5007 | if (!skipped_disk) { |
52c03291 N |
5008 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5009 | set_bit(STRIPE_HANDLE, &sh->state); | |
5010 | } | |
ab69ae12 | 5011 | list_add(&sh->lru, &stripes); |
52c03291 N |
5012 | } |
5013 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5014 | if (mddev->reshape_backwards) |
7a661381 | 5015 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5016 | else |
7a661381 | 5017 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5018 | spin_unlock_irq(&conf->device_lock); |
5019 | /* Ok, those stripe are ready. We can start scheduling | |
5020 | * reads on the source stripes. | |
5021 | * The source stripes are determined by mapping the first and last | |
5022 | * block on the destination stripes. | |
5023 | */ | |
52c03291 | 5024 | first_sector = |
ec32a2bd | 5025 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5026 | 1, &dd_idx, NULL); |
52c03291 | 5027 | last_sector = |
0e6e0271 | 5028 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5029 | * new_data_disks - 1), |
911d4ee8 | 5030 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5031 | if (last_sector >= mddev->dev_sectors) |
5032 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5033 | while (first_sector <= last_sector) { |
a8c906ca | 5034 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5035 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5036 | set_bit(STRIPE_HANDLE, &sh->state); | |
5037 | release_stripe(sh); | |
5038 | first_sector += STRIPE_SECTORS; | |
5039 | } | |
ab69ae12 N |
5040 | /* Now that the sources are clearly marked, we can release |
5041 | * the destination stripes | |
5042 | */ | |
5043 | while (!list_empty(&stripes)) { | |
5044 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5045 | list_del_init(&sh->lru); | |
5046 | release_stripe(sh); | |
5047 | } | |
c6207277 N |
5048 | /* If this takes us to the resync_max point where we have to pause, |
5049 | * then we need to write out the superblock. | |
5050 | */ | |
7a661381 | 5051 | sector_nr += reshape_sectors; |
c03f6a19 N |
5052 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
5053 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
5054 | /* Cannot proceed until we've updated the superblock... */ |
5055 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5056 | atomic_read(&conf->reshape_stripes) == 0 |
5057 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5058 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5059 | goto ret; | |
fef9c61f | 5060 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5061 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5062 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
5063 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
5064 | md_wakeup_thread(mddev->thread); | |
5065 | wait_event(mddev->sb_wait, | |
5066 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
c91abf5a N |
5067 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5068 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5069 | goto ret; | |
c6207277 | 5070 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5071 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5072 | spin_unlock_irq(&conf->device_lock); |
5073 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5074 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5075 | } |
c91abf5a | 5076 | ret: |
7a661381 | 5077 | return reshape_sectors; |
52c03291 N |
5078 | } |
5079 | ||
09314799 | 5080 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
52c03291 | 5081 | { |
d1688a6d | 5082 | struct r5conf *conf = mddev->private; |
52c03291 | 5083 | struct stripe_head *sh; |
58c0fed4 | 5084 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5085 | sector_t sync_blocks; |
16a53ecc N |
5086 | int still_degraded = 0; |
5087 | int i; | |
1da177e4 | 5088 | |
72626685 | 5089 | if (sector_nr >= max_sector) { |
1da177e4 | 5090 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5091 | |
29269553 N |
5092 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5093 | end_reshape(conf); | |
5094 | return 0; | |
5095 | } | |
72626685 N |
5096 | |
5097 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5098 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5099 | &sync_blocks, 1); | |
16a53ecc | 5100 | else /* completed sync */ |
72626685 N |
5101 | conf->fullsync = 0; |
5102 | bitmap_close_sync(mddev->bitmap); | |
5103 | ||
1da177e4 LT |
5104 | return 0; |
5105 | } | |
ccfcc3c1 | 5106 | |
64bd660b N |
5107 | /* Allow raid5_quiesce to complete */ |
5108 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5109 | ||
52c03291 N |
5110 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5111 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5112 | |
c6207277 N |
5113 | /* No need to check resync_max as we never do more than one |
5114 | * stripe, and as resync_max will always be on a chunk boundary, | |
5115 | * if the check in md_do_sync didn't fire, there is no chance | |
5116 | * of overstepping resync_max here | |
5117 | */ | |
5118 | ||
16a53ecc | 5119 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5120 | * to resync, then assert that we are finished, because there is |
5121 | * nothing we can do. | |
5122 | */ | |
3285edf1 | 5123 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5124 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5125 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5126 | *skipped = 1; |
1da177e4 LT |
5127 | return rv; |
5128 | } | |
6f608040 | 5129 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5130 | !conf->fullsync && | |
5131 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5132 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5133 | /* we can skip this block, and probably more */ |
5134 | sync_blocks /= STRIPE_SECTORS; | |
5135 | *skipped = 1; | |
5136 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5137 | } | |
1da177e4 | 5138 | |
b47490c9 N |
5139 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
5140 | ||
a8c906ca | 5141 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5142 | if (sh == NULL) { |
a8c906ca | 5143 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5144 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5145 | * is trying to get access |
1da177e4 | 5146 | */ |
66c006a5 | 5147 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5148 | } |
16a53ecc | 5149 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5150 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5151 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5152 | */ |
16d9cfab EM |
5153 | rcu_read_lock(); |
5154 | for (i = 0; i < conf->raid_disks; i++) { | |
5155 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5156 | ||
5157 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5158 | still_degraded = 1; |
16d9cfab EM |
5159 | } |
5160 | rcu_read_unlock(); | |
16a53ecc N |
5161 | |
5162 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5163 | ||
83206d66 | 5164 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5165 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5166 | |
1da177e4 LT |
5167 | release_stripe(sh); |
5168 | ||
5169 | return STRIPE_SECTORS; | |
5170 | } | |
5171 | ||
d1688a6d | 5172 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5173 | { |
5174 | /* We may not be able to submit a whole bio at once as there | |
5175 | * may not be enough stripe_heads available. | |
5176 | * We cannot pre-allocate enough stripe_heads as we may need | |
5177 | * more than exist in the cache (if we allow ever large chunks). | |
5178 | * So we do one stripe head at a time and record in | |
5179 | * ->bi_hw_segments how many have been done. | |
5180 | * | |
5181 | * We *know* that this entire raid_bio is in one chunk, so | |
5182 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5183 | */ | |
5184 | struct stripe_head *sh; | |
911d4ee8 | 5185 | int dd_idx; |
46031f9a RBJ |
5186 | sector_t sector, logical_sector, last_sector; |
5187 | int scnt = 0; | |
5188 | int remaining; | |
5189 | int handled = 0; | |
5190 | ||
4f024f37 KO |
5191 | logical_sector = raid_bio->bi_iter.bi_sector & |
5192 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5193 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5194 | 0, &dd_idx, NULL); |
f73a1c7d | 5195 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5196 | |
5197 | for (; logical_sector < last_sector; | |
387bb173 NB |
5198 | logical_sector += STRIPE_SECTORS, |
5199 | sector += STRIPE_SECTORS, | |
5200 | scnt++) { | |
46031f9a | 5201 | |
e7836bd6 | 5202 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5203 | /* already done this stripe */ |
5204 | continue; | |
5205 | ||
2844dc32 | 5206 | sh = get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5207 | |
5208 | if (!sh) { | |
5209 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5210 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5211 | conf->retry_read_aligned = raid_bio; |
5212 | return handled; | |
5213 | } | |
5214 | ||
da41ba65 | 5215 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
387bb173 | 5216 | release_stripe(sh); |
e7836bd6 | 5217 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
5218 | conf->retry_read_aligned = raid_bio; |
5219 | return handled; | |
5220 | } | |
5221 | ||
3f9e7c14 | 5222 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 5223 | handle_stripe(sh); |
46031f9a RBJ |
5224 | release_stripe(sh); |
5225 | handled++; | |
5226 | } | |
e7836bd6 | 5227 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
5228 | if (remaining == 0) { |
5229 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
5230 | raid_bio, 0); | |
0e13fe23 | 5231 | bio_endio(raid_bio, 0); |
0a82a8d1 | 5232 | } |
46031f9a RBJ |
5233 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
5234 | wake_up(&conf->wait_for_stripe); | |
5235 | return handled; | |
5236 | } | |
5237 | ||
bfc90cb0 | 5238 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
5239 | struct r5worker *worker, |
5240 | struct list_head *temp_inactive_list) | |
46a06401 SL |
5241 | { |
5242 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
5243 | int i, batch_size = 0, hash; |
5244 | bool release_inactive = false; | |
46a06401 SL |
5245 | |
5246 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 5247 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
5248 | batch[batch_size++] = sh; |
5249 | ||
566c09c5 SL |
5250 | if (batch_size == 0) { |
5251 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5252 | if (!list_empty(temp_inactive_list + i)) | |
5253 | break; | |
5254 | if (i == NR_STRIPE_HASH_LOCKS) | |
5255 | return batch_size; | |
5256 | release_inactive = true; | |
5257 | } | |
46a06401 SL |
5258 | spin_unlock_irq(&conf->device_lock); |
5259 | ||
566c09c5 SL |
5260 | release_inactive_stripe_list(conf, temp_inactive_list, |
5261 | NR_STRIPE_HASH_LOCKS); | |
5262 | ||
5263 | if (release_inactive) { | |
5264 | spin_lock_irq(&conf->device_lock); | |
5265 | return 0; | |
5266 | } | |
5267 | ||
46a06401 SL |
5268 | for (i = 0; i < batch_size; i++) |
5269 | handle_stripe(batch[i]); | |
5270 | ||
5271 | cond_resched(); | |
5272 | ||
5273 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
5274 | for (i = 0; i < batch_size; i++) { |
5275 | hash = batch[i]->hash_lock_index; | |
5276 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
5277 | } | |
46a06401 SL |
5278 | return batch_size; |
5279 | } | |
46031f9a | 5280 | |
851c30c9 SL |
5281 | static void raid5_do_work(struct work_struct *work) |
5282 | { | |
5283 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
5284 | struct r5worker_group *group = worker->group; | |
5285 | struct r5conf *conf = group->conf; | |
5286 | int group_id = group - conf->worker_groups; | |
5287 | int handled; | |
5288 | struct blk_plug plug; | |
5289 | ||
5290 | pr_debug("+++ raid5worker active\n"); | |
5291 | ||
5292 | blk_start_plug(&plug); | |
5293 | handled = 0; | |
5294 | spin_lock_irq(&conf->device_lock); | |
5295 | while (1) { | |
5296 | int batch_size, released; | |
5297 | ||
566c09c5 | 5298 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 5299 | |
566c09c5 SL |
5300 | batch_size = handle_active_stripes(conf, group_id, worker, |
5301 | worker->temp_inactive_list); | |
bfc90cb0 | 5302 | worker->working = false; |
851c30c9 SL |
5303 | if (!batch_size && !released) |
5304 | break; | |
5305 | handled += batch_size; | |
5306 | } | |
5307 | pr_debug("%d stripes handled\n", handled); | |
5308 | ||
5309 | spin_unlock_irq(&conf->device_lock); | |
5310 | blk_finish_plug(&plug); | |
5311 | ||
5312 | pr_debug("--- raid5worker inactive\n"); | |
5313 | } | |
5314 | ||
1da177e4 LT |
5315 | /* |
5316 | * This is our raid5 kernel thread. | |
5317 | * | |
5318 | * We scan the hash table for stripes which can be handled now. | |
5319 | * During the scan, completed stripes are saved for us by the interrupt | |
5320 | * handler, so that they will not have to wait for our next wakeup. | |
5321 | */ | |
4ed8731d | 5322 | static void raid5d(struct md_thread *thread) |
1da177e4 | 5323 | { |
4ed8731d | 5324 | struct mddev *mddev = thread->mddev; |
d1688a6d | 5325 | struct r5conf *conf = mddev->private; |
1da177e4 | 5326 | int handled; |
e1dfa0a2 | 5327 | struct blk_plug plug; |
1da177e4 | 5328 | |
45b4233c | 5329 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
5330 | |
5331 | md_check_recovery(mddev); | |
1da177e4 | 5332 | |
e1dfa0a2 | 5333 | blk_start_plug(&plug); |
1da177e4 LT |
5334 | handled = 0; |
5335 | spin_lock_irq(&conf->device_lock); | |
5336 | while (1) { | |
46031f9a | 5337 | struct bio *bio; |
773ca82f SL |
5338 | int batch_size, released; |
5339 | ||
566c09c5 | 5340 | released = release_stripe_list(conf, conf->temp_inactive_list); |
1da177e4 | 5341 | |
0021b7bc | 5342 | if ( |
7c13edc8 N |
5343 | !list_empty(&conf->bitmap_list)) { |
5344 | /* Now is a good time to flush some bitmap updates */ | |
5345 | conf->seq_flush++; | |
700e432d | 5346 | spin_unlock_irq(&conf->device_lock); |
72626685 | 5347 | bitmap_unplug(mddev->bitmap); |
700e432d | 5348 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 5349 | conf->seq_write = conf->seq_flush; |
566c09c5 | 5350 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 5351 | } |
0021b7bc | 5352 | raid5_activate_delayed(conf); |
72626685 | 5353 | |
46031f9a RBJ |
5354 | while ((bio = remove_bio_from_retry(conf))) { |
5355 | int ok; | |
5356 | spin_unlock_irq(&conf->device_lock); | |
5357 | ok = retry_aligned_read(conf, bio); | |
5358 | spin_lock_irq(&conf->device_lock); | |
5359 | if (!ok) | |
5360 | break; | |
5361 | handled++; | |
5362 | } | |
5363 | ||
566c09c5 SL |
5364 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
5365 | conf->temp_inactive_list); | |
773ca82f | 5366 | if (!batch_size && !released) |
1da177e4 | 5367 | break; |
46a06401 | 5368 | handled += batch_size; |
1da177e4 | 5369 | |
46a06401 SL |
5370 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
5371 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 5372 | md_check_recovery(mddev); |
46a06401 SL |
5373 | spin_lock_irq(&conf->device_lock); |
5374 | } | |
1da177e4 | 5375 | } |
45b4233c | 5376 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
5377 | |
5378 | spin_unlock_irq(&conf->device_lock); | |
5379 | ||
c9f21aaf | 5380 | async_tx_issue_pending_all(); |
e1dfa0a2 | 5381 | blk_finish_plug(&plug); |
1da177e4 | 5382 | |
45b4233c | 5383 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
5384 | } |
5385 | ||
3f294f4f | 5386 | static ssize_t |
fd01b88c | 5387 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 5388 | { |
7b1485ba N |
5389 | struct r5conf *conf; |
5390 | int ret = 0; | |
5391 | spin_lock(&mddev->lock); | |
5392 | conf = mddev->private; | |
96de1e66 | 5393 | if (conf) |
7b1485ba N |
5394 | ret = sprintf(page, "%d\n", conf->max_nr_stripes); |
5395 | spin_unlock(&mddev->lock); | |
5396 | return ret; | |
3f294f4f N |
5397 | } |
5398 | ||
c41d4ac4 | 5399 | int |
fd01b88c | 5400 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 5401 | { |
d1688a6d | 5402 | struct r5conf *conf = mddev->private; |
b5470dc5 | 5403 | int err; |
566c09c5 | 5404 | int hash; |
b5470dc5 | 5405 | |
c41d4ac4 | 5406 | if (size <= 16 || size > 32768) |
3f294f4f | 5407 | return -EINVAL; |
566c09c5 | 5408 | hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS; |
c41d4ac4 | 5409 | while (size < conf->max_nr_stripes) { |
566c09c5 | 5410 | if (drop_one_stripe(conf, hash)) |
3f294f4f N |
5411 | conf->max_nr_stripes--; |
5412 | else | |
5413 | break; | |
566c09c5 SL |
5414 | hash--; |
5415 | if (hash < 0) | |
5416 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
3f294f4f | 5417 | } |
b5470dc5 DW |
5418 | err = md_allow_write(mddev); |
5419 | if (err) | |
5420 | return err; | |
566c09c5 | 5421 | hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; |
c41d4ac4 | 5422 | while (size > conf->max_nr_stripes) { |
566c09c5 | 5423 | if (grow_one_stripe(conf, hash)) |
3f294f4f N |
5424 | conf->max_nr_stripes++; |
5425 | else break; | |
566c09c5 | 5426 | hash = (hash + 1) % NR_STRIPE_HASH_LOCKS; |
3f294f4f | 5427 | } |
c41d4ac4 N |
5428 | return 0; |
5429 | } | |
5430 | EXPORT_SYMBOL(raid5_set_cache_size); | |
5431 | ||
5432 | static ssize_t | |
fd01b88c | 5433 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 5434 | { |
6791875e | 5435 | struct r5conf *conf; |
c41d4ac4 N |
5436 | unsigned long new; |
5437 | int err; | |
5438 | ||
5439 | if (len >= PAGE_SIZE) | |
5440 | return -EINVAL; | |
b29bebd6 | 5441 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 5442 | return -EINVAL; |
6791875e | 5443 | err = mddev_lock(mddev); |
c41d4ac4 N |
5444 | if (err) |
5445 | return err; | |
6791875e N |
5446 | conf = mddev->private; |
5447 | if (!conf) | |
5448 | err = -ENODEV; | |
5449 | else | |
5450 | err = raid5_set_cache_size(mddev, new); | |
5451 | mddev_unlock(mddev); | |
5452 | ||
5453 | return err ?: len; | |
3f294f4f | 5454 | } |
007583c9 | 5455 | |
96de1e66 N |
5456 | static struct md_sysfs_entry |
5457 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
5458 | raid5_show_stripe_cache_size, | |
5459 | raid5_store_stripe_cache_size); | |
3f294f4f | 5460 | |
8b3e6cdc | 5461 | static ssize_t |
fd01b88c | 5462 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 5463 | { |
7b1485ba N |
5464 | struct r5conf *conf; |
5465 | int ret = 0; | |
5466 | spin_lock(&mddev->lock); | |
5467 | conf = mddev->private; | |
8b3e6cdc | 5468 | if (conf) |
7b1485ba N |
5469 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
5470 | spin_unlock(&mddev->lock); | |
5471 | return ret; | |
8b3e6cdc DW |
5472 | } |
5473 | ||
5474 | static ssize_t | |
fd01b88c | 5475 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 5476 | { |
6791875e | 5477 | struct r5conf *conf; |
4ef197d8 | 5478 | unsigned long new; |
6791875e N |
5479 | int err; |
5480 | ||
8b3e6cdc DW |
5481 | if (len >= PAGE_SIZE) |
5482 | return -EINVAL; | |
b29bebd6 | 5483 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 5484 | return -EINVAL; |
6791875e N |
5485 | |
5486 | err = mddev_lock(mddev); | |
5487 | if (err) | |
5488 | return err; | |
5489 | conf = mddev->private; | |
5490 | if (!conf) | |
5491 | err = -ENODEV; | |
5492 | else if (new > conf->max_nr_stripes) | |
5493 | err = -EINVAL; | |
5494 | else | |
5495 | conf->bypass_threshold = new; | |
5496 | mddev_unlock(mddev); | |
5497 | return err ?: len; | |
8b3e6cdc DW |
5498 | } |
5499 | ||
5500 | static struct md_sysfs_entry | |
5501 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
5502 | S_IRUGO | S_IWUSR, | |
5503 | raid5_show_preread_threshold, | |
5504 | raid5_store_preread_threshold); | |
5505 | ||
d592a996 SL |
5506 | static ssize_t |
5507 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
5508 | { | |
7b1485ba N |
5509 | struct r5conf *conf; |
5510 | int ret = 0; | |
5511 | spin_lock(&mddev->lock); | |
5512 | conf = mddev->private; | |
d592a996 | 5513 | if (conf) |
7b1485ba N |
5514 | ret = sprintf(page, "%d\n", conf->skip_copy); |
5515 | spin_unlock(&mddev->lock); | |
5516 | return ret; | |
d592a996 SL |
5517 | } |
5518 | ||
5519 | static ssize_t | |
5520 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
5521 | { | |
6791875e | 5522 | struct r5conf *conf; |
d592a996 | 5523 | unsigned long new; |
6791875e N |
5524 | int err; |
5525 | ||
d592a996 SL |
5526 | if (len >= PAGE_SIZE) |
5527 | return -EINVAL; | |
d592a996 SL |
5528 | if (kstrtoul(page, 10, &new)) |
5529 | return -EINVAL; | |
5530 | new = !!new; | |
6791875e N |
5531 | |
5532 | err = mddev_lock(mddev); | |
5533 | if (err) | |
5534 | return err; | |
5535 | conf = mddev->private; | |
5536 | if (!conf) | |
5537 | err = -ENODEV; | |
5538 | else if (new != conf->skip_copy) { | |
5539 | mddev_suspend(mddev); | |
5540 | conf->skip_copy = new; | |
5541 | if (new) | |
5542 | mddev->queue->backing_dev_info.capabilities |= | |
5543 | BDI_CAP_STABLE_WRITES; | |
5544 | else | |
5545 | mddev->queue->backing_dev_info.capabilities &= | |
5546 | ~BDI_CAP_STABLE_WRITES; | |
5547 | mddev_resume(mddev); | |
5548 | } | |
5549 | mddev_unlock(mddev); | |
5550 | return err ?: len; | |
d592a996 SL |
5551 | } |
5552 | ||
5553 | static struct md_sysfs_entry | |
5554 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
5555 | raid5_show_skip_copy, | |
5556 | raid5_store_skip_copy); | |
5557 | ||
3f294f4f | 5558 | static ssize_t |
fd01b88c | 5559 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 5560 | { |
d1688a6d | 5561 | struct r5conf *conf = mddev->private; |
96de1e66 N |
5562 | if (conf) |
5563 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
5564 | else | |
5565 | return 0; | |
3f294f4f N |
5566 | } |
5567 | ||
96de1e66 N |
5568 | static struct md_sysfs_entry |
5569 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 5570 | |
b721420e SL |
5571 | static ssize_t |
5572 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
5573 | { | |
7b1485ba N |
5574 | struct r5conf *conf; |
5575 | int ret = 0; | |
5576 | spin_lock(&mddev->lock); | |
5577 | conf = mddev->private; | |
b721420e | 5578 | if (conf) |
7b1485ba N |
5579 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
5580 | spin_unlock(&mddev->lock); | |
5581 | return ret; | |
b721420e SL |
5582 | } |
5583 | ||
60aaf933 | 5584 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
5585 | int *group_cnt, | |
5586 | int *worker_cnt_per_group, | |
5587 | struct r5worker_group **worker_groups); | |
b721420e SL |
5588 | static ssize_t |
5589 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
5590 | { | |
6791875e | 5591 | struct r5conf *conf; |
b721420e SL |
5592 | unsigned long new; |
5593 | int err; | |
60aaf933 | 5594 | struct r5worker_group *new_groups, *old_groups; |
5595 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
5596 | |
5597 | if (len >= PAGE_SIZE) | |
5598 | return -EINVAL; | |
b721420e SL |
5599 | if (kstrtoul(page, 10, &new)) |
5600 | return -EINVAL; | |
5601 | ||
6791875e N |
5602 | err = mddev_lock(mddev); |
5603 | if (err) | |
5604 | return err; | |
5605 | conf = mddev->private; | |
5606 | if (!conf) | |
5607 | err = -ENODEV; | |
5608 | else if (new != conf->worker_cnt_per_group) { | |
5609 | mddev_suspend(mddev); | |
b721420e | 5610 | |
6791875e N |
5611 | old_groups = conf->worker_groups; |
5612 | if (old_groups) | |
5613 | flush_workqueue(raid5_wq); | |
d206dcfa | 5614 | |
6791875e N |
5615 | err = alloc_thread_groups(conf, new, |
5616 | &group_cnt, &worker_cnt_per_group, | |
5617 | &new_groups); | |
5618 | if (!err) { | |
5619 | spin_lock_irq(&conf->device_lock); | |
5620 | conf->group_cnt = group_cnt; | |
5621 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
5622 | conf->worker_groups = new_groups; | |
5623 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 5624 | |
6791875e N |
5625 | if (old_groups) |
5626 | kfree(old_groups[0].workers); | |
5627 | kfree(old_groups); | |
5628 | } | |
5629 | mddev_resume(mddev); | |
b721420e | 5630 | } |
6791875e | 5631 | mddev_unlock(mddev); |
b721420e | 5632 | |
6791875e | 5633 | return err ?: len; |
b721420e SL |
5634 | } |
5635 | ||
5636 | static struct md_sysfs_entry | |
5637 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
5638 | raid5_show_group_thread_cnt, | |
5639 | raid5_store_group_thread_cnt); | |
5640 | ||
007583c9 | 5641 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
5642 | &raid5_stripecache_size.attr, |
5643 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 5644 | &raid5_preread_bypass_threshold.attr, |
b721420e | 5645 | &raid5_group_thread_cnt.attr, |
d592a996 | 5646 | &raid5_skip_copy.attr, |
3f294f4f N |
5647 | NULL, |
5648 | }; | |
007583c9 N |
5649 | static struct attribute_group raid5_attrs_group = { |
5650 | .name = NULL, | |
5651 | .attrs = raid5_attrs, | |
3f294f4f N |
5652 | }; |
5653 | ||
60aaf933 | 5654 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
5655 | int *group_cnt, | |
5656 | int *worker_cnt_per_group, | |
5657 | struct r5worker_group **worker_groups) | |
851c30c9 | 5658 | { |
566c09c5 | 5659 | int i, j, k; |
851c30c9 SL |
5660 | ssize_t size; |
5661 | struct r5worker *workers; | |
5662 | ||
60aaf933 | 5663 | *worker_cnt_per_group = cnt; |
851c30c9 | 5664 | if (cnt == 0) { |
60aaf933 | 5665 | *group_cnt = 0; |
5666 | *worker_groups = NULL; | |
851c30c9 SL |
5667 | return 0; |
5668 | } | |
60aaf933 | 5669 | *group_cnt = num_possible_nodes(); |
851c30c9 | 5670 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 5671 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
5672 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
5673 | *group_cnt, GFP_NOIO); | |
5674 | if (!*worker_groups || !workers) { | |
851c30c9 | 5675 | kfree(workers); |
60aaf933 | 5676 | kfree(*worker_groups); |
851c30c9 SL |
5677 | return -ENOMEM; |
5678 | } | |
5679 | ||
60aaf933 | 5680 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
5681 | struct r5worker_group *group; |
5682 | ||
0c775d52 | 5683 | group = &(*worker_groups)[i]; |
851c30c9 SL |
5684 | INIT_LIST_HEAD(&group->handle_list); |
5685 | group->conf = conf; | |
5686 | group->workers = workers + i * cnt; | |
5687 | ||
5688 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
5689 | struct r5worker *worker = group->workers + j; |
5690 | worker->group = group; | |
5691 | INIT_WORK(&worker->work, raid5_do_work); | |
5692 | ||
5693 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
5694 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
5695 | } |
5696 | } | |
5697 | ||
5698 | return 0; | |
5699 | } | |
5700 | ||
5701 | static void free_thread_groups(struct r5conf *conf) | |
5702 | { | |
5703 | if (conf->worker_groups) | |
5704 | kfree(conf->worker_groups[0].workers); | |
5705 | kfree(conf->worker_groups); | |
5706 | conf->worker_groups = NULL; | |
5707 | } | |
5708 | ||
80c3a6ce | 5709 | static sector_t |
fd01b88c | 5710 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 5711 | { |
d1688a6d | 5712 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
5713 | |
5714 | if (!sectors) | |
5715 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 5716 | if (!raid_disks) |
7ec05478 | 5717 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 5718 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 5719 | |
9d8f0363 | 5720 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 5721 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
5722 | return sectors * (raid_disks - conf->max_degraded); |
5723 | } | |
5724 | ||
789b5e03 ON |
5725 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
5726 | { | |
5727 | safe_put_page(percpu->spare_page); | |
46d5b785 | 5728 | if (percpu->scribble) |
5729 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
5730 | percpu->spare_page = NULL; |
5731 | percpu->scribble = NULL; | |
5732 | } | |
5733 | ||
5734 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
5735 | { | |
5736 | if (conf->level == 6 && !percpu->spare_page) | |
5737 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
5738 | if (!percpu->scribble) | |
46d5b785 | 5739 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
5740 | conf->previous_raid_disks), conf->chunk_sectors / | |
5741 | STRIPE_SECTORS, GFP_KERNEL); | |
789b5e03 ON |
5742 | |
5743 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
5744 | free_scratch_buffer(conf, percpu); | |
5745 | return -ENOMEM; | |
5746 | } | |
5747 | ||
5748 | return 0; | |
5749 | } | |
5750 | ||
d1688a6d | 5751 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 | 5752 | { |
36d1c647 DW |
5753 | unsigned long cpu; |
5754 | ||
5755 | if (!conf->percpu) | |
5756 | return; | |
5757 | ||
36d1c647 DW |
5758 | #ifdef CONFIG_HOTPLUG_CPU |
5759 | unregister_cpu_notifier(&conf->cpu_notify); | |
5760 | #endif | |
789b5e03 ON |
5761 | |
5762 | get_online_cpus(); | |
5763 | for_each_possible_cpu(cpu) | |
5764 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
36d1c647 DW |
5765 | put_online_cpus(); |
5766 | ||
5767 | free_percpu(conf->percpu); | |
5768 | } | |
5769 | ||
d1688a6d | 5770 | static void free_conf(struct r5conf *conf) |
95fc17aa | 5771 | { |
851c30c9 | 5772 | free_thread_groups(conf); |
95fc17aa | 5773 | shrink_stripes(conf); |
36d1c647 | 5774 | raid5_free_percpu(conf); |
95fc17aa DW |
5775 | kfree(conf->disks); |
5776 | kfree(conf->stripe_hashtbl); | |
5777 | kfree(conf); | |
5778 | } | |
5779 | ||
36d1c647 DW |
5780 | #ifdef CONFIG_HOTPLUG_CPU |
5781 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
5782 | void *hcpu) | |
5783 | { | |
d1688a6d | 5784 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
5785 | long cpu = (long)hcpu; |
5786 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
5787 | ||
5788 | switch (action) { | |
5789 | case CPU_UP_PREPARE: | |
5790 | case CPU_UP_PREPARE_FROZEN: | |
789b5e03 | 5791 | if (alloc_scratch_buffer(conf, percpu)) { |
36d1c647 DW |
5792 | pr_err("%s: failed memory allocation for cpu%ld\n", |
5793 | __func__, cpu); | |
55af6bb5 | 5794 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
5795 | } |
5796 | break; | |
5797 | case CPU_DEAD: | |
5798 | case CPU_DEAD_FROZEN: | |
789b5e03 | 5799 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
36d1c647 DW |
5800 | break; |
5801 | default: | |
5802 | break; | |
5803 | } | |
5804 | return NOTIFY_OK; | |
5805 | } | |
5806 | #endif | |
5807 | ||
d1688a6d | 5808 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
5809 | { |
5810 | unsigned long cpu; | |
789b5e03 | 5811 | int err = 0; |
36d1c647 | 5812 | |
789b5e03 ON |
5813 | conf->percpu = alloc_percpu(struct raid5_percpu); |
5814 | if (!conf->percpu) | |
36d1c647 | 5815 | return -ENOMEM; |
789b5e03 ON |
5816 | |
5817 | #ifdef CONFIG_HOTPLUG_CPU | |
5818 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
5819 | conf->cpu_notify.priority = 0; | |
5820 | err = register_cpu_notifier(&conf->cpu_notify); | |
5821 | if (err) | |
5822 | return err; | |
5823 | #endif | |
36d1c647 DW |
5824 | |
5825 | get_online_cpus(); | |
36d1c647 | 5826 | for_each_present_cpu(cpu) { |
789b5e03 ON |
5827 | err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
5828 | if (err) { | |
5829 | pr_err("%s: failed memory allocation for cpu%ld\n", | |
5830 | __func__, cpu); | |
36d1c647 DW |
5831 | break; |
5832 | } | |
36d1c647 | 5833 | } |
36d1c647 DW |
5834 | put_online_cpus(); |
5835 | ||
5836 | return err; | |
5837 | } | |
5838 | ||
d1688a6d | 5839 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 5840 | { |
d1688a6d | 5841 | struct r5conf *conf; |
5e5e3e78 | 5842 | int raid_disk, memory, max_disks; |
3cb03002 | 5843 | struct md_rdev *rdev; |
1da177e4 | 5844 | struct disk_info *disk; |
0232605d | 5845 | char pers_name[6]; |
566c09c5 | 5846 | int i; |
60aaf933 | 5847 | int group_cnt, worker_cnt_per_group; |
5848 | struct r5worker_group *new_group; | |
1da177e4 | 5849 | |
91adb564 N |
5850 | if (mddev->new_level != 5 |
5851 | && mddev->new_level != 4 | |
5852 | && mddev->new_level != 6) { | |
0c55e022 | 5853 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
5854 | mdname(mddev), mddev->new_level); |
5855 | return ERR_PTR(-EIO); | |
1da177e4 | 5856 | } |
91adb564 N |
5857 | if ((mddev->new_level == 5 |
5858 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
5859 | (mddev->new_level == 6 | |
5860 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 5861 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
5862 | mdname(mddev), mddev->new_layout); |
5863 | return ERR_PTR(-EIO); | |
99c0fb5f | 5864 | } |
91adb564 | 5865 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 5866 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
5867 | mdname(mddev), mddev->raid_disks); |
5868 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
5869 | } |
5870 | ||
664e7c41 AN |
5871 | if (!mddev->new_chunk_sectors || |
5872 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
5873 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
5874 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
5875 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 5876 | return ERR_PTR(-EINVAL); |
f6705578 N |
5877 | } |
5878 | ||
d1688a6d | 5879 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 5880 | if (conf == NULL) |
1da177e4 | 5881 | goto abort; |
851c30c9 | 5882 | /* Don't enable multi-threading by default*/ |
60aaf933 | 5883 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
5884 | &new_group)) { | |
5885 | conf->group_cnt = group_cnt; | |
5886 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
5887 | conf->worker_groups = new_group; | |
5888 | } else | |
851c30c9 | 5889 | goto abort; |
f5efd45a | 5890 | spin_lock_init(&conf->device_lock); |
c46501b2 | 5891 | seqcount_init(&conf->gen_lock); |
f5efd45a DW |
5892 | init_waitqueue_head(&conf->wait_for_stripe); |
5893 | init_waitqueue_head(&conf->wait_for_overlap); | |
5894 | INIT_LIST_HEAD(&conf->handle_list); | |
5895 | INIT_LIST_HEAD(&conf->hold_list); | |
5896 | INIT_LIST_HEAD(&conf->delayed_list); | |
5897 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 5898 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
5899 | atomic_set(&conf->active_stripes, 0); |
5900 | atomic_set(&conf->preread_active_stripes, 0); | |
5901 | atomic_set(&conf->active_aligned_reads, 0); | |
5902 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 5903 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
5904 | |
5905 | conf->raid_disks = mddev->raid_disks; | |
5906 | if (mddev->reshape_position == MaxSector) | |
5907 | conf->previous_raid_disks = mddev->raid_disks; | |
5908 | else | |
f6705578 | 5909 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 5910 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 5911 | |
5e5e3e78 | 5912 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
5913 | GFP_KERNEL); |
5914 | if (!conf->disks) | |
5915 | goto abort; | |
9ffae0cf | 5916 | |
1da177e4 LT |
5917 | conf->mddev = mddev; |
5918 | ||
fccddba0 | 5919 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 5920 | goto abort; |
1da177e4 | 5921 | |
566c09c5 SL |
5922 | /* We init hash_locks[0] separately to that it can be used |
5923 | * as the reference lock in the spin_lock_nest_lock() call | |
5924 | * in lock_all_device_hash_locks_irq in order to convince | |
5925 | * lockdep that we know what we are doing. | |
5926 | */ | |
5927 | spin_lock_init(conf->hash_locks); | |
5928 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
5929 | spin_lock_init(conf->hash_locks + i); | |
5930 | ||
5931 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5932 | INIT_LIST_HEAD(conf->inactive_list + i); | |
5933 | ||
5934 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5935 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
5936 | ||
36d1c647 | 5937 | conf->level = mddev->new_level; |
46d5b785 | 5938 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
5939 | if (raid5_alloc_percpu(conf) != 0) |
5940 | goto abort; | |
5941 | ||
0c55e022 | 5942 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 5943 | |
dafb20fa | 5944 | rdev_for_each(rdev, mddev) { |
1da177e4 | 5945 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 5946 | if (raid_disk >= max_disks |
1da177e4 LT |
5947 | || raid_disk < 0) |
5948 | continue; | |
5949 | disk = conf->disks + raid_disk; | |
5950 | ||
17045f52 N |
5951 | if (test_bit(Replacement, &rdev->flags)) { |
5952 | if (disk->replacement) | |
5953 | goto abort; | |
5954 | disk->replacement = rdev; | |
5955 | } else { | |
5956 | if (disk->rdev) | |
5957 | goto abort; | |
5958 | disk->rdev = rdev; | |
5959 | } | |
1da177e4 | 5960 | |
b2d444d7 | 5961 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 5962 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
5963 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
5964 | " disk %d\n", | |
5965 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 5966 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
5967 | /* Cannot rely on bitmap to complete recovery */ |
5968 | conf->fullsync = 1; | |
1da177e4 LT |
5969 | } |
5970 | ||
91adb564 | 5971 | conf->level = mddev->new_level; |
16a53ecc N |
5972 | if (conf->level == 6) |
5973 | conf->max_degraded = 2; | |
5974 | else | |
5975 | conf->max_degraded = 1; | |
91adb564 | 5976 | conf->algorithm = mddev->new_layout; |
fef9c61f | 5977 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 5978 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 5979 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
5980 | conf->prev_algo = mddev->layout; |
5981 | } | |
1da177e4 | 5982 | |
91adb564 | 5983 | memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 5984 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 5985 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
566c09c5 | 5986 | if (grow_stripes(conf, NR_STRIPES)) { |
91adb564 | 5987 | printk(KERN_ERR |
0c55e022 N |
5988 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
5989 | mdname(mddev), memory); | |
91adb564 N |
5990 | goto abort; |
5991 | } else | |
0c55e022 N |
5992 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
5993 | mdname(mddev), memory); | |
1da177e4 | 5994 | |
0232605d N |
5995 | sprintf(pers_name, "raid%d", mddev->new_level); |
5996 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
5997 | if (!conf->thread) { |
5998 | printk(KERN_ERR | |
0c55e022 | 5999 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 6000 | mdname(mddev)); |
16a53ecc N |
6001 | goto abort; |
6002 | } | |
91adb564 N |
6003 | |
6004 | return conf; | |
6005 | ||
6006 | abort: | |
6007 | if (conf) { | |
95fc17aa | 6008 | free_conf(conf); |
91adb564 N |
6009 | return ERR_PTR(-EIO); |
6010 | } else | |
6011 | return ERR_PTR(-ENOMEM); | |
6012 | } | |
6013 | ||
c148ffdc N |
6014 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6015 | { | |
6016 | switch (algo) { | |
6017 | case ALGORITHM_PARITY_0: | |
6018 | if (raid_disk < max_degraded) | |
6019 | return 1; | |
6020 | break; | |
6021 | case ALGORITHM_PARITY_N: | |
6022 | if (raid_disk >= raid_disks - max_degraded) | |
6023 | return 1; | |
6024 | break; | |
6025 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6026 | if (raid_disk == 0 || |
c148ffdc N |
6027 | raid_disk == raid_disks - 1) |
6028 | return 1; | |
6029 | break; | |
6030 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6031 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6032 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6033 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6034 | if (raid_disk == raid_disks - 1) | |
6035 | return 1; | |
6036 | } | |
6037 | return 0; | |
6038 | } | |
6039 | ||
fd01b88c | 6040 | static int run(struct mddev *mddev) |
91adb564 | 6041 | { |
d1688a6d | 6042 | struct r5conf *conf; |
9f7c2220 | 6043 | int working_disks = 0; |
c148ffdc | 6044 | int dirty_parity_disks = 0; |
3cb03002 | 6045 | struct md_rdev *rdev; |
c148ffdc | 6046 | sector_t reshape_offset = 0; |
17045f52 | 6047 | int i; |
b5254dd5 N |
6048 | long long min_offset_diff = 0; |
6049 | int first = 1; | |
91adb564 | 6050 | |
8c6ac868 | 6051 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 6052 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
6053 | " -- starting background reconstruction\n", |
6054 | mdname(mddev)); | |
b5254dd5 N |
6055 | |
6056 | rdev_for_each(rdev, mddev) { | |
6057 | long long diff; | |
6058 | if (rdev->raid_disk < 0) | |
6059 | continue; | |
6060 | diff = (rdev->new_data_offset - rdev->data_offset); | |
6061 | if (first) { | |
6062 | min_offset_diff = diff; | |
6063 | first = 0; | |
6064 | } else if (mddev->reshape_backwards && | |
6065 | diff < min_offset_diff) | |
6066 | min_offset_diff = diff; | |
6067 | else if (!mddev->reshape_backwards && | |
6068 | diff > min_offset_diff) | |
6069 | min_offset_diff = diff; | |
6070 | } | |
6071 | ||
91adb564 N |
6072 | if (mddev->reshape_position != MaxSector) { |
6073 | /* Check that we can continue the reshape. | |
b5254dd5 N |
6074 | * Difficulties arise if the stripe we would write to |
6075 | * next is at or after the stripe we would read from next. | |
6076 | * For a reshape that changes the number of devices, this | |
6077 | * is only possible for a very short time, and mdadm makes | |
6078 | * sure that time appears to have past before assembling | |
6079 | * the array. So we fail if that time hasn't passed. | |
6080 | * For a reshape that keeps the number of devices the same | |
6081 | * mdadm must be monitoring the reshape can keeping the | |
6082 | * critical areas read-only and backed up. It will start | |
6083 | * the array in read-only mode, so we check for that. | |
91adb564 N |
6084 | */ |
6085 | sector_t here_new, here_old; | |
6086 | int old_disks; | |
18b00334 | 6087 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 6088 | |
88ce4930 | 6089 | if (mddev->new_level != mddev->level) { |
0c55e022 | 6090 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
6091 | "required - aborting.\n", |
6092 | mdname(mddev)); | |
6093 | return -EINVAL; | |
6094 | } | |
91adb564 N |
6095 | old_disks = mddev->raid_disks - mddev->delta_disks; |
6096 | /* reshape_position must be on a new-stripe boundary, and one | |
6097 | * further up in new geometry must map after here in old | |
6098 | * geometry. | |
6099 | */ | |
6100 | here_new = mddev->reshape_position; | |
664e7c41 | 6101 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 6102 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
6103 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
6104 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
6105 | return -EINVAL; |
6106 | } | |
c148ffdc | 6107 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
6108 | /* here_new is the stripe we will write to */ |
6109 | here_old = mddev->reshape_position; | |
9d8f0363 | 6110 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
6111 | (old_disks-max_degraded)); |
6112 | /* here_old is the first stripe that we might need to read | |
6113 | * from */ | |
67ac6011 | 6114 | if (mddev->delta_disks == 0) { |
b5254dd5 N |
6115 | if ((here_new * mddev->new_chunk_sectors != |
6116 | here_old * mddev->chunk_sectors)) { | |
6117 | printk(KERN_ERR "md/raid:%s: reshape position is" | |
6118 | " confused - aborting\n", mdname(mddev)); | |
6119 | return -EINVAL; | |
6120 | } | |
67ac6011 | 6121 | /* We cannot be sure it is safe to start an in-place |
b5254dd5 | 6122 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
6123 | * and taking constant backups. |
6124 | * mdadm always starts a situation like this in | |
6125 | * readonly mode so it can take control before | |
6126 | * allowing any writes. So just check for that. | |
6127 | */ | |
b5254dd5 N |
6128 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
6129 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
6130 | /* not really in-place - so OK */; | |
6131 | else if (mddev->ro == 0) { | |
6132 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
6133 | "must be started in read-only mode " | |
6134 | "- aborting\n", | |
0c55e022 | 6135 | mdname(mddev)); |
67ac6011 N |
6136 | return -EINVAL; |
6137 | } | |
2c810cdd | 6138 | } else if (mddev->reshape_backwards |
b5254dd5 | 6139 | ? (here_new * mddev->new_chunk_sectors + min_offset_diff <= |
67ac6011 N |
6140 | here_old * mddev->chunk_sectors) |
6141 | : (here_new * mddev->new_chunk_sectors >= | |
b5254dd5 | 6142 | here_old * mddev->chunk_sectors + (-min_offset_diff))) { |
91adb564 | 6143 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
6144 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
6145 | "auto-recovery - aborting.\n", | |
6146 | mdname(mddev)); | |
91adb564 N |
6147 | return -EINVAL; |
6148 | } | |
0c55e022 N |
6149 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
6150 | mdname(mddev)); | |
91adb564 N |
6151 | /* OK, we should be able to continue; */ |
6152 | } else { | |
6153 | BUG_ON(mddev->level != mddev->new_level); | |
6154 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 6155 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 6156 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 6157 | } |
91adb564 | 6158 | |
245f46c2 N |
6159 | if (mddev->private == NULL) |
6160 | conf = setup_conf(mddev); | |
6161 | else | |
6162 | conf = mddev->private; | |
6163 | ||
91adb564 N |
6164 | if (IS_ERR(conf)) |
6165 | return PTR_ERR(conf); | |
6166 | ||
b5254dd5 | 6167 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
6168 | mddev->thread = conf->thread; |
6169 | conf->thread = NULL; | |
6170 | mddev->private = conf; | |
6171 | ||
17045f52 N |
6172 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
6173 | i++) { | |
6174 | rdev = conf->disks[i].rdev; | |
6175 | if (!rdev && conf->disks[i].replacement) { | |
6176 | /* The replacement is all we have yet */ | |
6177 | rdev = conf->disks[i].replacement; | |
6178 | conf->disks[i].replacement = NULL; | |
6179 | clear_bit(Replacement, &rdev->flags); | |
6180 | conf->disks[i].rdev = rdev; | |
6181 | } | |
6182 | if (!rdev) | |
c148ffdc | 6183 | continue; |
17045f52 N |
6184 | if (conf->disks[i].replacement && |
6185 | conf->reshape_progress != MaxSector) { | |
6186 | /* replacements and reshape simply do not mix. */ | |
6187 | printk(KERN_ERR "md: cannot handle concurrent " | |
6188 | "replacement and reshape.\n"); | |
6189 | goto abort; | |
6190 | } | |
2f115882 | 6191 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 6192 | working_disks++; |
2f115882 N |
6193 | continue; |
6194 | } | |
c148ffdc N |
6195 | /* This disc is not fully in-sync. However if it |
6196 | * just stored parity (beyond the recovery_offset), | |
6197 | * when we don't need to be concerned about the | |
6198 | * array being dirty. | |
6199 | * When reshape goes 'backwards', we never have | |
6200 | * partially completed devices, so we only need | |
6201 | * to worry about reshape going forwards. | |
6202 | */ | |
6203 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
6204 | if (mddev->major_version == 0 && | |
6205 | mddev->minor_version > 90) | |
6206 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 6207 | |
c148ffdc N |
6208 | if (rdev->recovery_offset < reshape_offset) { |
6209 | /* We need to check old and new layout */ | |
6210 | if (!only_parity(rdev->raid_disk, | |
6211 | conf->algorithm, | |
6212 | conf->raid_disks, | |
6213 | conf->max_degraded)) | |
6214 | continue; | |
6215 | } | |
6216 | if (!only_parity(rdev->raid_disk, | |
6217 | conf->prev_algo, | |
6218 | conf->previous_raid_disks, | |
6219 | conf->max_degraded)) | |
6220 | continue; | |
6221 | dirty_parity_disks++; | |
6222 | } | |
91adb564 | 6223 | |
17045f52 N |
6224 | /* |
6225 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
6226 | */ | |
908f4fbd | 6227 | mddev->degraded = calc_degraded(conf); |
91adb564 | 6228 | |
674806d6 | 6229 | if (has_failed(conf)) { |
0c55e022 | 6230 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 6231 | " (%d/%d failed)\n", |
02c2de8c | 6232 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
6233 | goto abort; |
6234 | } | |
6235 | ||
91adb564 | 6236 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 6237 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
6238 | mddev->resync_max_sectors = mddev->dev_sectors; |
6239 | ||
c148ffdc | 6240 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 6241 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
6242 | if (mddev->ok_start_degraded) |
6243 | printk(KERN_WARNING | |
0c55e022 N |
6244 | "md/raid:%s: starting dirty degraded array" |
6245 | " - data corruption possible.\n", | |
6ff8d8ec N |
6246 | mdname(mddev)); |
6247 | else { | |
6248 | printk(KERN_ERR | |
0c55e022 | 6249 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
6250 | mdname(mddev)); |
6251 | goto abort; | |
6252 | } | |
1da177e4 LT |
6253 | } |
6254 | ||
1da177e4 | 6255 | if (mddev->degraded == 0) |
0c55e022 N |
6256 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
6257 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
6258 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
6259 | mddev->new_layout); | |
1da177e4 | 6260 | else |
0c55e022 N |
6261 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
6262 | " out of %d devices, algorithm %d\n", | |
6263 | mdname(mddev), conf->level, | |
6264 | mddev->raid_disks - mddev->degraded, | |
6265 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
6266 | |
6267 | print_raid5_conf(conf); | |
6268 | ||
fef9c61f | 6269 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 6270 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
6271 | atomic_set(&conf->reshape_stripes, 0); |
6272 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
6273 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6274 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6275 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6276 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6277 | "reshape"); |
f6705578 N |
6278 | } |
6279 | ||
1da177e4 | 6280 | /* Ok, everything is just fine now */ |
a64c876f N |
6281 | if (mddev->to_remove == &raid5_attrs_group) |
6282 | mddev->to_remove = NULL; | |
00bcb4ac N |
6283 | else if (mddev->kobj.sd && |
6284 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 6285 | printk(KERN_WARNING |
4a5add49 | 6286 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 6287 | mdname(mddev)); |
4a5add49 | 6288 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 6289 | |
4a5add49 | 6290 | if (mddev->queue) { |
9f7c2220 | 6291 | int chunk_size; |
620125f2 | 6292 | bool discard_supported = true; |
4a5add49 N |
6293 | /* read-ahead size must cover two whole stripes, which |
6294 | * is 2 * (datadisks) * chunksize where 'n' is the | |
6295 | * number of raid devices | |
6296 | */ | |
6297 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
6298 | int stripe = data_disks * | |
6299 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
6300 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
6301 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 6302 | |
9f7c2220 N |
6303 | chunk_size = mddev->chunk_sectors << 9; |
6304 | blk_queue_io_min(mddev->queue, chunk_size); | |
6305 | blk_queue_io_opt(mddev->queue, chunk_size * | |
6306 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 6307 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
6308 | /* |
6309 | * We can only discard a whole stripe. It doesn't make sense to | |
6310 | * discard data disk but write parity disk | |
6311 | */ | |
6312 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
6313 | /* Round up to power of 2, as discard handling |
6314 | * currently assumes that */ | |
6315 | while ((stripe-1) & stripe) | |
6316 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
6317 | mddev->queue->limits.discard_alignment = stripe; |
6318 | mddev->queue->limits.discard_granularity = stripe; | |
6319 | /* | |
6320 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 6321 | * guarantee discard_zeroes_data |
620125f2 SL |
6322 | */ |
6323 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 6324 | |
5026d7a9 PA |
6325 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
6326 | ||
05616be5 | 6327 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
6328 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6329 | rdev->data_offset << 9); | |
05616be5 N |
6330 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6331 | rdev->new_data_offset << 9); | |
620125f2 SL |
6332 | /* |
6333 | * discard_zeroes_data is required, otherwise data | |
6334 | * could be lost. Consider a scenario: discard a stripe | |
6335 | * (the stripe could be inconsistent if | |
6336 | * discard_zeroes_data is 0); write one disk of the | |
6337 | * stripe (the stripe could be inconsistent again | |
6338 | * depending on which disks are used to calculate | |
6339 | * parity); the disk is broken; The stripe data of this | |
6340 | * disk is lost. | |
6341 | */ | |
6342 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
6343 | !bdev_get_queue(rdev->bdev)-> | |
6344 | limits.discard_zeroes_data) | |
6345 | discard_supported = false; | |
8e0e99ba N |
6346 | /* Unfortunately, discard_zeroes_data is not currently |
6347 | * a guarantee - just a hint. So we only allow DISCARD | |
6348 | * if the sysadmin has confirmed that only safe devices | |
6349 | * are in use by setting a module parameter. | |
6350 | */ | |
6351 | if (!devices_handle_discard_safely) { | |
6352 | if (discard_supported) { | |
6353 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
6354 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
6355 | } | |
6356 | discard_supported = false; | |
6357 | } | |
05616be5 | 6358 | } |
620125f2 SL |
6359 | |
6360 | if (discard_supported && | |
6361 | mddev->queue->limits.max_discard_sectors >= stripe && | |
6362 | mddev->queue->limits.discard_granularity >= stripe) | |
6363 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
6364 | mddev->queue); | |
6365 | else | |
6366 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
6367 | mddev->queue); | |
9f7c2220 | 6368 | } |
23032a0e | 6369 | |
1da177e4 LT |
6370 | return 0; |
6371 | abort: | |
01f96c0a | 6372 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
6373 | print_raid5_conf(conf); |
6374 | free_conf(conf); | |
1da177e4 | 6375 | mddev->private = NULL; |
0c55e022 | 6376 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
6377 | return -EIO; |
6378 | } | |
6379 | ||
afa0f557 | 6380 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 6381 | { |
afa0f557 | 6382 | struct r5conf *conf = priv; |
1da177e4 | 6383 | |
95fc17aa | 6384 | free_conf(conf); |
a64c876f | 6385 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
6386 | } |
6387 | ||
fd01b88c | 6388 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6389 | { |
d1688a6d | 6390 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
6391 | int i; |
6392 | ||
9d8f0363 AN |
6393 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
6394 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 6395 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
6396 | for (i = 0; i < conf->raid_disks; i++) |
6397 | seq_printf (seq, "%s", | |
6398 | conf->disks[i].rdev && | |
b2d444d7 | 6399 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 6400 | seq_printf (seq, "]"); |
1da177e4 LT |
6401 | } |
6402 | ||
d1688a6d | 6403 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
6404 | { |
6405 | int i; | |
6406 | struct disk_info *tmp; | |
6407 | ||
0c55e022 | 6408 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
6409 | if (!conf) { |
6410 | printk("(conf==NULL)\n"); | |
6411 | return; | |
6412 | } | |
0c55e022 N |
6413 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
6414 | conf->raid_disks, | |
6415 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
6416 | |
6417 | for (i = 0; i < conf->raid_disks; i++) { | |
6418 | char b[BDEVNAME_SIZE]; | |
6419 | tmp = conf->disks + i; | |
6420 | if (tmp->rdev) | |
0c55e022 N |
6421 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
6422 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
6423 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
6424 | } |
6425 | } | |
6426 | ||
fd01b88c | 6427 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
6428 | { |
6429 | int i; | |
d1688a6d | 6430 | struct r5conf *conf = mddev->private; |
1da177e4 | 6431 | struct disk_info *tmp; |
6b965620 N |
6432 | int count = 0; |
6433 | unsigned long flags; | |
1da177e4 LT |
6434 | |
6435 | for (i = 0; i < conf->raid_disks; i++) { | |
6436 | tmp = conf->disks + i; | |
dd054fce N |
6437 | if (tmp->replacement |
6438 | && tmp->replacement->recovery_offset == MaxSector | |
6439 | && !test_bit(Faulty, &tmp->replacement->flags) | |
6440 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
6441 | /* Replacement has just become active. */ | |
6442 | if (!tmp->rdev | |
6443 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
6444 | count++; | |
6445 | if (tmp->rdev) { | |
6446 | /* Replaced device not technically faulty, | |
6447 | * but we need to be sure it gets removed | |
6448 | * and never re-added. | |
6449 | */ | |
6450 | set_bit(Faulty, &tmp->rdev->flags); | |
6451 | sysfs_notify_dirent_safe( | |
6452 | tmp->rdev->sysfs_state); | |
6453 | } | |
6454 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
6455 | } else if (tmp->rdev | |
70fffd0b | 6456 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 6457 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 6458 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 6459 | count++; |
43c73ca4 | 6460 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
6461 | } |
6462 | } | |
6b965620 | 6463 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 6464 | mddev->degraded = calc_degraded(conf); |
6b965620 | 6465 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 6466 | print_raid5_conf(conf); |
6b965620 | 6467 | return count; |
1da177e4 LT |
6468 | } |
6469 | ||
b8321b68 | 6470 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 6471 | { |
d1688a6d | 6472 | struct r5conf *conf = mddev->private; |
1da177e4 | 6473 | int err = 0; |
b8321b68 | 6474 | int number = rdev->raid_disk; |
657e3e4d | 6475 | struct md_rdev **rdevp; |
1da177e4 LT |
6476 | struct disk_info *p = conf->disks + number; |
6477 | ||
6478 | print_raid5_conf(conf); | |
657e3e4d N |
6479 | if (rdev == p->rdev) |
6480 | rdevp = &p->rdev; | |
6481 | else if (rdev == p->replacement) | |
6482 | rdevp = &p->replacement; | |
6483 | else | |
6484 | return 0; | |
6485 | ||
6486 | if (number >= conf->raid_disks && | |
6487 | conf->reshape_progress == MaxSector) | |
6488 | clear_bit(In_sync, &rdev->flags); | |
6489 | ||
6490 | if (test_bit(In_sync, &rdev->flags) || | |
6491 | atomic_read(&rdev->nr_pending)) { | |
6492 | err = -EBUSY; | |
6493 | goto abort; | |
6494 | } | |
6495 | /* Only remove non-faulty devices if recovery | |
6496 | * isn't possible. | |
6497 | */ | |
6498 | if (!test_bit(Faulty, &rdev->flags) && | |
6499 | mddev->recovery_disabled != conf->recovery_disabled && | |
6500 | !has_failed(conf) && | |
dd054fce | 6501 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
6502 | number < conf->raid_disks) { |
6503 | err = -EBUSY; | |
6504 | goto abort; | |
6505 | } | |
6506 | *rdevp = NULL; | |
6507 | synchronize_rcu(); | |
6508 | if (atomic_read(&rdev->nr_pending)) { | |
6509 | /* lost the race, try later */ | |
6510 | err = -EBUSY; | |
6511 | *rdevp = rdev; | |
dd054fce N |
6512 | } else if (p->replacement) { |
6513 | /* We must have just cleared 'rdev' */ | |
6514 | p->rdev = p->replacement; | |
6515 | clear_bit(Replacement, &p->replacement->flags); | |
6516 | smp_mb(); /* Make sure other CPUs may see both as identical | |
6517 | * but will never see neither - if they are careful | |
6518 | */ | |
6519 | p->replacement = NULL; | |
6520 | clear_bit(WantReplacement, &rdev->flags); | |
6521 | } else | |
6522 | /* We might have just removed the Replacement as faulty- | |
6523 | * clear the bit just in case | |
6524 | */ | |
6525 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
6526 | abort: |
6527 | ||
6528 | print_raid5_conf(conf); | |
6529 | return err; | |
6530 | } | |
6531 | ||
fd01b88c | 6532 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 6533 | { |
d1688a6d | 6534 | struct r5conf *conf = mddev->private; |
199050ea | 6535 | int err = -EEXIST; |
1da177e4 LT |
6536 | int disk; |
6537 | struct disk_info *p; | |
6c2fce2e NB |
6538 | int first = 0; |
6539 | int last = conf->raid_disks - 1; | |
1da177e4 | 6540 | |
7f0da59b N |
6541 | if (mddev->recovery_disabled == conf->recovery_disabled) |
6542 | return -EBUSY; | |
6543 | ||
dc10c643 | 6544 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 6545 | /* no point adding a device */ |
199050ea | 6546 | return -EINVAL; |
1da177e4 | 6547 | |
6c2fce2e NB |
6548 | if (rdev->raid_disk >= 0) |
6549 | first = last = rdev->raid_disk; | |
1da177e4 LT |
6550 | |
6551 | /* | |
16a53ecc N |
6552 | * find the disk ... but prefer rdev->saved_raid_disk |
6553 | * if possible. | |
1da177e4 | 6554 | */ |
16a53ecc | 6555 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 6556 | rdev->saved_raid_disk >= first && |
16a53ecc | 6557 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
6558 | first = rdev->saved_raid_disk; |
6559 | ||
6560 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
6561 | p = conf->disks + disk; |
6562 | if (p->rdev == NULL) { | |
b2d444d7 | 6563 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 6564 | rdev->raid_disk = disk; |
199050ea | 6565 | err = 0; |
72626685 N |
6566 | if (rdev->saved_raid_disk != disk) |
6567 | conf->fullsync = 1; | |
d6065f7b | 6568 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 6569 | goto out; |
1da177e4 | 6570 | } |
5cfb22a1 N |
6571 | } |
6572 | for (disk = first; disk <= last; disk++) { | |
6573 | p = conf->disks + disk; | |
7bfec5f3 N |
6574 | if (test_bit(WantReplacement, &p->rdev->flags) && |
6575 | p->replacement == NULL) { | |
6576 | clear_bit(In_sync, &rdev->flags); | |
6577 | set_bit(Replacement, &rdev->flags); | |
6578 | rdev->raid_disk = disk; | |
6579 | err = 0; | |
6580 | conf->fullsync = 1; | |
6581 | rcu_assign_pointer(p->replacement, rdev); | |
6582 | break; | |
6583 | } | |
6584 | } | |
5cfb22a1 | 6585 | out: |
1da177e4 | 6586 | print_raid5_conf(conf); |
199050ea | 6587 | return err; |
1da177e4 LT |
6588 | } |
6589 | ||
fd01b88c | 6590 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
6591 | { |
6592 | /* no resync is happening, and there is enough space | |
6593 | * on all devices, so we can resize. | |
6594 | * We need to make sure resync covers any new space. | |
6595 | * If the array is shrinking we should possibly wait until | |
6596 | * any io in the removed space completes, but it hardly seems | |
6597 | * worth it. | |
6598 | */ | |
a4a6125a | 6599 | sector_t newsize; |
9d8f0363 | 6600 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
a4a6125a N |
6601 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
6602 | if (mddev->external_size && | |
6603 | mddev->array_sectors > newsize) | |
b522adcd | 6604 | return -EINVAL; |
a4a6125a N |
6605 | if (mddev->bitmap) { |
6606 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
6607 | if (ret) | |
6608 | return ret; | |
6609 | } | |
6610 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 6611 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 6612 | revalidate_disk(mddev->gendisk); |
b098636c N |
6613 | if (sectors > mddev->dev_sectors && |
6614 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 6615 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
6616 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
6617 | } | |
58c0fed4 | 6618 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 6619 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
6620 | return 0; |
6621 | } | |
6622 | ||
fd01b88c | 6623 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
6624 | { |
6625 | /* Can only proceed if there are plenty of stripe_heads. | |
6626 | * We need a minimum of one full stripe,, and for sensible progress | |
6627 | * it is best to have about 4 times that. | |
6628 | * If we require 4 times, then the default 256 4K stripe_heads will | |
6629 | * allow for chunk sizes up to 256K, which is probably OK. | |
6630 | * If the chunk size is greater, user-space should request more | |
6631 | * stripe_heads first. | |
6632 | */ | |
d1688a6d | 6633 | struct r5conf *conf = mddev->private; |
01ee22b4 N |
6634 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
6635 | > conf->max_nr_stripes || | |
6636 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 | |
6637 | > conf->max_nr_stripes) { | |
0c55e022 N |
6638 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
6639 | mdname(mddev), | |
01ee22b4 N |
6640 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
6641 | / STRIPE_SIZE)*4); | |
6642 | return 0; | |
6643 | } | |
6644 | return 1; | |
6645 | } | |
6646 | ||
fd01b88c | 6647 | static int check_reshape(struct mddev *mddev) |
29269553 | 6648 | { |
d1688a6d | 6649 | struct r5conf *conf = mddev->private; |
29269553 | 6650 | |
88ce4930 N |
6651 | if (mddev->delta_disks == 0 && |
6652 | mddev->new_layout == mddev->layout && | |
664e7c41 | 6653 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 6654 | return 0; /* nothing to do */ |
674806d6 | 6655 | if (has_failed(conf)) |
ec32a2bd | 6656 | return -EINVAL; |
fdcfbbb6 | 6657 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
6658 | /* We might be able to shrink, but the devices must |
6659 | * be made bigger first. | |
6660 | * For raid6, 4 is the minimum size. | |
6661 | * Otherwise 2 is the minimum | |
6662 | */ | |
6663 | int min = 2; | |
6664 | if (mddev->level == 6) | |
6665 | min = 4; | |
6666 | if (mddev->raid_disks + mddev->delta_disks < min) | |
6667 | return -EINVAL; | |
6668 | } | |
29269553 | 6669 | |
01ee22b4 | 6670 | if (!check_stripe_cache(mddev)) |
29269553 | 6671 | return -ENOSPC; |
29269553 | 6672 | |
e56108d6 N |
6673 | return resize_stripes(conf, (conf->previous_raid_disks |
6674 | + mddev->delta_disks)); | |
63c70c4f N |
6675 | } |
6676 | ||
fd01b88c | 6677 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 6678 | { |
d1688a6d | 6679 | struct r5conf *conf = mddev->private; |
3cb03002 | 6680 | struct md_rdev *rdev; |
63c70c4f | 6681 | int spares = 0; |
c04be0aa | 6682 | unsigned long flags; |
63c70c4f | 6683 | |
f416885e | 6684 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
6685 | return -EBUSY; |
6686 | ||
01ee22b4 N |
6687 | if (!check_stripe_cache(mddev)) |
6688 | return -ENOSPC; | |
6689 | ||
30b67645 N |
6690 | if (has_failed(conf)) |
6691 | return -EINVAL; | |
6692 | ||
c6563a8c | 6693 | rdev_for_each(rdev, mddev) { |
469518a3 N |
6694 | if (!test_bit(In_sync, &rdev->flags) |
6695 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 6696 | spares++; |
c6563a8c | 6697 | } |
63c70c4f | 6698 | |
f416885e | 6699 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
6700 | /* Not enough devices even to make a degraded array |
6701 | * of that size | |
6702 | */ | |
6703 | return -EINVAL; | |
6704 | ||
ec32a2bd N |
6705 | /* Refuse to reduce size of the array. Any reductions in |
6706 | * array size must be through explicit setting of array_size | |
6707 | * attribute. | |
6708 | */ | |
6709 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
6710 | < mddev->array_sectors) { | |
0c55e022 | 6711 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
6712 | "before number of disks\n", mdname(mddev)); |
6713 | return -EINVAL; | |
6714 | } | |
6715 | ||
f6705578 | 6716 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 6717 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 6718 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 6719 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 6720 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
6721 | conf->prev_chunk_sectors = conf->chunk_sectors; |
6722 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
6723 | conf->prev_algo = conf->algorithm; |
6724 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
6725 | conf->generation++; |
6726 | /* Code that selects data_offset needs to see the generation update | |
6727 | * if reshape_progress has been set - so a memory barrier needed. | |
6728 | */ | |
6729 | smp_mb(); | |
2c810cdd | 6730 | if (mddev->reshape_backwards) |
fef9c61f N |
6731 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
6732 | else | |
6733 | conf->reshape_progress = 0; | |
6734 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 6735 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
6736 | spin_unlock_irq(&conf->device_lock); |
6737 | ||
4d77e3ba N |
6738 | /* Now make sure any requests that proceeded on the assumption |
6739 | * the reshape wasn't running - like Discard or Read - have | |
6740 | * completed. | |
6741 | */ | |
6742 | mddev_suspend(mddev); | |
6743 | mddev_resume(mddev); | |
6744 | ||
29269553 N |
6745 | /* Add some new drives, as many as will fit. |
6746 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
6747 | * Don't add devices if we are reducing the number of |
6748 | * devices in the array. This is because it is not possible | |
6749 | * to correctly record the "partially reconstructed" state of | |
6750 | * such devices during the reshape and confusion could result. | |
29269553 | 6751 | */ |
87a8dec9 | 6752 | if (mddev->delta_disks >= 0) { |
dafb20fa | 6753 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
6754 | if (rdev->raid_disk < 0 && |
6755 | !test_bit(Faulty, &rdev->flags)) { | |
6756 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 6757 | if (rdev->raid_disk |
9d4c7d87 | 6758 | >= conf->previous_raid_disks) |
87a8dec9 | 6759 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 6760 | else |
87a8dec9 | 6761 | rdev->recovery_offset = 0; |
36fad858 NK |
6762 | |
6763 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 6764 | /* Failure here is OK */; |
50da0840 | 6765 | } |
87a8dec9 N |
6766 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
6767 | && !test_bit(Faulty, &rdev->flags)) { | |
6768 | /* This is a spare that was manually added */ | |
6769 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 6770 | } |
29269553 | 6771 | |
87a8dec9 N |
6772 | /* When a reshape changes the number of devices, |
6773 | * ->degraded is measured against the larger of the | |
6774 | * pre and post number of devices. | |
6775 | */ | |
ec32a2bd | 6776 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 6777 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
6778 | spin_unlock_irqrestore(&conf->device_lock, flags); |
6779 | } | |
63c70c4f | 6780 | mddev->raid_disks = conf->raid_disks; |
e516402c | 6781 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 6782 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 6783 | |
29269553 N |
6784 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
6785 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6786 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6787 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6788 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6789 | "reshape"); |
29269553 N |
6790 | if (!mddev->sync_thread) { |
6791 | mddev->recovery = 0; | |
6792 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 6793 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 6794 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
6795 | mddev->new_chunk_sectors = |
6796 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
6797 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
6798 | rdev_for_each(rdev, mddev) |
6799 | rdev->new_data_offset = rdev->data_offset; | |
6800 | smp_wmb(); | |
ba8805b9 | 6801 | conf->generation --; |
fef9c61f | 6802 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 6803 | mddev->reshape_position = MaxSector; |
ba8805b9 | 6804 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
6805 | spin_unlock_irq(&conf->device_lock); |
6806 | return -EAGAIN; | |
6807 | } | |
c8f517c4 | 6808 | conf->reshape_checkpoint = jiffies; |
29269553 N |
6809 | md_wakeup_thread(mddev->sync_thread); |
6810 | md_new_event(mddev); | |
6811 | return 0; | |
6812 | } | |
29269553 | 6813 | |
ec32a2bd N |
6814 | /* This is called from the reshape thread and should make any |
6815 | * changes needed in 'conf' | |
6816 | */ | |
d1688a6d | 6817 | static void end_reshape(struct r5conf *conf) |
29269553 | 6818 | { |
29269553 | 6819 | |
f6705578 | 6820 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 6821 | struct md_rdev *rdev; |
f6705578 | 6822 | |
f6705578 | 6823 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 6824 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
6825 | rdev_for_each(rdev, conf->mddev) |
6826 | rdev->data_offset = rdev->new_data_offset; | |
6827 | smp_wmb(); | |
fef9c61f | 6828 | conf->reshape_progress = MaxSector; |
f6705578 | 6829 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 6830 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
6831 | |
6832 | /* read-ahead size must cover two whole stripes, which is | |
6833 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
6834 | */ | |
4a5add49 | 6835 | if (conf->mddev->queue) { |
cea9c228 | 6836 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 6837 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 6838 | / PAGE_SIZE); |
16a53ecc N |
6839 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
6840 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
6841 | } | |
29269553 | 6842 | } |
29269553 N |
6843 | } |
6844 | ||
ec32a2bd N |
6845 | /* This is called from the raid5d thread with mddev_lock held. |
6846 | * It makes config changes to the device. | |
6847 | */ | |
fd01b88c | 6848 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 6849 | { |
d1688a6d | 6850 | struct r5conf *conf = mddev->private; |
cea9c228 N |
6851 | |
6852 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
6853 | ||
ec32a2bd N |
6854 | if (mddev->delta_disks > 0) { |
6855 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
6856 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 6857 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
6858 | } else { |
6859 | int d; | |
908f4fbd N |
6860 | spin_lock_irq(&conf->device_lock); |
6861 | mddev->degraded = calc_degraded(conf); | |
6862 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
6863 | for (d = conf->raid_disks ; |
6864 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 6865 | d++) { |
3cb03002 | 6866 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
6867 | if (rdev) |
6868 | clear_bit(In_sync, &rdev->flags); | |
6869 | rdev = conf->disks[d].replacement; | |
6870 | if (rdev) | |
6871 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 6872 | } |
cea9c228 | 6873 | } |
88ce4930 | 6874 | mddev->layout = conf->algorithm; |
09c9e5fa | 6875 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
6876 | mddev->reshape_position = MaxSector; |
6877 | mddev->delta_disks = 0; | |
2c810cdd | 6878 | mddev->reshape_backwards = 0; |
cea9c228 N |
6879 | } |
6880 | } | |
6881 | ||
fd01b88c | 6882 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 6883 | { |
d1688a6d | 6884 | struct r5conf *conf = mddev->private; |
72626685 N |
6885 | |
6886 | switch(state) { | |
e464eafd N |
6887 | case 2: /* resume for a suspend */ |
6888 | wake_up(&conf->wait_for_overlap); | |
6889 | break; | |
6890 | ||
72626685 | 6891 | case 1: /* stop all writes */ |
566c09c5 | 6892 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
6893 | /* '2' tells resync/reshape to pause so that all |
6894 | * active stripes can drain | |
6895 | */ | |
6896 | conf->quiesce = 2; | |
566c09c5 | 6897 | wait_event_cmd(conf->wait_for_stripe, |
46031f9a RBJ |
6898 | atomic_read(&conf->active_stripes) == 0 && |
6899 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
6900 | unlock_all_device_hash_locks_irq(conf), |
6901 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 6902 | conf->quiesce = 1; |
566c09c5 | 6903 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
6904 | /* allow reshape to continue */ |
6905 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
6906 | break; |
6907 | ||
6908 | case 0: /* re-enable writes */ | |
566c09c5 | 6909 | lock_all_device_hash_locks_irq(conf); |
72626685 N |
6910 | conf->quiesce = 0; |
6911 | wake_up(&conf->wait_for_stripe); | |
e464eafd | 6912 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 6913 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
6914 | break; |
6915 | } | |
72626685 | 6916 | } |
b15c2e57 | 6917 | |
fd01b88c | 6918 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 6919 | { |
e373ab10 | 6920 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 6921 | sector_t sectors; |
54071b38 | 6922 | |
f1b29bca | 6923 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 6924 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
6925 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
6926 | mdname(mddev)); | |
f1b29bca DW |
6927 | return ERR_PTR(-EINVAL); |
6928 | } | |
6929 | ||
e373ab10 N |
6930 | sectors = raid0_conf->strip_zone[0].zone_end; |
6931 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 6932 | mddev->dev_sectors = sectors; |
f1b29bca | 6933 | mddev->new_level = level; |
54071b38 TM |
6934 | mddev->new_layout = ALGORITHM_PARITY_N; |
6935 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
6936 | mddev->raid_disks += 1; | |
6937 | mddev->delta_disks = 1; | |
6938 | /* make sure it will be not marked as dirty */ | |
6939 | mddev->recovery_cp = MaxSector; | |
6940 | ||
6941 | return setup_conf(mddev); | |
6942 | } | |
6943 | ||
fd01b88c | 6944 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
6945 | { |
6946 | int chunksect; | |
6947 | ||
6948 | if (mddev->raid_disks != 2 || | |
6949 | mddev->degraded > 1) | |
6950 | return ERR_PTR(-EINVAL); | |
6951 | ||
6952 | /* Should check if there are write-behind devices? */ | |
6953 | ||
6954 | chunksect = 64*2; /* 64K by default */ | |
6955 | ||
6956 | /* The array must be an exact multiple of chunksize */ | |
6957 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
6958 | chunksect >>= 1; | |
6959 | ||
6960 | if ((chunksect<<9) < STRIPE_SIZE) | |
6961 | /* array size does not allow a suitable chunk size */ | |
6962 | return ERR_PTR(-EINVAL); | |
6963 | ||
6964 | mddev->new_level = 5; | |
6965 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 6966 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
6967 | |
6968 | return setup_conf(mddev); | |
6969 | } | |
6970 | ||
fd01b88c | 6971 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
6972 | { |
6973 | int new_layout; | |
6974 | ||
6975 | switch (mddev->layout) { | |
6976 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6977 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
6978 | break; | |
6979 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6980 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
6981 | break; | |
6982 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6983 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
6984 | break; | |
6985 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6986 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
6987 | break; | |
6988 | case ALGORITHM_PARITY_0_6: | |
6989 | new_layout = ALGORITHM_PARITY_0; | |
6990 | break; | |
6991 | case ALGORITHM_PARITY_N: | |
6992 | new_layout = ALGORITHM_PARITY_N; | |
6993 | break; | |
6994 | default: | |
6995 | return ERR_PTR(-EINVAL); | |
6996 | } | |
6997 | mddev->new_level = 5; | |
6998 | mddev->new_layout = new_layout; | |
6999 | mddev->delta_disks = -1; | |
7000 | mddev->raid_disks -= 1; | |
7001 | return setup_conf(mddev); | |
7002 | } | |
7003 | ||
fd01b88c | 7004 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 7005 | { |
88ce4930 N |
7006 | /* For a 2-drive array, the layout and chunk size can be changed |
7007 | * immediately as not restriping is needed. | |
7008 | * For larger arrays we record the new value - after validation | |
7009 | * to be used by a reshape pass. | |
b3546035 | 7010 | */ |
d1688a6d | 7011 | struct r5conf *conf = mddev->private; |
597a711b | 7012 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 7013 | |
597a711b | 7014 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
7015 | return -EINVAL; |
7016 | if (new_chunk > 0) { | |
0ba459d2 | 7017 | if (!is_power_of_2(new_chunk)) |
b3546035 | 7018 | return -EINVAL; |
597a711b | 7019 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 7020 | return -EINVAL; |
597a711b | 7021 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
7022 | /* not factor of array size */ |
7023 | return -EINVAL; | |
7024 | } | |
7025 | ||
7026 | /* They look valid */ | |
7027 | ||
88ce4930 | 7028 | if (mddev->raid_disks == 2) { |
597a711b N |
7029 | /* can make the change immediately */ |
7030 | if (mddev->new_layout >= 0) { | |
7031 | conf->algorithm = mddev->new_layout; | |
7032 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
7033 | } |
7034 | if (new_chunk > 0) { | |
597a711b N |
7035 | conf->chunk_sectors = new_chunk ; |
7036 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
7037 | } |
7038 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7039 | md_wakeup_thread(mddev->thread); | |
b3546035 | 7040 | } |
50ac168a | 7041 | return check_reshape(mddev); |
88ce4930 N |
7042 | } |
7043 | ||
fd01b88c | 7044 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 7045 | { |
597a711b | 7046 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 7047 | |
597a711b | 7048 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 7049 | return -EINVAL; |
b3546035 | 7050 | if (new_chunk > 0) { |
0ba459d2 | 7051 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 7052 | return -EINVAL; |
597a711b | 7053 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 7054 | return -EINVAL; |
597a711b | 7055 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
7056 | /* not factor of array size */ |
7057 | return -EINVAL; | |
b3546035 | 7058 | } |
88ce4930 N |
7059 | |
7060 | /* They look valid */ | |
50ac168a | 7061 | return check_reshape(mddev); |
b3546035 N |
7062 | } |
7063 | ||
fd01b88c | 7064 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
7065 | { |
7066 | /* raid5 can take over: | |
f1b29bca | 7067 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
7068 | * raid1 - if there are two drives. We need to know the chunk size |
7069 | * raid4 - trivial - just use a raid4 layout. | |
7070 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 7071 | */ |
f1b29bca DW |
7072 | if (mddev->level == 0) |
7073 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
7074 | if (mddev->level == 1) |
7075 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
7076 | if (mddev->level == 4) { |
7077 | mddev->new_layout = ALGORITHM_PARITY_N; | |
7078 | mddev->new_level = 5; | |
7079 | return setup_conf(mddev); | |
7080 | } | |
fc9739c6 N |
7081 | if (mddev->level == 6) |
7082 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
7083 | |
7084 | return ERR_PTR(-EINVAL); | |
7085 | } | |
7086 | ||
fd01b88c | 7087 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 7088 | { |
f1b29bca DW |
7089 | /* raid4 can take over: |
7090 | * raid0 - if there is only one strip zone | |
7091 | * raid5 - if layout is right | |
a78d38a1 | 7092 | */ |
f1b29bca DW |
7093 | if (mddev->level == 0) |
7094 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
7095 | if (mddev->level == 5 && |
7096 | mddev->layout == ALGORITHM_PARITY_N) { | |
7097 | mddev->new_layout = 0; | |
7098 | mddev->new_level = 4; | |
7099 | return setup_conf(mddev); | |
7100 | } | |
7101 | return ERR_PTR(-EINVAL); | |
7102 | } | |
d562b0c4 | 7103 | |
84fc4b56 | 7104 | static struct md_personality raid5_personality; |
245f46c2 | 7105 | |
fd01b88c | 7106 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
7107 | { |
7108 | /* Currently can only take over a raid5. We map the | |
7109 | * personality to an equivalent raid6 personality | |
7110 | * with the Q block at the end. | |
7111 | */ | |
7112 | int new_layout; | |
7113 | ||
7114 | if (mddev->pers != &raid5_personality) | |
7115 | return ERR_PTR(-EINVAL); | |
7116 | if (mddev->degraded > 1) | |
7117 | return ERR_PTR(-EINVAL); | |
7118 | if (mddev->raid_disks > 253) | |
7119 | return ERR_PTR(-EINVAL); | |
7120 | if (mddev->raid_disks < 3) | |
7121 | return ERR_PTR(-EINVAL); | |
7122 | ||
7123 | switch (mddev->layout) { | |
7124 | case ALGORITHM_LEFT_ASYMMETRIC: | |
7125 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
7126 | break; | |
7127 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
7128 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
7129 | break; | |
7130 | case ALGORITHM_LEFT_SYMMETRIC: | |
7131 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
7132 | break; | |
7133 | case ALGORITHM_RIGHT_SYMMETRIC: | |
7134 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
7135 | break; | |
7136 | case ALGORITHM_PARITY_0: | |
7137 | new_layout = ALGORITHM_PARITY_0_6; | |
7138 | break; | |
7139 | case ALGORITHM_PARITY_N: | |
7140 | new_layout = ALGORITHM_PARITY_N; | |
7141 | break; | |
7142 | default: | |
7143 | return ERR_PTR(-EINVAL); | |
7144 | } | |
7145 | mddev->new_level = 6; | |
7146 | mddev->new_layout = new_layout; | |
7147 | mddev->delta_disks = 1; | |
7148 | mddev->raid_disks += 1; | |
7149 | return setup_conf(mddev); | |
7150 | } | |
7151 | ||
84fc4b56 | 7152 | static struct md_personality raid6_personality = |
16a53ecc N |
7153 | { |
7154 | .name = "raid6", | |
7155 | .level = 6, | |
7156 | .owner = THIS_MODULE, | |
7157 | .make_request = make_request, | |
7158 | .run = run, | |
afa0f557 | 7159 | .free = raid5_free, |
16a53ecc N |
7160 | .status = status, |
7161 | .error_handler = error, | |
7162 | .hot_add_disk = raid5_add_disk, | |
7163 | .hot_remove_disk= raid5_remove_disk, | |
7164 | .spare_active = raid5_spare_active, | |
7165 | .sync_request = sync_request, | |
7166 | .resize = raid5_resize, | |
80c3a6ce | 7167 | .size = raid5_size, |
50ac168a | 7168 | .check_reshape = raid6_check_reshape, |
f416885e | 7169 | .start_reshape = raid5_start_reshape, |
cea9c228 | 7170 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 7171 | .quiesce = raid5_quiesce, |
245f46c2 | 7172 | .takeover = raid6_takeover, |
5c675f83 | 7173 | .congested = raid5_congested, |
64590f45 | 7174 | .mergeable_bvec = raid5_mergeable_bvec, |
16a53ecc | 7175 | }; |
84fc4b56 | 7176 | static struct md_personality raid5_personality = |
1da177e4 LT |
7177 | { |
7178 | .name = "raid5", | |
2604b703 | 7179 | .level = 5, |
1da177e4 LT |
7180 | .owner = THIS_MODULE, |
7181 | .make_request = make_request, | |
7182 | .run = run, | |
afa0f557 | 7183 | .free = raid5_free, |
1da177e4 LT |
7184 | .status = status, |
7185 | .error_handler = error, | |
7186 | .hot_add_disk = raid5_add_disk, | |
7187 | .hot_remove_disk= raid5_remove_disk, | |
7188 | .spare_active = raid5_spare_active, | |
7189 | .sync_request = sync_request, | |
7190 | .resize = raid5_resize, | |
80c3a6ce | 7191 | .size = raid5_size, |
63c70c4f N |
7192 | .check_reshape = raid5_check_reshape, |
7193 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7194 | .finish_reshape = raid5_finish_reshape, |
72626685 | 7195 | .quiesce = raid5_quiesce, |
d562b0c4 | 7196 | .takeover = raid5_takeover, |
5c675f83 | 7197 | .congested = raid5_congested, |
64590f45 | 7198 | .mergeable_bvec = raid5_mergeable_bvec, |
1da177e4 LT |
7199 | }; |
7200 | ||
84fc4b56 | 7201 | static struct md_personality raid4_personality = |
1da177e4 | 7202 | { |
2604b703 N |
7203 | .name = "raid4", |
7204 | .level = 4, | |
7205 | .owner = THIS_MODULE, | |
7206 | .make_request = make_request, | |
7207 | .run = run, | |
afa0f557 | 7208 | .free = raid5_free, |
2604b703 N |
7209 | .status = status, |
7210 | .error_handler = error, | |
7211 | .hot_add_disk = raid5_add_disk, | |
7212 | .hot_remove_disk= raid5_remove_disk, | |
7213 | .spare_active = raid5_spare_active, | |
7214 | .sync_request = sync_request, | |
7215 | .resize = raid5_resize, | |
80c3a6ce | 7216 | .size = raid5_size, |
3d37890b N |
7217 | .check_reshape = raid5_check_reshape, |
7218 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7219 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 7220 | .quiesce = raid5_quiesce, |
a78d38a1 | 7221 | .takeover = raid4_takeover, |
5c675f83 | 7222 | .congested = raid5_congested, |
64590f45 | 7223 | .mergeable_bvec = raid5_mergeable_bvec, |
2604b703 N |
7224 | }; |
7225 | ||
7226 | static int __init raid5_init(void) | |
7227 | { | |
851c30c9 SL |
7228 | raid5_wq = alloc_workqueue("raid5wq", |
7229 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
7230 | if (!raid5_wq) | |
7231 | return -ENOMEM; | |
16a53ecc | 7232 | register_md_personality(&raid6_personality); |
2604b703 N |
7233 | register_md_personality(&raid5_personality); |
7234 | register_md_personality(&raid4_personality); | |
7235 | return 0; | |
1da177e4 LT |
7236 | } |
7237 | ||
2604b703 | 7238 | static void raid5_exit(void) |
1da177e4 | 7239 | { |
16a53ecc | 7240 | unregister_md_personality(&raid6_personality); |
2604b703 N |
7241 | unregister_md_personality(&raid5_personality); |
7242 | unregister_md_personality(&raid4_personality); | |
851c30c9 | 7243 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
7244 | } |
7245 | ||
7246 | module_init(raid5_init); | |
7247 | module_exit(raid5_exit); | |
7248 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7249 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 7250 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
7251 | MODULE_ALIAS("md-raid5"); |
7252 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
7253 | MODULE_ALIAS("md-level-5"); |
7254 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
7255 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
7256 | MODULE_ALIAS("md-raid6"); | |
7257 | MODULE_ALIAS("md-level-6"); | |
7258 | ||
7259 | /* This used to be two separate modules, they were: */ | |
7260 | MODULE_ALIAS("raid5"); | |
7261 | MODULE_ALIAS("raid6"); |