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); | |
b1b46486 YL |
377 | if (atomic_read(&conf->active_stripes) == 0) |
378 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
379 | if (conf->retry_read_aligned) |
380 | md_wakeup_thread(conf->mddev->thread); | |
381 | } | |
4eb788df SL |
382 | } |
383 | ||
773ca82f | 384 | /* should hold conf->device_lock already */ |
566c09c5 SL |
385 | static int release_stripe_list(struct r5conf *conf, |
386 | struct list_head *temp_inactive_list) | |
773ca82f SL |
387 | { |
388 | struct stripe_head *sh; | |
389 | int count = 0; | |
390 | struct llist_node *head; | |
391 | ||
392 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 393 | head = llist_reverse_order(head); |
773ca82f | 394 | while (head) { |
566c09c5 SL |
395 | int hash; |
396 | ||
773ca82f SL |
397 | sh = llist_entry(head, struct stripe_head, release_list); |
398 | head = llist_next(head); | |
399 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ | |
400 | smp_mb(); | |
401 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
402 | /* | |
403 | * Don't worry the bit is set here, because if the bit is set | |
404 | * again, the count is always > 1. This is true for | |
405 | * STRIPE_ON_UNPLUG_LIST bit too. | |
406 | */ | |
566c09c5 SL |
407 | hash = sh->hash_lock_index; |
408 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
409 | count++; |
410 | } | |
411 | ||
412 | return count; | |
413 | } | |
414 | ||
1da177e4 LT |
415 | static void release_stripe(struct stripe_head *sh) |
416 | { | |
d1688a6d | 417 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 418 | unsigned long flags; |
566c09c5 SL |
419 | struct list_head list; |
420 | int hash; | |
773ca82f | 421 | bool wakeup; |
16a53ecc | 422 | |
cf170f3f ES |
423 | /* Avoid release_list until the last reference. |
424 | */ | |
425 | if (atomic_add_unless(&sh->count, -1, 1)) | |
426 | return; | |
427 | ||
ad4068de | 428 | if (unlikely(!conf->mddev->thread) || |
429 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
430 | goto slow_path; |
431 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
432 | if (wakeup) | |
433 | md_wakeup_thread(conf->mddev->thread); | |
434 | return; | |
435 | slow_path: | |
4eb788df | 436 | local_irq_save(flags); |
773ca82f | 437 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 438 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
439 | INIT_LIST_HEAD(&list); |
440 | hash = sh->hash_lock_index; | |
441 | do_release_stripe(conf, sh, &list); | |
4eb788df | 442 | spin_unlock(&conf->device_lock); |
566c09c5 | 443 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
444 | } |
445 | local_irq_restore(flags); | |
1da177e4 LT |
446 | } |
447 | ||
fccddba0 | 448 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 449 | { |
45b4233c DW |
450 | pr_debug("remove_hash(), stripe %llu\n", |
451 | (unsigned long long)sh->sector); | |
1da177e4 | 452 | |
fccddba0 | 453 | hlist_del_init(&sh->hash); |
1da177e4 LT |
454 | } |
455 | ||
d1688a6d | 456 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 457 | { |
fccddba0 | 458 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 459 | |
45b4233c DW |
460 | pr_debug("insert_hash(), stripe %llu\n", |
461 | (unsigned long long)sh->sector); | |
1da177e4 | 462 | |
fccddba0 | 463 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
464 | } |
465 | ||
1da177e4 | 466 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 467 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
468 | { |
469 | struct stripe_head *sh = NULL; | |
470 | struct list_head *first; | |
471 | ||
566c09c5 | 472 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 473 | goto out; |
566c09c5 | 474 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
475 | sh = list_entry(first, struct stripe_head, lru); |
476 | list_del_init(first); | |
477 | remove_hash(sh); | |
478 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 479 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
480 | if (list_empty(conf->inactive_list + hash)) |
481 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
482 | out: |
483 | return sh; | |
484 | } | |
485 | ||
e4e11e38 | 486 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
487 | { |
488 | struct page *p; | |
489 | int i; | |
e4e11e38 | 490 | int num = sh->raid_conf->pool_size; |
1da177e4 | 491 | |
e4e11e38 | 492 | for (i = 0; i < num ; i++) { |
d592a996 | 493 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
494 | p = sh->dev[i].page; |
495 | if (!p) | |
496 | continue; | |
497 | sh->dev[i].page = NULL; | |
2d1f3b5d | 498 | put_page(p); |
1da177e4 LT |
499 | } |
500 | } | |
501 | ||
a9683a79 | 502 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
503 | { |
504 | int i; | |
e4e11e38 | 505 | int num = sh->raid_conf->pool_size; |
1da177e4 | 506 | |
e4e11e38 | 507 | for (i = 0; i < num; i++) { |
1da177e4 LT |
508 | struct page *page; |
509 | ||
a9683a79 | 510 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
511 | return 1; |
512 | } | |
513 | sh->dev[i].page = page; | |
d592a996 | 514 | sh->dev[i].orig_page = page; |
1da177e4 LT |
515 | } |
516 | return 0; | |
517 | } | |
518 | ||
784052ec | 519 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 520 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 521 | struct stripe_head *sh); |
1da177e4 | 522 | |
b5663ba4 | 523 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 524 | { |
d1688a6d | 525 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 526 | int i, seq; |
1da177e4 | 527 | |
78bafebd ES |
528 | BUG_ON(atomic_read(&sh->count) != 0); |
529 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 530 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 531 | BUG_ON(sh->batch_head); |
d84e0f10 | 532 | |
45b4233c | 533 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 534 | (unsigned long long)sector); |
566c09c5 SL |
535 | retry: |
536 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 537 | sh->generation = conf->generation - previous; |
b5663ba4 | 538 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 539 | sh->sector = sector; |
911d4ee8 | 540 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
541 | sh->state = 0; |
542 | ||
7ecaa1e6 | 543 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
544 | struct r5dev *dev = &sh->dev[i]; |
545 | ||
d84e0f10 | 546 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 547 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 548 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 549 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 550 | dev->read, dev->towrite, dev->written, |
1da177e4 | 551 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 552 | WARN_ON(1); |
1da177e4 LT |
553 | } |
554 | dev->flags = 0; | |
784052ec | 555 | raid5_build_block(sh, i, previous); |
1da177e4 | 556 | } |
566c09c5 SL |
557 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
558 | goto retry; | |
7a87f434 | 559 | sh->overwrite_disks = 0; |
1da177e4 | 560 | insert_hash(conf, sh); |
851c30c9 | 561 | sh->cpu = smp_processor_id(); |
da41ba65 | 562 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
563 | } |
564 | ||
d1688a6d | 565 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 566 | short generation) |
1da177e4 LT |
567 | { |
568 | struct stripe_head *sh; | |
569 | ||
45b4233c | 570 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 571 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 572 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 573 | return sh; |
45b4233c | 574 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
575 | return NULL; |
576 | } | |
577 | ||
674806d6 N |
578 | /* |
579 | * Need to check if array has failed when deciding whether to: | |
580 | * - start an array | |
581 | * - remove non-faulty devices | |
582 | * - add a spare | |
583 | * - allow a reshape | |
584 | * This determination is simple when no reshape is happening. | |
585 | * However if there is a reshape, we need to carefully check | |
586 | * both the before and after sections. | |
587 | * This is because some failed devices may only affect one | |
588 | * of the two sections, and some non-in_sync devices may | |
589 | * be insync in the section most affected by failed devices. | |
590 | */ | |
908f4fbd | 591 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 592 | { |
908f4fbd | 593 | int degraded, degraded2; |
674806d6 | 594 | int i; |
674806d6 N |
595 | |
596 | rcu_read_lock(); | |
597 | degraded = 0; | |
598 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 599 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
600 | if (rdev && test_bit(Faulty, &rdev->flags)) |
601 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
602 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
603 | degraded++; | |
604 | else if (test_bit(In_sync, &rdev->flags)) | |
605 | ; | |
606 | else | |
607 | /* not in-sync or faulty. | |
608 | * If the reshape increases the number of devices, | |
609 | * this is being recovered by the reshape, so | |
610 | * this 'previous' section is not in_sync. | |
611 | * If the number of devices is being reduced however, | |
612 | * the device can only be part of the array if | |
613 | * we are reverting a reshape, so this section will | |
614 | * be in-sync. | |
615 | */ | |
616 | if (conf->raid_disks >= conf->previous_raid_disks) | |
617 | degraded++; | |
618 | } | |
619 | rcu_read_unlock(); | |
908f4fbd N |
620 | if (conf->raid_disks == conf->previous_raid_disks) |
621 | return degraded; | |
674806d6 | 622 | rcu_read_lock(); |
908f4fbd | 623 | degraded2 = 0; |
674806d6 | 624 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 625 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
626 | if (rdev && test_bit(Faulty, &rdev->flags)) |
627 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 628 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 629 | degraded2++; |
674806d6 N |
630 | else if (test_bit(In_sync, &rdev->flags)) |
631 | ; | |
632 | else | |
633 | /* not in-sync or faulty. | |
634 | * If reshape increases the number of devices, this | |
635 | * section has already been recovered, else it | |
636 | * almost certainly hasn't. | |
637 | */ | |
638 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 639 | degraded2++; |
674806d6 N |
640 | } |
641 | rcu_read_unlock(); | |
908f4fbd N |
642 | if (degraded2 > degraded) |
643 | return degraded2; | |
644 | return degraded; | |
645 | } | |
646 | ||
647 | static int has_failed(struct r5conf *conf) | |
648 | { | |
649 | int degraded; | |
650 | ||
651 | if (conf->mddev->reshape_position == MaxSector) | |
652 | return conf->mddev->degraded > conf->max_degraded; | |
653 | ||
654 | degraded = calc_degraded(conf); | |
674806d6 N |
655 | if (degraded > conf->max_degraded) |
656 | return 1; | |
657 | return 0; | |
658 | } | |
659 | ||
b5663ba4 | 660 | static struct stripe_head * |
d1688a6d | 661 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 662 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
663 | { |
664 | struct stripe_head *sh; | |
566c09c5 | 665 | int hash = stripe_hash_locks_hash(sector); |
1da177e4 | 666 | |
45b4233c | 667 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 668 | |
566c09c5 | 669 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
670 | |
671 | do { | |
b1b46486 | 672 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 673 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 674 | *(conf->hash_locks + hash)); |
86b42c71 | 675 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 676 | if (!sh) { |
edbe83ab | 677 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 678 | sh = get_free_stripe(conf, hash); |
edbe83ab N |
679 | if (!sh && llist_empty(&conf->released_stripes) && |
680 | !test_bit(R5_DID_ALLOC, &conf->cache_state)) | |
681 | set_bit(R5_ALLOC_MORE, | |
682 | &conf->cache_state); | |
683 | } | |
1da177e4 LT |
684 | if (noblock && sh == NULL) |
685 | break; | |
686 | if (!sh) { | |
5423399a N |
687 | set_bit(R5_INACTIVE_BLOCKED, |
688 | &conf->cache_state); | |
566c09c5 SL |
689 | wait_event_lock_irq( |
690 | conf->wait_for_stripe, | |
691 | !list_empty(conf->inactive_list + hash) && | |
692 | (atomic_read(&conf->active_stripes) | |
693 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
694 | || !test_bit(R5_INACTIVE_BLOCKED, |
695 | &conf->cache_state)), | |
566c09c5 | 696 | *(conf->hash_locks + hash)); |
5423399a N |
697 | clear_bit(R5_INACTIVE_BLOCKED, |
698 | &conf->cache_state); | |
7da9d450 | 699 | } else { |
b5663ba4 | 700 | init_stripe(sh, sector, previous); |
7da9d450 N |
701 | atomic_inc(&sh->count); |
702 | } | |
e240c183 | 703 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 704 | spin_lock(&conf->device_lock); |
e240c183 | 705 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
706 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
707 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
708 | BUG_ON(list_empty(&sh->lru) && |
709 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
16a53ecc | 710 | list_del_init(&sh->lru); |
bfc90cb0 SL |
711 | if (sh->group) { |
712 | sh->group->stripes_cnt--; | |
713 | sh->group = NULL; | |
714 | } | |
1da177e4 | 715 | } |
7da9d450 | 716 | atomic_inc(&sh->count); |
6d183de4 | 717 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
718 | } |
719 | } while (sh == NULL); | |
720 | ||
566c09c5 | 721 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
722 | return sh; |
723 | } | |
724 | ||
7a87f434 | 725 | static bool is_full_stripe_write(struct stripe_head *sh) |
726 | { | |
727 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
728 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
729 | } | |
730 | ||
59fc630b | 731 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
732 | { | |
733 | local_irq_disable(); | |
734 | if (sh1 > sh2) { | |
735 | spin_lock(&sh2->stripe_lock); | |
736 | spin_lock_nested(&sh1->stripe_lock, 1); | |
737 | } else { | |
738 | spin_lock(&sh1->stripe_lock); | |
739 | spin_lock_nested(&sh2->stripe_lock, 1); | |
740 | } | |
741 | } | |
742 | ||
743 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
744 | { | |
745 | spin_unlock(&sh1->stripe_lock); | |
746 | spin_unlock(&sh2->stripe_lock); | |
747 | local_irq_enable(); | |
748 | } | |
749 | ||
750 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
751 | static bool stripe_can_batch(struct stripe_head *sh) | |
752 | { | |
753 | return test_bit(STRIPE_BATCH_READY, &sh->state) && | |
d0852df5 | 754 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 755 | is_full_stripe_write(sh); |
756 | } | |
757 | ||
758 | /* we only do back search */ | |
759 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
760 | { | |
761 | struct stripe_head *head; | |
762 | sector_t head_sector, tmp_sec; | |
763 | int hash; | |
764 | int dd_idx; | |
765 | ||
766 | if (!stripe_can_batch(sh)) | |
767 | return; | |
768 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ | |
769 | tmp_sec = sh->sector; | |
770 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
771 | return; | |
772 | head_sector = sh->sector - STRIPE_SECTORS; | |
773 | ||
774 | hash = stripe_hash_locks_hash(head_sector); | |
775 | spin_lock_irq(conf->hash_locks + hash); | |
776 | head = __find_stripe(conf, head_sector, conf->generation); | |
777 | if (head && !atomic_inc_not_zero(&head->count)) { | |
778 | spin_lock(&conf->device_lock); | |
779 | if (!atomic_read(&head->count)) { | |
780 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
781 | atomic_inc(&conf->active_stripes); | |
782 | BUG_ON(list_empty(&head->lru) && | |
783 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
784 | list_del_init(&head->lru); | |
785 | if (head->group) { | |
786 | head->group->stripes_cnt--; | |
787 | head->group = NULL; | |
788 | } | |
789 | } | |
790 | atomic_inc(&head->count); | |
791 | spin_unlock(&conf->device_lock); | |
792 | } | |
793 | spin_unlock_irq(conf->hash_locks + hash); | |
794 | ||
795 | if (!head) | |
796 | return; | |
797 | if (!stripe_can_batch(head)) | |
798 | goto out; | |
799 | ||
800 | lock_two_stripes(head, sh); | |
801 | /* clear_batch_ready clear the flag */ | |
802 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
803 | goto unlock_out; | |
804 | ||
805 | if (sh->batch_head) | |
806 | goto unlock_out; | |
807 | ||
808 | dd_idx = 0; | |
809 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
810 | dd_idx++; | |
811 | if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw) | |
812 | goto unlock_out; | |
813 | ||
814 | if (head->batch_head) { | |
815 | spin_lock(&head->batch_head->batch_lock); | |
816 | /* This batch list is already running */ | |
817 | if (!stripe_can_batch(head)) { | |
818 | spin_unlock(&head->batch_head->batch_lock); | |
819 | goto unlock_out; | |
820 | } | |
821 | ||
822 | /* | |
823 | * at this point, head's BATCH_READY could be cleared, but we | |
824 | * can still add the stripe to batch list | |
825 | */ | |
826 | list_add(&sh->batch_list, &head->batch_list); | |
827 | spin_unlock(&head->batch_head->batch_lock); | |
828 | ||
829 | sh->batch_head = head->batch_head; | |
830 | } else { | |
831 | head->batch_head = head; | |
832 | sh->batch_head = head->batch_head; | |
833 | spin_lock(&head->batch_lock); | |
834 | list_add_tail(&sh->batch_list, &head->batch_list); | |
835 | spin_unlock(&head->batch_lock); | |
836 | } | |
837 | ||
838 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
839 | if (atomic_dec_return(&conf->preread_active_stripes) | |
840 | < IO_THRESHOLD) | |
841 | md_wakeup_thread(conf->mddev->thread); | |
842 | ||
2b6b2457 N |
843 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
844 | int seq = sh->bm_seq; | |
845 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
846 | sh->batch_head->bm_seq > seq) | |
847 | seq = sh->batch_head->bm_seq; | |
848 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
849 | sh->batch_head->bm_seq = seq; | |
850 | } | |
851 | ||
59fc630b | 852 | atomic_inc(&sh->count); |
853 | unlock_out: | |
854 | unlock_two_stripes(head, sh); | |
855 | out: | |
856 | release_stripe(head); | |
857 | } | |
858 | ||
05616be5 N |
859 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
860 | * in this stripe_head. | |
861 | */ | |
862 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
863 | { | |
864 | sector_t progress = conf->reshape_progress; | |
865 | /* Need a memory barrier to make sure we see the value | |
866 | * of conf->generation, or ->data_offset that was set before | |
867 | * reshape_progress was updated. | |
868 | */ | |
869 | smp_rmb(); | |
870 | if (progress == MaxSector) | |
871 | return 0; | |
872 | if (sh->generation == conf->generation - 1) | |
873 | return 0; | |
874 | /* We are in a reshape, and this is a new-generation stripe, | |
875 | * so use new_data_offset. | |
876 | */ | |
877 | return 1; | |
878 | } | |
879 | ||
6712ecf8 N |
880 | static void |
881 | raid5_end_read_request(struct bio *bi, int error); | |
882 | static void | |
883 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 884 | |
c4e5ac0a | 885 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 886 | { |
d1688a6d | 887 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 888 | int i, disks = sh->disks; |
59fc630b | 889 | struct stripe_head *head_sh = sh; |
91c00924 DW |
890 | |
891 | might_sleep(); | |
892 | ||
893 | for (i = disks; i--; ) { | |
894 | int rw; | |
9a3e1101 | 895 | int replace_only = 0; |
977df362 N |
896 | struct bio *bi, *rbi; |
897 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 898 | |
899 | sh = head_sh; | |
e9c7469b TH |
900 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
901 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
902 | rw = WRITE_FUA; | |
903 | else | |
904 | rw = WRITE; | |
9e444768 | 905 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
620125f2 | 906 | rw |= REQ_DISCARD; |
e9c7469b | 907 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
91c00924 | 908 | rw = READ; |
9a3e1101 N |
909 | else if (test_and_clear_bit(R5_WantReplace, |
910 | &sh->dev[i].flags)) { | |
911 | rw = WRITE; | |
912 | replace_only = 1; | |
913 | } else | |
91c00924 | 914 | continue; |
bc0934f0 SL |
915 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
916 | rw |= REQ_SYNC; | |
91c00924 | 917 | |
59fc630b | 918 | again: |
91c00924 | 919 | bi = &sh->dev[i].req; |
977df362 | 920 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 921 | |
91c00924 | 922 | rcu_read_lock(); |
9a3e1101 | 923 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
924 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
925 | rdev = rcu_dereference(conf->disks[i].rdev); | |
926 | if (!rdev) { | |
927 | rdev = rrdev; | |
928 | rrdev = NULL; | |
929 | } | |
9a3e1101 N |
930 | if (rw & WRITE) { |
931 | if (replace_only) | |
932 | rdev = NULL; | |
dd054fce N |
933 | if (rdev == rrdev) |
934 | /* We raced and saw duplicates */ | |
935 | rrdev = NULL; | |
9a3e1101 | 936 | } else { |
59fc630b | 937 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
938 | rdev = rrdev; |
939 | rrdev = NULL; | |
940 | } | |
977df362 | 941 | |
91c00924 DW |
942 | if (rdev && test_bit(Faulty, &rdev->flags)) |
943 | rdev = NULL; | |
944 | if (rdev) | |
945 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
946 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
947 | rrdev = NULL; | |
948 | if (rrdev) | |
949 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
950 | rcu_read_unlock(); |
951 | ||
73e92e51 | 952 | /* We have already checked bad blocks for reads. Now |
977df362 N |
953 | * need to check for writes. We never accept write errors |
954 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
955 | */ |
956 | while ((rw & WRITE) && rdev && | |
957 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
958 | sector_t first_bad; | |
959 | int bad_sectors; | |
960 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
961 | &first_bad, &bad_sectors); | |
962 | if (!bad) | |
963 | break; | |
964 | ||
965 | if (bad < 0) { | |
966 | set_bit(BlockedBadBlocks, &rdev->flags); | |
967 | if (!conf->mddev->external && | |
968 | conf->mddev->flags) { | |
969 | /* It is very unlikely, but we might | |
970 | * still need to write out the | |
971 | * bad block log - better give it | |
972 | * a chance*/ | |
973 | md_check_recovery(conf->mddev); | |
974 | } | |
1850753d | 975 | /* |
976 | * Because md_wait_for_blocked_rdev | |
977 | * will dec nr_pending, we must | |
978 | * increment it first. | |
979 | */ | |
980 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
981 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
982 | } else { | |
983 | /* Acknowledged bad block - skip the write */ | |
984 | rdev_dec_pending(rdev, conf->mddev); | |
985 | rdev = NULL; | |
986 | } | |
987 | } | |
988 | ||
91c00924 | 989 | if (rdev) { |
9a3e1101 N |
990 | if (s->syncing || s->expanding || s->expanded |
991 | || s->replacing) | |
91c00924 DW |
992 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
993 | ||
2b7497f0 DW |
994 | set_bit(STRIPE_IO_STARTED, &sh->state); |
995 | ||
2f6db2a7 | 996 | bio_reset(bi); |
91c00924 | 997 | bi->bi_bdev = rdev->bdev; |
2f6db2a7 KO |
998 | bi->bi_rw = rw; |
999 | bi->bi_end_io = (rw & WRITE) | |
1000 | ? raid5_end_write_request | |
1001 | : raid5_end_read_request; | |
1002 | bi->bi_private = sh; | |
1003 | ||
91c00924 | 1004 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", |
e46b272b | 1005 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
1006 | bi->bi_rw, i); |
1007 | atomic_inc(&sh->count); | |
59fc630b | 1008 | if (sh != head_sh) |
1009 | atomic_inc(&head_sh->count); | |
05616be5 | 1010 | if (use_new_offset(conf, sh)) |
4f024f37 | 1011 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1012 | + rdev->new_data_offset); |
1013 | else | |
4f024f37 | 1014 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1015 | + rdev->data_offset); |
59fc630b | 1016 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
e59aa23f | 1017 | bi->bi_rw |= REQ_NOMERGE; |
3f9e7c14 | 1018 | |
d592a996 SL |
1019 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1020 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1021 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1022 | bi->bi_vcnt = 1; |
91c00924 DW |
1023 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1024 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1025 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1026 | /* |
1027 | * If this is discard request, set bi_vcnt 0. We don't | |
1028 | * want to confuse SCSI because SCSI will replace payload | |
1029 | */ | |
1030 | if (rw & REQ_DISCARD) | |
1031 | bi->bi_vcnt = 0; | |
977df362 N |
1032 | if (rrdev) |
1033 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1034 | |
1035 | if (conf->mddev->gendisk) | |
1036 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1037 | bi, disk_devt(conf->mddev->gendisk), | |
1038 | sh->dev[i].sector); | |
91c00924 | 1039 | generic_make_request(bi); |
977df362 N |
1040 | } |
1041 | if (rrdev) { | |
9a3e1101 N |
1042 | if (s->syncing || s->expanding || s->expanded |
1043 | || s->replacing) | |
977df362 N |
1044 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1045 | ||
1046 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1047 | ||
2f6db2a7 | 1048 | bio_reset(rbi); |
977df362 | 1049 | rbi->bi_bdev = rrdev->bdev; |
2f6db2a7 KO |
1050 | rbi->bi_rw = rw; |
1051 | BUG_ON(!(rw & WRITE)); | |
1052 | rbi->bi_end_io = raid5_end_write_request; | |
1053 | rbi->bi_private = sh; | |
1054 | ||
977df362 N |
1055 | pr_debug("%s: for %llu schedule op %ld on " |
1056 | "replacement disc %d\n", | |
1057 | __func__, (unsigned long long)sh->sector, | |
1058 | rbi->bi_rw, i); | |
1059 | atomic_inc(&sh->count); | |
59fc630b | 1060 | if (sh != head_sh) |
1061 | atomic_inc(&head_sh->count); | |
05616be5 | 1062 | if (use_new_offset(conf, sh)) |
4f024f37 | 1063 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1064 | + rrdev->new_data_offset); |
1065 | else | |
4f024f37 | 1066 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1067 | + rrdev->data_offset); |
d592a996 SL |
1068 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1069 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1070 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1071 | rbi->bi_vcnt = 1; |
977df362 N |
1072 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1073 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1074 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1075 | /* |
1076 | * If this is discard request, set bi_vcnt 0. We don't | |
1077 | * want to confuse SCSI because SCSI will replace payload | |
1078 | */ | |
1079 | if (rw & REQ_DISCARD) | |
1080 | rbi->bi_vcnt = 0; | |
e3620a3a JB |
1081 | if (conf->mddev->gendisk) |
1082 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1083 | rbi, disk_devt(conf->mddev->gendisk), | |
1084 | sh->dev[i].sector); | |
977df362 N |
1085 | generic_make_request(rbi); |
1086 | } | |
1087 | if (!rdev && !rrdev) { | |
b062962e | 1088 | if (rw & WRITE) |
91c00924 DW |
1089 | set_bit(STRIPE_DEGRADED, &sh->state); |
1090 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
1091 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
1092 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1093 | set_bit(STRIPE_HANDLE, &sh->state); | |
1094 | } | |
59fc630b | 1095 | |
1096 | if (!head_sh->batch_head) | |
1097 | continue; | |
1098 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1099 | batch_list); | |
1100 | if (sh != head_sh) | |
1101 | goto again; | |
91c00924 DW |
1102 | } |
1103 | } | |
1104 | ||
1105 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1106 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1107 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1108 | struct stripe_head *sh) | |
91c00924 | 1109 | { |
7988613b KO |
1110 | struct bio_vec bvl; |
1111 | struct bvec_iter iter; | |
91c00924 | 1112 | struct page *bio_page; |
91c00924 | 1113 | int page_offset; |
a08abd8c | 1114 | struct async_submit_ctl submit; |
0403e382 | 1115 | enum async_tx_flags flags = 0; |
91c00924 | 1116 | |
4f024f37 KO |
1117 | if (bio->bi_iter.bi_sector >= sector) |
1118 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1119 | else |
4f024f37 | 1120 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1121 | |
0403e382 DW |
1122 | if (frombio) |
1123 | flags |= ASYNC_TX_FENCE; | |
1124 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1125 | ||
7988613b KO |
1126 | bio_for_each_segment(bvl, bio, iter) { |
1127 | int len = bvl.bv_len; | |
91c00924 DW |
1128 | int clen; |
1129 | int b_offset = 0; | |
1130 | ||
1131 | if (page_offset < 0) { | |
1132 | b_offset = -page_offset; | |
1133 | page_offset += b_offset; | |
1134 | len -= b_offset; | |
1135 | } | |
1136 | ||
1137 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1138 | clen = STRIPE_SIZE - page_offset; | |
1139 | else | |
1140 | clen = len; | |
1141 | ||
1142 | if (clen > 0) { | |
7988613b KO |
1143 | b_offset += bvl.bv_offset; |
1144 | bio_page = bvl.bv_page; | |
d592a996 SL |
1145 | if (frombio) { |
1146 | if (sh->raid_conf->skip_copy && | |
1147 | b_offset == 0 && page_offset == 0 && | |
1148 | clen == STRIPE_SIZE) | |
1149 | *page = bio_page; | |
1150 | else | |
1151 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1152 | b_offset, clen, &submit); |
d592a996 SL |
1153 | } else |
1154 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1155 | page_offset, clen, &submit); |
91c00924 | 1156 | } |
a08abd8c DW |
1157 | /* chain the operations */ |
1158 | submit.depend_tx = tx; | |
1159 | ||
91c00924 DW |
1160 | if (clen < len) /* hit end of page */ |
1161 | break; | |
1162 | page_offset += len; | |
1163 | } | |
1164 | ||
1165 | return tx; | |
1166 | } | |
1167 | ||
1168 | static void ops_complete_biofill(void *stripe_head_ref) | |
1169 | { | |
1170 | struct stripe_head *sh = stripe_head_ref; | |
1171 | struct bio *return_bi = NULL; | |
e4d84909 | 1172 | int i; |
91c00924 | 1173 | |
e46b272b | 1174 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1175 | (unsigned long long)sh->sector); |
1176 | ||
1177 | /* clear completed biofills */ | |
1178 | for (i = sh->disks; i--; ) { | |
1179 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1180 | |
1181 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1182 | /* and check if we need to reply to a read request, |
1183 | * new R5_Wantfill requests are held off until | |
83de75cc | 1184 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1185 | */ |
1186 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1187 | struct bio *rbi, *rbi2; |
91c00924 | 1188 | |
91c00924 DW |
1189 | BUG_ON(!dev->read); |
1190 | rbi = dev->read; | |
1191 | dev->read = NULL; | |
4f024f37 | 1192 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1193 | dev->sector + STRIPE_SECTORS) { |
1194 | rbi2 = r5_next_bio(rbi, dev->sector); | |
e7836bd6 | 1195 | if (!raid5_dec_bi_active_stripes(rbi)) { |
91c00924 DW |
1196 | rbi->bi_next = return_bi; |
1197 | return_bi = rbi; | |
1198 | } | |
91c00924 DW |
1199 | rbi = rbi2; |
1200 | } | |
1201 | } | |
1202 | } | |
83de75cc | 1203 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 DW |
1204 | |
1205 | return_io(return_bi); | |
1206 | ||
e4d84909 | 1207 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
1208 | release_stripe(sh); |
1209 | } | |
1210 | ||
1211 | static void ops_run_biofill(struct stripe_head *sh) | |
1212 | { | |
1213 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1214 | struct async_submit_ctl submit; |
91c00924 DW |
1215 | int i; |
1216 | ||
59fc630b | 1217 | BUG_ON(sh->batch_head); |
e46b272b | 1218 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1219 | (unsigned long long)sh->sector); |
1220 | ||
1221 | for (i = sh->disks; i--; ) { | |
1222 | struct r5dev *dev = &sh->dev[i]; | |
1223 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1224 | struct bio *rbi; | |
b17459c0 | 1225 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1226 | dev->read = rbi = dev->toread; |
1227 | dev->toread = NULL; | |
b17459c0 | 1228 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1229 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1230 | dev->sector + STRIPE_SECTORS) { |
d592a996 SL |
1231 | tx = async_copy_data(0, rbi, &dev->page, |
1232 | dev->sector, tx, sh); | |
91c00924 DW |
1233 | rbi = r5_next_bio(rbi, dev->sector); |
1234 | } | |
1235 | } | |
1236 | } | |
1237 | ||
1238 | atomic_inc(&sh->count); | |
a08abd8c DW |
1239 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1240 | async_trigger_callback(&submit); | |
91c00924 DW |
1241 | } |
1242 | ||
4e7d2c0a | 1243 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1244 | { |
4e7d2c0a | 1245 | struct r5dev *tgt; |
91c00924 | 1246 | |
4e7d2c0a DW |
1247 | if (target < 0) |
1248 | return; | |
91c00924 | 1249 | |
4e7d2c0a | 1250 | tgt = &sh->dev[target]; |
91c00924 DW |
1251 | set_bit(R5_UPTODATE, &tgt->flags); |
1252 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1253 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1254 | } |
1255 | ||
ac6b53b6 | 1256 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1257 | { |
1258 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1259 | |
e46b272b | 1260 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1261 | (unsigned long long)sh->sector); |
1262 | ||
ac6b53b6 | 1263 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1264 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1265 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1266 | |
ecc65c9b DW |
1267 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1268 | if (sh->check_state == check_state_compute_run) | |
1269 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
1270 | set_bit(STRIPE_HANDLE, &sh->state); |
1271 | release_stripe(sh); | |
1272 | } | |
1273 | ||
d6f38f31 DW |
1274 | /* return a pointer to the address conversion region of the scribble buffer */ |
1275 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1276 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1277 | { |
46d5b785 | 1278 | void *addr; |
1279 | ||
1280 | addr = flex_array_get(percpu->scribble, i); | |
1281 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1282 | } | |
1283 | ||
1284 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1285 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1286 | { | |
1287 | void *addr; | |
1288 | ||
1289 | addr = flex_array_get(percpu->scribble, i); | |
1290 | return addr; | |
d6f38f31 DW |
1291 | } |
1292 | ||
1293 | static struct dma_async_tx_descriptor * | |
1294 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1295 | { |
91c00924 | 1296 | int disks = sh->disks; |
46d5b785 | 1297 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1298 | int target = sh->ops.target; |
1299 | struct r5dev *tgt = &sh->dev[target]; | |
1300 | struct page *xor_dest = tgt->page; | |
1301 | int count = 0; | |
1302 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1303 | struct async_submit_ctl submit; |
91c00924 DW |
1304 | int i; |
1305 | ||
59fc630b | 1306 | BUG_ON(sh->batch_head); |
1307 | ||
91c00924 | 1308 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1309 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1310 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1311 | ||
1312 | for (i = disks; i--; ) | |
1313 | if (i != target) | |
1314 | xor_srcs[count++] = sh->dev[i].page; | |
1315 | ||
1316 | atomic_inc(&sh->count); | |
1317 | ||
0403e382 | 1318 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1319 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1320 | if (unlikely(count == 1)) |
a08abd8c | 1321 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1322 | else |
a08abd8c | 1323 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1324 | |
91c00924 DW |
1325 | return tx; |
1326 | } | |
1327 | ||
ac6b53b6 DW |
1328 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1329 | * @srcs - (struct page *) array of size sh->disks | |
1330 | * @sh - stripe_head to parse | |
1331 | * | |
1332 | * Populates srcs in proper layout order for the stripe and returns the | |
1333 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1334 | * destination buffer is recorded in srcs[count] and the Q destination | |
1335 | * is recorded in srcs[count+1]]. | |
1336 | */ | |
584acdd4 MS |
1337 | static int set_syndrome_sources(struct page **srcs, |
1338 | struct stripe_head *sh, | |
1339 | int srctype) | |
ac6b53b6 DW |
1340 | { |
1341 | int disks = sh->disks; | |
1342 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1343 | int d0_idx = raid6_d0(sh); | |
1344 | int count; | |
1345 | int i; | |
1346 | ||
1347 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1348 | srcs[i] = NULL; |
ac6b53b6 DW |
1349 | |
1350 | count = 0; | |
1351 | i = d0_idx; | |
1352 | do { | |
1353 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1354 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1355 | |
584acdd4 MS |
1356 | if (i == sh->qd_idx || i == sh->pd_idx || |
1357 | (srctype == SYNDROME_SRC_ALL) || | |
1358 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1359 | test_bit(R5_Wantdrain, &dev->flags)) || | |
1360 | (srctype == SYNDROME_SRC_WRITTEN && | |
1361 | dev->written)) | |
1362 | srcs[slot] = sh->dev[i].page; | |
ac6b53b6 DW |
1363 | i = raid6_next_disk(i, disks); |
1364 | } while (i != d0_idx); | |
ac6b53b6 | 1365 | |
e4424fee | 1366 | return syndrome_disks; |
ac6b53b6 DW |
1367 | } |
1368 | ||
1369 | static struct dma_async_tx_descriptor * | |
1370 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1371 | { | |
1372 | int disks = sh->disks; | |
46d5b785 | 1373 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1374 | int target; |
1375 | int qd_idx = sh->qd_idx; | |
1376 | struct dma_async_tx_descriptor *tx; | |
1377 | struct async_submit_ctl submit; | |
1378 | struct r5dev *tgt; | |
1379 | struct page *dest; | |
1380 | int i; | |
1381 | int count; | |
1382 | ||
59fc630b | 1383 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1384 | if (sh->ops.target < 0) |
1385 | target = sh->ops.target2; | |
1386 | else if (sh->ops.target2 < 0) | |
1387 | target = sh->ops.target; | |
91c00924 | 1388 | else |
ac6b53b6 DW |
1389 | /* we should only have one valid target */ |
1390 | BUG(); | |
1391 | BUG_ON(target < 0); | |
1392 | pr_debug("%s: stripe %llu block: %d\n", | |
1393 | __func__, (unsigned long long)sh->sector, target); | |
1394 | ||
1395 | tgt = &sh->dev[target]; | |
1396 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1397 | dest = tgt->page; | |
1398 | ||
1399 | atomic_inc(&sh->count); | |
1400 | ||
1401 | if (target == qd_idx) { | |
584acdd4 | 1402 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1403 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1404 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1405 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1406 | ops_complete_compute, sh, | |
46d5b785 | 1407 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1408 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1409 | } else { | |
1410 | /* Compute any data- or p-drive using XOR */ | |
1411 | count = 0; | |
1412 | for (i = disks; i-- ; ) { | |
1413 | if (i == target || i == qd_idx) | |
1414 | continue; | |
1415 | blocks[count++] = sh->dev[i].page; | |
1416 | } | |
1417 | ||
0403e382 DW |
1418 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1419 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1420 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1421 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1422 | } | |
91c00924 | 1423 | |
91c00924 DW |
1424 | return tx; |
1425 | } | |
1426 | ||
ac6b53b6 DW |
1427 | static struct dma_async_tx_descriptor * |
1428 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1429 | { | |
1430 | int i, count, disks = sh->disks; | |
1431 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1432 | int d0_idx = raid6_d0(sh); | |
1433 | int faila = -1, failb = -1; | |
1434 | int target = sh->ops.target; | |
1435 | int target2 = sh->ops.target2; | |
1436 | struct r5dev *tgt = &sh->dev[target]; | |
1437 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1438 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1439 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1440 | struct async_submit_ctl submit; |
1441 | ||
59fc630b | 1442 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1443 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1444 | __func__, (unsigned long long)sh->sector, target, target2); | |
1445 | BUG_ON(target < 0 || target2 < 0); | |
1446 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1447 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1448 | ||
6c910a78 | 1449 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1450 | * slot number conversion for 'faila' and 'failb' |
1451 | */ | |
1452 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1453 | blocks[i] = NULL; |
ac6b53b6 DW |
1454 | count = 0; |
1455 | i = d0_idx; | |
1456 | do { | |
1457 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1458 | ||
1459 | blocks[slot] = sh->dev[i].page; | |
1460 | ||
1461 | if (i == target) | |
1462 | faila = slot; | |
1463 | if (i == target2) | |
1464 | failb = slot; | |
1465 | i = raid6_next_disk(i, disks); | |
1466 | } while (i != d0_idx); | |
ac6b53b6 DW |
1467 | |
1468 | BUG_ON(faila == failb); | |
1469 | if (failb < faila) | |
1470 | swap(faila, failb); | |
1471 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1472 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1473 | ||
1474 | atomic_inc(&sh->count); | |
1475 | ||
1476 | if (failb == syndrome_disks+1) { | |
1477 | /* Q disk is one of the missing disks */ | |
1478 | if (faila == syndrome_disks) { | |
1479 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1480 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1481 | ops_complete_compute, sh, | |
46d5b785 | 1482 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1483 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1484 | STRIPE_SIZE, &submit); |
1485 | } else { | |
1486 | struct page *dest; | |
1487 | int data_target; | |
1488 | int qd_idx = sh->qd_idx; | |
1489 | ||
1490 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1491 | if (target == qd_idx) | |
1492 | data_target = target2; | |
1493 | else | |
1494 | data_target = target; | |
1495 | ||
1496 | count = 0; | |
1497 | for (i = disks; i-- ; ) { | |
1498 | if (i == data_target || i == qd_idx) | |
1499 | continue; | |
1500 | blocks[count++] = sh->dev[i].page; | |
1501 | } | |
1502 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1503 | init_async_submit(&submit, |
1504 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1505 | NULL, NULL, NULL, | |
46d5b785 | 1506 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1507 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1508 | &submit); | |
1509 | ||
584acdd4 | 1510 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1511 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1512 | ops_complete_compute, sh, | |
46d5b785 | 1513 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1514 | return async_gen_syndrome(blocks, 0, count+2, |
1515 | STRIPE_SIZE, &submit); | |
1516 | } | |
ac6b53b6 | 1517 | } else { |
6c910a78 DW |
1518 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1519 | ops_complete_compute, sh, | |
46d5b785 | 1520 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1521 | if (failb == syndrome_disks) { |
1522 | /* We're missing D+P. */ | |
1523 | return async_raid6_datap_recov(syndrome_disks+2, | |
1524 | STRIPE_SIZE, faila, | |
1525 | blocks, &submit); | |
1526 | } else { | |
1527 | /* We're missing D+D. */ | |
1528 | return async_raid6_2data_recov(syndrome_disks+2, | |
1529 | STRIPE_SIZE, faila, failb, | |
1530 | blocks, &submit); | |
1531 | } | |
ac6b53b6 DW |
1532 | } |
1533 | } | |
1534 | ||
91c00924 DW |
1535 | static void ops_complete_prexor(void *stripe_head_ref) |
1536 | { | |
1537 | struct stripe_head *sh = stripe_head_ref; | |
1538 | ||
e46b272b | 1539 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1540 | (unsigned long long)sh->sector); |
91c00924 DW |
1541 | } |
1542 | ||
1543 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1544 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1545 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1546 | { |
91c00924 | 1547 | int disks = sh->disks; |
46d5b785 | 1548 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1549 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1550 | struct async_submit_ctl submit; |
91c00924 DW |
1551 | |
1552 | /* existing parity data subtracted */ | |
1553 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1554 | ||
59fc630b | 1555 | BUG_ON(sh->batch_head); |
e46b272b | 1556 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1557 | (unsigned long long)sh->sector); |
1558 | ||
1559 | for (i = disks; i--; ) { | |
1560 | struct r5dev *dev = &sh->dev[i]; | |
1561 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1562 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1563 | xor_srcs[count++] = dev->page; |
1564 | } | |
1565 | ||
0403e382 | 1566 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1567 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1568 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1569 | |
1570 | return tx; | |
1571 | } | |
1572 | ||
584acdd4 MS |
1573 | static struct dma_async_tx_descriptor * |
1574 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1575 | struct dma_async_tx_descriptor *tx) | |
1576 | { | |
1577 | struct page **blocks = to_addr_page(percpu, 0); | |
1578 | int count; | |
1579 | struct async_submit_ctl submit; | |
1580 | ||
1581 | pr_debug("%s: stripe %llu\n", __func__, | |
1582 | (unsigned long long)sh->sector); | |
1583 | ||
1584 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1585 | ||
1586 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1587 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1588 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1589 | ||
1590 | return tx; | |
1591 | } | |
1592 | ||
91c00924 | 1593 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1594 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1595 | { |
1596 | int disks = sh->disks; | |
d8ee0728 | 1597 | int i; |
59fc630b | 1598 | struct stripe_head *head_sh = sh; |
91c00924 | 1599 | |
e46b272b | 1600 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1601 | (unsigned long long)sh->sector); |
1602 | ||
1603 | for (i = disks; i--; ) { | |
59fc630b | 1604 | struct r5dev *dev; |
91c00924 | 1605 | struct bio *chosen; |
91c00924 | 1606 | |
59fc630b | 1607 | sh = head_sh; |
1608 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1609 | struct bio *wbi; |
1610 | ||
59fc630b | 1611 | again: |
1612 | dev = &sh->dev[i]; | |
b17459c0 | 1613 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1614 | chosen = dev->towrite; |
1615 | dev->towrite = NULL; | |
7a87f434 | 1616 | sh->overwrite_disks = 0; |
91c00924 DW |
1617 | BUG_ON(dev->written); |
1618 | wbi = dev->written = chosen; | |
b17459c0 | 1619 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1620 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1621 | |
4f024f37 | 1622 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1623 | dev->sector + STRIPE_SECTORS) { |
e9c7469b TH |
1624 | if (wbi->bi_rw & REQ_FUA) |
1625 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1626 | if (wbi->bi_rw & REQ_SYNC) |
1627 | set_bit(R5_SyncIO, &dev->flags); | |
9e444768 | 1628 | if (wbi->bi_rw & REQ_DISCARD) |
620125f2 | 1629 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1630 | else { |
1631 | tx = async_copy_data(1, wbi, &dev->page, | |
1632 | dev->sector, tx, sh); | |
1633 | if (dev->page != dev->orig_page) { | |
1634 | set_bit(R5_SkipCopy, &dev->flags); | |
1635 | clear_bit(R5_UPTODATE, &dev->flags); | |
1636 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1637 | } | |
1638 | } | |
91c00924 DW |
1639 | wbi = r5_next_bio(wbi, dev->sector); |
1640 | } | |
59fc630b | 1641 | |
1642 | if (head_sh->batch_head) { | |
1643 | sh = list_first_entry(&sh->batch_list, | |
1644 | struct stripe_head, | |
1645 | batch_list); | |
1646 | if (sh == head_sh) | |
1647 | continue; | |
1648 | goto again; | |
1649 | } | |
91c00924 DW |
1650 | } |
1651 | } | |
1652 | ||
1653 | return tx; | |
1654 | } | |
1655 | ||
ac6b53b6 | 1656 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1657 | { |
1658 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1659 | int disks = sh->disks; |
1660 | int pd_idx = sh->pd_idx; | |
1661 | int qd_idx = sh->qd_idx; | |
1662 | int i; | |
9e444768 | 1663 | bool fua = false, sync = false, discard = false; |
91c00924 | 1664 | |
e46b272b | 1665 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1666 | (unsigned long long)sh->sector); |
1667 | ||
bc0934f0 | 1668 | for (i = disks; i--; ) { |
e9c7469b | 1669 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1670 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1671 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1672 | } |
e9c7469b | 1673 | |
91c00924 DW |
1674 | for (i = disks; i--; ) { |
1675 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1676 | |
e9c7469b | 1677 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1678 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1679 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1680 | if (fua) |
1681 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1682 | if (sync) |
1683 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1684 | } |
91c00924 DW |
1685 | } |
1686 | ||
d8ee0728 DW |
1687 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1688 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1689 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1690 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1691 | else { | |
1692 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1693 | sh->reconstruct_state = reconstruct_state_result; | |
1694 | } | |
91c00924 DW |
1695 | |
1696 | set_bit(STRIPE_HANDLE, &sh->state); | |
1697 | release_stripe(sh); | |
1698 | } | |
1699 | ||
1700 | static void | |
ac6b53b6 DW |
1701 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1702 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1703 | { |
91c00924 | 1704 | int disks = sh->disks; |
59fc630b | 1705 | struct page **xor_srcs; |
a08abd8c | 1706 | struct async_submit_ctl submit; |
59fc630b | 1707 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1708 | struct page *xor_dest; |
d8ee0728 | 1709 | int prexor = 0; |
91c00924 | 1710 | unsigned long flags; |
59fc630b | 1711 | int j = 0; |
1712 | struct stripe_head *head_sh = sh; | |
1713 | int last_stripe; | |
91c00924 | 1714 | |
e46b272b | 1715 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1716 | (unsigned long long)sh->sector); |
1717 | ||
620125f2 SL |
1718 | for (i = 0; i < sh->disks; i++) { |
1719 | if (pd_idx == i) | |
1720 | continue; | |
1721 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1722 | break; | |
1723 | } | |
1724 | if (i >= sh->disks) { | |
1725 | atomic_inc(&sh->count); | |
620125f2 SL |
1726 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1727 | ops_complete_reconstruct(sh); | |
1728 | return; | |
1729 | } | |
59fc630b | 1730 | again: |
1731 | count = 0; | |
1732 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1733 | /* check if prexor is active which means only process blocks |
1734 | * that are part of a read-modify-write (written) | |
1735 | */ | |
59fc630b | 1736 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1737 | prexor = 1; |
91c00924 DW |
1738 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1739 | for (i = disks; i--; ) { | |
1740 | struct r5dev *dev = &sh->dev[i]; | |
59fc630b | 1741 | if (head_sh->dev[i].written) |
91c00924 DW |
1742 | xor_srcs[count++] = dev->page; |
1743 | } | |
1744 | } else { | |
1745 | xor_dest = sh->dev[pd_idx].page; | |
1746 | for (i = disks; i--; ) { | |
1747 | struct r5dev *dev = &sh->dev[i]; | |
1748 | if (i != pd_idx) | |
1749 | xor_srcs[count++] = dev->page; | |
1750 | } | |
1751 | } | |
1752 | ||
91c00924 DW |
1753 | /* 1/ if we prexor'd then the dest is reused as a source |
1754 | * 2/ if we did not prexor then we are redoing the parity | |
1755 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1756 | * for the synchronous xor case | |
1757 | */ | |
59fc630b | 1758 | last_stripe = !head_sh->batch_head || |
1759 | list_first_entry(&sh->batch_list, | |
1760 | struct stripe_head, batch_list) == head_sh; | |
1761 | if (last_stripe) { | |
1762 | flags = ASYNC_TX_ACK | | |
1763 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1764 | ||
1765 | atomic_inc(&head_sh->count); | |
1766 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1767 | to_addr_conv(sh, percpu, j)); | |
1768 | } else { | |
1769 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1770 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1771 | to_addr_conv(sh, percpu, j)); | |
1772 | } | |
91c00924 | 1773 | |
a08abd8c DW |
1774 | if (unlikely(count == 1)) |
1775 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1776 | else | |
1777 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1778 | if (!last_stripe) { |
1779 | j++; | |
1780 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1781 | batch_list); | |
1782 | goto again; | |
1783 | } | |
91c00924 DW |
1784 | } |
1785 | ||
ac6b53b6 DW |
1786 | static void |
1787 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1788 | struct dma_async_tx_descriptor *tx) | |
1789 | { | |
1790 | struct async_submit_ctl submit; | |
59fc630b | 1791 | struct page **blocks; |
1792 | int count, i, j = 0; | |
1793 | struct stripe_head *head_sh = sh; | |
1794 | int last_stripe; | |
584acdd4 MS |
1795 | int synflags; |
1796 | unsigned long txflags; | |
ac6b53b6 DW |
1797 | |
1798 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1799 | ||
620125f2 SL |
1800 | for (i = 0; i < sh->disks; i++) { |
1801 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1802 | continue; | |
1803 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1804 | break; | |
1805 | } | |
1806 | if (i >= sh->disks) { | |
1807 | atomic_inc(&sh->count); | |
620125f2 SL |
1808 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1809 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1810 | ops_complete_reconstruct(sh); | |
1811 | return; | |
1812 | } | |
1813 | ||
59fc630b | 1814 | again: |
1815 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1816 | |
1817 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1818 | synflags = SYNDROME_SRC_WRITTEN; | |
1819 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1820 | } else { | |
1821 | synflags = SYNDROME_SRC_ALL; | |
1822 | txflags = ASYNC_TX_ACK; | |
1823 | } | |
1824 | ||
1825 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1826 | last_stripe = !head_sh->batch_head || |
1827 | list_first_entry(&sh->batch_list, | |
1828 | struct stripe_head, batch_list) == head_sh; | |
1829 | ||
1830 | if (last_stripe) { | |
1831 | atomic_inc(&head_sh->count); | |
584acdd4 | 1832 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1833 | head_sh, to_addr_conv(sh, percpu, j)); |
1834 | } else | |
1835 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1836 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1837 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1838 | if (!last_stripe) { |
1839 | j++; | |
1840 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1841 | batch_list); | |
1842 | goto again; | |
1843 | } | |
91c00924 DW |
1844 | } |
1845 | ||
1846 | static void ops_complete_check(void *stripe_head_ref) | |
1847 | { | |
1848 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1849 | |
e46b272b | 1850 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1851 | (unsigned long long)sh->sector); |
1852 | ||
ecc65c9b | 1853 | sh->check_state = check_state_check_result; |
91c00924 DW |
1854 | set_bit(STRIPE_HANDLE, &sh->state); |
1855 | release_stripe(sh); | |
1856 | } | |
1857 | ||
ac6b53b6 | 1858 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1859 | { |
91c00924 | 1860 | int disks = sh->disks; |
ac6b53b6 DW |
1861 | int pd_idx = sh->pd_idx; |
1862 | int qd_idx = sh->qd_idx; | |
1863 | struct page *xor_dest; | |
46d5b785 | 1864 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1865 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1866 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1867 | int count; |
1868 | int i; | |
91c00924 | 1869 | |
e46b272b | 1870 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1871 | (unsigned long long)sh->sector); |
1872 | ||
59fc630b | 1873 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1874 | count = 0; |
1875 | xor_dest = sh->dev[pd_idx].page; | |
1876 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1877 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1878 | if (i == pd_idx || i == qd_idx) |
1879 | continue; | |
1880 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1881 | } |
1882 | ||
d6f38f31 | 1883 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1884 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1885 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1886 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1887 | |
91c00924 | 1888 | atomic_inc(&sh->count); |
a08abd8c DW |
1889 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1890 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1891 | } |
1892 | ||
ac6b53b6 DW |
1893 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1894 | { | |
46d5b785 | 1895 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1896 | struct async_submit_ctl submit; |
1897 | int count; | |
1898 | ||
1899 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1900 | (unsigned long long)sh->sector, checkp); | |
1901 | ||
59fc630b | 1902 | BUG_ON(sh->batch_head); |
584acdd4 | 1903 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1904 | if (!checkp) |
1905 | srcs[count] = NULL; | |
91c00924 | 1906 | |
91c00924 | 1907 | atomic_inc(&sh->count); |
ac6b53b6 | 1908 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1909 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1910 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1911 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1912 | } |
1913 | ||
51acbcec | 1914 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1915 | { |
1916 | int overlap_clear = 0, i, disks = sh->disks; | |
1917 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1918 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1919 | int level = conf->level; |
d6f38f31 DW |
1920 | struct raid5_percpu *percpu; |
1921 | unsigned long cpu; | |
91c00924 | 1922 | |
d6f38f31 DW |
1923 | cpu = get_cpu(); |
1924 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1925 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1926 | ops_run_biofill(sh); |
1927 | overlap_clear++; | |
1928 | } | |
1929 | ||
7b3a871e | 1930 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1931 | if (level < 6) |
1932 | tx = ops_run_compute5(sh, percpu); | |
1933 | else { | |
1934 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1935 | tx = ops_run_compute6_1(sh, percpu); | |
1936 | else | |
1937 | tx = ops_run_compute6_2(sh, percpu); | |
1938 | } | |
1939 | /* terminate the chain if reconstruct is not set to be run */ | |
1940 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1941 | async_tx_ack(tx); |
1942 | } | |
91c00924 | 1943 | |
584acdd4 MS |
1944 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
1945 | if (level < 6) | |
1946 | tx = ops_run_prexor5(sh, percpu, tx); | |
1947 | else | |
1948 | tx = ops_run_prexor6(sh, percpu, tx); | |
1949 | } | |
91c00924 | 1950 | |
600aa109 | 1951 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1952 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1953 | overlap_clear++; |
1954 | } | |
1955 | ||
ac6b53b6 DW |
1956 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1957 | if (level < 6) | |
1958 | ops_run_reconstruct5(sh, percpu, tx); | |
1959 | else | |
1960 | ops_run_reconstruct6(sh, percpu, tx); | |
1961 | } | |
91c00924 | 1962 | |
ac6b53b6 DW |
1963 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1964 | if (sh->check_state == check_state_run) | |
1965 | ops_run_check_p(sh, percpu); | |
1966 | else if (sh->check_state == check_state_run_q) | |
1967 | ops_run_check_pq(sh, percpu, 0); | |
1968 | else if (sh->check_state == check_state_run_pq) | |
1969 | ops_run_check_pq(sh, percpu, 1); | |
1970 | else | |
1971 | BUG(); | |
1972 | } | |
91c00924 | 1973 | |
59fc630b | 1974 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
1975 | for (i = disks; i--; ) { |
1976 | struct r5dev *dev = &sh->dev[i]; | |
1977 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1978 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1979 | } | |
d6f38f31 | 1980 | put_cpu(); |
91c00924 DW |
1981 | } |
1982 | ||
f18c1a35 N |
1983 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp) |
1984 | { | |
1985 | struct stripe_head *sh; | |
1986 | ||
1987 | sh = kmem_cache_zalloc(sc, gfp); | |
1988 | if (sh) { | |
1989 | spin_lock_init(&sh->stripe_lock); | |
1990 | spin_lock_init(&sh->batch_lock); | |
1991 | INIT_LIST_HEAD(&sh->batch_list); | |
1992 | INIT_LIST_HEAD(&sh->lru); | |
1993 | atomic_set(&sh->count, 1); | |
1994 | } | |
1995 | return sh; | |
1996 | } | |
486f0644 | 1997 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
1998 | { |
1999 | struct stripe_head *sh; | |
f18c1a35 N |
2000 | |
2001 | sh = alloc_stripe(conf->slab_cache, gfp); | |
3f294f4f N |
2002 | if (!sh) |
2003 | return 0; | |
6ce32846 | 2004 | |
3f294f4f | 2005 | sh->raid_conf = conf; |
3f294f4f | 2006 | |
a9683a79 | 2007 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2008 | shrink_buffers(sh); |
3f294f4f N |
2009 | kmem_cache_free(conf->slab_cache, sh); |
2010 | return 0; | |
2011 | } | |
486f0644 N |
2012 | sh->hash_lock_index = |
2013 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2014 | /* we just created an active stripe so... */ |
3f294f4f | 2015 | atomic_inc(&conf->active_stripes); |
59fc630b | 2016 | |
3f294f4f | 2017 | release_stripe(sh); |
486f0644 | 2018 | conf->max_nr_stripes++; |
3f294f4f N |
2019 | return 1; |
2020 | } | |
2021 | ||
d1688a6d | 2022 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2023 | { |
e18b890b | 2024 | struct kmem_cache *sc; |
5e5e3e78 | 2025 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2026 | |
f4be6b43 N |
2027 | if (conf->mddev->gendisk) |
2028 | sprintf(conf->cache_name[0], | |
2029 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2030 | else | |
2031 | sprintf(conf->cache_name[0], | |
2032 | "raid%d-%p", conf->level, conf->mddev); | |
2033 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2034 | ||
ad01c9e3 N |
2035 | conf->active_name = 0; |
2036 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2037 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2038 | 0, 0, NULL); |
1da177e4 LT |
2039 | if (!sc) |
2040 | return 1; | |
2041 | conf->slab_cache = sc; | |
ad01c9e3 | 2042 | conf->pool_size = devs; |
486f0644 N |
2043 | while (num--) |
2044 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2045 | return 1; |
486f0644 | 2046 | |
1da177e4 LT |
2047 | return 0; |
2048 | } | |
29269553 | 2049 | |
d6f38f31 DW |
2050 | /** |
2051 | * scribble_len - return the required size of the scribble region | |
2052 | * @num - total number of disks in the array | |
2053 | * | |
2054 | * The size must be enough to contain: | |
2055 | * 1/ a struct page pointer for each device in the array +2 | |
2056 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2057 | * (dma_map_page()) or page (page_address()) address. | |
2058 | * | |
2059 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2060 | * calculate over all devices (not just the data blocks), using zeros in place | |
2061 | * of the P and Q blocks. | |
2062 | */ | |
46d5b785 | 2063 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2064 | { |
46d5b785 | 2065 | struct flex_array *ret; |
d6f38f31 DW |
2066 | size_t len; |
2067 | ||
2068 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2069 | ret = flex_array_alloc(len, cnt, flags); |
2070 | if (!ret) | |
2071 | return NULL; | |
2072 | /* always prealloc all elements, so no locking is required */ | |
2073 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2074 | flex_array_free(ret); | |
2075 | return NULL; | |
2076 | } | |
2077 | return ret; | |
d6f38f31 DW |
2078 | } |
2079 | ||
738a2738 N |
2080 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2081 | { | |
2082 | unsigned long cpu; | |
2083 | int err = 0; | |
2084 | ||
2085 | mddev_suspend(conf->mddev); | |
2086 | get_online_cpus(); | |
2087 | for_each_present_cpu(cpu) { | |
2088 | struct raid5_percpu *percpu; | |
2089 | struct flex_array *scribble; | |
2090 | ||
2091 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2092 | scribble = scribble_alloc(new_disks, | |
2093 | new_sectors / STRIPE_SECTORS, | |
2094 | GFP_NOIO); | |
2095 | ||
2096 | if (scribble) { | |
2097 | flex_array_free(percpu->scribble); | |
2098 | percpu->scribble = scribble; | |
2099 | } else { | |
2100 | err = -ENOMEM; | |
2101 | break; | |
2102 | } | |
2103 | } | |
2104 | put_online_cpus(); | |
2105 | mddev_resume(conf->mddev); | |
2106 | return err; | |
2107 | } | |
2108 | ||
d1688a6d | 2109 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2110 | { |
2111 | /* Make all the stripes able to hold 'newsize' devices. | |
2112 | * New slots in each stripe get 'page' set to a new page. | |
2113 | * | |
2114 | * This happens in stages: | |
2115 | * 1/ create a new kmem_cache and allocate the required number of | |
2116 | * stripe_heads. | |
83f0d77a | 2117 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2118 | * to the new stripe_heads. This will have the side effect of |
2119 | * freezing the array as once all stripe_heads have been collected, | |
2120 | * no IO will be possible. Old stripe heads are freed once their | |
2121 | * pages have been transferred over, and the old kmem_cache is | |
2122 | * freed when all stripes are done. | |
2123 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2124 | * we simple return a failre status - no need to clean anything up. | |
2125 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2126 | * If this fails, we don't bother trying the shrink the | |
2127 | * stripe_heads down again, we just leave them as they are. | |
2128 | * As each stripe_head is processed the new one is released into | |
2129 | * active service. | |
2130 | * | |
2131 | * Once step2 is started, we cannot afford to wait for a write, | |
2132 | * so we use GFP_NOIO allocations. | |
2133 | */ | |
2134 | struct stripe_head *osh, *nsh; | |
2135 | LIST_HEAD(newstripes); | |
2136 | struct disk_info *ndisks; | |
b5470dc5 | 2137 | int err; |
e18b890b | 2138 | struct kmem_cache *sc; |
ad01c9e3 | 2139 | int i; |
566c09c5 | 2140 | int hash, cnt; |
ad01c9e3 N |
2141 | |
2142 | if (newsize <= conf->pool_size) | |
2143 | return 0; /* never bother to shrink */ | |
2144 | ||
b5470dc5 DW |
2145 | err = md_allow_write(conf->mddev); |
2146 | if (err) | |
2147 | return err; | |
2a2275d6 | 2148 | |
ad01c9e3 N |
2149 | /* Step 1 */ |
2150 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2151 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2152 | 0, 0, NULL); |
ad01c9e3 N |
2153 | if (!sc) |
2154 | return -ENOMEM; | |
2155 | ||
2156 | for (i = conf->max_nr_stripes; i; i--) { | |
f18c1a35 | 2157 | nsh = alloc_stripe(sc, GFP_KERNEL); |
ad01c9e3 N |
2158 | if (!nsh) |
2159 | break; | |
2160 | ||
ad01c9e3 | 2161 | nsh->raid_conf = conf; |
ad01c9e3 N |
2162 | list_add(&nsh->lru, &newstripes); |
2163 | } | |
2164 | if (i) { | |
2165 | /* didn't get enough, give up */ | |
2166 | while (!list_empty(&newstripes)) { | |
2167 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2168 | list_del(&nsh->lru); | |
2169 | kmem_cache_free(sc, nsh); | |
2170 | } | |
2171 | kmem_cache_destroy(sc); | |
2172 | return -ENOMEM; | |
2173 | } | |
2174 | /* Step 2 - Must use GFP_NOIO now. | |
2175 | * OK, we have enough stripes, start collecting inactive | |
2176 | * stripes and copying them over | |
2177 | */ | |
566c09c5 SL |
2178 | hash = 0; |
2179 | cnt = 0; | |
ad01c9e3 | 2180 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 SL |
2181 | lock_device_hash_lock(conf, hash); |
2182 | wait_event_cmd(conf->wait_for_stripe, | |
2183 | !list_empty(conf->inactive_list + hash), | |
2184 | unlock_device_hash_lock(conf, hash), | |
2185 | lock_device_hash_lock(conf, hash)); | |
2186 | osh = get_free_stripe(conf, hash); | |
2187 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2188 | |
d592a996 | 2189 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2190 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2191 | nsh->dev[i].orig_page = osh->dev[i].page; |
2192 | } | |
566c09c5 | 2193 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2194 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2195 | cnt++; |
2196 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2197 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2198 | hash++; | |
2199 | cnt = 0; | |
2200 | } | |
ad01c9e3 N |
2201 | } |
2202 | kmem_cache_destroy(conf->slab_cache); | |
2203 | ||
2204 | /* Step 3. | |
2205 | * At this point, we are holding all the stripes so the array | |
2206 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2207 | * conf->disks and the scribble region |
ad01c9e3 N |
2208 | */ |
2209 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2210 | if (ndisks) { | |
2211 | for (i=0; i<conf->raid_disks; i++) | |
2212 | ndisks[i] = conf->disks[i]; | |
2213 | kfree(conf->disks); | |
2214 | conf->disks = ndisks; | |
2215 | } else | |
2216 | err = -ENOMEM; | |
2217 | ||
2218 | /* Step 4, return new stripes to service */ | |
2219 | while(!list_empty(&newstripes)) { | |
2220 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2221 | list_del_init(&nsh->lru); | |
d6f38f31 | 2222 | |
ad01c9e3 N |
2223 | for (i=conf->raid_disks; i < newsize; i++) |
2224 | if (nsh->dev[i].page == NULL) { | |
2225 | struct page *p = alloc_page(GFP_NOIO); | |
2226 | nsh->dev[i].page = p; | |
d592a996 | 2227 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2228 | if (!p) |
2229 | err = -ENOMEM; | |
2230 | } | |
2231 | release_stripe(nsh); | |
2232 | } | |
2233 | /* critical section pass, GFP_NOIO no longer needed */ | |
2234 | ||
2235 | conf->slab_cache = sc; | |
2236 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2237 | if (!err) |
2238 | conf->pool_size = newsize; | |
ad01c9e3 N |
2239 | return err; |
2240 | } | |
1da177e4 | 2241 | |
486f0644 | 2242 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2243 | { |
2244 | struct stripe_head *sh; | |
486f0644 | 2245 | int hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS; |
1da177e4 | 2246 | |
566c09c5 SL |
2247 | spin_lock_irq(conf->hash_locks + hash); |
2248 | sh = get_free_stripe(conf, hash); | |
2249 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2250 | if (!sh) |
2251 | return 0; | |
78bafebd | 2252 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2253 | shrink_buffers(sh); |
3f294f4f N |
2254 | kmem_cache_free(conf->slab_cache, sh); |
2255 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2256 | conf->max_nr_stripes--; |
3f294f4f N |
2257 | return 1; |
2258 | } | |
2259 | ||
d1688a6d | 2260 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2261 | { |
486f0644 N |
2262 | while (conf->max_nr_stripes && |
2263 | drop_one_stripe(conf)) | |
2264 | ; | |
3f294f4f | 2265 | |
29fc7e3e N |
2266 | if (conf->slab_cache) |
2267 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
2268 | conf->slab_cache = NULL; |
2269 | } | |
2270 | ||
6712ecf8 | 2271 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 2272 | { |
99c0fb5f | 2273 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2274 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2275 | int disks = sh->disks, i; |
1da177e4 | 2276 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 2277 | char b[BDEVNAME_SIZE]; |
dd054fce | 2278 | struct md_rdev *rdev = NULL; |
05616be5 | 2279 | sector_t s; |
1da177e4 LT |
2280 | |
2281 | for (i=0 ; i<disks; i++) | |
2282 | if (bi == &sh->dev[i].req) | |
2283 | break; | |
2284 | ||
45b4233c DW |
2285 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
2286 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
2287 | uptodate); |
2288 | if (i == disks) { | |
2289 | BUG(); | |
6712ecf8 | 2290 | return; |
1da177e4 | 2291 | } |
14a75d3e | 2292 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2293 | /* If replacement finished while this request was outstanding, |
2294 | * 'replacement' might be NULL already. | |
2295 | * In that case it moved down to 'rdev'. | |
2296 | * rdev is not removed until all requests are finished. | |
2297 | */ | |
14a75d3e | 2298 | rdev = conf->disks[i].replacement; |
dd054fce | 2299 | if (!rdev) |
14a75d3e | 2300 | rdev = conf->disks[i].rdev; |
1da177e4 | 2301 | |
05616be5 N |
2302 | if (use_new_offset(conf, sh)) |
2303 | s = sh->sector + rdev->new_data_offset; | |
2304 | else | |
2305 | s = sh->sector + rdev->data_offset; | |
1da177e4 | 2306 | if (uptodate) { |
1da177e4 | 2307 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2308 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2309 | /* Note that this cannot happen on a |
2310 | * replacement device. We just fail those on | |
2311 | * any error | |
2312 | */ | |
8bda470e CD |
2313 | printk_ratelimited( |
2314 | KERN_INFO | |
2315 | "md/raid:%s: read error corrected" | |
2316 | " (%lu sectors at %llu on %s)\n", | |
2317 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 2318 | (unsigned long long)s, |
8bda470e | 2319 | bdevname(rdev->bdev, b)); |
ddd5115f | 2320 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2321 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2322 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2323 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2324 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2325 | ||
14a75d3e N |
2326 | if (atomic_read(&rdev->read_errors)) |
2327 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2328 | } else { |
14a75d3e | 2329 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2330 | int retry = 0; |
2e8ac303 | 2331 | int set_bad = 0; |
d6950432 | 2332 | |
1da177e4 | 2333 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2334 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
2335 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
2336 | printk_ratelimited( | |
2337 | KERN_WARNING | |
2338 | "md/raid:%s: read error on replacement device " | |
2339 | "(sector %llu on %s).\n", | |
2340 | mdname(conf->mddev), | |
05616be5 | 2341 | (unsigned long long)s, |
14a75d3e | 2342 | bdn); |
2e8ac303 | 2343 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2344 | set_bad = 1; | |
8bda470e CD |
2345 | printk_ratelimited( |
2346 | KERN_WARNING | |
2347 | "md/raid:%s: read error not correctable " | |
2348 | "(sector %llu on %s).\n", | |
2349 | mdname(conf->mddev), | |
05616be5 | 2350 | (unsigned long long)s, |
8bda470e | 2351 | bdn); |
2e8ac303 | 2352 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2353 | /* Oh, no!!! */ |
2e8ac303 | 2354 | set_bad = 1; |
8bda470e CD |
2355 | printk_ratelimited( |
2356 | KERN_WARNING | |
2357 | "md/raid:%s: read error NOT corrected!! " | |
2358 | "(sector %llu on %s).\n", | |
2359 | mdname(conf->mddev), | |
05616be5 | 2360 | (unsigned long long)s, |
8bda470e | 2361 | bdn); |
2e8ac303 | 2362 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2363 | > conf->max_nr_stripes) |
14f8d26b | 2364 | printk(KERN_WARNING |
0c55e022 | 2365 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2366 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2367 | else |
2368 | retry = 1; | |
edfa1f65 BY |
2369 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2370 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2371 | retry = 1; | |
ba22dcbf | 2372 | if (retry) |
3f9e7c14 | 2373 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2374 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2375 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2376 | } else | |
2377 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2378 | else { |
4e5314b5 N |
2379 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2380 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2381 | if (!(set_bad |
2382 | && test_bit(In_sync, &rdev->flags) | |
2383 | && rdev_set_badblocks( | |
2384 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2385 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2386 | } |
1da177e4 | 2387 | } |
14a75d3e | 2388 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
2389 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2390 | set_bit(STRIPE_HANDLE, &sh->state); | |
2391 | release_stripe(sh); | |
1da177e4 LT |
2392 | } |
2393 | ||
d710e138 | 2394 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 2395 | { |
99c0fb5f | 2396 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2397 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2398 | int disks = sh->disks, i; |
977df362 | 2399 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 2400 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
2401 | sector_t first_bad; |
2402 | int bad_sectors; | |
977df362 | 2403 | int replacement = 0; |
1da177e4 | 2404 | |
977df362 N |
2405 | for (i = 0 ; i < disks; i++) { |
2406 | if (bi == &sh->dev[i].req) { | |
2407 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2408 | break; |
977df362 N |
2409 | } |
2410 | if (bi == &sh->dev[i].rreq) { | |
2411 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2412 | if (rdev) |
2413 | replacement = 1; | |
2414 | else | |
2415 | /* rdev was removed and 'replacement' | |
2416 | * replaced it. rdev is not removed | |
2417 | * until all requests are finished. | |
2418 | */ | |
2419 | rdev = conf->disks[i].rdev; | |
977df362 N |
2420 | break; |
2421 | } | |
2422 | } | |
45b4233c | 2423 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
2424 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
2425 | uptodate); | |
2426 | if (i == disks) { | |
2427 | BUG(); | |
6712ecf8 | 2428 | return; |
1da177e4 LT |
2429 | } |
2430 | ||
977df362 N |
2431 | if (replacement) { |
2432 | if (!uptodate) | |
2433 | md_error(conf->mddev, rdev); | |
2434 | else if (is_badblock(rdev, sh->sector, | |
2435 | STRIPE_SECTORS, | |
2436 | &first_bad, &bad_sectors)) | |
2437 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2438 | } else { | |
2439 | if (!uptodate) { | |
9f97e4b1 | 2440 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2441 | set_bit(WriteErrorSeen, &rdev->flags); |
2442 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2443 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2444 | set_bit(MD_RECOVERY_NEEDED, | |
2445 | &rdev->mddev->recovery); | |
977df362 N |
2446 | } else if (is_badblock(rdev, sh->sector, |
2447 | STRIPE_SECTORS, | |
c0b32972 | 2448 | &first_bad, &bad_sectors)) { |
977df362 | 2449 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2450 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2451 | /* That was a successful write so make | |
2452 | * sure it looks like we already did | |
2453 | * a re-write. | |
2454 | */ | |
2455 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2456 | } | |
977df362 N |
2457 | } |
2458 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2459 | |
bb27051f | 2460 | if (sh->batch_head && !uptodate && !replacement) |
72ac7330 | 2461 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2462 | ||
977df362 N |
2463 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2464 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2465 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 2466 | release_stripe(sh); |
59fc630b | 2467 | |
2468 | if (sh->batch_head && sh != sh->batch_head) | |
2469 | release_stripe(sh->batch_head); | |
1da177e4 LT |
2470 | } |
2471 | ||
784052ec | 2472 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
d592a996 | 2473 | |
784052ec | 2474 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2475 | { |
2476 | struct r5dev *dev = &sh->dev[i]; | |
2477 | ||
2478 | bio_init(&dev->req); | |
2479 | dev->req.bi_io_vec = &dev->vec; | |
d592a996 | 2480 | dev->req.bi_max_vecs = 1; |
1da177e4 LT |
2481 | dev->req.bi_private = sh; |
2482 | ||
977df362 N |
2483 | bio_init(&dev->rreq); |
2484 | dev->rreq.bi_io_vec = &dev->rvec; | |
d592a996 | 2485 | dev->rreq.bi_max_vecs = 1; |
977df362 | 2486 | dev->rreq.bi_private = sh; |
977df362 | 2487 | |
1da177e4 | 2488 | dev->flags = 0; |
784052ec | 2489 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
2490 | } |
2491 | ||
fd01b88c | 2492 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2493 | { |
2494 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2495 | struct r5conf *conf = mddev->private; |
908f4fbd | 2496 | unsigned long flags; |
0c55e022 | 2497 | pr_debug("raid456: error called\n"); |
1da177e4 | 2498 | |
908f4fbd N |
2499 | spin_lock_irqsave(&conf->device_lock, flags); |
2500 | clear_bit(In_sync, &rdev->flags); | |
2501 | mddev->degraded = calc_degraded(conf); | |
2502 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2503 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2504 | ||
de393cde | 2505 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
2506 | set_bit(Faulty, &rdev->flags); |
2507 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2508 | printk(KERN_ALERT | |
2509 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
2510 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2511 | mdname(mddev), | |
2512 | bdevname(rdev->bdev, b), | |
2513 | mdname(mddev), | |
2514 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 2515 | } |
1da177e4 LT |
2516 | |
2517 | /* | |
2518 | * Input: a 'big' sector number, | |
2519 | * Output: index of the data and parity disk, and the sector # in them. | |
2520 | */ | |
d1688a6d | 2521 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
2522 | int previous, int *dd_idx, |
2523 | struct stripe_head *sh) | |
1da177e4 | 2524 | { |
6e3b96ed | 2525 | sector_t stripe, stripe2; |
35f2a591 | 2526 | sector_t chunk_number; |
1da177e4 | 2527 | unsigned int chunk_offset; |
911d4ee8 | 2528 | int pd_idx, qd_idx; |
67cc2b81 | 2529 | int ddf_layout = 0; |
1da177e4 | 2530 | sector_t new_sector; |
e183eaed N |
2531 | int algorithm = previous ? conf->prev_algo |
2532 | : conf->algorithm; | |
09c9e5fa AN |
2533 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2534 | : conf->chunk_sectors; | |
112bf897 N |
2535 | int raid_disks = previous ? conf->previous_raid_disks |
2536 | : conf->raid_disks; | |
2537 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2538 | |
2539 | /* First compute the information on this sector */ | |
2540 | ||
2541 | /* | |
2542 | * Compute the chunk number and the sector offset inside the chunk | |
2543 | */ | |
2544 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2545 | chunk_number = r_sector; | |
1da177e4 LT |
2546 | |
2547 | /* | |
2548 | * Compute the stripe number | |
2549 | */ | |
35f2a591 N |
2550 | stripe = chunk_number; |
2551 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2552 | stripe2 = stripe; |
1da177e4 LT |
2553 | /* |
2554 | * Select the parity disk based on the user selected algorithm. | |
2555 | */ | |
84789554 | 2556 | pd_idx = qd_idx = -1; |
16a53ecc N |
2557 | switch(conf->level) { |
2558 | case 4: | |
911d4ee8 | 2559 | pd_idx = data_disks; |
16a53ecc N |
2560 | break; |
2561 | case 5: | |
e183eaed | 2562 | switch (algorithm) { |
1da177e4 | 2563 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2564 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2565 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2566 | (*dd_idx)++; |
2567 | break; | |
2568 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2569 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2570 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2571 | (*dd_idx)++; |
2572 | break; | |
2573 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2574 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2575 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2576 | break; |
2577 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2578 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2579 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2580 | break; |
99c0fb5f N |
2581 | case ALGORITHM_PARITY_0: |
2582 | pd_idx = 0; | |
2583 | (*dd_idx)++; | |
2584 | break; | |
2585 | case ALGORITHM_PARITY_N: | |
2586 | pd_idx = data_disks; | |
2587 | break; | |
1da177e4 | 2588 | default: |
99c0fb5f | 2589 | BUG(); |
16a53ecc N |
2590 | } |
2591 | break; | |
2592 | case 6: | |
2593 | ||
e183eaed | 2594 | switch (algorithm) { |
16a53ecc | 2595 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2596 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2597 | qd_idx = pd_idx + 1; |
2598 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2599 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2600 | qd_idx = 0; |
2601 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2602 | (*dd_idx) += 2; /* D D P Q D */ |
2603 | break; | |
2604 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2605 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2606 | qd_idx = pd_idx + 1; |
2607 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2608 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2609 | qd_idx = 0; |
2610 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2611 | (*dd_idx) += 2; /* D D P Q D */ |
2612 | break; | |
2613 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2614 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2615 | qd_idx = (pd_idx + 1) % raid_disks; |
2616 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2617 | break; |
2618 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2619 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2620 | qd_idx = (pd_idx + 1) % raid_disks; |
2621 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2622 | break; |
99c0fb5f N |
2623 | |
2624 | case ALGORITHM_PARITY_0: | |
2625 | pd_idx = 0; | |
2626 | qd_idx = 1; | |
2627 | (*dd_idx) += 2; | |
2628 | break; | |
2629 | case ALGORITHM_PARITY_N: | |
2630 | pd_idx = data_disks; | |
2631 | qd_idx = data_disks + 1; | |
2632 | break; | |
2633 | ||
2634 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2635 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2636 | * of blocks for computing Q is different. | |
2637 | */ | |
6e3b96ed | 2638 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2639 | qd_idx = pd_idx + 1; |
2640 | if (pd_idx == raid_disks-1) { | |
2641 | (*dd_idx)++; /* Q D D D P */ | |
2642 | qd_idx = 0; | |
2643 | } else if (*dd_idx >= pd_idx) | |
2644 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2645 | ddf_layout = 1; |
99c0fb5f N |
2646 | break; |
2647 | ||
2648 | case ALGORITHM_ROTATING_N_RESTART: | |
2649 | /* Same a left_asymmetric, by first stripe is | |
2650 | * D D D P Q rather than | |
2651 | * Q D D D P | |
2652 | */ | |
6e3b96ed N |
2653 | stripe2 += 1; |
2654 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2655 | qd_idx = pd_idx + 1; |
2656 | if (pd_idx == raid_disks-1) { | |
2657 | (*dd_idx)++; /* Q D D D P */ | |
2658 | qd_idx = 0; | |
2659 | } else if (*dd_idx >= pd_idx) | |
2660 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2661 | ddf_layout = 1; |
99c0fb5f N |
2662 | break; |
2663 | ||
2664 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2665 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2666 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2667 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2668 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2669 | ddf_layout = 1; |
99c0fb5f N |
2670 | break; |
2671 | ||
2672 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2673 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2674 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2675 | if (*dd_idx >= pd_idx) |
2676 | (*dd_idx)++; | |
2677 | qd_idx = raid_disks - 1; | |
2678 | break; | |
2679 | ||
2680 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2681 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2682 | if (*dd_idx >= pd_idx) |
2683 | (*dd_idx)++; | |
2684 | qd_idx = raid_disks - 1; | |
2685 | break; | |
2686 | ||
2687 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2688 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2689 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2690 | qd_idx = raid_disks - 1; | |
2691 | break; | |
2692 | ||
2693 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2694 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2695 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2696 | qd_idx = raid_disks - 1; | |
2697 | break; | |
2698 | ||
2699 | case ALGORITHM_PARITY_0_6: | |
2700 | pd_idx = 0; | |
2701 | (*dd_idx)++; | |
2702 | qd_idx = raid_disks - 1; | |
2703 | break; | |
2704 | ||
16a53ecc | 2705 | default: |
99c0fb5f | 2706 | BUG(); |
16a53ecc N |
2707 | } |
2708 | break; | |
1da177e4 LT |
2709 | } |
2710 | ||
911d4ee8 N |
2711 | if (sh) { |
2712 | sh->pd_idx = pd_idx; | |
2713 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2714 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2715 | } |
1da177e4 LT |
2716 | /* |
2717 | * Finally, compute the new sector number | |
2718 | */ | |
2719 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2720 | return new_sector; | |
2721 | } | |
2722 | ||
784052ec | 2723 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2724 | { |
d1688a6d | 2725 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2726 | int raid_disks = sh->disks; |
2727 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2728 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2729 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2730 | : conf->chunk_sectors; | |
e183eaed N |
2731 | int algorithm = previous ? conf->prev_algo |
2732 | : conf->algorithm; | |
1da177e4 LT |
2733 | sector_t stripe; |
2734 | int chunk_offset; | |
35f2a591 N |
2735 | sector_t chunk_number; |
2736 | int dummy1, dd_idx = i; | |
1da177e4 | 2737 | sector_t r_sector; |
911d4ee8 | 2738 | struct stripe_head sh2; |
1da177e4 LT |
2739 | |
2740 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2741 | stripe = new_sector; | |
1da177e4 | 2742 | |
16a53ecc N |
2743 | if (i == sh->pd_idx) |
2744 | return 0; | |
2745 | switch(conf->level) { | |
2746 | case 4: break; | |
2747 | case 5: | |
e183eaed | 2748 | switch (algorithm) { |
1da177e4 LT |
2749 | case ALGORITHM_LEFT_ASYMMETRIC: |
2750 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2751 | if (i > sh->pd_idx) | |
2752 | i--; | |
2753 | break; | |
2754 | case ALGORITHM_LEFT_SYMMETRIC: | |
2755 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2756 | if (i < sh->pd_idx) | |
2757 | i += raid_disks; | |
2758 | i -= (sh->pd_idx + 1); | |
2759 | break; | |
99c0fb5f N |
2760 | case ALGORITHM_PARITY_0: |
2761 | i -= 1; | |
2762 | break; | |
2763 | case ALGORITHM_PARITY_N: | |
2764 | break; | |
1da177e4 | 2765 | default: |
99c0fb5f | 2766 | BUG(); |
16a53ecc N |
2767 | } |
2768 | break; | |
2769 | case 6: | |
d0dabf7e | 2770 | if (i == sh->qd_idx) |
16a53ecc | 2771 | return 0; /* It is the Q disk */ |
e183eaed | 2772 | switch (algorithm) { |
16a53ecc N |
2773 | case ALGORITHM_LEFT_ASYMMETRIC: |
2774 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2775 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2776 | case ALGORITHM_ROTATING_N_RESTART: | |
2777 | if (sh->pd_idx == raid_disks-1) | |
2778 | i--; /* Q D D D P */ | |
16a53ecc N |
2779 | else if (i > sh->pd_idx) |
2780 | i -= 2; /* D D P Q D */ | |
2781 | break; | |
2782 | case ALGORITHM_LEFT_SYMMETRIC: | |
2783 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2784 | if (sh->pd_idx == raid_disks-1) | |
2785 | i--; /* Q D D D P */ | |
2786 | else { | |
2787 | /* D D P Q D */ | |
2788 | if (i < sh->pd_idx) | |
2789 | i += raid_disks; | |
2790 | i -= (sh->pd_idx + 2); | |
2791 | } | |
2792 | break; | |
99c0fb5f N |
2793 | case ALGORITHM_PARITY_0: |
2794 | i -= 2; | |
2795 | break; | |
2796 | case ALGORITHM_PARITY_N: | |
2797 | break; | |
2798 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2799 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2800 | if (sh->pd_idx == 0) |
2801 | i--; /* P D D D Q */ | |
e4424fee N |
2802 | else { |
2803 | /* D D Q P D */ | |
2804 | if (i < sh->pd_idx) | |
2805 | i += raid_disks; | |
2806 | i -= (sh->pd_idx + 1); | |
2807 | } | |
99c0fb5f N |
2808 | break; |
2809 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2810 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2811 | if (i > sh->pd_idx) | |
2812 | i--; | |
2813 | break; | |
2814 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2815 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2816 | if (i < sh->pd_idx) | |
2817 | i += data_disks + 1; | |
2818 | i -= (sh->pd_idx + 1); | |
2819 | break; | |
2820 | case ALGORITHM_PARITY_0_6: | |
2821 | i -= 1; | |
2822 | break; | |
16a53ecc | 2823 | default: |
99c0fb5f | 2824 | BUG(); |
16a53ecc N |
2825 | } |
2826 | break; | |
1da177e4 LT |
2827 | } |
2828 | ||
2829 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2830 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2831 | |
112bf897 | 2832 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2833 | previous, &dummy1, &sh2); |
911d4ee8 N |
2834 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2835 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2836 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2837 | mdname(conf->mddev)); | |
1da177e4 LT |
2838 | return 0; |
2839 | } | |
2840 | return r_sector; | |
2841 | } | |
2842 | ||
600aa109 | 2843 | static void |
c0f7bddb | 2844 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2845 | int rcw, int expand) |
e33129d8 | 2846 | { |
584acdd4 | 2847 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 2848 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2849 | int level = conf->level; |
e33129d8 DW |
2850 | |
2851 | if (rcw) { | |
e33129d8 DW |
2852 | |
2853 | for (i = disks; i--; ) { | |
2854 | struct r5dev *dev = &sh->dev[i]; | |
2855 | ||
2856 | if (dev->towrite) { | |
2857 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2858 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2859 | if (!expand) |
2860 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2861 | s->locked++; |
e33129d8 DW |
2862 | } |
2863 | } | |
ce7d363a N |
2864 | /* if we are not expanding this is a proper write request, and |
2865 | * there will be bios with new data to be drained into the | |
2866 | * stripe cache | |
2867 | */ | |
2868 | if (!expand) { | |
2869 | if (!s->locked) | |
2870 | /* False alarm, nothing to do */ | |
2871 | return; | |
2872 | sh->reconstruct_state = reconstruct_state_drain_run; | |
2873 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2874 | } else | |
2875 | sh->reconstruct_state = reconstruct_state_run; | |
2876 | ||
2877 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
2878 | ||
c0f7bddb | 2879 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2880 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2881 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
2882 | } else { |
2883 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
2884 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
2885 | BUG_ON(level == 6 && |
2886 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
2887 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 2888 | |
e33129d8 DW |
2889 | for (i = disks; i--; ) { |
2890 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 2891 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
2892 | continue; |
2893 | ||
e33129d8 DW |
2894 | if (dev->towrite && |
2895 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2896 | test_bit(R5_Wantcompute, &dev->flags))) { |
2897 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2898 | set_bit(R5_LOCKED, &dev->flags); |
2899 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2900 | s->locked++; |
e33129d8 DW |
2901 | } |
2902 | } | |
ce7d363a N |
2903 | if (!s->locked) |
2904 | /* False alarm - nothing to do */ | |
2905 | return; | |
2906 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
2907 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
2908 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2909 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
2910 | } |
2911 | ||
c0f7bddb | 2912 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2913 | * are in flight |
2914 | */ | |
2915 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2916 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2917 | s->locked++; |
e33129d8 | 2918 | |
c0f7bddb YT |
2919 | if (level == 6) { |
2920 | int qd_idx = sh->qd_idx; | |
2921 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2922 | ||
2923 | set_bit(R5_LOCKED, &dev->flags); | |
2924 | clear_bit(R5_UPTODATE, &dev->flags); | |
2925 | s->locked++; | |
2926 | } | |
2927 | ||
600aa109 | 2928 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2929 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2930 | s->locked, s->ops_request); |
e33129d8 | 2931 | } |
16a53ecc | 2932 | |
1da177e4 LT |
2933 | /* |
2934 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2935 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2936 | * The bi_next chain must be in order. |
2937 | */ | |
da41ba65 | 2938 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
2939 | int forwrite, int previous) | |
1da177e4 LT |
2940 | { |
2941 | struct bio **bip; | |
d1688a6d | 2942 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2943 | int firstwrite=0; |
1da177e4 | 2944 | |
cbe47ec5 | 2945 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 2946 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
2947 | (unsigned long long)sh->sector); |
2948 | ||
b17459c0 SL |
2949 | /* |
2950 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2951 | * reference count to avoid race. The reference count should already be | |
2952 | * increased before this function is called (for example, in | |
2953 | * make_request()), so other bio sharing this stripe will not free the | |
2954 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2955 | * protect it. | |
2956 | */ | |
2957 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 2958 | /* Don't allow new IO added to stripes in batch list */ |
2959 | if (sh->batch_head) | |
2960 | goto overlap; | |
72626685 | 2961 | if (forwrite) { |
1da177e4 | 2962 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2963 | if (*bip == NULL) |
72626685 N |
2964 | firstwrite = 1; |
2965 | } else | |
1da177e4 | 2966 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
2967 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
2968 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
2969 | goto overlap; |
2970 | bip = & (*bip)->bi_next; | |
2971 | } | |
4f024f37 | 2972 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
2973 | goto overlap; |
2974 | ||
da41ba65 | 2975 | if (!forwrite || previous) |
2976 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
2977 | ||
78bafebd | 2978 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2979 | if (*bip) |
2980 | bi->bi_next = *bip; | |
2981 | *bip = bi; | |
e7836bd6 | 2982 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2983 | |
1da177e4 LT |
2984 | if (forwrite) { |
2985 | /* check if page is covered */ | |
2986 | sector_t sector = sh->dev[dd_idx].sector; | |
2987 | for (bi=sh->dev[dd_idx].towrite; | |
2988 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 2989 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 2990 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
2991 | if (bio_end_sector(bi) >= sector) |
2992 | sector = bio_end_sector(bi); | |
1da177e4 LT |
2993 | } |
2994 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 2995 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
2996 | sh->overwrite_disks++; | |
1da177e4 | 2997 | } |
cbe47ec5 N |
2998 | |
2999 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3000 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3001 | (unsigned long long)sh->sector, dd_idx); |
3002 | ||
3003 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3004 | /* Cannot hold spinlock over bitmap_startwrite, |
3005 | * but must ensure this isn't added to a batch until | |
3006 | * we have added to the bitmap and set bm_seq. | |
3007 | * So set STRIPE_BITMAP_PENDING to prevent | |
3008 | * batching. | |
3009 | * If multiple add_stripe_bio() calls race here they | |
3010 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3011 | * to complete "bitmap_startwrite" gets to set | |
3012 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3013 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3014 | * any more. | |
3015 | */ | |
3016 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3017 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3018 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3019 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3020 | spin_lock_irq(&sh->stripe_lock); |
3021 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3022 | if (!sh->batch_head) { | |
3023 | sh->bm_seq = conf->seq_flush+1; | |
3024 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3025 | } | |
cbe47ec5 | 3026 | } |
d0852df5 | 3027 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3028 | |
3029 | if (stripe_can_batch(sh)) | |
3030 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3031 | return 1; |
3032 | ||
3033 | overlap: | |
3034 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3035 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3036 | return 0; |
3037 | } | |
3038 | ||
d1688a6d | 3039 | static void end_reshape(struct r5conf *conf); |
29269553 | 3040 | |
d1688a6d | 3041 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3042 | struct stripe_head *sh) |
ccfcc3c1 | 3043 | { |
784052ec | 3044 | int sectors_per_chunk = |
09c9e5fa | 3045 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3046 | int dd_idx; |
2d2063ce | 3047 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3048 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3049 | |
112bf897 N |
3050 | raid5_compute_sector(conf, |
3051 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3052 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3053 | previous, |
911d4ee8 | 3054 | &dd_idx, sh); |
ccfcc3c1 N |
3055 | } |
3056 | ||
a4456856 | 3057 | static void |
d1688a6d | 3058 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3059 | struct stripe_head_state *s, int disks, |
3060 | struct bio **return_bi) | |
3061 | { | |
3062 | int i; | |
59fc630b | 3063 | BUG_ON(sh->batch_head); |
a4456856 DW |
3064 | for (i = disks; i--; ) { |
3065 | struct bio *bi; | |
3066 | int bitmap_end = 0; | |
3067 | ||
3068 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3069 | struct md_rdev *rdev; |
a4456856 DW |
3070 | rcu_read_lock(); |
3071 | rdev = rcu_dereference(conf->disks[i].rdev); | |
3072 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
3073 | atomic_inc(&rdev->nr_pending); |
3074 | else | |
3075 | rdev = NULL; | |
a4456856 | 3076 | rcu_read_unlock(); |
7f0da59b N |
3077 | if (rdev) { |
3078 | if (!rdev_set_badblocks( | |
3079 | rdev, | |
3080 | sh->sector, | |
3081 | STRIPE_SECTORS, 0)) | |
3082 | md_error(conf->mddev, rdev); | |
3083 | rdev_dec_pending(rdev, conf->mddev); | |
3084 | } | |
a4456856 | 3085 | } |
b17459c0 | 3086 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3087 | /* fail all writes first */ |
3088 | bi = sh->dev[i].towrite; | |
3089 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3090 | sh->overwrite_disks = 0; |
b17459c0 | 3091 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3092 | if (bi) |
a4456856 | 3093 | bitmap_end = 1; |
a4456856 DW |
3094 | |
3095 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
3096 | wake_up(&conf->wait_for_overlap); | |
3097 | ||
4f024f37 | 3098 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3099 | sh->dev[i].sector + STRIPE_SECTORS) { |
3100 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
3101 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3102 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3103 | md_write_end(conf->mddev); |
3104 | bi->bi_next = *return_bi; | |
3105 | *return_bi = bi; | |
3106 | } | |
3107 | bi = nextbi; | |
3108 | } | |
7eaf7e8e SL |
3109 | if (bitmap_end) |
3110 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3111 | STRIPE_SECTORS, 0, 0); | |
3112 | bitmap_end = 0; | |
a4456856 DW |
3113 | /* and fail all 'written' */ |
3114 | bi = sh->dev[i].written; | |
3115 | sh->dev[i].written = NULL; | |
d592a996 SL |
3116 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3117 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3118 | sh->dev[i].page = sh->dev[i].orig_page; | |
3119 | } | |
3120 | ||
a4456856 | 3121 | if (bi) bitmap_end = 1; |
4f024f37 | 3122 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3123 | sh->dev[i].sector + STRIPE_SECTORS) { |
3124 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
3125 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3126 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3127 | md_write_end(conf->mddev); |
3128 | bi->bi_next = *return_bi; | |
3129 | *return_bi = bi; | |
3130 | } | |
3131 | bi = bi2; | |
3132 | } | |
3133 | ||
b5e98d65 DW |
3134 | /* fail any reads if this device is non-operational and |
3135 | * the data has not reached the cache yet. | |
3136 | */ | |
3137 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
3138 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
3139 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3140 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3141 | bi = sh->dev[i].toread; |
3142 | sh->dev[i].toread = NULL; | |
143c4d05 | 3143 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3144 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3145 | wake_up(&conf->wait_for_overlap); | |
4f024f37 | 3146 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3147 | sh->dev[i].sector + STRIPE_SECTORS) { |
3148 | struct bio *nextbi = | |
3149 | r5_next_bio(bi, sh->dev[i].sector); | |
3150 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3151 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3152 | bi->bi_next = *return_bi; |
3153 | *return_bi = bi; | |
3154 | } | |
3155 | bi = nextbi; | |
3156 | } | |
3157 | } | |
a4456856 DW |
3158 | if (bitmap_end) |
3159 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3160 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3161 | /* If we were in the middle of a write the parity block might |
3162 | * still be locked - so just clear all R5_LOCKED flags | |
3163 | */ | |
3164 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
3165 | } |
3166 | ||
8b3e6cdc DW |
3167 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3168 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3169 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3170 | } |
3171 | ||
7f0da59b | 3172 | static void |
d1688a6d | 3173 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3174 | struct stripe_head_state *s) |
3175 | { | |
3176 | int abort = 0; | |
3177 | int i; | |
3178 | ||
59fc630b | 3179 | BUG_ON(sh->batch_head); |
7f0da59b | 3180 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3181 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3182 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3183 | s->syncing = 0; |
9a3e1101 | 3184 | s->replacing = 0; |
7f0da59b | 3185 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3186 | * Don't even need to abort as that is handled elsewhere |
3187 | * if needed, and not always wanted e.g. if there is a known | |
3188 | * bad block here. | |
9a3e1101 | 3189 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3190 | * non-sync devices, or abort the recovery |
3191 | */ | |
18b9837e N |
3192 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3193 | /* During recovery devices cannot be removed, so | |
3194 | * locking and refcounting of rdevs is not needed | |
3195 | */ | |
3196 | for (i = 0; i < conf->raid_disks; i++) { | |
3197 | struct md_rdev *rdev = conf->disks[i].rdev; | |
3198 | if (rdev | |
3199 | && !test_bit(Faulty, &rdev->flags) | |
3200 | && !test_bit(In_sync, &rdev->flags) | |
3201 | && !rdev_set_badblocks(rdev, sh->sector, | |
3202 | STRIPE_SECTORS, 0)) | |
3203 | abort = 1; | |
3204 | rdev = conf->disks[i].replacement; | |
3205 | if (rdev | |
3206 | && !test_bit(Faulty, &rdev->flags) | |
3207 | && !test_bit(In_sync, &rdev->flags) | |
3208 | && !rdev_set_badblocks(rdev, sh->sector, | |
3209 | STRIPE_SECTORS, 0)) | |
3210 | abort = 1; | |
3211 | } | |
3212 | if (abort) | |
3213 | conf->recovery_disabled = | |
3214 | conf->mddev->recovery_disabled; | |
7f0da59b | 3215 | } |
18b9837e | 3216 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3217 | } |
3218 | ||
9a3e1101 N |
3219 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3220 | { | |
3221 | struct md_rdev *rdev; | |
3222 | int rv = 0; | |
3223 | /* Doing recovery so rcu locking not required */ | |
3224 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
3225 | if (rdev | |
3226 | && !test_bit(Faulty, &rdev->flags) | |
3227 | && !test_bit(In_sync, &rdev->flags) | |
3228 | && (rdev->recovery_offset <= sh->sector | |
3229 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3230 | rv = 1; | |
3231 | ||
3232 | return rv; | |
3233 | } | |
3234 | ||
93b3dbce | 3235 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
3236 | * to be read or computed to satisfy a request. |
3237 | * | |
3238 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 3239 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 3240 | */ |
2c58f06e N |
3241 | |
3242 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3243 | int disk_idx, int disks) | |
a4456856 | 3244 | { |
5599becc | 3245 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3246 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3247 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3248 | int i; |
5599becc | 3249 | |
a79cfe12 N |
3250 | |
3251 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3252 | test_bit(R5_UPTODATE, &dev->flags)) | |
3253 | /* No point reading this as we already have it or have | |
3254 | * decided to get it. | |
3255 | */ | |
3256 | return 0; | |
3257 | ||
3258 | if (dev->toread || | |
3259 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3260 | /* We need this block to directly satisfy a request */ | |
3261 | return 1; | |
3262 | ||
3263 | if (s->syncing || s->expanding || | |
3264 | (s->replacing && want_replace(sh, disk_idx))) | |
3265 | /* When syncing, or expanding we read everything. | |
3266 | * When replacing, we need the replaced block. | |
3267 | */ | |
3268 | return 1; | |
3269 | ||
3270 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3271 | (s->failed >= 2 && fdev[1]->toread)) | |
3272 | /* If we want to read from a failed device, then | |
3273 | * we need to actually read every other device. | |
3274 | */ | |
3275 | return 1; | |
3276 | ||
a9d56950 N |
3277 | /* Sometimes neither read-modify-write nor reconstruct-write |
3278 | * cycles can work. In those cases we read every block we | |
3279 | * can. Then the parity-update is certain to have enough to | |
3280 | * work with. | |
3281 | * This can only be a problem when we need to write something, | |
3282 | * and some device has failed. If either of those tests | |
3283 | * fail we need look no further. | |
3284 | */ | |
3285 | if (!s->failed || !s->to_write) | |
3286 | return 0; | |
3287 | ||
3288 | if (test_bit(R5_Insync, &dev->flags) && | |
3289 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3290 | /* Pre-reads at not permitted until after short delay | |
3291 | * to gather multiple requests. However if this | |
3292 | * device is no Insync, the block could only be be computed | |
3293 | * and there is no need to delay that. | |
3294 | */ | |
3295 | return 0; | |
ea664c82 N |
3296 | |
3297 | for (i = 0; i < s->failed; i++) { | |
3298 | if (fdev[i]->towrite && | |
3299 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3300 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3301 | /* If we have a partial write to a failed | |
3302 | * device, then we will need to reconstruct | |
3303 | * the content of that device, so all other | |
3304 | * devices must be read. | |
3305 | */ | |
3306 | return 1; | |
3307 | } | |
3308 | ||
3309 | /* If we are forced to do a reconstruct-write, either because | |
3310 | * the current RAID6 implementation only supports that, or | |
3311 | * or because parity cannot be trusted and we are currently | |
3312 | * recovering it, there is extra need to be careful. | |
3313 | * If one of the devices that we would need to read, because | |
3314 | * it is not being overwritten (and maybe not written at all) | |
3315 | * is missing/faulty, then we need to read everything we can. | |
3316 | */ | |
3317 | if (sh->raid_conf->level != 6 && | |
3318 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3319 | /* reconstruct-write isn't being forced */ | |
3320 | return 0; | |
3321 | for (i = 0; i < s->failed; i++) { | |
10d82c5f N |
3322 | if (s->failed_num[i] != sh->pd_idx && |
3323 | s->failed_num[i] != sh->qd_idx && | |
3324 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3325 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3326 | return 1; | |
3327 | } | |
3328 | ||
2c58f06e N |
3329 | return 0; |
3330 | } | |
3331 | ||
3332 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3333 | int disk_idx, int disks) | |
3334 | { | |
3335 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3336 | ||
3337 | /* is the data in this block needed, and can we get it? */ | |
3338 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3339 | /* we would like to get this block, possibly by computing it, |
3340 | * otherwise read it if the backing disk is insync | |
3341 | */ | |
3342 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3343 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3344 | BUG_ON(sh->batch_head); |
5599becc | 3345 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3346 | (s->failed && (disk_idx == s->failed_num[0] || |
3347 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3348 | /* have disk failed, and we're requested to fetch it; |
3349 | * do compute it | |
a4456856 | 3350 | */ |
5599becc YT |
3351 | pr_debug("Computing stripe %llu block %d\n", |
3352 | (unsigned long long)sh->sector, disk_idx); | |
3353 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3354 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3355 | set_bit(R5_Wantcompute, &dev->flags); | |
3356 | sh->ops.target = disk_idx; | |
3357 | sh->ops.target2 = -1; /* no 2nd target */ | |
3358 | s->req_compute = 1; | |
93b3dbce N |
3359 | /* Careful: from this point on 'uptodate' is in the eye |
3360 | * of raid_run_ops which services 'compute' operations | |
3361 | * before writes. R5_Wantcompute flags a block that will | |
3362 | * be R5_UPTODATE by the time it is needed for a | |
3363 | * subsequent operation. | |
3364 | */ | |
5599becc YT |
3365 | s->uptodate++; |
3366 | return 1; | |
3367 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3368 | /* Computing 2-failure is *very* expensive; only | |
3369 | * do it if failed >= 2 | |
3370 | */ | |
3371 | int other; | |
3372 | for (other = disks; other--; ) { | |
3373 | if (other == disk_idx) | |
3374 | continue; | |
3375 | if (!test_bit(R5_UPTODATE, | |
3376 | &sh->dev[other].flags)) | |
3377 | break; | |
a4456856 | 3378 | } |
5599becc YT |
3379 | BUG_ON(other < 0); |
3380 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3381 | (unsigned long long)sh->sector, | |
3382 | disk_idx, other); | |
3383 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3384 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3385 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3386 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3387 | sh->ops.target = disk_idx; | |
3388 | sh->ops.target2 = other; | |
3389 | s->uptodate += 2; | |
3390 | s->req_compute = 1; | |
3391 | return 1; | |
3392 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3393 | set_bit(R5_LOCKED, &dev->flags); | |
3394 | set_bit(R5_Wantread, &dev->flags); | |
3395 | s->locked++; | |
3396 | pr_debug("Reading block %d (sync=%d)\n", | |
3397 | disk_idx, s->syncing); | |
a4456856 DW |
3398 | } |
3399 | } | |
5599becc YT |
3400 | |
3401 | return 0; | |
3402 | } | |
3403 | ||
3404 | /** | |
93b3dbce | 3405 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3406 | */ |
93b3dbce N |
3407 | static void handle_stripe_fill(struct stripe_head *sh, |
3408 | struct stripe_head_state *s, | |
3409 | int disks) | |
5599becc YT |
3410 | { |
3411 | int i; | |
3412 | ||
3413 | /* look for blocks to read/compute, skip this if a compute | |
3414 | * is already in flight, or if the stripe contents are in the | |
3415 | * midst of changing due to a write | |
3416 | */ | |
3417 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
3418 | !sh->reconstruct_state) | |
3419 | for (i = disks; i--; ) | |
93b3dbce | 3420 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3421 | break; |
a4456856 DW |
3422 | set_bit(STRIPE_HANDLE, &sh->state); |
3423 | } | |
3424 | ||
787b76fa N |
3425 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3426 | unsigned long handle_flags); | |
1fe797e6 | 3427 | /* handle_stripe_clean_event |
a4456856 DW |
3428 | * any written block on an uptodate or failed drive can be returned. |
3429 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3430 | * never LOCKED, so we don't need to test 'failed' directly. | |
3431 | */ | |
d1688a6d | 3432 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
3433 | struct stripe_head *sh, int disks, struct bio **return_bi) |
3434 | { | |
3435 | int i; | |
3436 | struct r5dev *dev; | |
f8dfcffd | 3437 | int discard_pending = 0; |
59fc630b | 3438 | struct stripe_head *head_sh = sh; |
3439 | bool do_endio = false; | |
a4456856 DW |
3440 | |
3441 | for (i = disks; i--; ) | |
3442 | if (sh->dev[i].written) { | |
3443 | dev = &sh->dev[i]; | |
3444 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3445 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3446 | test_bit(R5_Discard, &dev->flags) || |
3447 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3448 | /* We can return any write requests */ |
3449 | struct bio *wbi, *wbi2; | |
45b4233c | 3450 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3451 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3452 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3453 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3454 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3455 | } |
59fc630b | 3456 | do_endio = true; |
3457 | ||
3458 | returnbi: | |
3459 | dev->page = dev->orig_page; | |
a4456856 DW |
3460 | wbi = dev->written; |
3461 | dev->written = NULL; | |
4f024f37 | 3462 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3463 | dev->sector + STRIPE_SECTORS) { |
3464 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3465 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 DW |
3466 | md_write_end(conf->mddev); |
3467 | wbi->bi_next = *return_bi; | |
3468 | *return_bi = wbi; | |
3469 | } | |
3470 | wbi = wbi2; | |
3471 | } | |
7eaf7e8e SL |
3472 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3473 | STRIPE_SECTORS, | |
a4456856 | 3474 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3475 | 0); |
59fc630b | 3476 | if (head_sh->batch_head) { |
3477 | sh = list_first_entry(&sh->batch_list, | |
3478 | struct stripe_head, | |
3479 | batch_list); | |
3480 | if (sh != head_sh) { | |
3481 | dev = &sh->dev[i]; | |
3482 | goto returnbi; | |
3483 | } | |
3484 | } | |
3485 | sh = head_sh; | |
3486 | dev = &sh->dev[i]; | |
f8dfcffd N |
3487 | } else if (test_bit(R5_Discard, &dev->flags)) |
3488 | discard_pending = 1; | |
d592a996 SL |
3489 | WARN_ON(test_bit(R5_SkipCopy, &dev->flags)); |
3490 | WARN_ON(dev->page != dev->orig_page); | |
f8dfcffd N |
3491 | } |
3492 | if (!discard_pending && | |
3493 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
3494 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); | |
3495 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3496 | if (sh->qd_idx >= 0) { | |
3497 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3498 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3499 | } | |
3500 | /* now that discard is done we can proceed with any sync */ | |
3501 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3502 | /* |
3503 | * SCSI discard will change some bio fields and the stripe has | |
3504 | * no updated data, so remove it from hash list and the stripe | |
3505 | * will be reinitialized | |
3506 | */ | |
3507 | spin_lock_irq(&conf->device_lock); | |
59fc630b | 3508 | unhash: |
d47648fc | 3509 | remove_hash(sh); |
59fc630b | 3510 | if (head_sh->batch_head) { |
3511 | sh = list_first_entry(&sh->batch_list, | |
3512 | struct stripe_head, batch_list); | |
3513 | if (sh != head_sh) | |
3514 | goto unhash; | |
3515 | } | |
d47648fc | 3516 | spin_unlock_irq(&conf->device_lock); |
59fc630b | 3517 | sh = head_sh; |
3518 | ||
f8dfcffd N |
3519 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3520 | set_bit(STRIPE_HANDLE, &sh->state); | |
3521 | ||
3522 | } | |
8b3e6cdc DW |
3523 | |
3524 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3525 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3526 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3527 | |
787b76fa N |
3528 | if (head_sh->batch_head && do_endio) |
3529 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3530 | } |
3531 | ||
d1688a6d | 3532 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
3533 | struct stripe_head *sh, |
3534 | struct stripe_head_state *s, | |
3535 | int disks) | |
a4456856 DW |
3536 | { |
3537 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3538 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3539 | ||
584acdd4 | 3540 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3541 | * If yes, then the array is dirty (after unclean shutdown or |
3542 | * initial creation), so parity in some stripes might be inconsistent. | |
3543 | * In this case, we need to always do reconstruct-write, to ensure | |
3544 | * that in case of drive failure or read-error correction, we | |
3545 | * generate correct data from the parity. | |
3546 | */ | |
584acdd4 | 3547 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3548 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3549 | s->failed == 0)) { | |
a7854487 | 3550 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3551 | * look like rcw is cheaper |
3552 | */ | |
3553 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3554 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3555 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3556 | (unsigned long long)sh->sector); |
c8ac1803 | 3557 | } else for (i = disks; i--; ) { |
a4456856 DW |
3558 | /* would I have to read this buffer for read_modify_write */ |
3559 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3560 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3561 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3562 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3563 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
3564 | if (test_bit(R5_Insync, &dev->flags)) |
3565 | rmw++; | |
3566 | else | |
3567 | rmw += 2*disks; /* cannot read it */ | |
3568 | } | |
3569 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3570 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3571 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3572 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3573 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3574 | test_bit(R5_Wantcompute, &dev->flags))) { | |
67f45548 N |
3575 | if (test_bit(R5_Insync, &dev->flags)) |
3576 | rcw++; | |
a4456856 DW |
3577 | else |
3578 | rcw += 2*disks; | |
3579 | } | |
3580 | } | |
45b4233c | 3581 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
3582 | (unsigned long long)sh->sector, rmw, rcw); |
3583 | set_bit(STRIPE_HANDLE, &sh->state); | |
584acdd4 | 3584 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) { |
a4456856 | 3585 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3586 | if (conf->mddev->queue) |
3587 | blk_add_trace_msg(conf->mddev->queue, | |
3588 | "raid5 rmw %llu %d", | |
3589 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3590 | for (i = disks; i--; ) { |
3591 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3592 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3593 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3594 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3595 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 | 3596 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3597 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3598 | &sh->state)) { | |
3599 | pr_debug("Read_old block %d for r-m-w\n", | |
3600 | i); | |
a4456856 DW |
3601 | set_bit(R5_LOCKED, &dev->flags); |
3602 | set_bit(R5_Wantread, &dev->flags); | |
3603 | s->locked++; | |
3604 | } else { | |
3605 | set_bit(STRIPE_DELAYED, &sh->state); | |
3606 | set_bit(STRIPE_HANDLE, &sh->state); | |
3607 | } | |
3608 | } | |
3609 | } | |
a9add5d9 | 3610 | } |
584acdd4 | 3611 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) { |
a4456856 | 3612 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3613 | int qread =0; |
c8ac1803 | 3614 | rcw = 0; |
a4456856 DW |
3615 | for (i = disks; i--; ) { |
3616 | struct r5dev *dev = &sh->dev[i]; | |
3617 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3618 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3619 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3620 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3621 | test_bit(R5_Wantcompute, &dev->flags))) { |
3622 | rcw++; | |
67f45548 N |
3623 | if (test_bit(R5_Insync, &dev->flags) && |
3624 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3625 | &sh->state)) { | |
45b4233c | 3626 | pr_debug("Read_old block " |
a4456856 DW |
3627 | "%d for Reconstruct\n", i); |
3628 | set_bit(R5_LOCKED, &dev->flags); | |
3629 | set_bit(R5_Wantread, &dev->flags); | |
3630 | s->locked++; | |
a9add5d9 | 3631 | qread++; |
a4456856 DW |
3632 | } else { |
3633 | set_bit(STRIPE_DELAYED, &sh->state); | |
3634 | set_bit(STRIPE_HANDLE, &sh->state); | |
3635 | } | |
3636 | } | |
3637 | } | |
e3620a3a | 3638 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3639 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3640 | (unsigned long long)sh->sector, | |
3641 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3642 | } |
b1b02fe9 N |
3643 | |
3644 | if (rcw > disks && rmw > disks && | |
3645 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3646 | set_bit(STRIPE_DELAYED, &sh->state); | |
3647 | ||
a4456856 DW |
3648 | /* now if nothing is locked, and if we have enough data, |
3649 | * we can start a write request | |
3650 | */ | |
f38e1219 DW |
3651 | /* since handle_stripe can be called at any time we need to handle the |
3652 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3653 | * subsequent call wants to start a write request. raid_run_ops only |
3654 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3655 | * simultaneously. If this is not the case then new writes need to be |
3656 | * held off until the compute completes. | |
3657 | */ | |
976ea8d4 DW |
3658 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3659 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
3660 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 3661 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
3662 | } |
3663 | ||
d1688a6d | 3664 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3665 | struct stripe_head_state *s, int disks) |
3666 | { | |
ecc65c9b | 3667 | struct r5dev *dev = NULL; |
bd2ab670 | 3668 | |
59fc630b | 3669 | BUG_ON(sh->batch_head); |
a4456856 | 3670 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3671 | |
ecc65c9b DW |
3672 | switch (sh->check_state) { |
3673 | case check_state_idle: | |
3674 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3675 | if (s->failed == 0) { |
bd2ab670 | 3676 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3677 | sh->check_state = check_state_run; |
3678 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3679 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3680 | s->uptodate--; |
ecc65c9b | 3681 | break; |
bd2ab670 | 3682 | } |
f2b3b44d | 3683 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3684 | /* fall through */ |
3685 | case check_state_compute_result: | |
3686 | sh->check_state = check_state_idle; | |
3687 | if (!dev) | |
3688 | dev = &sh->dev[sh->pd_idx]; | |
3689 | ||
3690 | /* check that a write has not made the stripe insync */ | |
3691 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3692 | break; | |
c8894419 | 3693 | |
a4456856 | 3694 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3695 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3696 | BUG_ON(s->uptodate != disks); | |
3697 | ||
3698 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3699 | s->locked++; |
a4456856 | 3700 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3701 | |
a4456856 | 3702 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3703 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3704 | break; |
3705 | case check_state_run: | |
3706 | break; /* we will be called again upon completion */ | |
3707 | case check_state_check_result: | |
3708 | sh->check_state = check_state_idle; | |
3709 | ||
3710 | /* if a failure occurred during the check operation, leave | |
3711 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3712 | */ | |
3713 | if (s->failed) | |
3714 | break; | |
3715 | ||
3716 | /* handle a successful check operation, if parity is correct | |
3717 | * we are done. Otherwise update the mismatch count and repair | |
3718 | * parity if !MD_RECOVERY_CHECK | |
3719 | */ | |
ad283ea4 | 3720 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3721 | /* parity is correct (on disc, |
3722 | * not in buffer any more) | |
3723 | */ | |
3724 | set_bit(STRIPE_INSYNC, &sh->state); | |
3725 | else { | |
7f7583d4 | 3726 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3727 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3728 | /* don't try to repair!! */ | |
3729 | set_bit(STRIPE_INSYNC, &sh->state); | |
3730 | else { | |
3731 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3732 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3733 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3734 | set_bit(R5_Wantcompute, | |
3735 | &sh->dev[sh->pd_idx].flags); | |
3736 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3737 | sh->ops.target2 = -1; |
ecc65c9b DW |
3738 | s->uptodate++; |
3739 | } | |
3740 | } | |
3741 | break; | |
3742 | case check_state_compute_run: | |
3743 | break; | |
3744 | default: | |
3745 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3746 | __func__, sh->check_state, | |
3747 | (unsigned long long) sh->sector); | |
3748 | BUG(); | |
a4456856 DW |
3749 | } |
3750 | } | |
3751 | ||
d1688a6d | 3752 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3753 | struct stripe_head_state *s, |
f2b3b44d | 3754 | int disks) |
a4456856 | 3755 | { |
a4456856 | 3756 | int pd_idx = sh->pd_idx; |
34e04e87 | 3757 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3758 | struct r5dev *dev; |
a4456856 | 3759 | |
59fc630b | 3760 | BUG_ON(sh->batch_head); |
a4456856 DW |
3761 | set_bit(STRIPE_HANDLE, &sh->state); |
3762 | ||
3763 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3764 | |
a4456856 DW |
3765 | /* Want to check and possibly repair P and Q. |
3766 | * However there could be one 'failed' device, in which | |
3767 | * case we can only check one of them, possibly using the | |
3768 | * other to generate missing data | |
3769 | */ | |
3770 | ||
d82dfee0 DW |
3771 | switch (sh->check_state) { |
3772 | case check_state_idle: | |
3773 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 3774 | if (s->failed == s->q_failed) { |
d82dfee0 | 3775 | /* The only possible failed device holds Q, so it |
a4456856 DW |
3776 | * makes sense to check P (If anything else were failed, |
3777 | * we would have used P to recreate it). | |
3778 | */ | |
d82dfee0 | 3779 | sh->check_state = check_state_run; |
a4456856 | 3780 | } |
f2b3b44d | 3781 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 3782 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
3783 | * anything, so it makes sense to check it |
3784 | */ | |
d82dfee0 DW |
3785 | if (sh->check_state == check_state_run) |
3786 | sh->check_state = check_state_run_pq; | |
3787 | else | |
3788 | sh->check_state = check_state_run_q; | |
a4456856 | 3789 | } |
a4456856 | 3790 | |
d82dfee0 DW |
3791 | /* discard potentially stale zero_sum_result */ |
3792 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3793 | |
d82dfee0 DW |
3794 | if (sh->check_state == check_state_run) { |
3795 | /* async_xor_zero_sum destroys the contents of P */ | |
3796 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3797 | s->uptodate--; | |
a4456856 | 3798 | } |
d82dfee0 DW |
3799 | if (sh->check_state >= check_state_run && |
3800 | sh->check_state <= check_state_run_pq) { | |
3801 | /* async_syndrome_zero_sum preserves P and Q, so | |
3802 | * no need to mark them !uptodate here | |
3803 | */ | |
3804 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3805 | break; | |
a4456856 DW |
3806 | } |
3807 | ||
d82dfee0 DW |
3808 | /* we have 2-disk failure */ |
3809 | BUG_ON(s->failed != 2); | |
3810 | /* fall through */ | |
3811 | case check_state_compute_result: | |
3812 | sh->check_state = check_state_idle; | |
a4456856 | 3813 | |
d82dfee0 DW |
3814 | /* check that a write has not made the stripe insync */ |
3815 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3816 | break; | |
a4456856 DW |
3817 | |
3818 | /* now write out any block on a failed drive, | |
d82dfee0 | 3819 | * or P or Q if they were recomputed |
a4456856 | 3820 | */ |
d82dfee0 | 3821 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3822 | if (s->failed == 2) { |
f2b3b44d | 3823 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3824 | s->locked++; |
3825 | set_bit(R5_LOCKED, &dev->flags); | |
3826 | set_bit(R5_Wantwrite, &dev->flags); | |
3827 | } | |
3828 | if (s->failed >= 1) { | |
f2b3b44d | 3829 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3830 | s->locked++; |
3831 | set_bit(R5_LOCKED, &dev->flags); | |
3832 | set_bit(R5_Wantwrite, &dev->flags); | |
3833 | } | |
d82dfee0 | 3834 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3835 | dev = &sh->dev[pd_idx]; |
3836 | s->locked++; | |
3837 | set_bit(R5_LOCKED, &dev->flags); | |
3838 | set_bit(R5_Wantwrite, &dev->flags); | |
3839 | } | |
d82dfee0 | 3840 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3841 | dev = &sh->dev[qd_idx]; |
3842 | s->locked++; | |
3843 | set_bit(R5_LOCKED, &dev->flags); | |
3844 | set_bit(R5_Wantwrite, &dev->flags); | |
3845 | } | |
3846 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3847 | ||
3848 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3849 | break; |
3850 | case check_state_run: | |
3851 | case check_state_run_q: | |
3852 | case check_state_run_pq: | |
3853 | break; /* we will be called again upon completion */ | |
3854 | case check_state_check_result: | |
3855 | sh->check_state = check_state_idle; | |
3856 | ||
3857 | /* handle a successful check operation, if parity is correct | |
3858 | * we are done. Otherwise update the mismatch count and repair | |
3859 | * parity if !MD_RECOVERY_CHECK | |
3860 | */ | |
3861 | if (sh->ops.zero_sum_result == 0) { | |
3862 | /* both parities are correct */ | |
3863 | if (!s->failed) | |
3864 | set_bit(STRIPE_INSYNC, &sh->state); | |
3865 | else { | |
3866 | /* in contrast to the raid5 case we can validate | |
3867 | * parity, but still have a failure to write | |
3868 | * back | |
3869 | */ | |
3870 | sh->check_state = check_state_compute_result; | |
3871 | /* Returning at this point means that we may go | |
3872 | * off and bring p and/or q uptodate again so | |
3873 | * we make sure to check zero_sum_result again | |
3874 | * to verify if p or q need writeback | |
3875 | */ | |
3876 | } | |
3877 | } else { | |
7f7583d4 | 3878 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
3879 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3880 | /* don't try to repair!! */ | |
3881 | set_bit(STRIPE_INSYNC, &sh->state); | |
3882 | else { | |
3883 | int *target = &sh->ops.target; | |
3884 | ||
3885 | sh->ops.target = -1; | |
3886 | sh->ops.target2 = -1; | |
3887 | sh->check_state = check_state_compute_run; | |
3888 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3889 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3890 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3891 | set_bit(R5_Wantcompute, | |
3892 | &sh->dev[pd_idx].flags); | |
3893 | *target = pd_idx; | |
3894 | target = &sh->ops.target2; | |
3895 | s->uptodate++; | |
3896 | } | |
3897 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3898 | set_bit(R5_Wantcompute, | |
3899 | &sh->dev[qd_idx].flags); | |
3900 | *target = qd_idx; | |
3901 | s->uptodate++; | |
3902 | } | |
3903 | } | |
3904 | } | |
3905 | break; | |
3906 | case check_state_compute_run: | |
3907 | break; | |
3908 | default: | |
3909 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3910 | __func__, sh->check_state, | |
3911 | (unsigned long long) sh->sector); | |
3912 | BUG(); | |
a4456856 DW |
3913 | } |
3914 | } | |
3915 | ||
d1688a6d | 3916 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3917 | { |
3918 | int i; | |
3919 | ||
3920 | /* We have read all the blocks in this stripe and now we need to | |
3921 | * copy some of them into a target stripe for expand. | |
3922 | */ | |
f0a50d37 | 3923 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 3924 | BUG_ON(sh->batch_head); |
a4456856 DW |
3925 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3926 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3927 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3928 | int dd_idx, j; |
a4456856 | 3929 | struct stripe_head *sh2; |
a08abd8c | 3930 | struct async_submit_ctl submit; |
a4456856 | 3931 | |
784052ec | 3932 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3933 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3934 | &dd_idx, NULL); | |
a8c906ca | 3935 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3936 | if (sh2 == NULL) |
3937 | /* so far only the early blocks of this stripe | |
3938 | * have been requested. When later blocks | |
3939 | * get requested, we will try again | |
3940 | */ | |
3941 | continue; | |
3942 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3943 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3944 | /* must have already done this block */ | |
3945 | release_stripe(sh2); | |
3946 | continue; | |
3947 | } | |
f0a50d37 DW |
3948 | |
3949 | /* place all the copies on one channel */ | |
a08abd8c | 3950 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3951 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3952 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3953 | &submit); |
f0a50d37 | 3954 | |
a4456856 DW |
3955 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3956 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3957 | for (j = 0; j < conf->raid_disks; j++) | |
3958 | if (j != sh2->pd_idx && | |
86c374ba | 3959 | j != sh2->qd_idx && |
a4456856 DW |
3960 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3961 | break; | |
3962 | if (j == conf->raid_disks) { | |
3963 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3964 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3965 | } | |
3966 | release_stripe(sh2); | |
f0a50d37 | 3967 | |
a4456856 | 3968 | } |
a2e08551 | 3969 | /* done submitting copies, wait for them to complete */ |
749586b7 | 3970 | async_tx_quiesce(&tx); |
a4456856 | 3971 | } |
1da177e4 LT |
3972 | |
3973 | /* | |
3974 | * handle_stripe - do things to a stripe. | |
3975 | * | |
9a3e1101 N |
3976 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3977 | * state of various bits to see what needs to be done. | |
1da177e4 | 3978 | * Possible results: |
9a3e1101 N |
3979 | * return some read requests which now have data |
3980 | * return some write requests which are safely on storage | |
1da177e4 LT |
3981 | * schedule a read on some buffers |
3982 | * schedule a write of some buffers | |
3983 | * return confirmation of parity correctness | |
3984 | * | |
1da177e4 | 3985 | */ |
a4456856 | 3986 | |
acfe726b | 3987 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3988 | { |
d1688a6d | 3989 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3990 | int disks = sh->disks; |
474af965 N |
3991 | struct r5dev *dev; |
3992 | int i; | |
9a3e1101 | 3993 | int do_recovery = 0; |
1da177e4 | 3994 | |
acfe726b N |
3995 | memset(s, 0, sizeof(*s)); |
3996 | ||
dabc4ec6 | 3997 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
3998 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
3999 | s->failed_num[0] = -1; |
4000 | s->failed_num[1] = -1; | |
1da177e4 | 4001 | |
acfe726b | 4002 | /* Now to look around and see what can be done */ |
1da177e4 | 4003 | rcu_read_lock(); |
16a53ecc | 4004 | for (i=disks; i--; ) { |
3cb03002 | 4005 | struct md_rdev *rdev; |
31c176ec N |
4006 | sector_t first_bad; |
4007 | int bad_sectors; | |
4008 | int is_bad = 0; | |
acfe726b | 4009 | |
16a53ecc | 4010 | dev = &sh->dev[i]; |
1da177e4 | 4011 | |
45b4233c | 4012 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4013 | i, dev->flags, |
4014 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4015 | /* maybe we can reply to a read |
4016 | * | |
4017 | * new wantfill requests are only permitted while | |
4018 | * ops_complete_biofill is guaranteed to be inactive | |
4019 | */ | |
4020 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4021 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4022 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4023 | |
16a53ecc | 4024 | /* now count some things */ |
cc94015a N |
4025 | if (test_bit(R5_LOCKED, &dev->flags)) |
4026 | s->locked++; | |
4027 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4028 | s->uptodate++; | |
2d6e4ecc | 4029 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4030 | s->compute++; |
4031 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4032 | } |
1da177e4 | 4033 | |
acfe726b | 4034 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4035 | s->to_fill++; |
acfe726b | 4036 | else if (dev->toread) |
cc94015a | 4037 | s->to_read++; |
16a53ecc | 4038 | if (dev->towrite) { |
cc94015a | 4039 | s->to_write++; |
16a53ecc | 4040 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4041 | s->non_overwrite++; |
16a53ecc | 4042 | } |
a4456856 | 4043 | if (dev->written) |
cc94015a | 4044 | s->written++; |
14a75d3e N |
4045 | /* Prefer to use the replacement for reads, but only |
4046 | * if it is recovered enough and has no bad blocks. | |
4047 | */ | |
4048 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4049 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4050 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4051 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4052 | &first_bad, &bad_sectors)) | |
4053 | set_bit(R5_ReadRepl, &dev->flags); | |
4054 | else { | |
9a3e1101 N |
4055 | if (rdev) |
4056 | set_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4057 | rdev = rcu_dereference(conf->disks[i].rdev); |
4058 | clear_bit(R5_ReadRepl, &dev->flags); | |
4059 | } | |
9283d8c5 N |
4060 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4061 | rdev = NULL; | |
31c176ec N |
4062 | if (rdev) { |
4063 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4064 | &first_bad, &bad_sectors); | |
4065 | if (s->blocked_rdev == NULL | |
4066 | && (test_bit(Blocked, &rdev->flags) | |
4067 | || is_bad < 0)) { | |
4068 | if (is_bad < 0) | |
4069 | set_bit(BlockedBadBlocks, | |
4070 | &rdev->flags); | |
4071 | s->blocked_rdev = rdev; | |
4072 | atomic_inc(&rdev->nr_pending); | |
4073 | } | |
6bfe0b49 | 4074 | } |
415e72d0 N |
4075 | clear_bit(R5_Insync, &dev->flags); |
4076 | if (!rdev) | |
4077 | /* Not in-sync */; | |
31c176ec N |
4078 | else if (is_bad) { |
4079 | /* also not in-sync */ | |
18b9837e N |
4080 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4081 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4082 | /* treat as in-sync, but with a read error |
4083 | * which we can now try to correct | |
4084 | */ | |
4085 | set_bit(R5_Insync, &dev->flags); | |
4086 | set_bit(R5_ReadError, &dev->flags); | |
4087 | } | |
4088 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4089 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4090 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4091 | /* in sync if before recovery_offset */ |
30d7a483 N |
4092 | set_bit(R5_Insync, &dev->flags); |
4093 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4094 | test_bit(R5_Expanded, &dev->flags)) | |
4095 | /* If we've reshaped into here, we assume it is Insync. | |
4096 | * We will shortly update recovery_offset to make | |
4097 | * it official. | |
4098 | */ | |
4099 | set_bit(R5_Insync, &dev->flags); | |
4100 | ||
1cc03eb9 | 4101 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4102 | /* This flag does not apply to '.replacement' |
4103 | * only to .rdev, so make sure to check that*/ | |
4104 | struct md_rdev *rdev2 = rcu_dereference( | |
4105 | conf->disks[i].rdev); | |
4106 | if (rdev2 == rdev) | |
4107 | clear_bit(R5_Insync, &dev->flags); | |
4108 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4109 | s->handle_bad_blocks = 1; |
14a75d3e | 4110 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4111 | } else |
4112 | clear_bit(R5_WriteError, &dev->flags); | |
4113 | } | |
1cc03eb9 | 4114 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4115 | /* This flag does not apply to '.replacement' |
4116 | * only to .rdev, so make sure to check that*/ | |
4117 | struct md_rdev *rdev2 = rcu_dereference( | |
4118 | conf->disks[i].rdev); | |
4119 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4120 | s->handle_bad_blocks = 1; |
14a75d3e | 4121 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4122 | } else |
4123 | clear_bit(R5_MadeGood, &dev->flags); | |
4124 | } | |
977df362 N |
4125 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4126 | struct md_rdev *rdev2 = rcu_dereference( | |
4127 | conf->disks[i].replacement); | |
4128 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4129 | s->handle_bad_blocks = 1; | |
4130 | atomic_inc(&rdev2->nr_pending); | |
4131 | } else | |
4132 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4133 | } | |
415e72d0 | 4134 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4135 | /* The ReadError flag will just be confusing now */ |
4136 | clear_bit(R5_ReadError, &dev->flags); | |
4137 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4138 | } |
415e72d0 N |
4139 | if (test_bit(R5_ReadError, &dev->flags)) |
4140 | clear_bit(R5_Insync, &dev->flags); | |
4141 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4142 | if (s->failed < 2) |
4143 | s->failed_num[s->failed] = i; | |
4144 | s->failed++; | |
9a3e1101 N |
4145 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4146 | do_recovery = 1; | |
415e72d0 | 4147 | } |
1da177e4 | 4148 | } |
9a3e1101 N |
4149 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4150 | /* If there is a failed device being replaced, | |
4151 | * we must be recovering. | |
4152 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4153 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4154 | * else we can only be replacing |
4155 | * sync and recovery both need to read all devices, and so | |
4156 | * use the same flag. | |
4157 | */ | |
4158 | if (do_recovery || | |
c6d2e084 | 4159 | sh->sector >= conf->mddev->recovery_cp || |
4160 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4161 | s->syncing = 1; |
4162 | else | |
4163 | s->replacing = 1; | |
4164 | } | |
1da177e4 | 4165 | rcu_read_unlock(); |
cc94015a N |
4166 | } |
4167 | ||
59fc630b | 4168 | static int clear_batch_ready(struct stripe_head *sh) |
4169 | { | |
b15a9dbd N |
4170 | /* Return '1' if this is a member of batch, or |
4171 | * '0' if it is a lone stripe or a head which can now be | |
4172 | * handled. | |
4173 | */ | |
59fc630b | 4174 | struct stripe_head *tmp; |
4175 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4176 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4177 | spin_lock(&sh->stripe_lock); |
4178 | if (!sh->batch_head) { | |
4179 | spin_unlock(&sh->stripe_lock); | |
4180 | return 0; | |
4181 | } | |
4182 | ||
4183 | /* | |
4184 | * this stripe could be added to a batch list before we check | |
4185 | * BATCH_READY, skips it | |
4186 | */ | |
4187 | if (sh->batch_head != sh) { | |
4188 | spin_unlock(&sh->stripe_lock); | |
4189 | return 1; | |
4190 | } | |
4191 | spin_lock(&sh->batch_lock); | |
4192 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4193 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4194 | spin_unlock(&sh->batch_lock); | |
4195 | spin_unlock(&sh->stripe_lock); | |
4196 | ||
4197 | /* | |
4198 | * BATCH_READY is cleared, no new stripes can be added. | |
4199 | * batch_list can be accessed without lock | |
4200 | */ | |
4201 | return 0; | |
4202 | } | |
4203 | ||
3960ce79 N |
4204 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4205 | unsigned long handle_flags) | |
72ac7330 | 4206 | { |
4e3d62ff | 4207 | struct stripe_head *sh, *next; |
72ac7330 | 4208 | int i; |
fb642b92 | 4209 | int do_wakeup = 0; |
72ac7330 | 4210 | |
bb27051f N |
4211 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4212 | ||
72ac7330 | 4213 | list_del_init(&sh->batch_list); |
4214 | ||
1b956f7a N |
4215 | WARN_ON_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
4216 | (1 << STRIPE_SYNCING) | | |
4217 | (1 << STRIPE_REPLACED) | | |
4218 | (1 << STRIPE_PREREAD_ACTIVE) | | |
4219 | (1 << STRIPE_DELAYED) | | |
4220 | (1 << STRIPE_BIT_DELAY) | | |
4221 | (1 << STRIPE_FULL_WRITE) | | |
4222 | (1 << STRIPE_BIOFILL_RUN) | | |
4223 | (1 << STRIPE_COMPUTE_RUN) | | |
4224 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4225 | (1 << STRIPE_DISCARD) | | |
4226 | (1 << STRIPE_BATCH_READY) | | |
4227 | (1 << STRIPE_BATCH_ERR) | | |
4228 | (1 << STRIPE_BITMAP_PENDING))); | |
4229 | WARN_ON_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4230 | (1 << STRIPE_REPLACED))); | |
4231 | ||
4232 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
4233 | (1 << STRIPE_DEGRADED)), | |
4234 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4235 | ||
72ac7330 | 4236 | sh->check_state = head_sh->check_state; |
4237 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4238 | for (i = 0; i < sh->disks; i++) { |
4239 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4240 | do_wakeup = 1; | |
72ac7330 | 4241 | sh->dev[i].flags = head_sh->dev[i].flags & |
4242 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4243 | } |
72ac7330 | 4244 | spin_lock_irq(&sh->stripe_lock); |
4245 | sh->batch_head = NULL; | |
4246 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4247 | if (handle_flags == 0 || |
4248 | sh->state & handle_flags) | |
4249 | set_bit(STRIPE_HANDLE, &sh->state); | |
72ac7330 | 4250 | release_stripe(sh); |
72ac7330 | 4251 | } |
fb642b92 N |
4252 | spin_lock_irq(&head_sh->stripe_lock); |
4253 | head_sh->batch_head = NULL; | |
4254 | spin_unlock_irq(&head_sh->stripe_lock); | |
4255 | for (i = 0; i < head_sh->disks; i++) | |
4256 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4257 | do_wakeup = 1; | |
3960ce79 N |
4258 | if (head_sh->state & handle_flags) |
4259 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4260 | |
4261 | if (do_wakeup) | |
4262 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4263 | } |
4264 | ||
cc94015a N |
4265 | static void handle_stripe(struct stripe_head *sh) |
4266 | { | |
4267 | struct stripe_head_state s; | |
d1688a6d | 4268 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4269 | int i; |
84789554 N |
4270 | int prexor; |
4271 | int disks = sh->disks; | |
474af965 | 4272 | struct r5dev *pdev, *qdev; |
cc94015a N |
4273 | |
4274 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4275 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4276 | /* already being handled, ensure it gets handled |
4277 | * again when current action finishes */ | |
4278 | set_bit(STRIPE_HANDLE, &sh->state); | |
4279 | return; | |
4280 | } | |
4281 | ||
59fc630b | 4282 | if (clear_batch_ready(sh) ) { |
4283 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4284 | return; | |
4285 | } | |
4286 | ||
4e3d62ff | 4287 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4288 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4289 | |
dabc4ec6 | 4290 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4291 | spin_lock(&sh->stripe_lock); |
4292 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4293 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4294 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4295 | set_bit(STRIPE_SYNCING, &sh->state); | |
4296 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4297 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4298 | } |
4299 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4300 | } |
4301 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4302 | ||
4303 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4304 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4305 | (unsigned long long)sh->sector, sh->state, | |
4306 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4307 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4308 | |
acfe726b | 4309 | analyse_stripe(sh, &s); |
c5a31000 | 4310 | |
bc2607f3 N |
4311 | if (s.handle_bad_blocks) { |
4312 | set_bit(STRIPE_HANDLE, &sh->state); | |
4313 | goto finish; | |
4314 | } | |
4315 | ||
474af965 N |
4316 | if (unlikely(s.blocked_rdev)) { |
4317 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4318 | s.replacing || s.to_write || s.written) { |
474af965 N |
4319 | set_bit(STRIPE_HANDLE, &sh->state); |
4320 | goto finish; | |
4321 | } | |
4322 | /* There is nothing for the blocked_rdev to block */ | |
4323 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4324 | s.blocked_rdev = NULL; | |
4325 | } | |
4326 | ||
4327 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4328 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4329 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4330 | } | |
4331 | ||
4332 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4333 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4334 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4335 | s.failed_num[0], s.failed_num[1]); | |
4336 | /* check if the array has lost more than max_degraded devices and, | |
4337 | * if so, some requests might need to be failed. | |
4338 | */ | |
9a3f530f N |
4339 | if (s.failed > conf->max_degraded) { |
4340 | sh->check_state = 0; | |
4341 | sh->reconstruct_state = 0; | |
626f2092 | 4342 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4343 | if (s.to_read+s.to_write+s.written) |
4344 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4345 | if (s.syncing + s.replacing) |
9a3f530f N |
4346 | handle_failed_sync(conf, sh, &s); |
4347 | } | |
474af965 | 4348 | |
84789554 N |
4349 | /* Now we check to see if any write operations have recently |
4350 | * completed | |
4351 | */ | |
4352 | prexor = 0; | |
4353 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4354 | prexor = 1; | |
4355 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4356 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4357 | sh->reconstruct_state = reconstruct_state_idle; | |
4358 | ||
4359 | /* All the 'written' buffers and the parity block are ready to | |
4360 | * be written back to disk | |
4361 | */ | |
9e444768 SL |
4362 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4363 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4364 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4365 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4366 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4367 | for (i = disks; i--; ) { |
4368 | struct r5dev *dev = &sh->dev[i]; | |
4369 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4370 | (i == sh->pd_idx || i == sh->qd_idx || | |
4371 | dev->written)) { | |
4372 | pr_debug("Writing block %d\n", i); | |
4373 | set_bit(R5_Wantwrite, &dev->flags); | |
4374 | if (prexor) | |
4375 | continue; | |
9c4bdf69 N |
4376 | if (s.failed > 1) |
4377 | continue; | |
84789554 N |
4378 | if (!test_bit(R5_Insync, &dev->flags) || |
4379 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4380 | s.failed == 0)) | |
4381 | set_bit(STRIPE_INSYNC, &sh->state); | |
4382 | } | |
4383 | } | |
4384 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4385 | s.dec_preread_active = 1; | |
4386 | } | |
4387 | ||
ef5b7c69 N |
4388 | /* |
4389 | * might be able to return some write requests if the parity blocks | |
4390 | * are safe, or on a failed drive | |
4391 | */ | |
4392 | pdev = &sh->dev[sh->pd_idx]; | |
4393 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4394 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4395 | qdev = &sh->dev[sh->qd_idx]; | |
4396 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4397 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4398 | || conf->level < 6; | |
4399 | ||
4400 | if (s.written && | |
4401 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4402 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4403 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4404 | test_bit(R5_Discard, &pdev->flags))))) && | |
4405 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4406 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4407 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4408 | test_bit(R5_Discard, &qdev->flags)))))) | |
4409 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4410 | ||
4411 | /* Now we might consider reading some blocks, either to check/generate | |
4412 | * parity, or to satisfy requests | |
4413 | * or to load a block that is being partially written. | |
4414 | */ | |
4415 | if (s.to_read || s.non_overwrite | |
4416 | || (conf->level == 6 && s.to_write && s.failed) | |
4417 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4418 | || s.replacing | |
4419 | || s.expanding) | |
4420 | handle_stripe_fill(sh, &s, disks); | |
4421 | ||
84789554 N |
4422 | /* Now to consider new write requests and what else, if anything |
4423 | * should be read. We do not handle new writes when: | |
4424 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
4425 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4426 | * block. | |
4427 | */ | |
4428 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
4429 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4430 | ||
4431 | /* maybe we need to check and possibly fix the parity for this stripe | |
4432 | * Any reads will already have been scheduled, so we just see if enough | |
4433 | * data is available. The parity check is held off while parity | |
4434 | * dependent operations are in flight. | |
4435 | */ | |
4436 | if (sh->check_state || | |
4437 | (s.syncing && s.locked == 0 && | |
4438 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4439 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4440 | if (conf->level == 6) | |
4441 | handle_parity_checks6(conf, sh, &s, disks); | |
4442 | else | |
4443 | handle_parity_checks5(conf, sh, &s, disks); | |
4444 | } | |
c5a31000 | 4445 | |
f94c0b66 N |
4446 | if ((s.replacing || s.syncing) && s.locked == 0 |
4447 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4448 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4449 | /* Write out to replacement devices where possible */ |
4450 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4451 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4452 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4453 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4454 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4455 | s.locked++; | |
4456 | } | |
f94c0b66 N |
4457 | if (s.replacing) |
4458 | set_bit(STRIPE_INSYNC, &sh->state); | |
4459 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4460 | } |
4461 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4462 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4463 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4464 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4465 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4466 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4467 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4468 | } |
4469 | ||
4470 | /* If the failed drives are just a ReadError, then we might need | |
4471 | * to progress the repair/check process | |
4472 | */ | |
4473 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4474 | for (i = 0; i < s.failed; i++) { | |
4475 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4476 | if (test_bit(R5_ReadError, &dev->flags) | |
4477 | && !test_bit(R5_LOCKED, &dev->flags) | |
4478 | && test_bit(R5_UPTODATE, &dev->flags) | |
4479 | ) { | |
4480 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4481 | set_bit(R5_Wantwrite, &dev->flags); | |
4482 | set_bit(R5_ReWrite, &dev->flags); | |
4483 | set_bit(R5_LOCKED, &dev->flags); | |
4484 | s.locked++; | |
4485 | } else { | |
4486 | /* let's read it back */ | |
4487 | set_bit(R5_Wantread, &dev->flags); | |
4488 | set_bit(R5_LOCKED, &dev->flags); | |
4489 | s.locked++; | |
4490 | } | |
4491 | } | |
4492 | } | |
4493 | ||
3687c061 N |
4494 | /* Finish reconstruct operations initiated by the expansion process */ |
4495 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4496 | struct stripe_head *sh_src | |
4497 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
4498 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
4499 | /* sh cannot be written until sh_src has been read. | |
4500 | * so arrange for sh to be delayed a little | |
4501 | */ | |
4502 | set_bit(STRIPE_DELAYED, &sh->state); | |
4503 | set_bit(STRIPE_HANDLE, &sh->state); | |
4504 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4505 | &sh_src->state)) | |
4506 | atomic_inc(&conf->preread_active_stripes); | |
4507 | release_stripe(sh_src); | |
4508 | goto finish; | |
4509 | } | |
4510 | if (sh_src) | |
4511 | release_stripe(sh_src); | |
4512 | ||
4513 | sh->reconstruct_state = reconstruct_state_idle; | |
4514 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4515 | for (i = conf->raid_disks; i--; ) { | |
4516 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4517 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4518 | s.locked++; | |
4519 | } | |
4520 | } | |
f416885e | 4521 | |
3687c061 N |
4522 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4523 | !sh->reconstruct_state) { | |
4524 | /* Need to write out all blocks after computing parity */ | |
4525 | sh->disks = conf->raid_disks; | |
4526 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4527 | schedule_reconstruction(sh, &s, 1, 1); | |
4528 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4529 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4530 | atomic_dec(&conf->reshape_stripes); | |
4531 | wake_up(&conf->wait_for_overlap); | |
4532 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4533 | } | |
4534 | ||
4535 | if (s.expanding && s.locked == 0 && | |
4536 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4537 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4538 | |
3687c061 | 4539 | finish: |
6bfe0b49 | 4540 | /* wait for this device to become unblocked */ |
5f066c63 N |
4541 | if (unlikely(s.blocked_rdev)) { |
4542 | if (conf->mddev->external) | |
4543 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4544 | conf->mddev); | |
4545 | else | |
4546 | /* Internal metadata will immediately | |
4547 | * be written by raid5d, so we don't | |
4548 | * need to wait here. | |
4549 | */ | |
4550 | rdev_dec_pending(s.blocked_rdev, | |
4551 | conf->mddev); | |
4552 | } | |
6bfe0b49 | 4553 | |
bc2607f3 N |
4554 | if (s.handle_bad_blocks) |
4555 | for (i = disks; i--; ) { | |
3cb03002 | 4556 | struct md_rdev *rdev; |
bc2607f3 N |
4557 | struct r5dev *dev = &sh->dev[i]; |
4558 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4559 | /* We own a safe reference to the rdev */ | |
4560 | rdev = conf->disks[i].rdev; | |
4561 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4562 | STRIPE_SECTORS, 0)) | |
4563 | md_error(conf->mddev, rdev); | |
4564 | rdev_dec_pending(rdev, conf->mddev); | |
4565 | } | |
b84db560 N |
4566 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4567 | rdev = conf->disks[i].rdev; | |
4568 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4569 | STRIPE_SECTORS, 0); |
b84db560 N |
4570 | rdev_dec_pending(rdev, conf->mddev); |
4571 | } | |
977df362 N |
4572 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4573 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4574 | if (!rdev) |
4575 | /* rdev have been moved down */ | |
4576 | rdev = conf->disks[i].rdev; | |
977df362 | 4577 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4578 | STRIPE_SECTORS, 0); |
977df362 N |
4579 | rdev_dec_pending(rdev, conf->mddev); |
4580 | } | |
bc2607f3 N |
4581 | } |
4582 | ||
6c0069c0 YT |
4583 | if (s.ops_request) |
4584 | raid_run_ops(sh, s.ops_request); | |
4585 | ||
f0e43bcd | 4586 | ops_run_io(sh, &s); |
16a53ecc | 4587 | |
c5709ef6 | 4588 | if (s.dec_preread_active) { |
729a1866 | 4589 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4590 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4591 | * have actually been submitted. |
4592 | */ | |
4593 | atomic_dec(&conf->preread_active_stripes); | |
4594 | if (atomic_read(&conf->preread_active_stripes) < | |
4595 | IO_THRESHOLD) | |
4596 | md_wakeup_thread(conf->mddev->thread); | |
4597 | } | |
4598 | ||
c5709ef6 | 4599 | return_io(s.return_bi); |
16a53ecc | 4600 | |
257a4b42 | 4601 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4602 | } |
4603 | ||
d1688a6d | 4604 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4605 | { |
4606 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4607 | while (!list_empty(&conf->delayed_list)) { | |
4608 | struct list_head *l = conf->delayed_list.next; | |
4609 | struct stripe_head *sh; | |
4610 | sh = list_entry(l, struct stripe_head, lru); | |
4611 | list_del_init(l); | |
4612 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4613 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4614 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4615 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4616 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4617 | } |
482c0834 | 4618 | } |
16a53ecc N |
4619 | } |
4620 | ||
566c09c5 SL |
4621 | static void activate_bit_delay(struct r5conf *conf, |
4622 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4623 | { |
4624 | /* device_lock is held */ | |
4625 | struct list_head head; | |
4626 | list_add(&head, &conf->bitmap_list); | |
4627 | list_del_init(&conf->bitmap_list); | |
4628 | while (!list_empty(&head)) { | |
4629 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4630 | int hash; |
16a53ecc N |
4631 | list_del_init(&sh->lru); |
4632 | atomic_inc(&sh->count); | |
566c09c5 SL |
4633 | hash = sh->hash_lock_index; |
4634 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4635 | } |
4636 | } | |
4637 | ||
5c675f83 | 4638 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4639 | { |
d1688a6d | 4640 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4641 | |
4642 | /* No difference between reads and writes. Just check | |
4643 | * how busy the stripe_cache is | |
4644 | */ | |
3fa841d7 | 4645 | |
5423399a | 4646 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd N |
4647 | return 1; |
4648 | if (conf->quiesce) | |
4649 | return 1; | |
4bda556a | 4650 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4651 | return 1; |
4652 | ||
4653 | return 0; | |
4654 | } | |
4655 | ||
23032a0e RBJ |
4656 | /* We want read requests to align with chunks where possible, |
4657 | * but write requests don't need to. | |
4658 | */ | |
64590f45 | 4659 | static int raid5_mergeable_bvec(struct mddev *mddev, |
cc371e66 AK |
4660 | struct bvec_merge_data *bvm, |
4661 | struct bio_vec *biovec) | |
23032a0e | 4662 | { |
cc371e66 | 4663 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 4664 | int max; |
9d8f0363 | 4665 | unsigned int chunk_sectors = mddev->chunk_sectors; |
cc371e66 | 4666 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 4667 | |
9ffc8f7c EM |
4668 | /* |
4669 | * always allow writes to be mergeable, read as well if array | |
4670 | * is degraded as we'll go through stripe cache anyway. | |
4671 | */ | |
4672 | if ((bvm->bi_rw & 1) == WRITE || mddev->degraded) | |
4673 | return biovec->bv_len; | |
23032a0e | 4674 | |
664e7c41 AN |
4675 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
4676 | chunk_sectors = mddev->new_chunk_sectors; | |
23032a0e RBJ |
4677 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
4678 | if (max < 0) max = 0; | |
4679 | if (max <= biovec->bv_len && bio_sectors == 0) | |
4680 | return biovec->bv_len; | |
4681 | else | |
4682 | return max; | |
4683 | } | |
4684 | ||
fd01b88c | 4685 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4686 | { |
4f024f37 | 4687 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
9d8f0363 | 4688 | unsigned int chunk_sectors = mddev->chunk_sectors; |
aa8b57aa | 4689 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4690 | |
664e7c41 AN |
4691 | if (mddev->new_chunk_sectors < mddev->chunk_sectors) |
4692 | chunk_sectors = mddev->new_chunk_sectors; | |
f679623f RBJ |
4693 | return chunk_sectors >= |
4694 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4695 | } | |
4696 | ||
46031f9a RBJ |
4697 | /* |
4698 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4699 | * later sampled by raid5d. | |
4700 | */ | |
d1688a6d | 4701 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4702 | { |
4703 | unsigned long flags; | |
4704 | ||
4705 | spin_lock_irqsave(&conf->device_lock, flags); | |
4706 | ||
4707 | bi->bi_next = conf->retry_read_aligned_list; | |
4708 | conf->retry_read_aligned_list = bi; | |
4709 | ||
4710 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4711 | md_wakeup_thread(conf->mddev->thread); | |
4712 | } | |
4713 | ||
d1688a6d | 4714 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4715 | { |
4716 | struct bio *bi; | |
4717 | ||
4718 | bi = conf->retry_read_aligned; | |
4719 | if (bi) { | |
4720 | conf->retry_read_aligned = NULL; | |
4721 | return bi; | |
4722 | } | |
4723 | bi = conf->retry_read_aligned_list; | |
4724 | if(bi) { | |
387bb173 | 4725 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4726 | bi->bi_next = NULL; |
960e739d JA |
4727 | /* |
4728 | * this sets the active strip count to 1 and the processed | |
4729 | * strip count to zero (upper 8 bits) | |
4730 | */ | |
e7836bd6 | 4731 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
4732 | } |
4733 | ||
4734 | return bi; | |
4735 | } | |
4736 | ||
f679623f RBJ |
4737 | /* |
4738 | * The "raid5_align_endio" should check if the read succeeded and if it | |
4739 | * did, call bio_endio on the original bio (having bio_put the new bio | |
4740 | * first). | |
4741 | * If the read failed.. | |
4742 | */ | |
6712ecf8 | 4743 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
4744 | { |
4745 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 4746 | struct mddev *mddev; |
d1688a6d | 4747 | struct r5conf *conf; |
46031f9a | 4748 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 4749 | struct md_rdev *rdev; |
46031f9a | 4750 | |
f679623f | 4751 | bio_put(bi); |
46031f9a | 4752 | |
46031f9a RBJ |
4753 | rdev = (void*)raid_bi->bi_next; |
4754 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
4755 | mddev = rdev->mddev; |
4756 | conf = mddev->private; | |
46031f9a RBJ |
4757 | |
4758 | rdev_dec_pending(rdev, conf->mddev); | |
4759 | ||
4760 | if (!error && uptodate) { | |
0a82a8d1 LT |
4761 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
4762 | raid_bi, 0); | |
6712ecf8 | 4763 | bio_endio(raid_bi, 0); |
46031f9a | 4764 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 4765 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 4766 | return; |
46031f9a RBJ |
4767 | } |
4768 | ||
45b4233c | 4769 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
4770 | |
4771 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
4772 | } |
4773 | ||
387bb173 NB |
4774 | static int bio_fits_rdev(struct bio *bi) |
4775 | { | |
165125e1 | 4776 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 4777 | |
aa8b57aa | 4778 | if (bio_sectors(bi) > queue_max_sectors(q)) |
387bb173 NB |
4779 | return 0; |
4780 | blk_recount_segments(q, bi); | |
8a78362c | 4781 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
4782 | return 0; |
4783 | ||
4784 | if (q->merge_bvec_fn) | |
4785 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
4786 | * just just give up | |
4787 | */ | |
4788 | return 0; | |
4789 | ||
4790 | return 1; | |
4791 | } | |
4792 | ||
fd01b88c | 4793 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 4794 | { |
d1688a6d | 4795 | struct r5conf *conf = mddev->private; |
8553fe7e | 4796 | int dd_idx; |
f679623f | 4797 | struct bio* align_bi; |
3cb03002 | 4798 | struct md_rdev *rdev; |
671488cc | 4799 | sector_t end_sector; |
f679623f RBJ |
4800 | |
4801 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 4802 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
4803 | return 0; |
4804 | } | |
4805 | /* | |
a167f663 | 4806 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 4807 | */ |
a167f663 | 4808 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
4809 | if (!align_bi) |
4810 | return 0; | |
4811 | /* | |
4812 | * set bi_end_io to a new function, and set bi_private to the | |
4813 | * original bio. | |
4814 | */ | |
4815 | align_bi->bi_end_io = raid5_align_endio; | |
4816 | align_bi->bi_private = raid_bio; | |
4817 | /* | |
4818 | * compute position | |
4819 | */ | |
4f024f37 KO |
4820 | align_bi->bi_iter.bi_sector = |
4821 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
4822 | 0, &dd_idx, NULL); | |
f679623f | 4823 | |
f73a1c7d | 4824 | end_sector = bio_end_sector(align_bi); |
f679623f | 4825 | rcu_read_lock(); |
671488cc N |
4826 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
4827 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
4828 | rdev->recovery_offset < end_sector) { | |
4829 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
4830 | if (rdev && | |
4831 | (test_bit(Faulty, &rdev->flags) || | |
4832 | !(test_bit(In_sync, &rdev->flags) || | |
4833 | rdev->recovery_offset >= end_sector))) | |
4834 | rdev = NULL; | |
4835 | } | |
4836 | if (rdev) { | |
31c176ec N |
4837 | sector_t first_bad; |
4838 | int bad_sectors; | |
4839 | ||
f679623f RBJ |
4840 | atomic_inc(&rdev->nr_pending); |
4841 | rcu_read_unlock(); | |
46031f9a RBJ |
4842 | raid_bio->bi_next = (void*)rdev; |
4843 | align_bi->bi_bdev = rdev->bdev; | |
3fd83717 | 4844 | __clear_bit(BIO_SEG_VALID, &align_bi->bi_flags); |
46031f9a | 4845 | |
31c176ec | 4846 | if (!bio_fits_rdev(align_bi) || |
4f024f37 KO |
4847 | is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4848 | bio_sectors(align_bi), | |
31c176ec N |
4849 | &first_bad, &bad_sectors)) { |
4850 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
4851 | bio_put(align_bi); |
4852 | rdev_dec_pending(rdev, mddev); | |
4853 | return 0; | |
4854 | } | |
4855 | ||
6c0544e2 | 4856 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 4857 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 4858 | |
46031f9a | 4859 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 4860 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 4861 | conf->quiesce == 0, |
eed8c02e | 4862 | conf->device_lock); |
46031f9a RBJ |
4863 | atomic_inc(&conf->active_aligned_reads); |
4864 | spin_unlock_irq(&conf->device_lock); | |
4865 | ||
e3620a3a JB |
4866 | if (mddev->gendisk) |
4867 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
4868 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 4869 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
4870 | generic_make_request(align_bi); |
4871 | return 1; | |
4872 | } else { | |
4873 | rcu_read_unlock(); | |
46031f9a | 4874 | bio_put(align_bi); |
f679623f RBJ |
4875 | return 0; |
4876 | } | |
4877 | } | |
4878 | ||
8b3e6cdc DW |
4879 | /* __get_priority_stripe - get the next stripe to process |
4880 | * | |
4881 | * Full stripe writes are allowed to pass preread active stripes up until | |
4882 | * the bypass_threshold is exceeded. In general the bypass_count | |
4883 | * increments when the handle_list is handled before the hold_list; however, it | |
4884 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
4885 | * stripe with in flight i/o. The bypass_count will be reset when the | |
4886 | * head of the hold_list has changed, i.e. the head was promoted to the | |
4887 | * handle_list. | |
4888 | */ | |
851c30c9 | 4889 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 4890 | { |
851c30c9 SL |
4891 | struct stripe_head *sh = NULL, *tmp; |
4892 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 4893 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
4894 | |
4895 | if (conf->worker_cnt_per_group == 0) { | |
4896 | handle_list = &conf->handle_list; | |
4897 | } else if (group != ANY_GROUP) { | |
4898 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 4899 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
4900 | } else { |
4901 | int i; | |
4902 | for (i = 0; i < conf->group_cnt; i++) { | |
4903 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 4904 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
4905 | if (!list_empty(handle_list)) |
4906 | break; | |
4907 | } | |
4908 | } | |
8b3e6cdc DW |
4909 | |
4910 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
4911 | __func__, | |
851c30c9 | 4912 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
4913 | list_empty(&conf->hold_list) ? "empty" : "busy", |
4914 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
4915 | ||
851c30c9 SL |
4916 | if (!list_empty(handle_list)) { |
4917 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
4918 | |
4919 | if (list_empty(&conf->hold_list)) | |
4920 | conf->bypass_count = 0; | |
4921 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
4922 | if (conf->hold_list.next == conf->last_hold) | |
4923 | conf->bypass_count++; | |
4924 | else { | |
4925 | conf->last_hold = conf->hold_list.next; | |
4926 | conf->bypass_count -= conf->bypass_threshold; | |
4927 | if (conf->bypass_count < 0) | |
4928 | conf->bypass_count = 0; | |
4929 | } | |
4930 | } | |
4931 | } else if (!list_empty(&conf->hold_list) && | |
4932 | ((conf->bypass_threshold && | |
4933 | conf->bypass_count > conf->bypass_threshold) || | |
4934 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
4935 | |
4936 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
4937 | if (conf->worker_cnt_per_group == 0 || | |
4938 | group == ANY_GROUP || | |
4939 | !cpu_online(tmp->cpu) || | |
4940 | cpu_to_group(tmp->cpu) == group) { | |
4941 | sh = tmp; | |
4942 | break; | |
4943 | } | |
4944 | } | |
4945 | ||
4946 | if (sh) { | |
4947 | conf->bypass_count -= conf->bypass_threshold; | |
4948 | if (conf->bypass_count < 0) | |
4949 | conf->bypass_count = 0; | |
4950 | } | |
bfc90cb0 | 4951 | wg = NULL; |
851c30c9 SL |
4952 | } |
4953 | ||
4954 | if (!sh) | |
8b3e6cdc DW |
4955 | return NULL; |
4956 | ||
bfc90cb0 SL |
4957 | if (wg) { |
4958 | wg->stripes_cnt--; | |
4959 | sh->group = NULL; | |
4960 | } | |
8b3e6cdc | 4961 | list_del_init(&sh->lru); |
c7a6d35e | 4962 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
4963 | return sh; |
4964 | } | |
f679623f | 4965 | |
8811b596 SL |
4966 | struct raid5_plug_cb { |
4967 | struct blk_plug_cb cb; | |
4968 | struct list_head list; | |
566c09c5 | 4969 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
4970 | }; |
4971 | ||
4972 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4973 | { | |
4974 | struct raid5_plug_cb *cb = container_of( | |
4975 | blk_cb, struct raid5_plug_cb, cb); | |
4976 | struct stripe_head *sh; | |
4977 | struct mddev *mddev = cb->cb.data; | |
4978 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 4979 | int cnt = 0; |
566c09c5 | 4980 | int hash; |
8811b596 SL |
4981 | |
4982 | if (cb->list.next && !list_empty(&cb->list)) { | |
4983 | spin_lock_irq(&conf->device_lock); | |
4984 | while (!list_empty(&cb->list)) { | |
4985 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
4986 | list_del_init(&sh->lru); | |
4987 | /* | |
4988 | * avoid race release_stripe_plug() sees | |
4989 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
4990 | * is still in our list | |
4991 | */ | |
4e857c58 | 4992 | smp_mb__before_atomic(); |
8811b596 | 4993 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
4994 | /* |
4995 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
4996 | * case, the count is always > 1 here | |
4997 | */ | |
566c09c5 SL |
4998 | hash = sh->hash_lock_index; |
4999 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5000 | cnt++; |
8811b596 SL |
5001 | } |
5002 | spin_unlock_irq(&conf->device_lock); | |
5003 | } | |
566c09c5 SL |
5004 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5005 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5006 | if (mddev->queue) |
5007 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5008 | kfree(cb); |
5009 | } | |
5010 | ||
5011 | static void release_stripe_plug(struct mddev *mddev, | |
5012 | struct stripe_head *sh) | |
5013 | { | |
5014 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5015 | raid5_unplug, mddev, | |
5016 | sizeof(struct raid5_plug_cb)); | |
5017 | struct raid5_plug_cb *cb; | |
5018 | ||
5019 | if (!blk_cb) { | |
5020 | release_stripe(sh); | |
5021 | return; | |
5022 | } | |
5023 | ||
5024 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5025 | ||
566c09c5 SL |
5026 | if (cb->list.next == NULL) { |
5027 | int i; | |
8811b596 | 5028 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5029 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5030 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5031 | } | |
8811b596 SL |
5032 | |
5033 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5034 | list_add_tail(&sh->lru, &cb->list); | |
5035 | else | |
5036 | release_stripe(sh); | |
5037 | } | |
5038 | ||
620125f2 SL |
5039 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5040 | { | |
5041 | struct r5conf *conf = mddev->private; | |
5042 | sector_t logical_sector, last_sector; | |
5043 | struct stripe_head *sh; | |
5044 | int remaining; | |
5045 | int stripe_sectors; | |
5046 | ||
5047 | if (mddev->reshape_position != MaxSector) | |
5048 | /* Skip discard while reshape is happening */ | |
5049 | return; | |
5050 | ||
4f024f37 KO |
5051 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5052 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5053 | |
5054 | bi->bi_next = NULL; | |
5055 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5056 | ||
5057 | stripe_sectors = conf->chunk_sectors * | |
5058 | (conf->raid_disks - conf->max_degraded); | |
5059 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5060 | stripe_sectors); | |
5061 | sector_div(last_sector, stripe_sectors); | |
5062 | ||
5063 | logical_sector *= conf->chunk_sectors; | |
5064 | last_sector *= conf->chunk_sectors; | |
5065 | ||
5066 | for (; logical_sector < last_sector; | |
5067 | logical_sector += STRIPE_SECTORS) { | |
5068 | DEFINE_WAIT(w); | |
5069 | int d; | |
5070 | again: | |
5071 | sh = get_active_stripe(conf, logical_sector, 0, 0, 0); | |
5072 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5073 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5074 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5075 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
5076 | release_stripe(sh); | |
5077 | schedule(); | |
5078 | goto again; | |
5079 | } | |
5080 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5081 | spin_lock_irq(&sh->stripe_lock); |
5082 | for (d = 0; d < conf->raid_disks; d++) { | |
5083 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5084 | continue; | |
5085 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5086 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5087 | spin_unlock_irq(&sh->stripe_lock); | |
5088 | release_stripe(sh); | |
5089 | schedule(); | |
5090 | goto again; | |
5091 | } | |
5092 | } | |
f8dfcffd | 5093 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5094 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5095 | sh->overwrite_disks = 0; |
620125f2 SL |
5096 | for (d = 0; d < conf->raid_disks; d++) { |
5097 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5098 | continue; | |
5099 | sh->dev[d].towrite = bi; | |
5100 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5101 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5102 | sh->overwrite_disks++; |
620125f2 SL |
5103 | } |
5104 | spin_unlock_irq(&sh->stripe_lock); | |
5105 | if (conf->mddev->bitmap) { | |
5106 | for (d = 0; | |
5107 | d < conf->raid_disks - conf->max_degraded; | |
5108 | d++) | |
5109 | bitmap_startwrite(mddev->bitmap, | |
5110 | sh->sector, | |
5111 | STRIPE_SECTORS, | |
5112 | 0); | |
5113 | sh->bm_seq = conf->seq_flush + 1; | |
5114 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5115 | } | |
5116 | ||
5117 | set_bit(STRIPE_HANDLE, &sh->state); | |
5118 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5119 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5120 | atomic_inc(&conf->preread_active_stripes); | |
5121 | release_stripe_plug(mddev, sh); | |
5122 | } | |
5123 | ||
5124 | remaining = raid5_dec_bi_active_stripes(bi); | |
5125 | if (remaining == 0) { | |
5126 | md_write_end(mddev); | |
5127 | bio_endio(bi, 0); | |
5128 | } | |
5129 | } | |
5130 | ||
b4fdcb02 | 5131 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5132 | { |
d1688a6d | 5133 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5134 | int dd_idx; |
1da177e4 LT |
5135 | sector_t new_sector; |
5136 | sector_t logical_sector, last_sector; | |
5137 | struct stripe_head *sh; | |
a362357b | 5138 | const int rw = bio_data_dir(bi); |
49077326 | 5139 | int remaining; |
27c0f68f SL |
5140 | DEFINE_WAIT(w); |
5141 | bool do_prepare; | |
1da177e4 | 5142 | |
e9c7469b TH |
5143 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
5144 | md_flush_request(mddev, bi); | |
5a7bbad2 | 5145 | return; |
e5dcdd80 N |
5146 | } |
5147 | ||
3d310eb7 | 5148 | md_write_start(mddev, bi); |
06d91a5f | 5149 | |
9ffc8f7c EM |
5150 | /* |
5151 | * If array is degraded, better not do chunk aligned read because | |
5152 | * later we might have to read it again in order to reconstruct | |
5153 | * data on failed drives. | |
5154 | */ | |
5155 | if (rw == READ && mddev->degraded == 0 && | |
52488615 | 5156 | mddev->reshape_position == MaxSector && |
21a52c6d | 5157 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 5158 | return; |
52488615 | 5159 | |
620125f2 SL |
5160 | if (unlikely(bi->bi_rw & REQ_DISCARD)) { |
5161 | make_discard_request(mddev, bi); | |
5162 | return; | |
5163 | } | |
5164 | ||
4f024f37 | 5165 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5166 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5167 | bi->bi_next = NULL; |
5168 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5169 | |
27c0f68f | 5170 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5171 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5172 | int previous; |
c46501b2 | 5173 | int seq; |
b578d55f | 5174 | |
27c0f68f | 5175 | do_prepare = false; |
7ecaa1e6 | 5176 | retry: |
c46501b2 | 5177 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5178 | previous = 0; |
27c0f68f SL |
5179 | if (do_prepare) |
5180 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5181 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5182 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5183 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5184 | * 64bit on a 32bit platform, and so it might be |
5185 | * possible to see a half-updated value | |
aeb878b0 | 5186 | * Of course reshape_progress could change after |
df8e7f76 N |
5187 | * the lock is dropped, so once we get a reference |
5188 | * to the stripe that we think it is, we will have | |
5189 | * to check again. | |
5190 | */ | |
7ecaa1e6 | 5191 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5192 | if (mddev->reshape_backwards |
fef9c61f N |
5193 | ? logical_sector < conf->reshape_progress |
5194 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5195 | previous = 1; |
5196 | } else { | |
2c810cdd | 5197 | if (mddev->reshape_backwards |
fef9c61f N |
5198 | ? logical_sector < conf->reshape_safe |
5199 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5200 | spin_unlock_irq(&conf->device_lock); |
5201 | schedule(); | |
27c0f68f | 5202 | do_prepare = true; |
b578d55f N |
5203 | goto retry; |
5204 | } | |
5205 | } | |
7ecaa1e6 N |
5206 | spin_unlock_irq(&conf->device_lock); |
5207 | } | |
16a53ecc | 5208 | |
112bf897 N |
5209 | new_sector = raid5_compute_sector(conf, logical_sector, |
5210 | previous, | |
911d4ee8 | 5211 | &dd_idx, NULL); |
0c55e022 | 5212 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
c46501b2 | 5213 | (unsigned long long)new_sector, |
1da177e4 LT |
5214 | (unsigned long long)logical_sector); |
5215 | ||
b5663ba4 | 5216 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 5217 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 5218 | if (sh) { |
b0f9ec04 | 5219 | if (unlikely(previous)) { |
7ecaa1e6 | 5220 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5221 | * stripe, so we must do the range check again. |
5222 | * Expansion could still move past after this | |
5223 | * test, but as we are holding a reference to | |
5224 | * 'sh', we know that if that happens, | |
5225 | * STRIPE_EXPANDING will get set and the expansion | |
5226 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5227 | */ |
5228 | int must_retry = 0; | |
5229 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5230 | if (mddev->reshape_backwards |
b0f9ec04 N |
5231 | ? logical_sector >= conf->reshape_progress |
5232 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5233 | /* mismatch, need to try again */ |
5234 | must_retry = 1; | |
5235 | spin_unlock_irq(&conf->device_lock); | |
5236 | if (must_retry) { | |
5237 | release_stripe(sh); | |
7a3ab908 | 5238 | schedule(); |
27c0f68f | 5239 | do_prepare = true; |
7ecaa1e6 N |
5240 | goto retry; |
5241 | } | |
5242 | } | |
c46501b2 N |
5243 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5244 | /* Might have got the wrong stripe_head | |
5245 | * by accident | |
5246 | */ | |
5247 | release_stripe(sh); | |
5248 | goto retry; | |
5249 | } | |
e62e58a5 | 5250 | |
ffd96e35 | 5251 | if (rw == WRITE && |
a5c308d4 | 5252 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
5253 | logical_sector < mddev->suspend_hi) { |
5254 | release_stripe(sh); | |
e62e58a5 N |
5255 | /* As the suspend_* range is controlled by |
5256 | * userspace, we want an interruptible | |
5257 | * wait. | |
5258 | */ | |
5259 | flush_signals(current); | |
5260 | prepare_to_wait(&conf->wait_for_overlap, | |
5261 | &w, TASK_INTERRUPTIBLE); | |
5262 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5263 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5264 | schedule(); |
27c0f68f SL |
5265 | do_prepare = true; |
5266 | } | |
e464eafd N |
5267 | goto retry; |
5268 | } | |
7ecaa1e6 N |
5269 | |
5270 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5271 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5272 | /* Stripe is busy expanding or |
5273 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5274 | * and wait a while |
5275 | */ | |
482c0834 | 5276 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5277 | release_stripe(sh); |
5278 | schedule(); | |
27c0f68f | 5279 | do_prepare = true; |
1da177e4 LT |
5280 | goto retry; |
5281 | } | |
6ed3003c N |
5282 | set_bit(STRIPE_HANDLE, &sh->state); |
5283 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5284 | if ((!sh->batch_head || sh == sh->batch_head) && |
5285 | (bi->bi_rw & REQ_SYNC) && | |
729a1866 N |
5286 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5287 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5288 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5289 | } else { |
5290 | /* cannot get stripe for read-ahead, just give-up */ | |
5291 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1da177e4 LT |
5292 | break; |
5293 | } | |
1da177e4 | 5294 | } |
27c0f68f | 5295 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5296 | |
e7836bd6 | 5297 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5298 | if (remaining == 0) { |
1da177e4 | 5299 | |
16a53ecc | 5300 | if ( rw == WRITE ) |
1da177e4 | 5301 | md_write_end(mddev); |
6712ecf8 | 5302 | |
0a82a8d1 LT |
5303 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5304 | bi, 0); | |
0e13fe23 | 5305 | bio_endio(bi, 0); |
1da177e4 | 5306 | } |
1da177e4 LT |
5307 | } |
5308 | ||
fd01b88c | 5309 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5310 | |
fd01b88c | 5311 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5312 | { |
52c03291 N |
5313 | /* reshaping is quite different to recovery/resync so it is |
5314 | * handled quite separately ... here. | |
5315 | * | |
5316 | * On each call to sync_request, we gather one chunk worth of | |
5317 | * destination stripes and flag them as expanding. | |
5318 | * Then we find all the source stripes and request reads. | |
5319 | * As the reads complete, handle_stripe will copy the data | |
5320 | * into the destination stripe and release that stripe. | |
5321 | */ | |
d1688a6d | 5322 | struct r5conf *conf = mddev->private; |
1da177e4 | 5323 | struct stripe_head *sh; |
ccfcc3c1 | 5324 | sector_t first_sector, last_sector; |
f416885e N |
5325 | int raid_disks = conf->previous_raid_disks; |
5326 | int data_disks = raid_disks - conf->max_degraded; | |
5327 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5328 | int i; |
5329 | int dd_idx; | |
c8f517c4 | 5330 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5331 | sector_t stripe_addr; |
7a661381 | 5332 | int reshape_sectors; |
ab69ae12 | 5333 | struct list_head stripes; |
52c03291 | 5334 | |
fef9c61f N |
5335 | if (sector_nr == 0) { |
5336 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5337 | if (mddev->reshape_backwards && |
fef9c61f N |
5338 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5339 | sector_nr = raid5_size(mddev, 0, 0) | |
5340 | - conf->reshape_progress; | |
2c810cdd | 5341 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5342 | conf->reshape_progress > 0) |
5343 | sector_nr = conf->reshape_progress; | |
f416885e | 5344 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5345 | if (sector_nr) { |
8dee7211 N |
5346 | mddev->curr_resync_completed = sector_nr; |
5347 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f N |
5348 | *skipped = 1; |
5349 | return sector_nr; | |
5350 | } | |
52c03291 N |
5351 | } |
5352 | ||
7a661381 N |
5353 | /* We need to process a full chunk at a time. |
5354 | * If old and new chunk sizes differ, we need to process the | |
5355 | * largest of these | |
5356 | */ | |
664e7c41 AN |
5357 | if (mddev->new_chunk_sectors > mddev->chunk_sectors) |
5358 | reshape_sectors = mddev->new_chunk_sectors; | |
7a661381 | 5359 | else |
9d8f0363 | 5360 | reshape_sectors = mddev->chunk_sectors; |
7a661381 | 5361 | |
b5254dd5 N |
5362 | /* We update the metadata at least every 10 seconds, or when |
5363 | * the data about to be copied would over-write the source of | |
5364 | * the data at the front of the range. i.e. one new_stripe | |
5365 | * along from reshape_progress new_maps to after where | |
5366 | * reshape_safe old_maps to | |
52c03291 | 5367 | */ |
fef9c61f | 5368 | writepos = conf->reshape_progress; |
f416885e | 5369 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5370 | readpos = conf->reshape_progress; |
5371 | sector_div(readpos, data_disks); | |
fef9c61f | 5372 | safepos = conf->reshape_safe; |
f416885e | 5373 | sector_div(safepos, data_disks); |
2c810cdd | 5374 | if (mddev->reshape_backwards) { |
ed37d83e | 5375 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 5376 | readpos += reshape_sectors; |
7a661381 | 5377 | safepos += reshape_sectors; |
fef9c61f | 5378 | } else { |
7a661381 | 5379 | writepos += reshape_sectors; |
ed37d83e N |
5380 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5381 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5382 | } |
52c03291 | 5383 | |
b5254dd5 N |
5384 | /* Having calculated the 'writepos' possibly use it |
5385 | * to set 'stripe_addr' which is where we will write to. | |
5386 | */ | |
5387 | if (mddev->reshape_backwards) { | |
5388 | BUG_ON(conf->reshape_progress == 0); | |
5389 | stripe_addr = writepos; | |
5390 | BUG_ON((mddev->dev_sectors & | |
5391 | ~((sector_t)reshape_sectors - 1)) | |
5392 | - reshape_sectors - stripe_addr | |
5393 | != sector_nr); | |
5394 | } else { | |
5395 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5396 | stripe_addr = sector_nr; | |
5397 | } | |
5398 | ||
c8f517c4 N |
5399 | /* 'writepos' is the most advanced device address we might write. |
5400 | * 'readpos' is the least advanced device address we might read. | |
5401 | * 'safepos' is the least address recorded in the metadata as having | |
5402 | * been reshaped. | |
b5254dd5 N |
5403 | * If there is a min_offset_diff, these are adjusted either by |
5404 | * increasing the safepos/readpos if diff is negative, or | |
5405 | * increasing writepos if diff is positive. | |
5406 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5407 | * ensure safety in the face of a crash - that must be done by userspace |
5408 | * making a backup of the data. So in that case there is no particular | |
5409 | * rush to update metadata. | |
5410 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5411 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5412 | * we can be safe in the event of a crash. | |
5413 | * So we insist on updating metadata if safepos is behind writepos and | |
5414 | * readpos is beyond writepos. | |
5415 | * In any case, update the metadata every 10 seconds. | |
5416 | * Maybe that number should be configurable, but I'm not sure it is | |
5417 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5418 | */ | |
b5254dd5 N |
5419 | if (conf->min_offset_diff < 0) { |
5420 | safepos += -conf->min_offset_diff; | |
5421 | readpos += -conf->min_offset_diff; | |
5422 | } else | |
5423 | writepos += conf->min_offset_diff; | |
5424 | ||
2c810cdd | 5425 | if ((mddev->reshape_backwards |
c8f517c4 N |
5426 | ? (safepos > writepos && readpos < writepos) |
5427 | : (safepos < writepos && readpos > writepos)) || | |
5428 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5429 | /* Cannot proceed until we've updated the superblock... */ |
5430 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5431 | atomic_read(&conf->reshape_stripes)==0 |
5432 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5433 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5434 | return 0; | |
fef9c61f | 5435 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5436 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5437 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 5438 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 5439 | md_wakeup_thread(mddev->thread); |
850b2b42 | 5440 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
c91abf5a N |
5441 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5442 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5443 | return 0; | |
52c03291 | 5444 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5445 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5446 | spin_unlock_irq(&conf->device_lock); |
5447 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5448 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5449 | } |
5450 | ||
ab69ae12 | 5451 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5452 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5453 | int j; |
a9f326eb | 5454 | int skipped_disk = 0; |
a8c906ca | 5455 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5456 | set_bit(STRIPE_EXPANDING, &sh->state); |
5457 | atomic_inc(&conf->reshape_stripes); | |
5458 | /* If any of this stripe is beyond the end of the old | |
5459 | * array, then we need to zero those blocks | |
5460 | */ | |
5461 | for (j=sh->disks; j--;) { | |
5462 | sector_t s; | |
5463 | if (j == sh->pd_idx) | |
5464 | continue; | |
f416885e | 5465 | if (conf->level == 6 && |
d0dabf7e | 5466 | j == sh->qd_idx) |
f416885e | 5467 | continue; |
784052ec | 5468 | s = compute_blocknr(sh, j, 0); |
b522adcd | 5469 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5470 | skipped_disk = 1; |
52c03291 N |
5471 | continue; |
5472 | } | |
5473 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5474 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5475 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5476 | } | |
a9f326eb | 5477 | if (!skipped_disk) { |
52c03291 N |
5478 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5479 | set_bit(STRIPE_HANDLE, &sh->state); | |
5480 | } | |
ab69ae12 | 5481 | list_add(&sh->lru, &stripes); |
52c03291 N |
5482 | } |
5483 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5484 | if (mddev->reshape_backwards) |
7a661381 | 5485 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5486 | else |
7a661381 | 5487 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5488 | spin_unlock_irq(&conf->device_lock); |
5489 | /* Ok, those stripe are ready. We can start scheduling | |
5490 | * reads on the source stripes. | |
5491 | * The source stripes are determined by mapping the first and last | |
5492 | * block on the destination stripes. | |
5493 | */ | |
52c03291 | 5494 | first_sector = |
ec32a2bd | 5495 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5496 | 1, &dd_idx, NULL); |
52c03291 | 5497 | last_sector = |
0e6e0271 | 5498 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5499 | * new_data_disks - 1), |
911d4ee8 | 5500 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5501 | if (last_sector >= mddev->dev_sectors) |
5502 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5503 | while (first_sector <= last_sector) { |
a8c906ca | 5504 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5505 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5506 | set_bit(STRIPE_HANDLE, &sh->state); | |
5507 | release_stripe(sh); | |
5508 | first_sector += STRIPE_SECTORS; | |
5509 | } | |
ab69ae12 N |
5510 | /* Now that the sources are clearly marked, we can release |
5511 | * the destination stripes | |
5512 | */ | |
5513 | while (!list_empty(&stripes)) { | |
5514 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5515 | list_del_init(&sh->lru); | |
5516 | release_stripe(sh); | |
5517 | } | |
c6207277 N |
5518 | /* If this takes us to the resync_max point where we have to pause, |
5519 | * then we need to write out the superblock. | |
5520 | */ | |
7a661381 | 5521 | sector_nr += reshape_sectors; |
c03f6a19 N |
5522 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
5523 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
5524 | /* Cannot proceed until we've updated the superblock... */ |
5525 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5526 | atomic_read(&conf->reshape_stripes) == 0 |
5527 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5528 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5529 | goto ret; | |
fef9c61f | 5530 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5531 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5532 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
5533 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
5534 | md_wakeup_thread(mddev->thread); | |
5535 | wait_event(mddev->sb_wait, | |
5536 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
c91abf5a N |
5537 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5538 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5539 | goto ret; | |
c6207277 | 5540 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5541 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5542 | spin_unlock_irq(&conf->device_lock); |
5543 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5544 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5545 | } |
c91abf5a | 5546 | ret: |
7a661381 | 5547 | return reshape_sectors; |
52c03291 N |
5548 | } |
5549 | ||
09314799 | 5550 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
52c03291 | 5551 | { |
d1688a6d | 5552 | struct r5conf *conf = mddev->private; |
52c03291 | 5553 | struct stripe_head *sh; |
58c0fed4 | 5554 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5555 | sector_t sync_blocks; |
16a53ecc N |
5556 | int still_degraded = 0; |
5557 | int i; | |
1da177e4 | 5558 | |
72626685 | 5559 | if (sector_nr >= max_sector) { |
1da177e4 | 5560 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5561 | |
29269553 N |
5562 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5563 | end_reshape(conf); | |
5564 | return 0; | |
5565 | } | |
72626685 N |
5566 | |
5567 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5568 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5569 | &sync_blocks, 1); | |
16a53ecc | 5570 | else /* completed sync */ |
72626685 N |
5571 | conf->fullsync = 0; |
5572 | bitmap_close_sync(mddev->bitmap); | |
5573 | ||
1da177e4 LT |
5574 | return 0; |
5575 | } | |
ccfcc3c1 | 5576 | |
64bd660b N |
5577 | /* Allow raid5_quiesce to complete */ |
5578 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5579 | ||
52c03291 N |
5580 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5581 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5582 | |
c6207277 N |
5583 | /* No need to check resync_max as we never do more than one |
5584 | * stripe, and as resync_max will always be on a chunk boundary, | |
5585 | * if the check in md_do_sync didn't fire, there is no chance | |
5586 | * of overstepping resync_max here | |
5587 | */ | |
5588 | ||
16a53ecc | 5589 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5590 | * to resync, then assert that we are finished, because there is |
5591 | * nothing we can do. | |
5592 | */ | |
3285edf1 | 5593 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5594 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5595 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5596 | *skipped = 1; |
1da177e4 LT |
5597 | return rv; |
5598 | } | |
6f608040 | 5599 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5600 | !conf->fullsync && | |
5601 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5602 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5603 | /* we can skip this block, and probably more */ |
5604 | sync_blocks /= STRIPE_SECTORS; | |
5605 | *skipped = 1; | |
5606 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5607 | } | |
1da177e4 | 5608 | |
b47490c9 N |
5609 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
5610 | ||
a8c906ca | 5611 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5612 | if (sh == NULL) { |
a8c906ca | 5613 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5614 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5615 | * is trying to get access |
1da177e4 | 5616 | */ |
66c006a5 | 5617 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5618 | } |
16a53ecc | 5619 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5620 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5621 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5622 | */ |
16d9cfab EM |
5623 | rcu_read_lock(); |
5624 | for (i = 0; i < conf->raid_disks; i++) { | |
5625 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5626 | ||
5627 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5628 | still_degraded = 1; |
16d9cfab EM |
5629 | } |
5630 | rcu_read_unlock(); | |
16a53ecc N |
5631 | |
5632 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5633 | ||
83206d66 | 5634 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5635 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5636 | |
1da177e4 LT |
5637 | release_stripe(sh); |
5638 | ||
5639 | return STRIPE_SECTORS; | |
5640 | } | |
5641 | ||
d1688a6d | 5642 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5643 | { |
5644 | /* We may not be able to submit a whole bio at once as there | |
5645 | * may not be enough stripe_heads available. | |
5646 | * We cannot pre-allocate enough stripe_heads as we may need | |
5647 | * more than exist in the cache (if we allow ever large chunks). | |
5648 | * So we do one stripe head at a time and record in | |
5649 | * ->bi_hw_segments how many have been done. | |
5650 | * | |
5651 | * We *know* that this entire raid_bio is in one chunk, so | |
5652 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5653 | */ | |
5654 | struct stripe_head *sh; | |
911d4ee8 | 5655 | int dd_idx; |
46031f9a RBJ |
5656 | sector_t sector, logical_sector, last_sector; |
5657 | int scnt = 0; | |
5658 | int remaining; | |
5659 | int handled = 0; | |
5660 | ||
4f024f37 KO |
5661 | logical_sector = raid_bio->bi_iter.bi_sector & |
5662 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5663 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5664 | 0, &dd_idx, NULL); |
f73a1c7d | 5665 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5666 | |
5667 | for (; logical_sector < last_sector; | |
387bb173 NB |
5668 | logical_sector += STRIPE_SECTORS, |
5669 | sector += STRIPE_SECTORS, | |
5670 | scnt++) { | |
46031f9a | 5671 | |
e7836bd6 | 5672 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5673 | /* already done this stripe */ |
5674 | continue; | |
5675 | ||
2844dc32 | 5676 | sh = get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5677 | |
5678 | if (!sh) { | |
5679 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5680 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5681 | conf->retry_read_aligned = raid_bio; |
5682 | return handled; | |
5683 | } | |
5684 | ||
da41ba65 | 5685 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
387bb173 | 5686 | release_stripe(sh); |
e7836bd6 | 5687 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
5688 | conf->retry_read_aligned = raid_bio; |
5689 | return handled; | |
5690 | } | |
5691 | ||
3f9e7c14 | 5692 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 5693 | handle_stripe(sh); |
46031f9a RBJ |
5694 | release_stripe(sh); |
5695 | handled++; | |
5696 | } | |
e7836bd6 | 5697 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
5698 | if (remaining == 0) { |
5699 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
5700 | raid_bio, 0); | |
0e13fe23 | 5701 | bio_endio(raid_bio, 0); |
0a82a8d1 | 5702 | } |
46031f9a | 5703 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5704 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
5705 | return handled; |
5706 | } | |
5707 | ||
bfc90cb0 | 5708 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
5709 | struct r5worker *worker, |
5710 | struct list_head *temp_inactive_list) | |
46a06401 SL |
5711 | { |
5712 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
5713 | int i, batch_size = 0, hash; |
5714 | bool release_inactive = false; | |
46a06401 SL |
5715 | |
5716 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 5717 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
5718 | batch[batch_size++] = sh; |
5719 | ||
566c09c5 SL |
5720 | if (batch_size == 0) { |
5721 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5722 | if (!list_empty(temp_inactive_list + i)) | |
5723 | break; | |
5724 | if (i == NR_STRIPE_HASH_LOCKS) | |
5725 | return batch_size; | |
5726 | release_inactive = true; | |
5727 | } | |
46a06401 SL |
5728 | spin_unlock_irq(&conf->device_lock); |
5729 | ||
566c09c5 SL |
5730 | release_inactive_stripe_list(conf, temp_inactive_list, |
5731 | NR_STRIPE_HASH_LOCKS); | |
5732 | ||
5733 | if (release_inactive) { | |
5734 | spin_lock_irq(&conf->device_lock); | |
5735 | return 0; | |
5736 | } | |
5737 | ||
46a06401 SL |
5738 | for (i = 0; i < batch_size; i++) |
5739 | handle_stripe(batch[i]); | |
5740 | ||
5741 | cond_resched(); | |
5742 | ||
5743 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
5744 | for (i = 0; i < batch_size; i++) { |
5745 | hash = batch[i]->hash_lock_index; | |
5746 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
5747 | } | |
46a06401 SL |
5748 | return batch_size; |
5749 | } | |
46031f9a | 5750 | |
851c30c9 SL |
5751 | static void raid5_do_work(struct work_struct *work) |
5752 | { | |
5753 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
5754 | struct r5worker_group *group = worker->group; | |
5755 | struct r5conf *conf = group->conf; | |
5756 | int group_id = group - conf->worker_groups; | |
5757 | int handled; | |
5758 | struct blk_plug plug; | |
5759 | ||
5760 | pr_debug("+++ raid5worker active\n"); | |
5761 | ||
5762 | blk_start_plug(&plug); | |
5763 | handled = 0; | |
5764 | spin_lock_irq(&conf->device_lock); | |
5765 | while (1) { | |
5766 | int batch_size, released; | |
5767 | ||
566c09c5 | 5768 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 5769 | |
566c09c5 SL |
5770 | batch_size = handle_active_stripes(conf, group_id, worker, |
5771 | worker->temp_inactive_list); | |
bfc90cb0 | 5772 | worker->working = false; |
851c30c9 SL |
5773 | if (!batch_size && !released) |
5774 | break; | |
5775 | handled += batch_size; | |
5776 | } | |
5777 | pr_debug("%d stripes handled\n", handled); | |
5778 | ||
5779 | spin_unlock_irq(&conf->device_lock); | |
5780 | blk_finish_plug(&plug); | |
5781 | ||
5782 | pr_debug("--- raid5worker inactive\n"); | |
5783 | } | |
5784 | ||
1da177e4 LT |
5785 | /* |
5786 | * This is our raid5 kernel thread. | |
5787 | * | |
5788 | * We scan the hash table for stripes which can be handled now. | |
5789 | * During the scan, completed stripes are saved for us by the interrupt | |
5790 | * handler, so that they will not have to wait for our next wakeup. | |
5791 | */ | |
4ed8731d | 5792 | static void raid5d(struct md_thread *thread) |
1da177e4 | 5793 | { |
4ed8731d | 5794 | struct mddev *mddev = thread->mddev; |
d1688a6d | 5795 | struct r5conf *conf = mddev->private; |
1da177e4 | 5796 | int handled; |
e1dfa0a2 | 5797 | struct blk_plug plug; |
1da177e4 | 5798 | |
45b4233c | 5799 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
5800 | |
5801 | md_check_recovery(mddev); | |
1da177e4 | 5802 | |
e1dfa0a2 | 5803 | blk_start_plug(&plug); |
1da177e4 LT |
5804 | handled = 0; |
5805 | spin_lock_irq(&conf->device_lock); | |
5806 | while (1) { | |
46031f9a | 5807 | struct bio *bio; |
773ca82f SL |
5808 | int batch_size, released; |
5809 | ||
566c09c5 | 5810 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
5811 | if (released) |
5812 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 5813 | |
0021b7bc | 5814 | if ( |
7c13edc8 N |
5815 | !list_empty(&conf->bitmap_list)) { |
5816 | /* Now is a good time to flush some bitmap updates */ | |
5817 | conf->seq_flush++; | |
700e432d | 5818 | spin_unlock_irq(&conf->device_lock); |
72626685 | 5819 | bitmap_unplug(mddev->bitmap); |
700e432d | 5820 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 5821 | conf->seq_write = conf->seq_flush; |
566c09c5 | 5822 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 5823 | } |
0021b7bc | 5824 | raid5_activate_delayed(conf); |
72626685 | 5825 | |
46031f9a RBJ |
5826 | while ((bio = remove_bio_from_retry(conf))) { |
5827 | int ok; | |
5828 | spin_unlock_irq(&conf->device_lock); | |
5829 | ok = retry_aligned_read(conf, bio); | |
5830 | spin_lock_irq(&conf->device_lock); | |
5831 | if (!ok) | |
5832 | break; | |
5833 | handled++; | |
5834 | } | |
5835 | ||
566c09c5 SL |
5836 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
5837 | conf->temp_inactive_list); | |
773ca82f | 5838 | if (!batch_size && !released) |
1da177e4 | 5839 | break; |
46a06401 | 5840 | handled += batch_size; |
1da177e4 | 5841 | |
46a06401 SL |
5842 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
5843 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 5844 | md_check_recovery(mddev); |
46a06401 SL |
5845 | spin_lock_irq(&conf->device_lock); |
5846 | } | |
1da177e4 | 5847 | } |
45b4233c | 5848 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
5849 | |
5850 | spin_unlock_irq(&conf->device_lock); | |
edbe83ab N |
5851 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state)) { |
5852 | grow_one_stripe(conf, __GFP_NOWARN); | |
5853 | /* Set flag even if allocation failed. This helps | |
5854 | * slow down allocation requests when mem is short | |
5855 | */ | |
5856 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
5857 | } | |
1da177e4 | 5858 | |
c9f21aaf | 5859 | async_tx_issue_pending_all(); |
e1dfa0a2 | 5860 | blk_finish_plug(&plug); |
1da177e4 | 5861 | |
45b4233c | 5862 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
5863 | } |
5864 | ||
3f294f4f | 5865 | static ssize_t |
fd01b88c | 5866 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 5867 | { |
7b1485ba N |
5868 | struct r5conf *conf; |
5869 | int ret = 0; | |
5870 | spin_lock(&mddev->lock); | |
5871 | conf = mddev->private; | |
96de1e66 | 5872 | if (conf) |
edbe83ab | 5873 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
5874 | spin_unlock(&mddev->lock); |
5875 | return ret; | |
3f294f4f N |
5876 | } |
5877 | ||
c41d4ac4 | 5878 | int |
fd01b88c | 5879 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 5880 | { |
d1688a6d | 5881 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
5882 | int err; |
5883 | ||
c41d4ac4 | 5884 | if (size <= 16 || size > 32768) |
3f294f4f | 5885 | return -EINVAL; |
486f0644 | 5886 | |
edbe83ab | 5887 | conf->min_nr_stripes = size; |
486f0644 N |
5888 | while (size < conf->max_nr_stripes && |
5889 | drop_one_stripe(conf)) | |
5890 | ; | |
5891 | ||
edbe83ab | 5892 | |
b5470dc5 DW |
5893 | err = md_allow_write(mddev); |
5894 | if (err) | |
5895 | return err; | |
486f0644 N |
5896 | |
5897 | while (size > conf->max_nr_stripes) | |
5898 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
5899 | break; | |
5900 | ||
c41d4ac4 N |
5901 | return 0; |
5902 | } | |
5903 | EXPORT_SYMBOL(raid5_set_cache_size); | |
5904 | ||
5905 | static ssize_t | |
fd01b88c | 5906 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 5907 | { |
6791875e | 5908 | struct r5conf *conf; |
c41d4ac4 N |
5909 | unsigned long new; |
5910 | int err; | |
5911 | ||
5912 | if (len >= PAGE_SIZE) | |
5913 | return -EINVAL; | |
b29bebd6 | 5914 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 5915 | return -EINVAL; |
6791875e | 5916 | err = mddev_lock(mddev); |
c41d4ac4 N |
5917 | if (err) |
5918 | return err; | |
6791875e N |
5919 | conf = mddev->private; |
5920 | if (!conf) | |
5921 | err = -ENODEV; | |
5922 | else | |
5923 | err = raid5_set_cache_size(mddev, new); | |
5924 | mddev_unlock(mddev); | |
5925 | ||
5926 | return err ?: len; | |
3f294f4f | 5927 | } |
007583c9 | 5928 | |
96de1e66 N |
5929 | static struct md_sysfs_entry |
5930 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
5931 | raid5_show_stripe_cache_size, | |
5932 | raid5_store_stripe_cache_size); | |
3f294f4f | 5933 | |
d06f191f MS |
5934 | static ssize_t |
5935 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
5936 | { | |
5937 | struct r5conf *conf = mddev->private; | |
5938 | if (conf) | |
5939 | return sprintf(page, "%d\n", conf->rmw_level); | |
5940 | else | |
5941 | return 0; | |
5942 | } | |
5943 | ||
5944 | static ssize_t | |
5945 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
5946 | { | |
5947 | struct r5conf *conf = mddev->private; | |
5948 | unsigned long new; | |
5949 | ||
5950 | if (!conf) | |
5951 | return -ENODEV; | |
5952 | ||
5953 | if (len >= PAGE_SIZE) | |
5954 | return -EINVAL; | |
5955 | ||
5956 | if (kstrtoul(page, 10, &new)) | |
5957 | return -EINVAL; | |
5958 | ||
5959 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
5960 | return -EINVAL; | |
5961 | ||
5962 | if (new != PARITY_DISABLE_RMW && | |
5963 | new != PARITY_ENABLE_RMW && | |
5964 | new != PARITY_PREFER_RMW) | |
5965 | return -EINVAL; | |
5966 | ||
5967 | conf->rmw_level = new; | |
5968 | return len; | |
5969 | } | |
5970 | ||
5971 | static struct md_sysfs_entry | |
5972 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
5973 | raid5_show_rmw_level, | |
5974 | raid5_store_rmw_level); | |
5975 | ||
5976 | ||
8b3e6cdc | 5977 | static ssize_t |
fd01b88c | 5978 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 5979 | { |
7b1485ba N |
5980 | struct r5conf *conf; |
5981 | int ret = 0; | |
5982 | spin_lock(&mddev->lock); | |
5983 | conf = mddev->private; | |
8b3e6cdc | 5984 | if (conf) |
7b1485ba N |
5985 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
5986 | spin_unlock(&mddev->lock); | |
5987 | return ret; | |
8b3e6cdc DW |
5988 | } |
5989 | ||
5990 | static ssize_t | |
fd01b88c | 5991 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 5992 | { |
6791875e | 5993 | struct r5conf *conf; |
4ef197d8 | 5994 | unsigned long new; |
6791875e N |
5995 | int err; |
5996 | ||
8b3e6cdc DW |
5997 | if (len >= PAGE_SIZE) |
5998 | return -EINVAL; | |
b29bebd6 | 5999 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6000 | return -EINVAL; |
6791875e N |
6001 | |
6002 | err = mddev_lock(mddev); | |
6003 | if (err) | |
6004 | return err; | |
6005 | conf = mddev->private; | |
6006 | if (!conf) | |
6007 | err = -ENODEV; | |
edbe83ab | 6008 | else if (new > conf->min_nr_stripes) |
6791875e N |
6009 | err = -EINVAL; |
6010 | else | |
6011 | conf->bypass_threshold = new; | |
6012 | mddev_unlock(mddev); | |
6013 | return err ?: len; | |
8b3e6cdc DW |
6014 | } |
6015 | ||
6016 | static struct md_sysfs_entry | |
6017 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6018 | S_IRUGO | S_IWUSR, | |
6019 | raid5_show_preread_threshold, | |
6020 | raid5_store_preread_threshold); | |
6021 | ||
d592a996 SL |
6022 | static ssize_t |
6023 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6024 | { | |
7b1485ba N |
6025 | struct r5conf *conf; |
6026 | int ret = 0; | |
6027 | spin_lock(&mddev->lock); | |
6028 | conf = mddev->private; | |
d592a996 | 6029 | if (conf) |
7b1485ba N |
6030 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6031 | spin_unlock(&mddev->lock); | |
6032 | return ret; | |
d592a996 SL |
6033 | } |
6034 | ||
6035 | static ssize_t | |
6036 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6037 | { | |
6791875e | 6038 | struct r5conf *conf; |
d592a996 | 6039 | unsigned long new; |
6791875e N |
6040 | int err; |
6041 | ||
d592a996 SL |
6042 | if (len >= PAGE_SIZE) |
6043 | return -EINVAL; | |
d592a996 SL |
6044 | if (kstrtoul(page, 10, &new)) |
6045 | return -EINVAL; | |
6046 | new = !!new; | |
6791875e N |
6047 | |
6048 | err = mddev_lock(mddev); | |
6049 | if (err) | |
6050 | return err; | |
6051 | conf = mddev->private; | |
6052 | if (!conf) | |
6053 | err = -ENODEV; | |
6054 | else if (new != conf->skip_copy) { | |
6055 | mddev_suspend(mddev); | |
6056 | conf->skip_copy = new; | |
6057 | if (new) | |
6058 | mddev->queue->backing_dev_info.capabilities |= | |
6059 | BDI_CAP_STABLE_WRITES; | |
6060 | else | |
6061 | mddev->queue->backing_dev_info.capabilities &= | |
6062 | ~BDI_CAP_STABLE_WRITES; | |
6063 | mddev_resume(mddev); | |
6064 | } | |
6065 | mddev_unlock(mddev); | |
6066 | return err ?: len; | |
d592a996 SL |
6067 | } |
6068 | ||
6069 | static struct md_sysfs_entry | |
6070 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6071 | raid5_show_skip_copy, | |
6072 | raid5_store_skip_copy); | |
6073 | ||
3f294f4f | 6074 | static ssize_t |
fd01b88c | 6075 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6076 | { |
d1688a6d | 6077 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6078 | if (conf) |
6079 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6080 | else | |
6081 | return 0; | |
3f294f4f N |
6082 | } |
6083 | ||
96de1e66 N |
6084 | static struct md_sysfs_entry |
6085 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6086 | |
b721420e SL |
6087 | static ssize_t |
6088 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6089 | { | |
7b1485ba N |
6090 | struct r5conf *conf; |
6091 | int ret = 0; | |
6092 | spin_lock(&mddev->lock); | |
6093 | conf = mddev->private; | |
b721420e | 6094 | if (conf) |
7b1485ba N |
6095 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6096 | spin_unlock(&mddev->lock); | |
6097 | return ret; | |
b721420e SL |
6098 | } |
6099 | ||
60aaf933 | 6100 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6101 | int *group_cnt, | |
6102 | int *worker_cnt_per_group, | |
6103 | struct r5worker_group **worker_groups); | |
b721420e SL |
6104 | static ssize_t |
6105 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6106 | { | |
6791875e | 6107 | struct r5conf *conf; |
b721420e SL |
6108 | unsigned long new; |
6109 | int err; | |
60aaf933 | 6110 | struct r5worker_group *new_groups, *old_groups; |
6111 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6112 | |
6113 | if (len >= PAGE_SIZE) | |
6114 | return -EINVAL; | |
b721420e SL |
6115 | if (kstrtoul(page, 10, &new)) |
6116 | return -EINVAL; | |
6117 | ||
6791875e N |
6118 | err = mddev_lock(mddev); |
6119 | if (err) | |
6120 | return err; | |
6121 | conf = mddev->private; | |
6122 | if (!conf) | |
6123 | err = -ENODEV; | |
6124 | else if (new != conf->worker_cnt_per_group) { | |
6125 | mddev_suspend(mddev); | |
b721420e | 6126 | |
6791875e N |
6127 | old_groups = conf->worker_groups; |
6128 | if (old_groups) | |
6129 | flush_workqueue(raid5_wq); | |
d206dcfa | 6130 | |
6791875e N |
6131 | err = alloc_thread_groups(conf, new, |
6132 | &group_cnt, &worker_cnt_per_group, | |
6133 | &new_groups); | |
6134 | if (!err) { | |
6135 | spin_lock_irq(&conf->device_lock); | |
6136 | conf->group_cnt = group_cnt; | |
6137 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6138 | conf->worker_groups = new_groups; | |
6139 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6140 | |
6791875e N |
6141 | if (old_groups) |
6142 | kfree(old_groups[0].workers); | |
6143 | kfree(old_groups); | |
6144 | } | |
6145 | mddev_resume(mddev); | |
b721420e | 6146 | } |
6791875e | 6147 | mddev_unlock(mddev); |
b721420e | 6148 | |
6791875e | 6149 | return err ?: len; |
b721420e SL |
6150 | } |
6151 | ||
6152 | static struct md_sysfs_entry | |
6153 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6154 | raid5_show_group_thread_cnt, | |
6155 | raid5_store_group_thread_cnt); | |
6156 | ||
007583c9 | 6157 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6158 | &raid5_stripecache_size.attr, |
6159 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6160 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6161 | &raid5_group_thread_cnt.attr, |
d592a996 | 6162 | &raid5_skip_copy.attr, |
d06f191f | 6163 | &raid5_rmw_level.attr, |
3f294f4f N |
6164 | NULL, |
6165 | }; | |
007583c9 N |
6166 | static struct attribute_group raid5_attrs_group = { |
6167 | .name = NULL, | |
6168 | .attrs = raid5_attrs, | |
3f294f4f N |
6169 | }; |
6170 | ||
60aaf933 | 6171 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6172 | int *group_cnt, | |
6173 | int *worker_cnt_per_group, | |
6174 | struct r5worker_group **worker_groups) | |
851c30c9 | 6175 | { |
566c09c5 | 6176 | int i, j, k; |
851c30c9 SL |
6177 | ssize_t size; |
6178 | struct r5worker *workers; | |
6179 | ||
60aaf933 | 6180 | *worker_cnt_per_group = cnt; |
851c30c9 | 6181 | if (cnt == 0) { |
60aaf933 | 6182 | *group_cnt = 0; |
6183 | *worker_groups = NULL; | |
851c30c9 SL |
6184 | return 0; |
6185 | } | |
60aaf933 | 6186 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6187 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6188 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6189 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6190 | *group_cnt, GFP_NOIO); | |
6191 | if (!*worker_groups || !workers) { | |
851c30c9 | 6192 | kfree(workers); |
60aaf933 | 6193 | kfree(*worker_groups); |
851c30c9 SL |
6194 | return -ENOMEM; |
6195 | } | |
6196 | ||
60aaf933 | 6197 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6198 | struct r5worker_group *group; |
6199 | ||
0c775d52 | 6200 | group = &(*worker_groups)[i]; |
851c30c9 SL |
6201 | INIT_LIST_HEAD(&group->handle_list); |
6202 | group->conf = conf; | |
6203 | group->workers = workers + i * cnt; | |
6204 | ||
6205 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6206 | struct r5worker *worker = group->workers + j; |
6207 | worker->group = group; | |
6208 | INIT_WORK(&worker->work, raid5_do_work); | |
6209 | ||
6210 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6211 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6212 | } |
6213 | } | |
6214 | ||
6215 | return 0; | |
6216 | } | |
6217 | ||
6218 | static void free_thread_groups(struct r5conf *conf) | |
6219 | { | |
6220 | if (conf->worker_groups) | |
6221 | kfree(conf->worker_groups[0].workers); | |
6222 | kfree(conf->worker_groups); | |
6223 | conf->worker_groups = NULL; | |
6224 | } | |
6225 | ||
80c3a6ce | 6226 | static sector_t |
fd01b88c | 6227 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6228 | { |
d1688a6d | 6229 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6230 | |
6231 | if (!sectors) | |
6232 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6233 | if (!raid_disks) |
7ec05478 | 6234 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6235 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6236 | |
9d8f0363 | 6237 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
664e7c41 | 6238 | sectors &= ~((sector_t)mddev->new_chunk_sectors - 1); |
80c3a6ce DW |
6239 | return sectors * (raid_disks - conf->max_degraded); |
6240 | } | |
6241 | ||
789b5e03 ON |
6242 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6243 | { | |
6244 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6245 | if (percpu->scribble) |
6246 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6247 | percpu->spare_page = NULL; |
6248 | percpu->scribble = NULL; | |
6249 | } | |
6250 | ||
6251 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6252 | { | |
6253 | if (conf->level == 6 && !percpu->spare_page) | |
6254 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6255 | if (!percpu->scribble) | |
46d5b785 | 6256 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6257 | conf->previous_raid_disks), |
6258 | max(conf->chunk_sectors, | |
6259 | conf->prev_chunk_sectors) | |
6260 | / STRIPE_SECTORS, | |
6261 | GFP_KERNEL); | |
789b5e03 ON |
6262 | |
6263 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6264 | free_scratch_buffer(conf, percpu); | |
6265 | return -ENOMEM; | |
6266 | } | |
6267 | ||
6268 | return 0; | |
6269 | } | |
6270 | ||
d1688a6d | 6271 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 | 6272 | { |
36d1c647 DW |
6273 | unsigned long cpu; |
6274 | ||
6275 | if (!conf->percpu) | |
6276 | return; | |
6277 | ||
36d1c647 DW |
6278 | #ifdef CONFIG_HOTPLUG_CPU |
6279 | unregister_cpu_notifier(&conf->cpu_notify); | |
6280 | #endif | |
789b5e03 ON |
6281 | |
6282 | get_online_cpus(); | |
6283 | for_each_possible_cpu(cpu) | |
6284 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
36d1c647 DW |
6285 | put_online_cpus(); |
6286 | ||
6287 | free_percpu(conf->percpu); | |
6288 | } | |
6289 | ||
d1688a6d | 6290 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6291 | { |
edbe83ab N |
6292 | if (conf->shrinker.seeks) |
6293 | unregister_shrinker(&conf->shrinker); | |
851c30c9 | 6294 | free_thread_groups(conf); |
95fc17aa | 6295 | shrink_stripes(conf); |
36d1c647 | 6296 | raid5_free_percpu(conf); |
95fc17aa DW |
6297 | kfree(conf->disks); |
6298 | kfree(conf->stripe_hashtbl); | |
6299 | kfree(conf); | |
6300 | } | |
6301 | ||
36d1c647 DW |
6302 | #ifdef CONFIG_HOTPLUG_CPU |
6303 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
6304 | void *hcpu) | |
6305 | { | |
d1688a6d | 6306 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
6307 | long cpu = (long)hcpu; |
6308 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
6309 | ||
6310 | switch (action) { | |
6311 | case CPU_UP_PREPARE: | |
6312 | case CPU_UP_PREPARE_FROZEN: | |
789b5e03 | 6313 | if (alloc_scratch_buffer(conf, percpu)) { |
36d1c647 DW |
6314 | pr_err("%s: failed memory allocation for cpu%ld\n", |
6315 | __func__, cpu); | |
55af6bb5 | 6316 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
6317 | } |
6318 | break; | |
6319 | case CPU_DEAD: | |
6320 | case CPU_DEAD_FROZEN: | |
789b5e03 | 6321 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
36d1c647 DW |
6322 | break; |
6323 | default: | |
6324 | break; | |
6325 | } | |
6326 | return NOTIFY_OK; | |
6327 | } | |
6328 | #endif | |
6329 | ||
d1688a6d | 6330 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
6331 | { |
6332 | unsigned long cpu; | |
789b5e03 | 6333 | int err = 0; |
36d1c647 | 6334 | |
789b5e03 ON |
6335 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6336 | if (!conf->percpu) | |
36d1c647 | 6337 | return -ENOMEM; |
789b5e03 ON |
6338 | |
6339 | #ifdef CONFIG_HOTPLUG_CPU | |
6340 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
6341 | conf->cpu_notify.priority = 0; | |
6342 | err = register_cpu_notifier(&conf->cpu_notify); | |
6343 | if (err) | |
6344 | return err; | |
6345 | #endif | |
36d1c647 DW |
6346 | |
6347 | get_online_cpus(); | |
36d1c647 | 6348 | for_each_present_cpu(cpu) { |
789b5e03 ON |
6349 | err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
6350 | if (err) { | |
6351 | pr_err("%s: failed memory allocation for cpu%ld\n", | |
6352 | __func__, cpu); | |
36d1c647 DW |
6353 | break; |
6354 | } | |
36d1c647 | 6355 | } |
36d1c647 DW |
6356 | put_online_cpus(); |
6357 | ||
6358 | return err; | |
6359 | } | |
6360 | ||
edbe83ab N |
6361 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6362 | struct shrink_control *sc) | |
6363 | { | |
6364 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6365 | int ret = 0; | |
6366 | while (ret < sc->nr_to_scan) { | |
6367 | if (drop_one_stripe(conf) == 0) | |
6368 | return SHRINK_STOP; | |
6369 | ret++; | |
6370 | } | |
6371 | return ret; | |
6372 | } | |
6373 | ||
6374 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6375 | struct shrink_control *sc) | |
6376 | { | |
6377 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6378 | ||
6379 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6380 | /* unlikely, but not impossible */ | |
6381 | return 0; | |
6382 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6383 | } | |
6384 | ||
d1688a6d | 6385 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6386 | { |
d1688a6d | 6387 | struct r5conf *conf; |
5e5e3e78 | 6388 | int raid_disk, memory, max_disks; |
3cb03002 | 6389 | struct md_rdev *rdev; |
1da177e4 | 6390 | struct disk_info *disk; |
0232605d | 6391 | char pers_name[6]; |
566c09c5 | 6392 | int i; |
60aaf933 | 6393 | int group_cnt, worker_cnt_per_group; |
6394 | struct r5worker_group *new_group; | |
1da177e4 | 6395 | |
91adb564 N |
6396 | if (mddev->new_level != 5 |
6397 | && mddev->new_level != 4 | |
6398 | && mddev->new_level != 6) { | |
0c55e022 | 6399 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
6400 | mdname(mddev), mddev->new_level); |
6401 | return ERR_PTR(-EIO); | |
1da177e4 | 6402 | } |
91adb564 N |
6403 | if ((mddev->new_level == 5 |
6404 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6405 | (mddev->new_level == 6 | |
6406 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 6407 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
6408 | mdname(mddev), mddev->new_layout); |
6409 | return ERR_PTR(-EIO); | |
99c0fb5f | 6410 | } |
91adb564 | 6411 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 6412 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
6413 | mdname(mddev), mddev->raid_disks); |
6414 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
6415 | } |
6416 | ||
664e7c41 AN |
6417 | if (!mddev->new_chunk_sectors || |
6418 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6419 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
6420 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
6421 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6422 | return ERR_PTR(-EINVAL); |
f6705578 N |
6423 | } |
6424 | ||
d1688a6d | 6425 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6426 | if (conf == NULL) |
1da177e4 | 6427 | goto abort; |
851c30c9 | 6428 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6429 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6430 | &new_group)) { | |
6431 | conf->group_cnt = group_cnt; | |
6432 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6433 | conf->worker_groups = new_group; | |
6434 | } else | |
851c30c9 | 6435 | goto abort; |
f5efd45a | 6436 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6437 | seqcount_init(&conf->gen_lock); |
b1b46486 | 6438 | init_waitqueue_head(&conf->wait_for_quiescent); |
f5efd45a DW |
6439 | init_waitqueue_head(&conf->wait_for_stripe); |
6440 | init_waitqueue_head(&conf->wait_for_overlap); | |
6441 | INIT_LIST_HEAD(&conf->handle_list); | |
6442 | INIT_LIST_HEAD(&conf->hold_list); | |
6443 | INIT_LIST_HEAD(&conf->delayed_list); | |
6444 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 6445 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6446 | atomic_set(&conf->active_stripes, 0); |
6447 | atomic_set(&conf->preread_active_stripes, 0); | |
6448 | atomic_set(&conf->active_aligned_reads, 0); | |
6449 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 6450 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6451 | |
6452 | conf->raid_disks = mddev->raid_disks; | |
6453 | if (mddev->reshape_position == MaxSector) | |
6454 | conf->previous_raid_disks = mddev->raid_disks; | |
6455 | else | |
f6705578 | 6456 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6457 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6458 | |
5e5e3e78 | 6459 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
6460 | GFP_KERNEL); |
6461 | if (!conf->disks) | |
6462 | goto abort; | |
9ffae0cf | 6463 | |
1da177e4 LT |
6464 | conf->mddev = mddev; |
6465 | ||
fccddba0 | 6466 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6467 | goto abort; |
1da177e4 | 6468 | |
566c09c5 SL |
6469 | /* We init hash_locks[0] separately to that it can be used |
6470 | * as the reference lock in the spin_lock_nest_lock() call | |
6471 | * in lock_all_device_hash_locks_irq in order to convince | |
6472 | * lockdep that we know what we are doing. | |
6473 | */ | |
6474 | spin_lock_init(conf->hash_locks); | |
6475 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6476 | spin_lock_init(conf->hash_locks + i); | |
6477 | ||
6478 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6479 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6480 | ||
6481 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6482 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6483 | ||
36d1c647 | 6484 | conf->level = mddev->new_level; |
46d5b785 | 6485 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6486 | if (raid5_alloc_percpu(conf) != 0) |
6487 | goto abort; | |
6488 | ||
0c55e022 | 6489 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6490 | |
dafb20fa | 6491 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6492 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6493 | if (raid_disk >= max_disks |
1da177e4 LT |
6494 | || raid_disk < 0) |
6495 | continue; | |
6496 | disk = conf->disks + raid_disk; | |
6497 | ||
17045f52 N |
6498 | if (test_bit(Replacement, &rdev->flags)) { |
6499 | if (disk->replacement) | |
6500 | goto abort; | |
6501 | disk->replacement = rdev; | |
6502 | } else { | |
6503 | if (disk->rdev) | |
6504 | goto abort; | |
6505 | disk->rdev = rdev; | |
6506 | } | |
1da177e4 | 6507 | |
b2d444d7 | 6508 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6509 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
6510 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
6511 | " disk %d\n", | |
6512 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6513 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6514 | /* Cannot rely on bitmap to complete recovery */ |
6515 | conf->fullsync = 1; | |
1da177e4 LT |
6516 | } |
6517 | ||
91adb564 | 6518 | conf->level = mddev->new_level; |
584acdd4 | 6519 | if (conf->level == 6) { |
16a53ecc | 6520 | conf->max_degraded = 2; |
584acdd4 MS |
6521 | if (raid6_call.xor_syndrome) |
6522 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6523 | else | |
6524 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6525 | } else { | |
16a53ecc | 6526 | conf->max_degraded = 1; |
584acdd4 MS |
6527 | conf->rmw_level = PARITY_ENABLE_RMW; |
6528 | } | |
91adb564 | 6529 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6530 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6531 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6532 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed N |
6533 | conf->prev_algo = mddev->layout; |
6534 | } | |
1da177e4 | 6535 | |
edbe83ab N |
6536 | conf->min_nr_stripes = NR_STRIPES; |
6537 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + | |
5e5e3e78 | 6538 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6539 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6540 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
91adb564 | 6541 | printk(KERN_ERR |
0c55e022 N |
6542 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
6543 | mdname(mddev), memory); | |
91adb564 N |
6544 | goto abort; |
6545 | } else | |
0c55e022 N |
6546 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
6547 | mdname(mddev), memory); | |
edbe83ab N |
6548 | /* |
6549 | * Losing a stripe head costs more than the time to refill it, | |
6550 | * it reduces the queue depth and so can hurt throughput. | |
6551 | * So set it rather large, scaled by number of devices. | |
6552 | */ | |
6553 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6554 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6555 | conf->shrinker.count_objects = raid5_cache_count; | |
6556 | conf->shrinker.batch = 128; | |
6557 | conf->shrinker.flags = 0; | |
6558 | register_shrinker(&conf->shrinker); | |
1da177e4 | 6559 | |
0232605d N |
6560 | sprintf(pers_name, "raid%d", mddev->new_level); |
6561 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
6562 | if (!conf->thread) { |
6563 | printk(KERN_ERR | |
0c55e022 | 6564 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 6565 | mdname(mddev)); |
16a53ecc N |
6566 | goto abort; |
6567 | } | |
91adb564 N |
6568 | |
6569 | return conf; | |
6570 | ||
6571 | abort: | |
6572 | if (conf) { | |
95fc17aa | 6573 | free_conf(conf); |
91adb564 N |
6574 | return ERR_PTR(-EIO); |
6575 | } else | |
6576 | return ERR_PTR(-ENOMEM); | |
6577 | } | |
6578 | ||
c148ffdc N |
6579 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6580 | { | |
6581 | switch (algo) { | |
6582 | case ALGORITHM_PARITY_0: | |
6583 | if (raid_disk < max_degraded) | |
6584 | return 1; | |
6585 | break; | |
6586 | case ALGORITHM_PARITY_N: | |
6587 | if (raid_disk >= raid_disks - max_degraded) | |
6588 | return 1; | |
6589 | break; | |
6590 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6591 | if (raid_disk == 0 || |
c148ffdc N |
6592 | raid_disk == raid_disks - 1) |
6593 | return 1; | |
6594 | break; | |
6595 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6596 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6597 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6598 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6599 | if (raid_disk == raid_disks - 1) | |
6600 | return 1; | |
6601 | } | |
6602 | return 0; | |
6603 | } | |
6604 | ||
fd01b88c | 6605 | static int run(struct mddev *mddev) |
91adb564 | 6606 | { |
d1688a6d | 6607 | struct r5conf *conf; |
9f7c2220 | 6608 | int working_disks = 0; |
c148ffdc | 6609 | int dirty_parity_disks = 0; |
3cb03002 | 6610 | struct md_rdev *rdev; |
c148ffdc | 6611 | sector_t reshape_offset = 0; |
17045f52 | 6612 | int i; |
b5254dd5 N |
6613 | long long min_offset_diff = 0; |
6614 | int first = 1; | |
91adb564 | 6615 | |
8c6ac868 | 6616 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 6617 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
6618 | " -- starting background reconstruction\n", |
6619 | mdname(mddev)); | |
b5254dd5 N |
6620 | |
6621 | rdev_for_each(rdev, mddev) { | |
6622 | long long diff; | |
6623 | if (rdev->raid_disk < 0) | |
6624 | continue; | |
6625 | diff = (rdev->new_data_offset - rdev->data_offset); | |
6626 | if (first) { | |
6627 | min_offset_diff = diff; | |
6628 | first = 0; | |
6629 | } else if (mddev->reshape_backwards && | |
6630 | diff < min_offset_diff) | |
6631 | min_offset_diff = diff; | |
6632 | else if (!mddev->reshape_backwards && | |
6633 | diff > min_offset_diff) | |
6634 | min_offset_diff = diff; | |
6635 | } | |
6636 | ||
91adb564 N |
6637 | if (mddev->reshape_position != MaxSector) { |
6638 | /* Check that we can continue the reshape. | |
b5254dd5 N |
6639 | * Difficulties arise if the stripe we would write to |
6640 | * next is at or after the stripe we would read from next. | |
6641 | * For a reshape that changes the number of devices, this | |
6642 | * is only possible for a very short time, and mdadm makes | |
6643 | * sure that time appears to have past before assembling | |
6644 | * the array. So we fail if that time hasn't passed. | |
6645 | * For a reshape that keeps the number of devices the same | |
6646 | * mdadm must be monitoring the reshape can keeping the | |
6647 | * critical areas read-only and backed up. It will start | |
6648 | * the array in read-only mode, so we check for that. | |
91adb564 N |
6649 | */ |
6650 | sector_t here_new, here_old; | |
6651 | int old_disks; | |
18b00334 | 6652 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 6653 | |
88ce4930 | 6654 | if (mddev->new_level != mddev->level) { |
0c55e022 | 6655 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
6656 | "required - aborting.\n", |
6657 | mdname(mddev)); | |
6658 | return -EINVAL; | |
6659 | } | |
91adb564 N |
6660 | old_disks = mddev->raid_disks - mddev->delta_disks; |
6661 | /* reshape_position must be on a new-stripe boundary, and one | |
6662 | * further up in new geometry must map after here in old | |
6663 | * geometry. | |
6664 | */ | |
6665 | here_new = mddev->reshape_position; | |
664e7c41 | 6666 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 6667 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
6668 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
6669 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
6670 | return -EINVAL; |
6671 | } | |
c148ffdc | 6672 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
6673 | /* here_new is the stripe we will write to */ |
6674 | here_old = mddev->reshape_position; | |
9d8f0363 | 6675 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
6676 | (old_disks-max_degraded)); |
6677 | /* here_old is the first stripe that we might need to read | |
6678 | * from */ | |
67ac6011 | 6679 | if (mddev->delta_disks == 0) { |
b5254dd5 N |
6680 | if ((here_new * mddev->new_chunk_sectors != |
6681 | here_old * mddev->chunk_sectors)) { | |
6682 | printk(KERN_ERR "md/raid:%s: reshape position is" | |
6683 | " confused - aborting\n", mdname(mddev)); | |
6684 | return -EINVAL; | |
6685 | } | |
67ac6011 | 6686 | /* We cannot be sure it is safe to start an in-place |
b5254dd5 | 6687 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
6688 | * and taking constant backups. |
6689 | * mdadm always starts a situation like this in | |
6690 | * readonly mode so it can take control before | |
6691 | * allowing any writes. So just check for that. | |
6692 | */ | |
b5254dd5 N |
6693 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
6694 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
6695 | /* not really in-place - so OK */; | |
6696 | else if (mddev->ro == 0) { | |
6697 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
6698 | "must be started in read-only mode " | |
6699 | "- aborting\n", | |
0c55e022 | 6700 | mdname(mddev)); |
67ac6011 N |
6701 | return -EINVAL; |
6702 | } | |
2c810cdd | 6703 | } else if (mddev->reshape_backwards |
b5254dd5 | 6704 | ? (here_new * mddev->new_chunk_sectors + min_offset_diff <= |
67ac6011 N |
6705 | here_old * mddev->chunk_sectors) |
6706 | : (here_new * mddev->new_chunk_sectors >= | |
b5254dd5 | 6707 | here_old * mddev->chunk_sectors + (-min_offset_diff))) { |
91adb564 | 6708 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
6709 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
6710 | "auto-recovery - aborting.\n", | |
6711 | mdname(mddev)); | |
91adb564 N |
6712 | return -EINVAL; |
6713 | } | |
0c55e022 N |
6714 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
6715 | mdname(mddev)); | |
91adb564 N |
6716 | /* OK, we should be able to continue; */ |
6717 | } else { | |
6718 | BUG_ON(mddev->level != mddev->new_level); | |
6719 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 6720 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 6721 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 6722 | } |
91adb564 | 6723 | |
245f46c2 N |
6724 | if (mddev->private == NULL) |
6725 | conf = setup_conf(mddev); | |
6726 | else | |
6727 | conf = mddev->private; | |
6728 | ||
91adb564 N |
6729 | if (IS_ERR(conf)) |
6730 | return PTR_ERR(conf); | |
6731 | ||
b5254dd5 | 6732 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
6733 | mddev->thread = conf->thread; |
6734 | conf->thread = NULL; | |
6735 | mddev->private = conf; | |
6736 | ||
17045f52 N |
6737 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
6738 | i++) { | |
6739 | rdev = conf->disks[i].rdev; | |
6740 | if (!rdev && conf->disks[i].replacement) { | |
6741 | /* The replacement is all we have yet */ | |
6742 | rdev = conf->disks[i].replacement; | |
6743 | conf->disks[i].replacement = NULL; | |
6744 | clear_bit(Replacement, &rdev->flags); | |
6745 | conf->disks[i].rdev = rdev; | |
6746 | } | |
6747 | if (!rdev) | |
c148ffdc | 6748 | continue; |
17045f52 N |
6749 | if (conf->disks[i].replacement && |
6750 | conf->reshape_progress != MaxSector) { | |
6751 | /* replacements and reshape simply do not mix. */ | |
6752 | printk(KERN_ERR "md: cannot handle concurrent " | |
6753 | "replacement and reshape.\n"); | |
6754 | goto abort; | |
6755 | } | |
2f115882 | 6756 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 6757 | working_disks++; |
2f115882 N |
6758 | continue; |
6759 | } | |
c148ffdc N |
6760 | /* This disc is not fully in-sync. However if it |
6761 | * just stored parity (beyond the recovery_offset), | |
6762 | * when we don't need to be concerned about the | |
6763 | * array being dirty. | |
6764 | * When reshape goes 'backwards', we never have | |
6765 | * partially completed devices, so we only need | |
6766 | * to worry about reshape going forwards. | |
6767 | */ | |
6768 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
6769 | if (mddev->major_version == 0 && | |
6770 | mddev->minor_version > 90) | |
6771 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 6772 | |
c148ffdc N |
6773 | if (rdev->recovery_offset < reshape_offset) { |
6774 | /* We need to check old and new layout */ | |
6775 | if (!only_parity(rdev->raid_disk, | |
6776 | conf->algorithm, | |
6777 | conf->raid_disks, | |
6778 | conf->max_degraded)) | |
6779 | continue; | |
6780 | } | |
6781 | if (!only_parity(rdev->raid_disk, | |
6782 | conf->prev_algo, | |
6783 | conf->previous_raid_disks, | |
6784 | conf->max_degraded)) | |
6785 | continue; | |
6786 | dirty_parity_disks++; | |
6787 | } | |
91adb564 | 6788 | |
17045f52 N |
6789 | /* |
6790 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
6791 | */ | |
908f4fbd | 6792 | mddev->degraded = calc_degraded(conf); |
91adb564 | 6793 | |
674806d6 | 6794 | if (has_failed(conf)) { |
0c55e022 | 6795 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 6796 | " (%d/%d failed)\n", |
02c2de8c | 6797 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
6798 | goto abort; |
6799 | } | |
6800 | ||
91adb564 | 6801 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 6802 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
6803 | mddev->resync_max_sectors = mddev->dev_sectors; |
6804 | ||
c148ffdc | 6805 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 6806 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
6807 | if (mddev->ok_start_degraded) |
6808 | printk(KERN_WARNING | |
0c55e022 N |
6809 | "md/raid:%s: starting dirty degraded array" |
6810 | " - data corruption possible.\n", | |
6ff8d8ec N |
6811 | mdname(mddev)); |
6812 | else { | |
6813 | printk(KERN_ERR | |
0c55e022 | 6814 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
6815 | mdname(mddev)); |
6816 | goto abort; | |
6817 | } | |
1da177e4 LT |
6818 | } |
6819 | ||
1da177e4 | 6820 | if (mddev->degraded == 0) |
0c55e022 N |
6821 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
6822 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
6823 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
6824 | mddev->new_layout); | |
1da177e4 | 6825 | else |
0c55e022 N |
6826 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
6827 | " out of %d devices, algorithm %d\n", | |
6828 | mdname(mddev), conf->level, | |
6829 | mddev->raid_disks - mddev->degraded, | |
6830 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
6831 | |
6832 | print_raid5_conf(conf); | |
6833 | ||
fef9c61f | 6834 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 6835 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
6836 | atomic_set(&conf->reshape_stripes, 0); |
6837 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
6838 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6839 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6840 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6841 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6842 | "reshape"); |
f6705578 N |
6843 | } |
6844 | ||
1da177e4 | 6845 | /* Ok, everything is just fine now */ |
a64c876f N |
6846 | if (mddev->to_remove == &raid5_attrs_group) |
6847 | mddev->to_remove = NULL; | |
00bcb4ac N |
6848 | else if (mddev->kobj.sd && |
6849 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 6850 | printk(KERN_WARNING |
4a5add49 | 6851 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 6852 | mdname(mddev)); |
4a5add49 | 6853 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 6854 | |
4a5add49 | 6855 | if (mddev->queue) { |
9f7c2220 | 6856 | int chunk_size; |
620125f2 | 6857 | bool discard_supported = true; |
4a5add49 N |
6858 | /* read-ahead size must cover two whole stripes, which |
6859 | * is 2 * (datadisks) * chunksize where 'n' is the | |
6860 | * number of raid devices | |
6861 | */ | |
6862 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
6863 | int stripe = data_disks * | |
6864 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
6865 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
6866 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 6867 | |
9f7c2220 N |
6868 | chunk_size = mddev->chunk_sectors << 9; |
6869 | blk_queue_io_min(mddev->queue, chunk_size); | |
6870 | blk_queue_io_opt(mddev->queue, chunk_size * | |
6871 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 6872 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
6873 | /* |
6874 | * We can only discard a whole stripe. It doesn't make sense to | |
6875 | * discard data disk but write parity disk | |
6876 | */ | |
6877 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
6878 | /* Round up to power of 2, as discard handling |
6879 | * currently assumes that */ | |
6880 | while ((stripe-1) & stripe) | |
6881 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
6882 | mddev->queue->limits.discard_alignment = stripe; |
6883 | mddev->queue->limits.discard_granularity = stripe; | |
6884 | /* | |
6885 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 6886 | * guarantee discard_zeroes_data |
620125f2 SL |
6887 | */ |
6888 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 6889 | |
5026d7a9 PA |
6890 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
6891 | ||
05616be5 | 6892 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
6893 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6894 | rdev->data_offset << 9); | |
05616be5 N |
6895 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6896 | rdev->new_data_offset << 9); | |
620125f2 SL |
6897 | /* |
6898 | * discard_zeroes_data is required, otherwise data | |
6899 | * could be lost. Consider a scenario: discard a stripe | |
6900 | * (the stripe could be inconsistent if | |
6901 | * discard_zeroes_data is 0); write one disk of the | |
6902 | * stripe (the stripe could be inconsistent again | |
6903 | * depending on which disks are used to calculate | |
6904 | * parity); the disk is broken; The stripe data of this | |
6905 | * disk is lost. | |
6906 | */ | |
6907 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
6908 | !bdev_get_queue(rdev->bdev)-> | |
6909 | limits.discard_zeroes_data) | |
6910 | discard_supported = false; | |
8e0e99ba N |
6911 | /* Unfortunately, discard_zeroes_data is not currently |
6912 | * a guarantee - just a hint. So we only allow DISCARD | |
6913 | * if the sysadmin has confirmed that only safe devices | |
6914 | * are in use by setting a module parameter. | |
6915 | */ | |
6916 | if (!devices_handle_discard_safely) { | |
6917 | if (discard_supported) { | |
6918 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
6919 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
6920 | } | |
6921 | discard_supported = false; | |
6922 | } | |
05616be5 | 6923 | } |
620125f2 SL |
6924 | |
6925 | if (discard_supported && | |
6926 | mddev->queue->limits.max_discard_sectors >= stripe && | |
6927 | mddev->queue->limits.discard_granularity >= stripe) | |
6928 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
6929 | mddev->queue); | |
6930 | else | |
6931 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
6932 | mddev->queue); | |
9f7c2220 | 6933 | } |
23032a0e | 6934 | |
1da177e4 LT |
6935 | return 0; |
6936 | abort: | |
01f96c0a | 6937 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
6938 | print_raid5_conf(conf); |
6939 | free_conf(conf); | |
1da177e4 | 6940 | mddev->private = NULL; |
0c55e022 | 6941 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
6942 | return -EIO; |
6943 | } | |
6944 | ||
afa0f557 | 6945 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 6946 | { |
afa0f557 | 6947 | struct r5conf *conf = priv; |
1da177e4 | 6948 | |
95fc17aa | 6949 | free_conf(conf); |
a64c876f | 6950 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
6951 | } |
6952 | ||
fd01b88c | 6953 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6954 | { |
d1688a6d | 6955 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
6956 | int i; |
6957 | ||
9d8f0363 AN |
6958 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
6959 | mddev->chunk_sectors / 2, mddev->layout); | |
02c2de8c | 6960 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
6961 | for (i = 0; i < conf->raid_disks; i++) |
6962 | seq_printf (seq, "%s", | |
6963 | conf->disks[i].rdev && | |
b2d444d7 | 6964 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 6965 | seq_printf (seq, "]"); |
1da177e4 LT |
6966 | } |
6967 | ||
d1688a6d | 6968 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
6969 | { |
6970 | int i; | |
6971 | struct disk_info *tmp; | |
6972 | ||
0c55e022 | 6973 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
6974 | if (!conf) { |
6975 | printk("(conf==NULL)\n"); | |
6976 | return; | |
6977 | } | |
0c55e022 N |
6978 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
6979 | conf->raid_disks, | |
6980 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
6981 | |
6982 | for (i = 0; i < conf->raid_disks; i++) { | |
6983 | char b[BDEVNAME_SIZE]; | |
6984 | tmp = conf->disks + i; | |
6985 | if (tmp->rdev) | |
0c55e022 N |
6986 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
6987 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
6988 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
6989 | } |
6990 | } | |
6991 | ||
fd01b88c | 6992 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
6993 | { |
6994 | int i; | |
d1688a6d | 6995 | struct r5conf *conf = mddev->private; |
1da177e4 | 6996 | struct disk_info *tmp; |
6b965620 N |
6997 | int count = 0; |
6998 | unsigned long flags; | |
1da177e4 LT |
6999 | |
7000 | for (i = 0; i < conf->raid_disks; i++) { | |
7001 | tmp = conf->disks + i; | |
dd054fce N |
7002 | if (tmp->replacement |
7003 | && tmp->replacement->recovery_offset == MaxSector | |
7004 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7005 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7006 | /* Replacement has just become active. */ | |
7007 | if (!tmp->rdev | |
7008 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7009 | count++; | |
7010 | if (tmp->rdev) { | |
7011 | /* Replaced device not technically faulty, | |
7012 | * but we need to be sure it gets removed | |
7013 | * and never re-added. | |
7014 | */ | |
7015 | set_bit(Faulty, &tmp->rdev->flags); | |
7016 | sysfs_notify_dirent_safe( | |
7017 | tmp->rdev->sysfs_state); | |
7018 | } | |
7019 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7020 | } else if (tmp->rdev | |
70fffd0b | 7021 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7022 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7023 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7024 | count++; |
43c73ca4 | 7025 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7026 | } |
7027 | } | |
6b965620 | 7028 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7029 | mddev->degraded = calc_degraded(conf); |
6b965620 | 7030 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7031 | print_raid5_conf(conf); |
6b965620 | 7032 | return count; |
1da177e4 LT |
7033 | } |
7034 | ||
b8321b68 | 7035 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7036 | { |
d1688a6d | 7037 | struct r5conf *conf = mddev->private; |
1da177e4 | 7038 | int err = 0; |
b8321b68 | 7039 | int number = rdev->raid_disk; |
657e3e4d | 7040 | struct md_rdev **rdevp; |
1da177e4 LT |
7041 | struct disk_info *p = conf->disks + number; |
7042 | ||
7043 | print_raid5_conf(conf); | |
657e3e4d N |
7044 | if (rdev == p->rdev) |
7045 | rdevp = &p->rdev; | |
7046 | else if (rdev == p->replacement) | |
7047 | rdevp = &p->replacement; | |
7048 | else | |
7049 | return 0; | |
7050 | ||
7051 | if (number >= conf->raid_disks && | |
7052 | conf->reshape_progress == MaxSector) | |
7053 | clear_bit(In_sync, &rdev->flags); | |
7054 | ||
7055 | if (test_bit(In_sync, &rdev->flags) || | |
7056 | atomic_read(&rdev->nr_pending)) { | |
7057 | err = -EBUSY; | |
7058 | goto abort; | |
7059 | } | |
7060 | /* Only remove non-faulty devices if recovery | |
7061 | * isn't possible. | |
7062 | */ | |
7063 | if (!test_bit(Faulty, &rdev->flags) && | |
7064 | mddev->recovery_disabled != conf->recovery_disabled && | |
7065 | !has_failed(conf) && | |
dd054fce | 7066 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7067 | number < conf->raid_disks) { |
7068 | err = -EBUSY; | |
7069 | goto abort; | |
7070 | } | |
7071 | *rdevp = NULL; | |
7072 | synchronize_rcu(); | |
7073 | if (atomic_read(&rdev->nr_pending)) { | |
7074 | /* lost the race, try later */ | |
7075 | err = -EBUSY; | |
7076 | *rdevp = rdev; | |
dd054fce N |
7077 | } else if (p->replacement) { |
7078 | /* We must have just cleared 'rdev' */ | |
7079 | p->rdev = p->replacement; | |
7080 | clear_bit(Replacement, &p->replacement->flags); | |
7081 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7082 | * but will never see neither - if they are careful | |
7083 | */ | |
7084 | p->replacement = NULL; | |
7085 | clear_bit(WantReplacement, &rdev->flags); | |
7086 | } else | |
7087 | /* We might have just removed the Replacement as faulty- | |
7088 | * clear the bit just in case | |
7089 | */ | |
7090 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7091 | abort: |
7092 | ||
7093 | print_raid5_conf(conf); | |
7094 | return err; | |
7095 | } | |
7096 | ||
fd01b88c | 7097 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7098 | { |
d1688a6d | 7099 | struct r5conf *conf = mddev->private; |
199050ea | 7100 | int err = -EEXIST; |
1da177e4 LT |
7101 | int disk; |
7102 | struct disk_info *p; | |
6c2fce2e NB |
7103 | int first = 0; |
7104 | int last = conf->raid_disks - 1; | |
1da177e4 | 7105 | |
7f0da59b N |
7106 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7107 | return -EBUSY; | |
7108 | ||
dc10c643 | 7109 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7110 | /* no point adding a device */ |
199050ea | 7111 | return -EINVAL; |
1da177e4 | 7112 | |
6c2fce2e NB |
7113 | if (rdev->raid_disk >= 0) |
7114 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7115 | |
7116 | /* | |
16a53ecc N |
7117 | * find the disk ... but prefer rdev->saved_raid_disk |
7118 | * if possible. | |
1da177e4 | 7119 | */ |
16a53ecc | 7120 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7121 | rdev->saved_raid_disk >= first && |
16a53ecc | 7122 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7123 | first = rdev->saved_raid_disk; |
7124 | ||
7125 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7126 | p = conf->disks + disk; |
7127 | if (p->rdev == NULL) { | |
b2d444d7 | 7128 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7129 | rdev->raid_disk = disk; |
199050ea | 7130 | err = 0; |
72626685 N |
7131 | if (rdev->saved_raid_disk != disk) |
7132 | conf->fullsync = 1; | |
d6065f7b | 7133 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7134 | goto out; |
1da177e4 | 7135 | } |
5cfb22a1 N |
7136 | } |
7137 | for (disk = first; disk <= last; disk++) { | |
7138 | p = conf->disks + disk; | |
7bfec5f3 N |
7139 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7140 | p->replacement == NULL) { | |
7141 | clear_bit(In_sync, &rdev->flags); | |
7142 | set_bit(Replacement, &rdev->flags); | |
7143 | rdev->raid_disk = disk; | |
7144 | err = 0; | |
7145 | conf->fullsync = 1; | |
7146 | rcu_assign_pointer(p->replacement, rdev); | |
7147 | break; | |
7148 | } | |
7149 | } | |
5cfb22a1 | 7150 | out: |
1da177e4 | 7151 | print_raid5_conf(conf); |
199050ea | 7152 | return err; |
1da177e4 LT |
7153 | } |
7154 | ||
fd01b88c | 7155 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7156 | { |
7157 | /* no resync is happening, and there is enough space | |
7158 | * on all devices, so we can resize. | |
7159 | * We need to make sure resync covers any new space. | |
7160 | * If the array is shrinking we should possibly wait until | |
7161 | * any io in the removed space completes, but it hardly seems | |
7162 | * worth it. | |
7163 | */ | |
a4a6125a | 7164 | sector_t newsize; |
9d8f0363 | 7165 | sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
a4a6125a N |
7166 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7167 | if (mddev->external_size && | |
7168 | mddev->array_sectors > newsize) | |
b522adcd | 7169 | return -EINVAL; |
a4a6125a N |
7170 | if (mddev->bitmap) { |
7171 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7172 | if (ret) | |
7173 | return ret; | |
7174 | } | |
7175 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 7176 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 7177 | revalidate_disk(mddev->gendisk); |
b098636c N |
7178 | if (sectors > mddev->dev_sectors && |
7179 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7180 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7181 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7182 | } | |
58c0fed4 | 7183 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7184 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7185 | return 0; |
7186 | } | |
7187 | ||
fd01b88c | 7188 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7189 | { |
7190 | /* Can only proceed if there are plenty of stripe_heads. | |
7191 | * We need a minimum of one full stripe,, and for sensible progress | |
7192 | * it is best to have about 4 times that. | |
7193 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7194 | * allow for chunk sizes up to 256K, which is probably OK. | |
7195 | * If the chunk size is greater, user-space should request more | |
7196 | * stripe_heads first. | |
7197 | */ | |
d1688a6d | 7198 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7199 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7200 | > conf->min_nr_stripes || |
01ee22b4 | 7201 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7202 | > conf->min_nr_stripes) { |
0c55e022 N |
7203 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7204 | mdname(mddev), | |
01ee22b4 N |
7205 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
7206 | / STRIPE_SIZE)*4); | |
7207 | return 0; | |
7208 | } | |
7209 | return 1; | |
7210 | } | |
7211 | ||
fd01b88c | 7212 | static int check_reshape(struct mddev *mddev) |
29269553 | 7213 | { |
d1688a6d | 7214 | struct r5conf *conf = mddev->private; |
29269553 | 7215 | |
88ce4930 N |
7216 | if (mddev->delta_disks == 0 && |
7217 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7218 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7219 | return 0; /* nothing to do */ |
674806d6 | 7220 | if (has_failed(conf)) |
ec32a2bd | 7221 | return -EINVAL; |
fdcfbbb6 | 7222 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7223 | /* We might be able to shrink, but the devices must |
7224 | * be made bigger first. | |
7225 | * For raid6, 4 is the minimum size. | |
7226 | * Otherwise 2 is the minimum | |
7227 | */ | |
7228 | int min = 2; | |
7229 | if (mddev->level == 6) | |
7230 | min = 4; | |
7231 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7232 | return -EINVAL; | |
7233 | } | |
29269553 | 7234 | |
01ee22b4 | 7235 | if (!check_stripe_cache(mddev)) |
29269553 | 7236 | return -ENOSPC; |
29269553 | 7237 | |
738a2738 N |
7238 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7239 | mddev->delta_disks > 0) | |
7240 | if (resize_chunks(conf, | |
7241 | conf->previous_raid_disks | |
7242 | + max(0, mddev->delta_disks), | |
7243 | max(mddev->new_chunk_sectors, | |
7244 | mddev->chunk_sectors) | |
7245 | ) < 0) | |
7246 | return -ENOMEM; | |
e56108d6 N |
7247 | return resize_stripes(conf, (conf->previous_raid_disks |
7248 | + mddev->delta_disks)); | |
63c70c4f N |
7249 | } |
7250 | ||
fd01b88c | 7251 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7252 | { |
d1688a6d | 7253 | struct r5conf *conf = mddev->private; |
3cb03002 | 7254 | struct md_rdev *rdev; |
63c70c4f | 7255 | int spares = 0; |
c04be0aa | 7256 | unsigned long flags; |
63c70c4f | 7257 | |
f416885e | 7258 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7259 | return -EBUSY; |
7260 | ||
01ee22b4 N |
7261 | if (!check_stripe_cache(mddev)) |
7262 | return -ENOSPC; | |
7263 | ||
30b67645 N |
7264 | if (has_failed(conf)) |
7265 | return -EINVAL; | |
7266 | ||
c6563a8c | 7267 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7268 | if (!test_bit(In_sync, &rdev->flags) |
7269 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7270 | spares++; |
c6563a8c | 7271 | } |
63c70c4f | 7272 | |
f416885e | 7273 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7274 | /* Not enough devices even to make a degraded array |
7275 | * of that size | |
7276 | */ | |
7277 | return -EINVAL; | |
7278 | ||
ec32a2bd N |
7279 | /* Refuse to reduce size of the array. Any reductions in |
7280 | * array size must be through explicit setting of array_size | |
7281 | * attribute. | |
7282 | */ | |
7283 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7284 | < mddev->array_sectors) { | |
0c55e022 | 7285 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
7286 | "before number of disks\n", mdname(mddev)); |
7287 | return -EINVAL; | |
7288 | } | |
7289 | ||
f6705578 | 7290 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7291 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7292 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7293 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7294 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7295 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7296 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7297 | conf->prev_algo = conf->algorithm; |
7298 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7299 | conf->generation++; |
7300 | /* Code that selects data_offset needs to see the generation update | |
7301 | * if reshape_progress has been set - so a memory barrier needed. | |
7302 | */ | |
7303 | smp_mb(); | |
2c810cdd | 7304 | if (mddev->reshape_backwards) |
fef9c61f N |
7305 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7306 | else | |
7307 | conf->reshape_progress = 0; | |
7308 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7309 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7310 | spin_unlock_irq(&conf->device_lock); |
7311 | ||
4d77e3ba N |
7312 | /* Now make sure any requests that proceeded on the assumption |
7313 | * the reshape wasn't running - like Discard or Read - have | |
7314 | * completed. | |
7315 | */ | |
7316 | mddev_suspend(mddev); | |
7317 | mddev_resume(mddev); | |
7318 | ||
29269553 N |
7319 | /* Add some new drives, as many as will fit. |
7320 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7321 | * Don't add devices if we are reducing the number of |
7322 | * devices in the array. This is because it is not possible | |
7323 | * to correctly record the "partially reconstructed" state of | |
7324 | * such devices during the reshape and confusion could result. | |
29269553 | 7325 | */ |
87a8dec9 | 7326 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7327 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7328 | if (rdev->raid_disk < 0 && |
7329 | !test_bit(Faulty, &rdev->flags)) { | |
7330 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7331 | if (rdev->raid_disk |
9d4c7d87 | 7332 | >= conf->previous_raid_disks) |
87a8dec9 | 7333 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7334 | else |
87a8dec9 | 7335 | rdev->recovery_offset = 0; |
36fad858 NK |
7336 | |
7337 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7338 | /* Failure here is OK */; |
50da0840 | 7339 | } |
87a8dec9 N |
7340 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7341 | && !test_bit(Faulty, &rdev->flags)) { | |
7342 | /* This is a spare that was manually added */ | |
7343 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7344 | } |
29269553 | 7345 | |
87a8dec9 N |
7346 | /* When a reshape changes the number of devices, |
7347 | * ->degraded is measured against the larger of the | |
7348 | * pre and post number of devices. | |
7349 | */ | |
ec32a2bd | 7350 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7351 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
7352 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7353 | } | |
63c70c4f | 7354 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7355 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 7356 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 7357 | |
29269553 N |
7358 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7359 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7360 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7361 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7362 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7363 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7364 | "reshape"); |
29269553 N |
7365 | if (!mddev->sync_thread) { |
7366 | mddev->recovery = 0; | |
7367 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7368 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7369 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7370 | mddev->new_chunk_sectors = |
7371 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7372 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7373 | rdev_for_each(rdev, mddev) |
7374 | rdev->new_data_offset = rdev->data_offset; | |
7375 | smp_wmb(); | |
ba8805b9 | 7376 | conf->generation --; |
fef9c61f | 7377 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7378 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7379 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7380 | spin_unlock_irq(&conf->device_lock); |
7381 | return -EAGAIN; | |
7382 | } | |
c8f517c4 | 7383 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7384 | md_wakeup_thread(mddev->sync_thread); |
7385 | md_new_event(mddev); | |
7386 | return 0; | |
7387 | } | |
29269553 | 7388 | |
ec32a2bd N |
7389 | /* This is called from the reshape thread and should make any |
7390 | * changes needed in 'conf' | |
7391 | */ | |
d1688a6d | 7392 | static void end_reshape(struct r5conf *conf) |
29269553 | 7393 | { |
29269553 | 7394 | |
f6705578 | 7395 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7396 | struct md_rdev *rdev; |
f6705578 | 7397 | |
f6705578 | 7398 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7399 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7400 | rdev_for_each(rdev, conf->mddev) |
7401 | rdev->data_offset = rdev->new_data_offset; | |
7402 | smp_wmb(); | |
fef9c61f | 7403 | conf->reshape_progress = MaxSector; |
f6705578 | 7404 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7405 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7406 | |
7407 | /* read-ahead size must cover two whole stripes, which is | |
7408 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7409 | */ | |
4a5add49 | 7410 | if (conf->mddev->queue) { |
cea9c228 | 7411 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7412 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7413 | / PAGE_SIZE); |
16a53ecc N |
7414 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
7415 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
7416 | } | |
29269553 | 7417 | } |
29269553 N |
7418 | } |
7419 | ||
ec32a2bd N |
7420 | /* This is called from the raid5d thread with mddev_lock held. |
7421 | * It makes config changes to the device. | |
7422 | */ | |
fd01b88c | 7423 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7424 | { |
d1688a6d | 7425 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7426 | |
7427 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7428 | ||
ec32a2bd N |
7429 | if (mddev->delta_disks > 0) { |
7430 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
7431 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 7432 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
7433 | } else { |
7434 | int d; | |
908f4fbd N |
7435 | spin_lock_irq(&conf->device_lock); |
7436 | mddev->degraded = calc_degraded(conf); | |
7437 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
7438 | for (d = conf->raid_disks ; |
7439 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7440 | d++) { |
3cb03002 | 7441 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7442 | if (rdev) |
7443 | clear_bit(In_sync, &rdev->flags); | |
7444 | rdev = conf->disks[d].replacement; | |
7445 | if (rdev) | |
7446 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7447 | } |
cea9c228 | 7448 | } |
88ce4930 | 7449 | mddev->layout = conf->algorithm; |
09c9e5fa | 7450 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7451 | mddev->reshape_position = MaxSector; |
7452 | mddev->delta_disks = 0; | |
2c810cdd | 7453 | mddev->reshape_backwards = 0; |
cea9c228 N |
7454 | } |
7455 | } | |
7456 | ||
fd01b88c | 7457 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7458 | { |
d1688a6d | 7459 | struct r5conf *conf = mddev->private; |
72626685 N |
7460 | |
7461 | switch(state) { | |
e464eafd N |
7462 | case 2: /* resume for a suspend */ |
7463 | wake_up(&conf->wait_for_overlap); | |
7464 | break; | |
7465 | ||
72626685 | 7466 | case 1: /* stop all writes */ |
566c09c5 | 7467 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7468 | /* '2' tells resync/reshape to pause so that all |
7469 | * active stripes can drain | |
7470 | */ | |
7471 | conf->quiesce = 2; | |
b1b46486 | 7472 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7473 | atomic_read(&conf->active_stripes) == 0 && |
7474 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7475 | unlock_all_device_hash_locks_irq(conf), |
7476 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7477 | conf->quiesce = 1; |
566c09c5 | 7478 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7479 | /* allow reshape to continue */ |
7480 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7481 | break; |
7482 | ||
7483 | case 0: /* re-enable writes */ | |
566c09c5 | 7484 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7485 | conf->quiesce = 0; |
b1b46486 | 7486 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7487 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7488 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7489 | break; |
7490 | } | |
72626685 | 7491 | } |
b15c2e57 | 7492 | |
fd01b88c | 7493 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7494 | { |
e373ab10 | 7495 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7496 | sector_t sectors; |
54071b38 | 7497 | |
f1b29bca | 7498 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7499 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
7500 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7501 | mdname(mddev)); | |
f1b29bca DW |
7502 | return ERR_PTR(-EINVAL); |
7503 | } | |
7504 | ||
e373ab10 N |
7505 | sectors = raid0_conf->strip_zone[0].zone_end; |
7506 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7507 | mddev->dev_sectors = sectors; |
f1b29bca | 7508 | mddev->new_level = level; |
54071b38 TM |
7509 | mddev->new_layout = ALGORITHM_PARITY_N; |
7510 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7511 | mddev->raid_disks += 1; | |
7512 | mddev->delta_disks = 1; | |
7513 | /* make sure it will be not marked as dirty */ | |
7514 | mddev->recovery_cp = MaxSector; | |
7515 | ||
7516 | return setup_conf(mddev); | |
7517 | } | |
7518 | ||
fd01b88c | 7519 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7520 | { |
7521 | int chunksect; | |
7522 | ||
7523 | if (mddev->raid_disks != 2 || | |
7524 | mddev->degraded > 1) | |
7525 | return ERR_PTR(-EINVAL); | |
7526 | ||
7527 | /* Should check if there are write-behind devices? */ | |
7528 | ||
7529 | chunksect = 64*2; /* 64K by default */ | |
7530 | ||
7531 | /* The array must be an exact multiple of chunksize */ | |
7532 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
7533 | chunksect >>= 1; | |
7534 | ||
7535 | if ((chunksect<<9) < STRIPE_SIZE) | |
7536 | /* array size does not allow a suitable chunk size */ | |
7537 | return ERR_PTR(-EINVAL); | |
7538 | ||
7539 | mddev->new_level = 5; | |
7540 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 7541 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
7542 | |
7543 | return setup_conf(mddev); | |
7544 | } | |
7545 | ||
fd01b88c | 7546 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
7547 | { |
7548 | int new_layout; | |
7549 | ||
7550 | switch (mddev->layout) { | |
7551 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7552 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7553 | break; | |
7554 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7555 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
7556 | break; | |
7557 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7558 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
7559 | break; | |
7560 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7561 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
7562 | break; | |
7563 | case ALGORITHM_PARITY_0_6: | |
7564 | new_layout = ALGORITHM_PARITY_0; | |
7565 | break; | |
7566 | case ALGORITHM_PARITY_N: | |
7567 | new_layout = ALGORITHM_PARITY_N; | |
7568 | break; | |
7569 | default: | |
7570 | return ERR_PTR(-EINVAL); | |
7571 | } | |
7572 | mddev->new_level = 5; | |
7573 | mddev->new_layout = new_layout; | |
7574 | mddev->delta_disks = -1; | |
7575 | mddev->raid_disks -= 1; | |
7576 | return setup_conf(mddev); | |
7577 | } | |
7578 | ||
fd01b88c | 7579 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 7580 | { |
88ce4930 N |
7581 | /* For a 2-drive array, the layout and chunk size can be changed |
7582 | * immediately as not restriping is needed. | |
7583 | * For larger arrays we record the new value - after validation | |
7584 | * to be used by a reshape pass. | |
b3546035 | 7585 | */ |
d1688a6d | 7586 | struct r5conf *conf = mddev->private; |
597a711b | 7587 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 7588 | |
597a711b | 7589 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
7590 | return -EINVAL; |
7591 | if (new_chunk > 0) { | |
0ba459d2 | 7592 | if (!is_power_of_2(new_chunk)) |
b3546035 | 7593 | return -EINVAL; |
597a711b | 7594 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 7595 | return -EINVAL; |
597a711b | 7596 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
7597 | /* not factor of array size */ |
7598 | return -EINVAL; | |
7599 | } | |
7600 | ||
7601 | /* They look valid */ | |
7602 | ||
88ce4930 | 7603 | if (mddev->raid_disks == 2) { |
597a711b N |
7604 | /* can make the change immediately */ |
7605 | if (mddev->new_layout >= 0) { | |
7606 | conf->algorithm = mddev->new_layout; | |
7607 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
7608 | } |
7609 | if (new_chunk > 0) { | |
597a711b N |
7610 | conf->chunk_sectors = new_chunk ; |
7611 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
7612 | } |
7613 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7614 | md_wakeup_thread(mddev->thread); | |
b3546035 | 7615 | } |
50ac168a | 7616 | return check_reshape(mddev); |
88ce4930 N |
7617 | } |
7618 | ||
fd01b88c | 7619 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 7620 | { |
597a711b | 7621 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 7622 | |
597a711b | 7623 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 7624 | return -EINVAL; |
b3546035 | 7625 | if (new_chunk > 0) { |
0ba459d2 | 7626 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 7627 | return -EINVAL; |
597a711b | 7628 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 7629 | return -EINVAL; |
597a711b | 7630 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
7631 | /* not factor of array size */ |
7632 | return -EINVAL; | |
b3546035 | 7633 | } |
88ce4930 N |
7634 | |
7635 | /* They look valid */ | |
50ac168a | 7636 | return check_reshape(mddev); |
b3546035 N |
7637 | } |
7638 | ||
fd01b88c | 7639 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
7640 | { |
7641 | /* raid5 can take over: | |
f1b29bca | 7642 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
7643 | * raid1 - if there are two drives. We need to know the chunk size |
7644 | * raid4 - trivial - just use a raid4 layout. | |
7645 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 7646 | */ |
f1b29bca DW |
7647 | if (mddev->level == 0) |
7648 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
7649 | if (mddev->level == 1) |
7650 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
7651 | if (mddev->level == 4) { |
7652 | mddev->new_layout = ALGORITHM_PARITY_N; | |
7653 | mddev->new_level = 5; | |
7654 | return setup_conf(mddev); | |
7655 | } | |
fc9739c6 N |
7656 | if (mddev->level == 6) |
7657 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
7658 | |
7659 | return ERR_PTR(-EINVAL); | |
7660 | } | |
7661 | ||
fd01b88c | 7662 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 7663 | { |
f1b29bca DW |
7664 | /* raid4 can take over: |
7665 | * raid0 - if there is only one strip zone | |
7666 | * raid5 - if layout is right | |
a78d38a1 | 7667 | */ |
f1b29bca DW |
7668 | if (mddev->level == 0) |
7669 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
7670 | if (mddev->level == 5 && |
7671 | mddev->layout == ALGORITHM_PARITY_N) { | |
7672 | mddev->new_layout = 0; | |
7673 | mddev->new_level = 4; | |
7674 | return setup_conf(mddev); | |
7675 | } | |
7676 | return ERR_PTR(-EINVAL); | |
7677 | } | |
d562b0c4 | 7678 | |
84fc4b56 | 7679 | static struct md_personality raid5_personality; |
245f46c2 | 7680 | |
fd01b88c | 7681 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
7682 | { |
7683 | /* Currently can only take over a raid5. We map the | |
7684 | * personality to an equivalent raid6 personality | |
7685 | * with the Q block at the end. | |
7686 | */ | |
7687 | int new_layout; | |
7688 | ||
7689 | if (mddev->pers != &raid5_personality) | |
7690 | return ERR_PTR(-EINVAL); | |
7691 | if (mddev->degraded > 1) | |
7692 | return ERR_PTR(-EINVAL); | |
7693 | if (mddev->raid_disks > 253) | |
7694 | return ERR_PTR(-EINVAL); | |
7695 | if (mddev->raid_disks < 3) | |
7696 | return ERR_PTR(-EINVAL); | |
7697 | ||
7698 | switch (mddev->layout) { | |
7699 | case ALGORITHM_LEFT_ASYMMETRIC: | |
7700 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
7701 | break; | |
7702 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
7703 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
7704 | break; | |
7705 | case ALGORITHM_LEFT_SYMMETRIC: | |
7706 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
7707 | break; | |
7708 | case ALGORITHM_RIGHT_SYMMETRIC: | |
7709 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
7710 | break; | |
7711 | case ALGORITHM_PARITY_0: | |
7712 | new_layout = ALGORITHM_PARITY_0_6; | |
7713 | break; | |
7714 | case ALGORITHM_PARITY_N: | |
7715 | new_layout = ALGORITHM_PARITY_N; | |
7716 | break; | |
7717 | default: | |
7718 | return ERR_PTR(-EINVAL); | |
7719 | } | |
7720 | mddev->new_level = 6; | |
7721 | mddev->new_layout = new_layout; | |
7722 | mddev->delta_disks = 1; | |
7723 | mddev->raid_disks += 1; | |
7724 | return setup_conf(mddev); | |
7725 | } | |
7726 | ||
84fc4b56 | 7727 | static struct md_personality raid6_personality = |
16a53ecc N |
7728 | { |
7729 | .name = "raid6", | |
7730 | .level = 6, | |
7731 | .owner = THIS_MODULE, | |
7732 | .make_request = make_request, | |
7733 | .run = run, | |
afa0f557 | 7734 | .free = raid5_free, |
16a53ecc N |
7735 | .status = status, |
7736 | .error_handler = error, | |
7737 | .hot_add_disk = raid5_add_disk, | |
7738 | .hot_remove_disk= raid5_remove_disk, | |
7739 | .spare_active = raid5_spare_active, | |
7740 | .sync_request = sync_request, | |
7741 | .resize = raid5_resize, | |
80c3a6ce | 7742 | .size = raid5_size, |
50ac168a | 7743 | .check_reshape = raid6_check_reshape, |
f416885e | 7744 | .start_reshape = raid5_start_reshape, |
cea9c228 | 7745 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 7746 | .quiesce = raid5_quiesce, |
245f46c2 | 7747 | .takeover = raid6_takeover, |
5c675f83 | 7748 | .congested = raid5_congested, |
64590f45 | 7749 | .mergeable_bvec = raid5_mergeable_bvec, |
16a53ecc | 7750 | }; |
84fc4b56 | 7751 | static struct md_personality raid5_personality = |
1da177e4 LT |
7752 | { |
7753 | .name = "raid5", | |
2604b703 | 7754 | .level = 5, |
1da177e4 LT |
7755 | .owner = THIS_MODULE, |
7756 | .make_request = make_request, | |
7757 | .run = run, | |
afa0f557 | 7758 | .free = raid5_free, |
1da177e4 LT |
7759 | .status = status, |
7760 | .error_handler = error, | |
7761 | .hot_add_disk = raid5_add_disk, | |
7762 | .hot_remove_disk= raid5_remove_disk, | |
7763 | .spare_active = raid5_spare_active, | |
7764 | .sync_request = sync_request, | |
7765 | .resize = raid5_resize, | |
80c3a6ce | 7766 | .size = raid5_size, |
63c70c4f N |
7767 | .check_reshape = raid5_check_reshape, |
7768 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7769 | .finish_reshape = raid5_finish_reshape, |
72626685 | 7770 | .quiesce = raid5_quiesce, |
d562b0c4 | 7771 | .takeover = raid5_takeover, |
5c675f83 | 7772 | .congested = raid5_congested, |
64590f45 | 7773 | .mergeable_bvec = raid5_mergeable_bvec, |
1da177e4 LT |
7774 | }; |
7775 | ||
84fc4b56 | 7776 | static struct md_personality raid4_personality = |
1da177e4 | 7777 | { |
2604b703 N |
7778 | .name = "raid4", |
7779 | .level = 4, | |
7780 | .owner = THIS_MODULE, | |
7781 | .make_request = make_request, | |
7782 | .run = run, | |
afa0f557 | 7783 | .free = raid5_free, |
2604b703 N |
7784 | .status = status, |
7785 | .error_handler = error, | |
7786 | .hot_add_disk = raid5_add_disk, | |
7787 | .hot_remove_disk= raid5_remove_disk, | |
7788 | .spare_active = raid5_spare_active, | |
7789 | .sync_request = sync_request, | |
7790 | .resize = raid5_resize, | |
80c3a6ce | 7791 | .size = raid5_size, |
3d37890b N |
7792 | .check_reshape = raid5_check_reshape, |
7793 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7794 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 7795 | .quiesce = raid5_quiesce, |
a78d38a1 | 7796 | .takeover = raid4_takeover, |
5c675f83 | 7797 | .congested = raid5_congested, |
64590f45 | 7798 | .mergeable_bvec = raid5_mergeable_bvec, |
2604b703 N |
7799 | }; |
7800 | ||
7801 | static int __init raid5_init(void) | |
7802 | { | |
851c30c9 SL |
7803 | raid5_wq = alloc_workqueue("raid5wq", |
7804 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
7805 | if (!raid5_wq) | |
7806 | return -ENOMEM; | |
16a53ecc | 7807 | register_md_personality(&raid6_personality); |
2604b703 N |
7808 | register_md_personality(&raid5_personality); |
7809 | register_md_personality(&raid4_personality); | |
7810 | return 0; | |
1da177e4 LT |
7811 | } |
7812 | ||
2604b703 | 7813 | static void raid5_exit(void) |
1da177e4 | 7814 | { |
16a53ecc | 7815 | unregister_md_personality(&raid6_personality); |
2604b703 N |
7816 | unregister_md_personality(&raid5_personality); |
7817 | unregister_md_personality(&raid4_personality); | |
851c30c9 | 7818 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
7819 | } |
7820 | ||
7821 | module_init(raid5_init); | |
7822 | module_exit(raid5_exit); | |
7823 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7824 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 7825 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
7826 | MODULE_ALIAS("md-raid5"); |
7827 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
7828 | MODULE_ALIAS("md-level-5"); |
7829 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
7830 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
7831 | MODULE_ALIAS("md-raid6"); | |
7832 | MODULE_ALIAS("md-level-6"); | |
7833 | ||
7834 | /* This used to be two separate modules, they were: */ | |
7835 | MODULE_ALIAS("raid5"); | |
7836 | MODULE_ALIAS("raid6"); |