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 | ||
34a6f80e | 226 | static void return_io(struct bio_list *return_bi) |
a4456856 | 227 | { |
34a6f80e N |
228 | struct bio *bi; |
229 | while ((bi = bio_list_pop(return_bi)) != NULL) { | |
4f024f37 | 230 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
231 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
232 | bi, 0); | |
4246a0b6 | 233 | bio_endio(bi); |
a4456856 DW |
234 | } |
235 | } | |
236 | ||
d1688a6d | 237 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 238 | |
600aa109 DW |
239 | static int stripe_operations_active(struct stripe_head *sh) |
240 | { | |
241 | return sh->check_state || sh->reconstruct_state || | |
242 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
243 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
244 | } | |
245 | ||
851c30c9 SL |
246 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
247 | { | |
248 | struct r5conf *conf = sh->raid_conf; | |
249 | struct r5worker_group *group; | |
bfc90cb0 | 250 | int thread_cnt; |
851c30c9 SL |
251 | int i, cpu = sh->cpu; |
252 | ||
253 | if (!cpu_online(cpu)) { | |
254 | cpu = cpumask_any(cpu_online_mask); | |
255 | sh->cpu = cpu; | |
256 | } | |
257 | ||
258 | if (list_empty(&sh->lru)) { | |
259 | struct r5worker_group *group; | |
260 | group = conf->worker_groups + cpu_to_group(cpu); | |
261 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
262 | group->stripes_cnt++; |
263 | sh->group = group; | |
851c30c9 SL |
264 | } |
265 | ||
266 | if (conf->worker_cnt_per_group == 0) { | |
267 | md_wakeup_thread(conf->mddev->thread); | |
268 | return; | |
269 | } | |
270 | ||
271 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
272 | ||
bfc90cb0 SL |
273 | group->workers[0].working = true; |
274 | /* at least one worker should run to avoid race */ | |
275 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
276 | ||
277 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
278 | /* wakeup more workers */ | |
279 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
280 | if (group->workers[i].working == false) { | |
281 | group->workers[i].working = true; | |
282 | queue_work_on(sh->cpu, raid5_wq, | |
283 | &group->workers[i].work); | |
284 | thread_cnt--; | |
285 | } | |
286 | } | |
851c30c9 SL |
287 | } |
288 | ||
566c09c5 SL |
289 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
290 | struct list_head *temp_inactive_list) | |
1da177e4 | 291 | { |
4eb788df SL |
292 | BUG_ON(!list_empty(&sh->lru)); |
293 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
294 | if (test_bit(STRIPE_HANDLE, &sh->state)) { | |
295 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 296 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 297 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 298 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
299 | sh->bm_seq - conf->seq_write > 0) |
300 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
301 | else { | |
302 | clear_bit(STRIPE_DELAYED, &sh->state); | |
303 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 SL |
304 | if (conf->worker_cnt_per_group == 0) { |
305 | list_add_tail(&sh->lru, &conf->handle_list); | |
306 | } else { | |
307 | raid5_wakeup_stripe_thread(sh); | |
308 | return; | |
309 | } | |
4eb788df SL |
310 | } |
311 | md_wakeup_thread(conf->mddev->thread); | |
312 | } else { | |
313 | BUG_ON(stripe_operations_active(sh)); | |
314 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
315 | if (atomic_dec_return(&conf->preread_active_stripes) | |
316 | < IO_THRESHOLD) | |
317 | md_wakeup_thread(conf->mddev->thread); | |
318 | atomic_dec(&conf->active_stripes); | |
566c09c5 SL |
319 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) |
320 | list_add_tail(&sh->lru, temp_inactive_list); | |
1da177e4 LT |
321 | } |
322 | } | |
d0dabf7e | 323 | |
566c09c5 SL |
324 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
325 | struct list_head *temp_inactive_list) | |
4eb788df SL |
326 | { |
327 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
328 | do_release_stripe(conf, sh, temp_inactive_list); |
329 | } | |
330 | ||
331 | /* | |
332 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
333 | * | |
334 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
335 | * given time. Adding stripes only takes device lock, while deleting stripes | |
336 | * only takes hash lock. | |
337 | */ | |
338 | static void release_inactive_stripe_list(struct r5conf *conf, | |
339 | struct list_head *temp_inactive_list, | |
340 | int hash) | |
341 | { | |
342 | int size; | |
e9e4c377 YL |
343 | unsigned long do_wakeup = 0; |
344 | int i = 0; | |
566c09c5 SL |
345 | unsigned long flags; |
346 | ||
347 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
348 | size = NR_STRIPE_HASH_LOCKS; | |
349 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
350 | } else | |
351 | size = 1; | |
352 | while (size) { | |
353 | struct list_head *list = &temp_inactive_list[size - 1]; | |
354 | ||
355 | /* | |
356 | * We don't hold any lock here yet, get_active_stripe() might | |
357 | * remove stripes from the list | |
358 | */ | |
359 | if (!list_empty_careful(list)) { | |
360 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
361 | if (list_empty(conf->inactive_list + hash) && |
362 | !list_empty(list)) | |
363 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 364 | list_splice_tail_init(list, conf->inactive_list + hash); |
e9e4c377 | 365 | do_wakeup |= 1 << hash; |
566c09c5 SL |
366 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
367 | } | |
368 | size--; | |
369 | hash--; | |
370 | } | |
371 | ||
e9e4c377 YL |
372 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) { |
373 | if (do_wakeup & (1 << i)) | |
374 | wake_up(&conf->wait_for_stripe[i]); | |
375 | } | |
376 | ||
566c09c5 | 377 | if (do_wakeup) { |
b1b46486 YL |
378 | if (atomic_read(&conf->active_stripes) == 0) |
379 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
380 | if (conf->retry_read_aligned) |
381 | md_wakeup_thread(conf->mddev->thread); | |
382 | } | |
4eb788df SL |
383 | } |
384 | ||
773ca82f | 385 | /* should hold conf->device_lock already */ |
566c09c5 SL |
386 | static int release_stripe_list(struct r5conf *conf, |
387 | struct list_head *temp_inactive_list) | |
773ca82f SL |
388 | { |
389 | struct stripe_head *sh; | |
390 | int count = 0; | |
391 | struct llist_node *head; | |
392 | ||
393 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 394 | head = llist_reverse_order(head); |
773ca82f | 395 | while (head) { |
566c09c5 SL |
396 | int hash; |
397 | ||
773ca82f SL |
398 | sh = llist_entry(head, struct stripe_head, release_list); |
399 | head = llist_next(head); | |
400 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ | |
401 | smp_mb(); | |
402 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
403 | /* | |
404 | * Don't worry the bit is set here, because if the bit is set | |
405 | * again, the count is always > 1. This is true for | |
406 | * STRIPE_ON_UNPLUG_LIST bit too. | |
407 | */ | |
566c09c5 SL |
408 | hash = sh->hash_lock_index; |
409 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
410 | count++; |
411 | } | |
412 | ||
413 | return count; | |
414 | } | |
415 | ||
1da177e4 LT |
416 | static void release_stripe(struct stripe_head *sh) |
417 | { | |
d1688a6d | 418 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 419 | unsigned long flags; |
566c09c5 SL |
420 | struct list_head list; |
421 | int hash; | |
773ca82f | 422 | bool wakeup; |
16a53ecc | 423 | |
cf170f3f ES |
424 | /* Avoid release_list until the last reference. |
425 | */ | |
426 | if (atomic_add_unless(&sh->count, -1, 1)) | |
427 | return; | |
428 | ||
ad4068de | 429 | if (unlikely(!conf->mddev->thread) || |
430 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
431 | goto slow_path; |
432 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
433 | if (wakeup) | |
434 | md_wakeup_thread(conf->mddev->thread); | |
435 | return; | |
436 | slow_path: | |
4eb788df | 437 | local_irq_save(flags); |
773ca82f | 438 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 439 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
440 | INIT_LIST_HEAD(&list); |
441 | hash = sh->hash_lock_index; | |
442 | do_release_stripe(conf, sh, &list); | |
4eb788df | 443 | spin_unlock(&conf->device_lock); |
566c09c5 | 444 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
445 | } |
446 | local_irq_restore(flags); | |
1da177e4 LT |
447 | } |
448 | ||
fccddba0 | 449 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 450 | { |
45b4233c DW |
451 | pr_debug("remove_hash(), stripe %llu\n", |
452 | (unsigned long long)sh->sector); | |
1da177e4 | 453 | |
fccddba0 | 454 | hlist_del_init(&sh->hash); |
1da177e4 LT |
455 | } |
456 | ||
d1688a6d | 457 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 458 | { |
fccddba0 | 459 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 460 | |
45b4233c DW |
461 | pr_debug("insert_hash(), stripe %llu\n", |
462 | (unsigned long long)sh->sector); | |
1da177e4 | 463 | |
fccddba0 | 464 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
465 | } |
466 | ||
1da177e4 | 467 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 468 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
469 | { |
470 | struct stripe_head *sh = NULL; | |
471 | struct list_head *first; | |
472 | ||
566c09c5 | 473 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 474 | goto out; |
566c09c5 | 475 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
476 | sh = list_entry(first, struct stripe_head, lru); |
477 | list_del_init(first); | |
478 | remove_hash(sh); | |
479 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 480 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
481 | if (list_empty(conf->inactive_list + hash)) |
482 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
483 | out: |
484 | return sh; | |
485 | } | |
486 | ||
e4e11e38 | 487 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
488 | { |
489 | struct page *p; | |
490 | int i; | |
e4e11e38 | 491 | int num = sh->raid_conf->pool_size; |
1da177e4 | 492 | |
e4e11e38 | 493 | for (i = 0; i < num ; i++) { |
d592a996 | 494 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
495 | p = sh->dev[i].page; |
496 | if (!p) | |
497 | continue; | |
498 | sh->dev[i].page = NULL; | |
2d1f3b5d | 499 | put_page(p); |
1da177e4 LT |
500 | } |
501 | } | |
502 | ||
a9683a79 | 503 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
504 | { |
505 | int i; | |
e4e11e38 | 506 | int num = sh->raid_conf->pool_size; |
1da177e4 | 507 | |
e4e11e38 | 508 | for (i = 0; i < num; i++) { |
1da177e4 LT |
509 | struct page *page; |
510 | ||
a9683a79 | 511 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
512 | return 1; |
513 | } | |
514 | sh->dev[i].page = page; | |
d592a996 | 515 | sh->dev[i].orig_page = page; |
1da177e4 LT |
516 | } |
517 | return 0; | |
518 | } | |
519 | ||
784052ec | 520 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 521 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 522 | struct stripe_head *sh); |
1da177e4 | 523 | |
b5663ba4 | 524 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 525 | { |
d1688a6d | 526 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 527 | int i, seq; |
1da177e4 | 528 | |
78bafebd ES |
529 | BUG_ON(atomic_read(&sh->count) != 0); |
530 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 531 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 532 | BUG_ON(sh->batch_head); |
d84e0f10 | 533 | |
45b4233c | 534 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 535 | (unsigned long long)sector); |
566c09c5 SL |
536 | retry: |
537 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 538 | sh->generation = conf->generation - previous; |
b5663ba4 | 539 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 540 | sh->sector = sector; |
911d4ee8 | 541 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
542 | sh->state = 0; |
543 | ||
7ecaa1e6 | 544 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
545 | struct r5dev *dev = &sh->dev[i]; |
546 | ||
d84e0f10 | 547 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 548 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 549 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 550 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 551 | dev->read, dev->towrite, dev->written, |
1da177e4 | 552 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 553 | WARN_ON(1); |
1da177e4 LT |
554 | } |
555 | dev->flags = 0; | |
784052ec | 556 | raid5_build_block(sh, i, previous); |
1da177e4 | 557 | } |
566c09c5 SL |
558 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
559 | goto retry; | |
7a87f434 | 560 | sh->overwrite_disks = 0; |
1da177e4 | 561 | insert_hash(conf, sh); |
851c30c9 | 562 | sh->cpu = smp_processor_id(); |
da41ba65 | 563 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
564 | } |
565 | ||
d1688a6d | 566 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 567 | short generation) |
1da177e4 LT |
568 | { |
569 | struct stripe_head *sh; | |
570 | ||
45b4233c | 571 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 572 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 573 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 574 | return sh; |
45b4233c | 575 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
576 | return NULL; |
577 | } | |
578 | ||
674806d6 N |
579 | /* |
580 | * Need to check if array has failed when deciding whether to: | |
581 | * - start an array | |
582 | * - remove non-faulty devices | |
583 | * - add a spare | |
584 | * - allow a reshape | |
585 | * This determination is simple when no reshape is happening. | |
586 | * However if there is a reshape, we need to carefully check | |
587 | * both the before and after sections. | |
588 | * This is because some failed devices may only affect one | |
589 | * of the two sections, and some non-in_sync devices may | |
590 | * be insync in the section most affected by failed devices. | |
591 | */ | |
908f4fbd | 592 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 593 | { |
908f4fbd | 594 | int degraded, degraded2; |
674806d6 | 595 | int i; |
674806d6 N |
596 | |
597 | rcu_read_lock(); | |
598 | degraded = 0; | |
599 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 600 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
601 | if (rdev && test_bit(Faulty, &rdev->flags)) |
602 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
603 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
604 | degraded++; | |
605 | else if (test_bit(In_sync, &rdev->flags)) | |
606 | ; | |
607 | else | |
608 | /* not in-sync or faulty. | |
609 | * If the reshape increases the number of devices, | |
610 | * this is being recovered by the reshape, so | |
611 | * this 'previous' section is not in_sync. | |
612 | * If the number of devices is being reduced however, | |
613 | * the device can only be part of the array if | |
614 | * we are reverting a reshape, so this section will | |
615 | * be in-sync. | |
616 | */ | |
617 | if (conf->raid_disks >= conf->previous_raid_disks) | |
618 | degraded++; | |
619 | } | |
620 | rcu_read_unlock(); | |
908f4fbd N |
621 | if (conf->raid_disks == conf->previous_raid_disks) |
622 | return degraded; | |
674806d6 | 623 | rcu_read_lock(); |
908f4fbd | 624 | degraded2 = 0; |
674806d6 | 625 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 626 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
627 | if (rdev && test_bit(Faulty, &rdev->flags)) |
628 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 629 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 630 | degraded2++; |
674806d6 N |
631 | else if (test_bit(In_sync, &rdev->flags)) |
632 | ; | |
633 | else | |
634 | /* not in-sync or faulty. | |
635 | * If reshape increases the number of devices, this | |
636 | * section has already been recovered, else it | |
637 | * almost certainly hasn't. | |
638 | */ | |
639 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 640 | degraded2++; |
674806d6 N |
641 | } |
642 | rcu_read_unlock(); | |
908f4fbd N |
643 | if (degraded2 > degraded) |
644 | return degraded2; | |
645 | return degraded; | |
646 | } | |
647 | ||
648 | static int has_failed(struct r5conf *conf) | |
649 | { | |
650 | int degraded; | |
651 | ||
652 | if (conf->mddev->reshape_position == MaxSector) | |
653 | return conf->mddev->degraded > conf->max_degraded; | |
654 | ||
655 | degraded = calc_degraded(conf); | |
674806d6 N |
656 | if (degraded > conf->max_degraded) |
657 | return 1; | |
658 | return 0; | |
659 | } | |
660 | ||
b5663ba4 | 661 | static struct stripe_head * |
d1688a6d | 662 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 663 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
664 | { |
665 | struct stripe_head *sh; | |
566c09c5 | 666 | int hash = stripe_hash_locks_hash(sector); |
1da177e4 | 667 | |
45b4233c | 668 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 669 | |
566c09c5 | 670 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
671 | |
672 | do { | |
b1b46486 | 673 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 674 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 675 | *(conf->hash_locks + hash)); |
86b42c71 | 676 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 677 | if (!sh) { |
edbe83ab | 678 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 679 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
680 | if (!sh && !test_bit(R5_DID_ALLOC, |
681 | &conf->cache_state)) | |
edbe83ab N |
682 | set_bit(R5_ALLOC_MORE, |
683 | &conf->cache_state); | |
684 | } | |
1da177e4 LT |
685 | if (noblock && sh == NULL) |
686 | break; | |
687 | if (!sh) { | |
5423399a N |
688 | set_bit(R5_INACTIVE_BLOCKED, |
689 | &conf->cache_state); | |
e9e4c377 YL |
690 | wait_event_exclusive_cmd( |
691 | conf->wait_for_stripe[hash], | |
566c09c5 SL |
692 | !list_empty(conf->inactive_list + hash) && |
693 | (atomic_read(&conf->active_stripes) | |
694 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
695 | || !test_bit(R5_INACTIVE_BLOCKED, |
696 | &conf->cache_state)), | |
e9e4c377 YL |
697 | spin_unlock_irq(conf->hash_locks + hash), |
698 | spin_lock_irq(conf->hash_locks + hash)); | |
5423399a N |
699 | clear_bit(R5_INACTIVE_BLOCKED, |
700 | &conf->cache_state); | |
7da9d450 | 701 | } else { |
b5663ba4 | 702 | init_stripe(sh, sector, previous); |
7da9d450 N |
703 | atomic_inc(&sh->count); |
704 | } | |
e240c183 | 705 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 706 | spin_lock(&conf->device_lock); |
e240c183 | 707 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
708 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
709 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
710 | BUG_ON(list_empty(&sh->lru) && |
711 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
16a53ecc | 712 | list_del_init(&sh->lru); |
bfc90cb0 SL |
713 | if (sh->group) { |
714 | sh->group->stripes_cnt--; | |
715 | sh->group = NULL; | |
716 | } | |
1da177e4 | 717 | } |
7da9d450 | 718 | atomic_inc(&sh->count); |
6d183de4 | 719 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
720 | } |
721 | } while (sh == NULL); | |
722 | ||
e9e4c377 YL |
723 | if (!list_empty(conf->inactive_list + hash)) |
724 | wake_up(&conf->wait_for_stripe[hash]); | |
725 | ||
566c09c5 | 726 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
727 | return sh; |
728 | } | |
729 | ||
7a87f434 | 730 | static bool is_full_stripe_write(struct stripe_head *sh) |
731 | { | |
732 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
733 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
734 | } | |
735 | ||
59fc630b | 736 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
737 | { | |
738 | local_irq_disable(); | |
739 | if (sh1 > sh2) { | |
740 | spin_lock(&sh2->stripe_lock); | |
741 | spin_lock_nested(&sh1->stripe_lock, 1); | |
742 | } else { | |
743 | spin_lock(&sh1->stripe_lock); | |
744 | spin_lock_nested(&sh2->stripe_lock, 1); | |
745 | } | |
746 | } | |
747 | ||
748 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
749 | { | |
750 | spin_unlock(&sh1->stripe_lock); | |
751 | spin_unlock(&sh2->stripe_lock); | |
752 | local_irq_enable(); | |
753 | } | |
754 | ||
755 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
756 | static bool stripe_can_batch(struct stripe_head *sh) | |
757 | { | |
758 | return test_bit(STRIPE_BATCH_READY, &sh->state) && | |
d0852df5 | 759 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 760 | is_full_stripe_write(sh); |
761 | } | |
762 | ||
763 | /* we only do back search */ | |
764 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
765 | { | |
766 | struct stripe_head *head; | |
767 | sector_t head_sector, tmp_sec; | |
768 | int hash; | |
769 | int dd_idx; | |
770 | ||
771 | if (!stripe_can_batch(sh)) | |
772 | return; | |
773 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ | |
774 | tmp_sec = sh->sector; | |
775 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
776 | return; | |
777 | head_sector = sh->sector - STRIPE_SECTORS; | |
778 | ||
779 | hash = stripe_hash_locks_hash(head_sector); | |
780 | spin_lock_irq(conf->hash_locks + hash); | |
781 | head = __find_stripe(conf, head_sector, conf->generation); | |
782 | if (head && !atomic_inc_not_zero(&head->count)) { | |
783 | spin_lock(&conf->device_lock); | |
784 | if (!atomic_read(&head->count)) { | |
785 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
786 | atomic_inc(&conf->active_stripes); | |
787 | BUG_ON(list_empty(&head->lru) && | |
788 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
789 | list_del_init(&head->lru); | |
790 | if (head->group) { | |
791 | head->group->stripes_cnt--; | |
792 | head->group = NULL; | |
793 | } | |
794 | } | |
795 | atomic_inc(&head->count); | |
796 | spin_unlock(&conf->device_lock); | |
797 | } | |
798 | spin_unlock_irq(conf->hash_locks + hash); | |
799 | ||
800 | if (!head) | |
801 | return; | |
802 | if (!stripe_can_batch(head)) | |
803 | goto out; | |
804 | ||
805 | lock_two_stripes(head, sh); | |
806 | /* clear_batch_ready clear the flag */ | |
807 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
808 | goto unlock_out; | |
809 | ||
810 | if (sh->batch_head) | |
811 | goto unlock_out; | |
812 | ||
813 | dd_idx = 0; | |
814 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
815 | dd_idx++; | |
816 | if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw) | |
817 | goto unlock_out; | |
818 | ||
819 | if (head->batch_head) { | |
820 | spin_lock(&head->batch_head->batch_lock); | |
821 | /* This batch list is already running */ | |
822 | if (!stripe_can_batch(head)) { | |
823 | spin_unlock(&head->batch_head->batch_lock); | |
824 | goto unlock_out; | |
825 | } | |
826 | ||
827 | /* | |
828 | * at this point, head's BATCH_READY could be cleared, but we | |
829 | * can still add the stripe to batch list | |
830 | */ | |
831 | list_add(&sh->batch_list, &head->batch_list); | |
832 | spin_unlock(&head->batch_head->batch_lock); | |
833 | ||
834 | sh->batch_head = head->batch_head; | |
835 | } else { | |
836 | head->batch_head = head; | |
837 | sh->batch_head = head->batch_head; | |
838 | spin_lock(&head->batch_lock); | |
839 | list_add_tail(&sh->batch_list, &head->batch_list); | |
840 | spin_unlock(&head->batch_lock); | |
841 | } | |
842 | ||
843 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
844 | if (atomic_dec_return(&conf->preread_active_stripes) | |
845 | < IO_THRESHOLD) | |
846 | md_wakeup_thread(conf->mddev->thread); | |
847 | ||
2b6b2457 N |
848 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
849 | int seq = sh->bm_seq; | |
850 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
851 | sh->batch_head->bm_seq > seq) | |
852 | seq = sh->batch_head->bm_seq; | |
853 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
854 | sh->batch_head->bm_seq = seq; | |
855 | } | |
856 | ||
59fc630b | 857 | atomic_inc(&sh->count); |
858 | unlock_out: | |
859 | unlock_two_stripes(head, sh); | |
860 | out: | |
861 | release_stripe(head); | |
862 | } | |
863 | ||
05616be5 N |
864 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
865 | * in this stripe_head. | |
866 | */ | |
867 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
868 | { | |
869 | sector_t progress = conf->reshape_progress; | |
870 | /* Need a memory barrier to make sure we see the value | |
871 | * of conf->generation, or ->data_offset that was set before | |
872 | * reshape_progress was updated. | |
873 | */ | |
874 | smp_rmb(); | |
875 | if (progress == MaxSector) | |
876 | return 0; | |
877 | if (sh->generation == conf->generation - 1) | |
878 | return 0; | |
879 | /* We are in a reshape, and this is a new-generation stripe, | |
880 | * so use new_data_offset. | |
881 | */ | |
882 | return 1; | |
883 | } | |
884 | ||
6712ecf8 | 885 | static void |
4246a0b6 | 886 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 887 | static void |
4246a0b6 | 888 | raid5_end_write_request(struct bio *bi); |
91c00924 | 889 | |
c4e5ac0a | 890 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 891 | { |
d1688a6d | 892 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 893 | int i, disks = sh->disks; |
59fc630b | 894 | struct stripe_head *head_sh = sh; |
91c00924 DW |
895 | |
896 | might_sleep(); | |
897 | ||
898 | for (i = disks; i--; ) { | |
899 | int rw; | |
9a3e1101 | 900 | int replace_only = 0; |
977df362 N |
901 | struct bio *bi, *rbi; |
902 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 903 | |
904 | sh = head_sh; | |
e9c7469b TH |
905 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
906 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
907 | rw = WRITE_FUA; | |
908 | else | |
909 | rw = WRITE; | |
9e444768 | 910 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
620125f2 | 911 | rw |= REQ_DISCARD; |
e9c7469b | 912 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
91c00924 | 913 | rw = READ; |
9a3e1101 N |
914 | else if (test_and_clear_bit(R5_WantReplace, |
915 | &sh->dev[i].flags)) { | |
916 | rw = WRITE; | |
917 | replace_only = 1; | |
918 | } else | |
91c00924 | 919 | continue; |
bc0934f0 SL |
920 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
921 | rw |= REQ_SYNC; | |
91c00924 | 922 | |
59fc630b | 923 | again: |
91c00924 | 924 | bi = &sh->dev[i].req; |
977df362 | 925 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 926 | |
91c00924 | 927 | rcu_read_lock(); |
9a3e1101 | 928 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
929 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
930 | rdev = rcu_dereference(conf->disks[i].rdev); | |
931 | if (!rdev) { | |
932 | rdev = rrdev; | |
933 | rrdev = NULL; | |
934 | } | |
9a3e1101 N |
935 | if (rw & WRITE) { |
936 | if (replace_only) | |
937 | rdev = NULL; | |
dd054fce N |
938 | if (rdev == rrdev) |
939 | /* We raced and saw duplicates */ | |
940 | rrdev = NULL; | |
9a3e1101 | 941 | } else { |
59fc630b | 942 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
943 | rdev = rrdev; |
944 | rrdev = NULL; | |
945 | } | |
977df362 | 946 | |
91c00924 DW |
947 | if (rdev && test_bit(Faulty, &rdev->flags)) |
948 | rdev = NULL; | |
949 | if (rdev) | |
950 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
951 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
952 | rrdev = NULL; | |
953 | if (rrdev) | |
954 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
955 | rcu_read_unlock(); |
956 | ||
73e92e51 | 957 | /* We have already checked bad blocks for reads. Now |
977df362 N |
958 | * need to check for writes. We never accept write errors |
959 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
960 | */ |
961 | while ((rw & WRITE) && rdev && | |
962 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
963 | sector_t first_bad; | |
964 | int bad_sectors; | |
965 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
966 | &first_bad, &bad_sectors); | |
967 | if (!bad) | |
968 | break; | |
969 | ||
970 | if (bad < 0) { | |
971 | set_bit(BlockedBadBlocks, &rdev->flags); | |
972 | if (!conf->mddev->external && | |
973 | conf->mddev->flags) { | |
974 | /* It is very unlikely, but we might | |
975 | * still need to write out the | |
976 | * bad block log - better give it | |
977 | * a chance*/ | |
978 | md_check_recovery(conf->mddev); | |
979 | } | |
1850753d | 980 | /* |
981 | * Because md_wait_for_blocked_rdev | |
982 | * will dec nr_pending, we must | |
983 | * increment it first. | |
984 | */ | |
985 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
986 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
987 | } else { | |
988 | /* Acknowledged bad block - skip the write */ | |
989 | rdev_dec_pending(rdev, conf->mddev); | |
990 | rdev = NULL; | |
991 | } | |
992 | } | |
993 | ||
91c00924 | 994 | if (rdev) { |
9a3e1101 N |
995 | if (s->syncing || s->expanding || s->expanded |
996 | || s->replacing) | |
91c00924 DW |
997 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
998 | ||
2b7497f0 DW |
999 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1000 | ||
2f6db2a7 | 1001 | bio_reset(bi); |
91c00924 | 1002 | bi->bi_bdev = rdev->bdev; |
2f6db2a7 KO |
1003 | bi->bi_rw = rw; |
1004 | bi->bi_end_io = (rw & WRITE) | |
1005 | ? raid5_end_write_request | |
1006 | : raid5_end_read_request; | |
1007 | bi->bi_private = sh; | |
1008 | ||
91c00924 | 1009 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", |
e46b272b | 1010 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
1011 | bi->bi_rw, i); |
1012 | atomic_inc(&sh->count); | |
59fc630b | 1013 | if (sh != head_sh) |
1014 | atomic_inc(&head_sh->count); | |
05616be5 | 1015 | if (use_new_offset(conf, sh)) |
4f024f37 | 1016 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1017 | + rdev->new_data_offset); |
1018 | else | |
4f024f37 | 1019 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1020 | + rdev->data_offset); |
59fc630b | 1021 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
e59aa23f | 1022 | bi->bi_rw |= REQ_NOMERGE; |
3f9e7c14 | 1023 | |
d592a996 SL |
1024 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1025 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1026 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1027 | bi->bi_vcnt = 1; |
91c00924 DW |
1028 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1029 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1030 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1031 | /* |
1032 | * If this is discard request, set bi_vcnt 0. We don't | |
1033 | * want to confuse SCSI because SCSI will replace payload | |
1034 | */ | |
1035 | if (rw & REQ_DISCARD) | |
1036 | bi->bi_vcnt = 0; | |
977df362 N |
1037 | if (rrdev) |
1038 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1039 | |
1040 | if (conf->mddev->gendisk) | |
1041 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1042 | bi, disk_devt(conf->mddev->gendisk), | |
1043 | sh->dev[i].sector); | |
91c00924 | 1044 | generic_make_request(bi); |
977df362 N |
1045 | } |
1046 | if (rrdev) { | |
9a3e1101 N |
1047 | if (s->syncing || s->expanding || s->expanded |
1048 | || s->replacing) | |
977df362 N |
1049 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1050 | ||
1051 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1052 | ||
2f6db2a7 | 1053 | bio_reset(rbi); |
977df362 | 1054 | rbi->bi_bdev = rrdev->bdev; |
2f6db2a7 KO |
1055 | rbi->bi_rw = rw; |
1056 | BUG_ON(!(rw & WRITE)); | |
1057 | rbi->bi_end_io = raid5_end_write_request; | |
1058 | rbi->bi_private = sh; | |
1059 | ||
977df362 N |
1060 | pr_debug("%s: for %llu schedule op %ld on " |
1061 | "replacement disc %d\n", | |
1062 | __func__, (unsigned long long)sh->sector, | |
1063 | rbi->bi_rw, i); | |
1064 | atomic_inc(&sh->count); | |
59fc630b | 1065 | if (sh != head_sh) |
1066 | atomic_inc(&head_sh->count); | |
05616be5 | 1067 | if (use_new_offset(conf, sh)) |
4f024f37 | 1068 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1069 | + rrdev->new_data_offset); |
1070 | else | |
4f024f37 | 1071 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1072 | + rrdev->data_offset); |
d592a996 SL |
1073 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1074 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1075 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1076 | rbi->bi_vcnt = 1; |
977df362 N |
1077 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1078 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1079 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1080 | /* |
1081 | * If this is discard request, set bi_vcnt 0. We don't | |
1082 | * want to confuse SCSI because SCSI will replace payload | |
1083 | */ | |
1084 | if (rw & REQ_DISCARD) | |
1085 | rbi->bi_vcnt = 0; | |
e3620a3a JB |
1086 | if (conf->mddev->gendisk) |
1087 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1088 | rbi, disk_devt(conf->mddev->gendisk), | |
1089 | sh->dev[i].sector); | |
977df362 N |
1090 | generic_make_request(rbi); |
1091 | } | |
1092 | if (!rdev && !rrdev) { | |
b062962e | 1093 | if (rw & WRITE) |
91c00924 DW |
1094 | set_bit(STRIPE_DEGRADED, &sh->state); |
1095 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
1096 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
1097 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1098 | set_bit(STRIPE_HANDLE, &sh->state); | |
1099 | } | |
59fc630b | 1100 | |
1101 | if (!head_sh->batch_head) | |
1102 | continue; | |
1103 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1104 | batch_list); | |
1105 | if (sh != head_sh) | |
1106 | goto again; | |
91c00924 DW |
1107 | } |
1108 | } | |
1109 | ||
1110 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1111 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1112 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1113 | struct stripe_head *sh) | |
91c00924 | 1114 | { |
7988613b KO |
1115 | struct bio_vec bvl; |
1116 | struct bvec_iter iter; | |
91c00924 | 1117 | struct page *bio_page; |
91c00924 | 1118 | int page_offset; |
a08abd8c | 1119 | struct async_submit_ctl submit; |
0403e382 | 1120 | enum async_tx_flags flags = 0; |
91c00924 | 1121 | |
4f024f37 KO |
1122 | if (bio->bi_iter.bi_sector >= sector) |
1123 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1124 | else |
4f024f37 | 1125 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1126 | |
0403e382 DW |
1127 | if (frombio) |
1128 | flags |= ASYNC_TX_FENCE; | |
1129 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1130 | ||
7988613b KO |
1131 | bio_for_each_segment(bvl, bio, iter) { |
1132 | int len = bvl.bv_len; | |
91c00924 DW |
1133 | int clen; |
1134 | int b_offset = 0; | |
1135 | ||
1136 | if (page_offset < 0) { | |
1137 | b_offset = -page_offset; | |
1138 | page_offset += b_offset; | |
1139 | len -= b_offset; | |
1140 | } | |
1141 | ||
1142 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1143 | clen = STRIPE_SIZE - page_offset; | |
1144 | else | |
1145 | clen = len; | |
1146 | ||
1147 | if (clen > 0) { | |
7988613b KO |
1148 | b_offset += bvl.bv_offset; |
1149 | bio_page = bvl.bv_page; | |
d592a996 SL |
1150 | if (frombio) { |
1151 | if (sh->raid_conf->skip_copy && | |
1152 | b_offset == 0 && page_offset == 0 && | |
1153 | clen == STRIPE_SIZE) | |
1154 | *page = bio_page; | |
1155 | else | |
1156 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1157 | b_offset, clen, &submit); |
d592a996 SL |
1158 | } else |
1159 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1160 | page_offset, clen, &submit); |
91c00924 | 1161 | } |
a08abd8c DW |
1162 | /* chain the operations */ |
1163 | submit.depend_tx = tx; | |
1164 | ||
91c00924 DW |
1165 | if (clen < len) /* hit end of page */ |
1166 | break; | |
1167 | page_offset += len; | |
1168 | } | |
1169 | ||
1170 | return tx; | |
1171 | } | |
1172 | ||
1173 | static void ops_complete_biofill(void *stripe_head_ref) | |
1174 | { | |
1175 | struct stripe_head *sh = stripe_head_ref; | |
34a6f80e | 1176 | struct bio_list return_bi = BIO_EMPTY_LIST; |
e4d84909 | 1177 | int i; |
91c00924 | 1178 | |
e46b272b | 1179 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1180 | (unsigned long long)sh->sector); |
1181 | ||
1182 | /* clear completed biofills */ | |
1183 | for (i = sh->disks; i--; ) { | |
1184 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1185 | |
1186 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1187 | /* and check if we need to reply to a read request, |
1188 | * new R5_Wantfill requests are held off until | |
83de75cc | 1189 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1190 | */ |
1191 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1192 | struct bio *rbi, *rbi2; |
91c00924 | 1193 | |
91c00924 DW |
1194 | BUG_ON(!dev->read); |
1195 | rbi = dev->read; | |
1196 | dev->read = NULL; | |
4f024f37 | 1197 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1198 | dev->sector + STRIPE_SECTORS) { |
1199 | rbi2 = r5_next_bio(rbi, dev->sector); | |
34a6f80e N |
1200 | if (!raid5_dec_bi_active_stripes(rbi)) |
1201 | bio_list_add(&return_bi, rbi); | |
91c00924 DW |
1202 | rbi = rbi2; |
1203 | } | |
1204 | } | |
1205 | } | |
83de75cc | 1206 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1207 | |
34a6f80e | 1208 | return_io(&return_bi); |
91c00924 | 1209 | |
e4d84909 | 1210 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
1211 | release_stripe(sh); |
1212 | } | |
1213 | ||
1214 | static void ops_run_biofill(struct stripe_head *sh) | |
1215 | { | |
1216 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1217 | struct async_submit_ctl submit; |
91c00924 DW |
1218 | int i; |
1219 | ||
59fc630b | 1220 | BUG_ON(sh->batch_head); |
e46b272b | 1221 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1222 | (unsigned long long)sh->sector); |
1223 | ||
1224 | for (i = sh->disks; i--; ) { | |
1225 | struct r5dev *dev = &sh->dev[i]; | |
1226 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1227 | struct bio *rbi; | |
b17459c0 | 1228 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1229 | dev->read = rbi = dev->toread; |
1230 | dev->toread = NULL; | |
b17459c0 | 1231 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1232 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1233 | dev->sector + STRIPE_SECTORS) { |
d592a996 SL |
1234 | tx = async_copy_data(0, rbi, &dev->page, |
1235 | dev->sector, tx, sh); | |
91c00924 DW |
1236 | rbi = r5_next_bio(rbi, dev->sector); |
1237 | } | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | atomic_inc(&sh->count); | |
a08abd8c DW |
1242 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1243 | async_trigger_callback(&submit); | |
91c00924 DW |
1244 | } |
1245 | ||
4e7d2c0a | 1246 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1247 | { |
4e7d2c0a | 1248 | struct r5dev *tgt; |
91c00924 | 1249 | |
4e7d2c0a DW |
1250 | if (target < 0) |
1251 | return; | |
91c00924 | 1252 | |
4e7d2c0a | 1253 | tgt = &sh->dev[target]; |
91c00924 DW |
1254 | set_bit(R5_UPTODATE, &tgt->flags); |
1255 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1256 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1257 | } |
1258 | ||
ac6b53b6 | 1259 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1260 | { |
1261 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1262 | |
e46b272b | 1263 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1264 | (unsigned long long)sh->sector); |
1265 | ||
ac6b53b6 | 1266 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1267 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1268 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1269 | |
ecc65c9b DW |
1270 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1271 | if (sh->check_state == check_state_compute_run) | |
1272 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
1273 | set_bit(STRIPE_HANDLE, &sh->state); |
1274 | release_stripe(sh); | |
1275 | } | |
1276 | ||
d6f38f31 DW |
1277 | /* return a pointer to the address conversion region of the scribble buffer */ |
1278 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1279 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1280 | { |
46d5b785 | 1281 | void *addr; |
1282 | ||
1283 | addr = flex_array_get(percpu->scribble, i); | |
1284 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1285 | } | |
1286 | ||
1287 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1288 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1289 | { | |
1290 | void *addr; | |
1291 | ||
1292 | addr = flex_array_get(percpu->scribble, i); | |
1293 | return addr; | |
d6f38f31 DW |
1294 | } |
1295 | ||
1296 | static struct dma_async_tx_descriptor * | |
1297 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1298 | { |
91c00924 | 1299 | int disks = sh->disks; |
46d5b785 | 1300 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1301 | int target = sh->ops.target; |
1302 | struct r5dev *tgt = &sh->dev[target]; | |
1303 | struct page *xor_dest = tgt->page; | |
1304 | int count = 0; | |
1305 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1306 | struct async_submit_ctl submit; |
91c00924 DW |
1307 | int i; |
1308 | ||
59fc630b | 1309 | BUG_ON(sh->batch_head); |
1310 | ||
91c00924 | 1311 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1312 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1313 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1314 | ||
1315 | for (i = disks; i--; ) | |
1316 | if (i != target) | |
1317 | xor_srcs[count++] = sh->dev[i].page; | |
1318 | ||
1319 | atomic_inc(&sh->count); | |
1320 | ||
0403e382 | 1321 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1322 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1323 | if (unlikely(count == 1)) |
a08abd8c | 1324 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1325 | else |
a08abd8c | 1326 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1327 | |
91c00924 DW |
1328 | return tx; |
1329 | } | |
1330 | ||
ac6b53b6 DW |
1331 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1332 | * @srcs - (struct page *) array of size sh->disks | |
1333 | * @sh - stripe_head to parse | |
1334 | * | |
1335 | * Populates srcs in proper layout order for the stripe and returns the | |
1336 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1337 | * destination buffer is recorded in srcs[count] and the Q destination | |
1338 | * is recorded in srcs[count+1]]. | |
1339 | */ | |
584acdd4 MS |
1340 | static int set_syndrome_sources(struct page **srcs, |
1341 | struct stripe_head *sh, | |
1342 | int srctype) | |
ac6b53b6 DW |
1343 | { |
1344 | int disks = sh->disks; | |
1345 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1346 | int d0_idx = raid6_d0(sh); | |
1347 | int count; | |
1348 | int i; | |
1349 | ||
1350 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1351 | srcs[i] = NULL; |
ac6b53b6 DW |
1352 | |
1353 | count = 0; | |
1354 | i = d0_idx; | |
1355 | do { | |
1356 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1357 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1358 | |
584acdd4 MS |
1359 | if (i == sh->qd_idx || i == sh->pd_idx || |
1360 | (srctype == SYNDROME_SRC_ALL) || | |
1361 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1362 | test_bit(R5_Wantdrain, &dev->flags)) || | |
1363 | (srctype == SYNDROME_SRC_WRITTEN && | |
1364 | dev->written)) | |
1365 | srcs[slot] = sh->dev[i].page; | |
ac6b53b6 DW |
1366 | i = raid6_next_disk(i, disks); |
1367 | } while (i != d0_idx); | |
ac6b53b6 | 1368 | |
e4424fee | 1369 | return syndrome_disks; |
ac6b53b6 DW |
1370 | } |
1371 | ||
1372 | static struct dma_async_tx_descriptor * | |
1373 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1374 | { | |
1375 | int disks = sh->disks; | |
46d5b785 | 1376 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1377 | int target; |
1378 | int qd_idx = sh->qd_idx; | |
1379 | struct dma_async_tx_descriptor *tx; | |
1380 | struct async_submit_ctl submit; | |
1381 | struct r5dev *tgt; | |
1382 | struct page *dest; | |
1383 | int i; | |
1384 | int count; | |
1385 | ||
59fc630b | 1386 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1387 | if (sh->ops.target < 0) |
1388 | target = sh->ops.target2; | |
1389 | else if (sh->ops.target2 < 0) | |
1390 | target = sh->ops.target; | |
91c00924 | 1391 | else |
ac6b53b6 DW |
1392 | /* we should only have one valid target */ |
1393 | BUG(); | |
1394 | BUG_ON(target < 0); | |
1395 | pr_debug("%s: stripe %llu block: %d\n", | |
1396 | __func__, (unsigned long long)sh->sector, target); | |
1397 | ||
1398 | tgt = &sh->dev[target]; | |
1399 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1400 | dest = tgt->page; | |
1401 | ||
1402 | atomic_inc(&sh->count); | |
1403 | ||
1404 | if (target == qd_idx) { | |
584acdd4 | 1405 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1406 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1407 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1408 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1409 | ops_complete_compute, sh, | |
46d5b785 | 1410 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1411 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1412 | } else { | |
1413 | /* Compute any data- or p-drive using XOR */ | |
1414 | count = 0; | |
1415 | for (i = disks; i-- ; ) { | |
1416 | if (i == target || i == qd_idx) | |
1417 | continue; | |
1418 | blocks[count++] = sh->dev[i].page; | |
1419 | } | |
1420 | ||
0403e382 DW |
1421 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1422 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1423 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1424 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1425 | } | |
91c00924 | 1426 | |
91c00924 DW |
1427 | return tx; |
1428 | } | |
1429 | ||
ac6b53b6 DW |
1430 | static struct dma_async_tx_descriptor * |
1431 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1432 | { | |
1433 | int i, count, disks = sh->disks; | |
1434 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1435 | int d0_idx = raid6_d0(sh); | |
1436 | int faila = -1, failb = -1; | |
1437 | int target = sh->ops.target; | |
1438 | int target2 = sh->ops.target2; | |
1439 | struct r5dev *tgt = &sh->dev[target]; | |
1440 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1441 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1442 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1443 | struct async_submit_ctl submit; |
1444 | ||
59fc630b | 1445 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1446 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1447 | __func__, (unsigned long long)sh->sector, target, target2); | |
1448 | BUG_ON(target < 0 || target2 < 0); | |
1449 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1450 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1451 | ||
6c910a78 | 1452 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1453 | * slot number conversion for 'faila' and 'failb' |
1454 | */ | |
1455 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1456 | blocks[i] = NULL; |
ac6b53b6 DW |
1457 | count = 0; |
1458 | i = d0_idx; | |
1459 | do { | |
1460 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1461 | ||
1462 | blocks[slot] = sh->dev[i].page; | |
1463 | ||
1464 | if (i == target) | |
1465 | faila = slot; | |
1466 | if (i == target2) | |
1467 | failb = slot; | |
1468 | i = raid6_next_disk(i, disks); | |
1469 | } while (i != d0_idx); | |
ac6b53b6 DW |
1470 | |
1471 | BUG_ON(faila == failb); | |
1472 | if (failb < faila) | |
1473 | swap(faila, failb); | |
1474 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1475 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1476 | ||
1477 | atomic_inc(&sh->count); | |
1478 | ||
1479 | if (failb == syndrome_disks+1) { | |
1480 | /* Q disk is one of the missing disks */ | |
1481 | if (faila == syndrome_disks) { | |
1482 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1483 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1484 | ops_complete_compute, sh, | |
46d5b785 | 1485 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1486 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1487 | STRIPE_SIZE, &submit); |
1488 | } else { | |
1489 | struct page *dest; | |
1490 | int data_target; | |
1491 | int qd_idx = sh->qd_idx; | |
1492 | ||
1493 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1494 | if (target == qd_idx) | |
1495 | data_target = target2; | |
1496 | else | |
1497 | data_target = target; | |
1498 | ||
1499 | count = 0; | |
1500 | for (i = disks; i-- ; ) { | |
1501 | if (i == data_target || i == qd_idx) | |
1502 | continue; | |
1503 | blocks[count++] = sh->dev[i].page; | |
1504 | } | |
1505 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1506 | init_async_submit(&submit, |
1507 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1508 | NULL, NULL, NULL, | |
46d5b785 | 1509 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1510 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1511 | &submit); | |
1512 | ||
584acdd4 | 1513 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1514 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1515 | ops_complete_compute, sh, | |
46d5b785 | 1516 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1517 | return async_gen_syndrome(blocks, 0, count+2, |
1518 | STRIPE_SIZE, &submit); | |
1519 | } | |
ac6b53b6 | 1520 | } else { |
6c910a78 DW |
1521 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1522 | ops_complete_compute, sh, | |
46d5b785 | 1523 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1524 | if (failb == syndrome_disks) { |
1525 | /* We're missing D+P. */ | |
1526 | return async_raid6_datap_recov(syndrome_disks+2, | |
1527 | STRIPE_SIZE, faila, | |
1528 | blocks, &submit); | |
1529 | } else { | |
1530 | /* We're missing D+D. */ | |
1531 | return async_raid6_2data_recov(syndrome_disks+2, | |
1532 | STRIPE_SIZE, faila, failb, | |
1533 | blocks, &submit); | |
1534 | } | |
ac6b53b6 DW |
1535 | } |
1536 | } | |
1537 | ||
91c00924 DW |
1538 | static void ops_complete_prexor(void *stripe_head_ref) |
1539 | { | |
1540 | struct stripe_head *sh = stripe_head_ref; | |
1541 | ||
e46b272b | 1542 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1543 | (unsigned long long)sh->sector); |
91c00924 DW |
1544 | } |
1545 | ||
1546 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1547 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1548 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1549 | { |
91c00924 | 1550 | int disks = sh->disks; |
46d5b785 | 1551 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1552 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1553 | struct async_submit_ctl submit; |
91c00924 DW |
1554 | |
1555 | /* existing parity data subtracted */ | |
1556 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1557 | ||
59fc630b | 1558 | BUG_ON(sh->batch_head); |
e46b272b | 1559 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1560 | (unsigned long long)sh->sector); |
1561 | ||
1562 | for (i = disks; i--; ) { | |
1563 | struct r5dev *dev = &sh->dev[i]; | |
1564 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1565 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1566 | xor_srcs[count++] = dev->page; |
1567 | } | |
1568 | ||
0403e382 | 1569 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1570 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1571 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1572 | |
1573 | return tx; | |
1574 | } | |
1575 | ||
584acdd4 MS |
1576 | static struct dma_async_tx_descriptor * |
1577 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1578 | struct dma_async_tx_descriptor *tx) | |
1579 | { | |
1580 | struct page **blocks = to_addr_page(percpu, 0); | |
1581 | int count; | |
1582 | struct async_submit_ctl submit; | |
1583 | ||
1584 | pr_debug("%s: stripe %llu\n", __func__, | |
1585 | (unsigned long long)sh->sector); | |
1586 | ||
1587 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1588 | ||
1589 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1590 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1591 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1592 | ||
1593 | return tx; | |
1594 | } | |
1595 | ||
91c00924 | 1596 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1597 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1598 | { |
1599 | int disks = sh->disks; | |
d8ee0728 | 1600 | int i; |
59fc630b | 1601 | struct stripe_head *head_sh = sh; |
91c00924 | 1602 | |
e46b272b | 1603 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1604 | (unsigned long long)sh->sector); |
1605 | ||
1606 | for (i = disks; i--; ) { | |
59fc630b | 1607 | struct r5dev *dev; |
91c00924 | 1608 | struct bio *chosen; |
91c00924 | 1609 | |
59fc630b | 1610 | sh = head_sh; |
1611 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1612 | struct bio *wbi; |
1613 | ||
59fc630b | 1614 | again: |
1615 | dev = &sh->dev[i]; | |
b17459c0 | 1616 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1617 | chosen = dev->towrite; |
1618 | dev->towrite = NULL; | |
7a87f434 | 1619 | sh->overwrite_disks = 0; |
91c00924 DW |
1620 | BUG_ON(dev->written); |
1621 | wbi = dev->written = chosen; | |
b17459c0 | 1622 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1623 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1624 | |
4f024f37 | 1625 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1626 | dev->sector + STRIPE_SECTORS) { |
e9c7469b TH |
1627 | if (wbi->bi_rw & REQ_FUA) |
1628 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1629 | if (wbi->bi_rw & REQ_SYNC) |
1630 | set_bit(R5_SyncIO, &dev->flags); | |
9e444768 | 1631 | if (wbi->bi_rw & REQ_DISCARD) |
620125f2 | 1632 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1633 | else { |
1634 | tx = async_copy_data(1, wbi, &dev->page, | |
1635 | dev->sector, tx, sh); | |
1636 | if (dev->page != dev->orig_page) { | |
1637 | set_bit(R5_SkipCopy, &dev->flags); | |
1638 | clear_bit(R5_UPTODATE, &dev->flags); | |
1639 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1640 | } | |
1641 | } | |
91c00924 DW |
1642 | wbi = r5_next_bio(wbi, dev->sector); |
1643 | } | |
59fc630b | 1644 | |
1645 | if (head_sh->batch_head) { | |
1646 | sh = list_first_entry(&sh->batch_list, | |
1647 | struct stripe_head, | |
1648 | batch_list); | |
1649 | if (sh == head_sh) | |
1650 | continue; | |
1651 | goto again; | |
1652 | } | |
91c00924 DW |
1653 | } |
1654 | } | |
1655 | ||
1656 | return tx; | |
1657 | } | |
1658 | ||
ac6b53b6 | 1659 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1660 | { |
1661 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1662 | int disks = sh->disks; |
1663 | int pd_idx = sh->pd_idx; | |
1664 | int qd_idx = sh->qd_idx; | |
1665 | int i; | |
9e444768 | 1666 | bool fua = false, sync = false, discard = false; |
91c00924 | 1667 | |
e46b272b | 1668 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1669 | (unsigned long long)sh->sector); |
1670 | ||
bc0934f0 | 1671 | for (i = disks; i--; ) { |
e9c7469b | 1672 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1673 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1674 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1675 | } |
e9c7469b | 1676 | |
91c00924 DW |
1677 | for (i = disks; i--; ) { |
1678 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1679 | |
e9c7469b | 1680 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1681 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1682 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1683 | if (fua) |
1684 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1685 | if (sync) |
1686 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1687 | } |
91c00924 DW |
1688 | } |
1689 | ||
d8ee0728 DW |
1690 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1691 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1692 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1693 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1694 | else { | |
1695 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1696 | sh->reconstruct_state = reconstruct_state_result; | |
1697 | } | |
91c00924 DW |
1698 | |
1699 | set_bit(STRIPE_HANDLE, &sh->state); | |
1700 | release_stripe(sh); | |
1701 | } | |
1702 | ||
1703 | static void | |
ac6b53b6 DW |
1704 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1705 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1706 | { |
91c00924 | 1707 | int disks = sh->disks; |
59fc630b | 1708 | struct page **xor_srcs; |
a08abd8c | 1709 | struct async_submit_ctl submit; |
59fc630b | 1710 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1711 | struct page *xor_dest; |
d8ee0728 | 1712 | int prexor = 0; |
91c00924 | 1713 | unsigned long flags; |
59fc630b | 1714 | int j = 0; |
1715 | struct stripe_head *head_sh = sh; | |
1716 | int last_stripe; | |
91c00924 | 1717 | |
e46b272b | 1718 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1719 | (unsigned long long)sh->sector); |
1720 | ||
620125f2 SL |
1721 | for (i = 0; i < sh->disks; i++) { |
1722 | if (pd_idx == i) | |
1723 | continue; | |
1724 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1725 | break; | |
1726 | } | |
1727 | if (i >= sh->disks) { | |
1728 | atomic_inc(&sh->count); | |
620125f2 SL |
1729 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1730 | ops_complete_reconstruct(sh); | |
1731 | return; | |
1732 | } | |
59fc630b | 1733 | again: |
1734 | count = 0; | |
1735 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1736 | /* check if prexor is active which means only process blocks |
1737 | * that are part of a read-modify-write (written) | |
1738 | */ | |
59fc630b | 1739 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1740 | prexor = 1; |
91c00924 DW |
1741 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1742 | for (i = disks; i--; ) { | |
1743 | struct r5dev *dev = &sh->dev[i]; | |
59fc630b | 1744 | if (head_sh->dev[i].written) |
91c00924 DW |
1745 | xor_srcs[count++] = dev->page; |
1746 | } | |
1747 | } else { | |
1748 | xor_dest = sh->dev[pd_idx].page; | |
1749 | for (i = disks; i--; ) { | |
1750 | struct r5dev *dev = &sh->dev[i]; | |
1751 | if (i != pd_idx) | |
1752 | xor_srcs[count++] = dev->page; | |
1753 | } | |
1754 | } | |
1755 | ||
91c00924 DW |
1756 | /* 1/ if we prexor'd then the dest is reused as a source |
1757 | * 2/ if we did not prexor then we are redoing the parity | |
1758 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1759 | * for the synchronous xor case | |
1760 | */ | |
59fc630b | 1761 | last_stripe = !head_sh->batch_head || |
1762 | list_first_entry(&sh->batch_list, | |
1763 | struct stripe_head, batch_list) == head_sh; | |
1764 | if (last_stripe) { | |
1765 | flags = ASYNC_TX_ACK | | |
1766 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1767 | ||
1768 | atomic_inc(&head_sh->count); | |
1769 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1770 | to_addr_conv(sh, percpu, j)); | |
1771 | } else { | |
1772 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1773 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1774 | to_addr_conv(sh, percpu, j)); | |
1775 | } | |
91c00924 | 1776 | |
a08abd8c DW |
1777 | if (unlikely(count == 1)) |
1778 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1779 | else | |
1780 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1781 | if (!last_stripe) { |
1782 | j++; | |
1783 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1784 | batch_list); | |
1785 | goto again; | |
1786 | } | |
91c00924 DW |
1787 | } |
1788 | ||
ac6b53b6 DW |
1789 | static void |
1790 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1791 | struct dma_async_tx_descriptor *tx) | |
1792 | { | |
1793 | struct async_submit_ctl submit; | |
59fc630b | 1794 | struct page **blocks; |
1795 | int count, i, j = 0; | |
1796 | struct stripe_head *head_sh = sh; | |
1797 | int last_stripe; | |
584acdd4 MS |
1798 | int synflags; |
1799 | unsigned long txflags; | |
ac6b53b6 DW |
1800 | |
1801 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1802 | ||
620125f2 SL |
1803 | for (i = 0; i < sh->disks; i++) { |
1804 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1805 | continue; | |
1806 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1807 | break; | |
1808 | } | |
1809 | if (i >= sh->disks) { | |
1810 | atomic_inc(&sh->count); | |
620125f2 SL |
1811 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1812 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1813 | ops_complete_reconstruct(sh); | |
1814 | return; | |
1815 | } | |
1816 | ||
59fc630b | 1817 | again: |
1818 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1819 | |
1820 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1821 | synflags = SYNDROME_SRC_WRITTEN; | |
1822 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1823 | } else { | |
1824 | synflags = SYNDROME_SRC_ALL; | |
1825 | txflags = ASYNC_TX_ACK; | |
1826 | } | |
1827 | ||
1828 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1829 | last_stripe = !head_sh->batch_head || |
1830 | list_first_entry(&sh->batch_list, | |
1831 | struct stripe_head, batch_list) == head_sh; | |
1832 | ||
1833 | if (last_stripe) { | |
1834 | atomic_inc(&head_sh->count); | |
584acdd4 | 1835 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1836 | head_sh, to_addr_conv(sh, percpu, j)); |
1837 | } else | |
1838 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1839 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1840 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1841 | if (!last_stripe) { |
1842 | j++; | |
1843 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1844 | batch_list); | |
1845 | goto again; | |
1846 | } | |
91c00924 DW |
1847 | } |
1848 | ||
1849 | static void ops_complete_check(void *stripe_head_ref) | |
1850 | { | |
1851 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1852 | |
e46b272b | 1853 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1854 | (unsigned long long)sh->sector); |
1855 | ||
ecc65c9b | 1856 | sh->check_state = check_state_check_result; |
91c00924 DW |
1857 | set_bit(STRIPE_HANDLE, &sh->state); |
1858 | release_stripe(sh); | |
1859 | } | |
1860 | ||
ac6b53b6 | 1861 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1862 | { |
91c00924 | 1863 | int disks = sh->disks; |
ac6b53b6 DW |
1864 | int pd_idx = sh->pd_idx; |
1865 | int qd_idx = sh->qd_idx; | |
1866 | struct page *xor_dest; | |
46d5b785 | 1867 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1868 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1869 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1870 | int count; |
1871 | int i; | |
91c00924 | 1872 | |
e46b272b | 1873 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1874 | (unsigned long long)sh->sector); |
1875 | ||
59fc630b | 1876 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1877 | count = 0; |
1878 | xor_dest = sh->dev[pd_idx].page; | |
1879 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1880 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1881 | if (i == pd_idx || i == qd_idx) |
1882 | continue; | |
1883 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1884 | } |
1885 | ||
d6f38f31 | 1886 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1887 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1888 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1889 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1890 | |
91c00924 | 1891 | atomic_inc(&sh->count); |
a08abd8c DW |
1892 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1893 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1894 | } |
1895 | ||
ac6b53b6 DW |
1896 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1897 | { | |
46d5b785 | 1898 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1899 | struct async_submit_ctl submit; |
1900 | int count; | |
1901 | ||
1902 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1903 | (unsigned long long)sh->sector, checkp); | |
1904 | ||
59fc630b | 1905 | BUG_ON(sh->batch_head); |
584acdd4 | 1906 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1907 | if (!checkp) |
1908 | srcs[count] = NULL; | |
91c00924 | 1909 | |
91c00924 | 1910 | atomic_inc(&sh->count); |
ac6b53b6 | 1911 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1912 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1913 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1914 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1915 | } |
1916 | ||
51acbcec | 1917 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1918 | { |
1919 | int overlap_clear = 0, i, disks = sh->disks; | |
1920 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1921 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1922 | int level = conf->level; |
d6f38f31 DW |
1923 | struct raid5_percpu *percpu; |
1924 | unsigned long cpu; | |
91c00924 | 1925 | |
d6f38f31 DW |
1926 | cpu = get_cpu(); |
1927 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1928 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1929 | ops_run_biofill(sh); |
1930 | overlap_clear++; | |
1931 | } | |
1932 | ||
7b3a871e | 1933 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1934 | if (level < 6) |
1935 | tx = ops_run_compute5(sh, percpu); | |
1936 | else { | |
1937 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1938 | tx = ops_run_compute6_1(sh, percpu); | |
1939 | else | |
1940 | tx = ops_run_compute6_2(sh, percpu); | |
1941 | } | |
1942 | /* terminate the chain if reconstruct is not set to be run */ | |
1943 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1944 | async_tx_ack(tx); |
1945 | } | |
91c00924 | 1946 | |
584acdd4 MS |
1947 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
1948 | if (level < 6) | |
1949 | tx = ops_run_prexor5(sh, percpu, tx); | |
1950 | else | |
1951 | tx = ops_run_prexor6(sh, percpu, tx); | |
1952 | } | |
91c00924 | 1953 | |
600aa109 | 1954 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1955 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1956 | overlap_clear++; |
1957 | } | |
1958 | ||
ac6b53b6 DW |
1959 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1960 | if (level < 6) | |
1961 | ops_run_reconstruct5(sh, percpu, tx); | |
1962 | else | |
1963 | ops_run_reconstruct6(sh, percpu, tx); | |
1964 | } | |
91c00924 | 1965 | |
ac6b53b6 DW |
1966 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1967 | if (sh->check_state == check_state_run) | |
1968 | ops_run_check_p(sh, percpu); | |
1969 | else if (sh->check_state == check_state_run_q) | |
1970 | ops_run_check_pq(sh, percpu, 0); | |
1971 | else if (sh->check_state == check_state_run_pq) | |
1972 | ops_run_check_pq(sh, percpu, 1); | |
1973 | else | |
1974 | BUG(); | |
1975 | } | |
91c00924 | 1976 | |
59fc630b | 1977 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
1978 | for (i = disks; i--; ) { |
1979 | struct r5dev *dev = &sh->dev[i]; | |
1980 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1981 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1982 | } | |
d6f38f31 | 1983 | put_cpu(); |
91c00924 DW |
1984 | } |
1985 | ||
f18c1a35 N |
1986 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp) |
1987 | { | |
1988 | struct stripe_head *sh; | |
1989 | ||
1990 | sh = kmem_cache_zalloc(sc, gfp); | |
1991 | if (sh) { | |
1992 | spin_lock_init(&sh->stripe_lock); | |
1993 | spin_lock_init(&sh->batch_lock); | |
1994 | INIT_LIST_HEAD(&sh->batch_list); | |
1995 | INIT_LIST_HEAD(&sh->lru); | |
1996 | atomic_set(&sh->count, 1); | |
1997 | } | |
1998 | return sh; | |
1999 | } | |
486f0644 | 2000 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2001 | { |
2002 | struct stripe_head *sh; | |
f18c1a35 N |
2003 | |
2004 | sh = alloc_stripe(conf->slab_cache, gfp); | |
3f294f4f N |
2005 | if (!sh) |
2006 | return 0; | |
6ce32846 | 2007 | |
3f294f4f | 2008 | sh->raid_conf = conf; |
3f294f4f | 2009 | |
a9683a79 | 2010 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2011 | shrink_buffers(sh); |
3f294f4f N |
2012 | kmem_cache_free(conf->slab_cache, sh); |
2013 | return 0; | |
2014 | } | |
486f0644 N |
2015 | sh->hash_lock_index = |
2016 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2017 | /* we just created an active stripe so... */ |
3f294f4f | 2018 | atomic_inc(&conf->active_stripes); |
59fc630b | 2019 | |
3f294f4f | 2020 | release_stripe(sh); |
486f0644 | 2021 | conf->max_nr_stripes++; |
3f294f4f N |
2022 | return 1; |
2023 | } | |
2024 | ||
d1688a6d | 2025 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2026 | { |
e18b890b | 2027 | struct kmem_cache *sc; |
5e5e3e78 | 2028 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2029 | |
f4be6b43 N |
2030 | if (conf->mddev->gendisk) |
2031 | sprintf(conf->cache_name[0], | |
2032 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2033 | else | |
2034 | sprintf(conf->cache_name[0], | |
2035 | "raid%d-%p", conf->level, conf->mddev); | |
2036 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2037 | ||
ad01c9e3 N |
2038 | conf->active_name = 0; |
2039 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2040 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2041 | 0, 0, NULL); |
1da177e4 LT |
2042 | if (!sc) |
2043 | return 1; | |
2044 | conf->slab_cache = sc; | |
ad01c9e3 | 2045 | conf->pool_size = devs; |
486f0644 N |
2046 | while (num--) |
2047 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2048 | return 1; |
486f0644 | 2049 | |
1da177e4 LT |
2050 | return 0; |
2051 | } | |
29269553 | 2052 | |
d6f38f31 DW |
2053 | /** |
2054 | * scribble_len - return the required size of the scribble region | |
2055 | * @num - total number of disks in the array | |
2056 | * | |
2057 | * The size must be enough to contain: | |
2058 | * 1/ a struct page pointer for each device in the array +2 | |
2059 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2060 | * (dma_map_page()) or page (page_address()) address. | |
2061 | * | |
2062 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2063 | * calculate over all devices (not just the data blocks), using zeros in place | |
2064 | * of the P and Q blocks. | |
2065 | */ | |
46d5b785 | 2066 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2067 | { |
46d5b785 | 2068 | struct flex_array *ret; |
d6f38f31 DW |
2069 | size_t len; |
2070 | ||
2071 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2072 | ret = flex_array_alloc(len, cnt, flags); |
2073 | if (!ret) | |
2074 | return NULL; | |
2075 | /* always prealloc all elements, so no locking is required */ | |
2076 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2077 | flex_array_free(ret); | |
2078 | return NULL; | |
2079 | } | |
2080 | return ret; | |
d6f38f31 DW |
2081 | } |
2082 | ||
738a2738 N |
2083 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2084 | { | |
2085 | unsigned long cpu; | |
2086 | int err = 0; | |
2087 | ||
2088 | mddev_suspend(conf->mddev); | |
2089 | get_online_cpus(); | |
2090 | for_each_present_cpu(cpu) { | |
2091 | struct raid5_percpu *percpu; | |
2092 | struct flex_array *scribble; | |
2093 | ||
2094 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2095 | scribble = scribble_alloc(new_disks, | |
2096 | new_sectors / STRIPE_SECTORS, | |
2097 | GFP_NOIO); | |
2098 | ||
2099 | if (scribble) { | |
2100 | flex_array_free(percpu->scribble); | |
2101 | percpu->scribble = scribble; | |
2102 | } else { | |
2103 | err = -ENOMEM; | |
2104 | break; | |
2105 | } | |
2106 | } | |
2107 | put_online_cpus(); | |
2108 | mddev_resume(conf->mddev); | |
2109 | return err; | |
2110 | } | |
2111 | ||
d1688a6d | 2112 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2113 | { |
2114 | /* Make all the stripes able to hold 'newsize' devices. | |
2115 | * New slots in each stripe get 'page' set to a new page. | |
2116 | * | |
2117 | * This happens in stages: | |
2118 | * 1/ create a new kmem_cache and allocate the required number of | |
2119 | * stripe_heads. | |
83f0d77a | 2120 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2121 | * to the new stripe_heads. This will have the side effect of |
2122 | * freezing the array as once all stripe_heads have been collected, | |
2123 | * no IO will be possible. Old stripe heads are freed once their | |
2124 | * pages have been transferred over, and the old kmem_cache is | |
2125 | * freed when all stripes are done. | |
2126 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2127 | * we simple return a failre status - no need to clean anything up. | |
2128 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2129 | * If this fails, we don't bother trying the shrink the | |
2130 | * stripe_heads down again, we just leave them as they are. | |
2131 | * As each stripe_head is processed the new one is released into | |
2132 | * active service. | |
2133 | * | |
2134 | * Once step2 is started, we cannot afford to wait for a write, | |
2135 | * so we use GFP_NOIO allocations. | |
2136 | */ | |
2137 | struct stripe_head *osh, *nsh; | |
2138 | LIST_HEAD(newstripes); | |
2139 | struct disk_info *ndisks; | |
b5470dc5 | 2140 | int err; |
e18b890b | 2141 | struct kmem_cache *sc; |
ad01c9e3 | 2142 | int i; |
566c09c5 | 2143 | int hash, cnt; |
ad01c9e3 N |
2144 | |
2145 | if (newsize <= conf->pool_size) | |
2146 | return 0; /* never bother to shrink */ | |
2147 | ||
b5470dc5 DW |
2148 | err = md_allow_write(conf->mddev); |
2149 | if (err) | |
2150 | return err; | |
2a2275d6 | 2151 | |
ad01c9e3 N |
2152 | /* Step 1 */ |
2153 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2154 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2155 | 0, 0, NULL); |
ad01c9e3 N |
2156 | if (!sc) |
2157 | return -ENOMEM; | |
2158 | ||
2d5b569b N |
2159 | /* Need to ensure auto-resizing doesn't interfere */ |
2160 | mutex_lock(&conf->cache_size_mutex); | |
2161 | ||
ad01c9e3 | 2162 | for (i = conf->max_nr_stripes; i; i--) { |
f18c1a35 | 2163 | nsh = alloc_stripe(sc, GFP_KERNEL); |
ad01c9e3 N |
2164 | if (!nsh) |
2165 | break; | |
2166 | ||
ad01c9e3 | 2167 | nsh->raid_conf = conf; |
ad01c9e3 N |
2168 | list_add(&nsh->lru, &newstripes); |
2169 | } | |
2170 | if (i) { | |
2171 | /* didn't get enough, give up */ | |
2172 | while (!list_empty(&newstripes)) { | |
2173 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2174 | list_del(&nsh->lru); | |
2175 | kmem_cache_free(sc, nsh); | |
2176 | } | |
2177 | kmem_cache_destroy(sc); | |
2d5b569b | 2178 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2179 | return -ENOMEM; |
2180 | } | |
2181 | /* Step 2 - Must use GFP_NOIO now. | |
2182 | * OK, we have enough stripes, start collecting inactive | |
2183 | * stripes and copying them over | |
2184 | */ | |
566c09c5 SL |
2185 | hash = 0; |
2186 | cnt = 0; | |
ad01c9e3 | 2187 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2188 | lock_device_hash_lock(conf, hash); |
e9e4c377 | 2189 | wait_event_exclusive_cmd(conf->wait_for_stripe[hash], |
566c09c5 SL |
2190 | !list_empty(conf->inactive_list + hash), |
2191 | unlock_device_hash_lock(conf, hash), | |
2192 | lock_device_hash_lock(conf, hash)); | |
2193 | osh = get_free_stripe(conf, hash); | |
2194 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2195 | |
d592a996 | 2196 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2197 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2198 | nsh->dev[i].orig_page = osh->dev[i].page; |
2199 | } | |
566c09c5 | 2200 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2201 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2202 | cnt++; |
2203 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2204 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2205 | hash++; | |
2206 | cnt = 0; | |
2207 | } | |
ad01c9e3 N |
2208 | } |
2209 | kmem_cache_destroy(conf->slab_cache); | |
2210 | ||
2211 | /* Step 3. | |
2212 | * At this point, we are holding all the stripes so the array | |
2213 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2214 | * conf->disks and the scribble region |
ad01c9e3 N |
2215 | */ |
2216 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2217 | if (ndisks) { | |
2218 | for (i=0; i<conf->raid_disks; i++) | |
2219 | ndisks[i] = conf->disks[i]; | |
2220 | kfree(conf->disks); | |
2221 | conf->disks = ndisks; | |
2222 | } else | |
2223 | err = -ENOMEM; | |
2224 | ||
2d5b569b | 2225 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2226 | /* Step 4, return new stripes to service */ |
2227 | while(!list_empty(&newstripes)) { | |
2228 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2229 | list_del_init(&nsh->lru); | |
d6f38f31 | 2230 | |
ad01c9e3 N |
2231 | for (i=conf->raid_disks; i < newsize; i++) |
2232 | if (nsh->dev[i].page == NULL) { | |
2233 | struct page *p = alloc_page(GFP_NOIO); | |
2234 | nsh->dev[i].page = p; | |
d592a996 | 2235 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2236 | if (!p) |
2237 | err = -ENOMEM; | |
2238 | } | |
2239 | release_stripe(nsh); | |
2240 | } | |
2241 | /* critical section pass, GFP_NOIO no longer needed */ | |
2242 | ||
2243 | conf->slab_cache = sc; | |
2244 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2245 | if (!err) |
2246 | conf->pool_size = newsize; | |
ad01c9e3 N |
2247 | return err; |
2248 | } | |
1da177e4 | 2249 | |
486f0644 | 2250 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2251 | { |
2252 | struct stripe_head *sh; | |
49895bcc | 2253 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2254 | |
566c09c5 SL |
2255 | spin_lock_irq(conf->hash_locks + hash); |
2256 | sh = get_free_stripe(conf, hash); | |
2257 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2258 | if (!sh) |
2259 | return 0; | |
78bafebd | 2260 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2261 | shrink_buffers(sh); |
3f294f4f N |
2262 | kmem_cache_free(conf->slab_cache, sh); |
2263 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2264 | conf->max_nr_stripes--; |
3f294f4f N |
2265 | return 1; |
2266 | } | |
2267 | ||
d1688a6d | 2268 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2269 | { |
486f0644 N |
2270 | while (conf->max_nr_stripes && |
2271 | drop_one_stripe(conf)) | |
2272 | ; | |
3f294f4f | 2273 | |
29fc7e3e N |
2274 | if (conf->slab_cache) |
2275 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
2276 | conf->slab_cache = NULL; |
2277 | } | |
2278 | ||
4246a0b6 | 2279 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2280 | { |
99c0fb5f | 2281 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2282 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2283 | int disks = sh->disks, i; |
d6950432 | 2284 | char b[BDEVNAME_SIZE]; |
dd054fce | 2285 | struct md_rdev *rdev = NULL; |
05616be5 | 2286 | sector_t s; |
1da177e4 LT |
2287 | |
2288 | for (i=0 ; i<disks; i++) | |
2289 | if (bi == &sh->dev[i].req) | |
2290 | break; | |
2291 | ||
4246a0b6 | 2292 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2293 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2294 | bi->bi_error); |
1da177e4 LT |
2295 | if (i == disks) { |
2296 | BUG(); | |
6712ecf8 | 2297 | return; |
1da177e4 | 2298 | } |
14a75d3e | 2299 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2300 | /* If replacement finished while this request was outstanding, |
2301 | * 'replacement' might be NULL already. | |
2302 | * In that case it moved down to 'rdev'. | |
2303 | * rdev is not removed until all requests are finished. | |
2304 | */ | |
14a75d3e | 2305 | rdev = conf->disks[i].replacement; |
dd054fce | 2306 | if (!rdev) |
14a75d3e | 2307 | rdev = conf->disks[i].rdev; |
1da177e4 | 2308 | |
05616be5 N |
2309 | if (use_new_offset(conf, sh)) |
2310 | s = sh->sector + rdev->new_data_offset; | |
2311 | else | |
2312 | s = sh->sector + rdev->data_offset; | |
4246a0b6 | 2313 | if (!bi->bi_error) { |
1da177e4 | 2314 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2315 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2316 | /* Note that this cannot happen on a |
2317 | * replacement device. We just fail those on | |
2318 | * any error | |
2319 | */ | |
8bda470e CD |
2320 | printk_ratelimited( |
2321 | KERN_INFO | |
2322 | "md/raid:%s: read error corrected" | |
2323 | " (%lu sectors at %llu on %s)\n", | |
2324 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 2325 | (unsigned long long)s, |
8bda470e | 2326 | bdevname(rdev->bdev, b)); |
ddd5115f | 2327 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2328 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2329 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2330 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2331 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2332 | ||
14a75d3e N |
2333 | if (atomic_read(&rdev->read_errors)) |
2334 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2335 | } else { |
14a75d3e | 2336 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2337 | int retry = 0; |
2e8ac303 | 2338 | int set_bad = 0; |
d6950432 | 2339 | |
1da177e4 | 2340 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2341 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
2342 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
2343 | printk_ratelimited( | |
2344 | KERN_WARNING | |
2345 | "md/raid:%s: read error on replacement device " | |
2346 | "(sector %llu on %s).\n", | |
2347 | mdname(conf->mddev), | |
05616be5 | 2348 | (unsigned long long)s, |
14a75d3e | 2349 | bdn); |
2e8ac303 | 2350 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2351 | set_bad = 1; | |
8bda470e CD |
2352 | printk_ratelimited( |
2353 | KERN_WARNING | |
2354 | "md/raid:%s: read error not correctable " | |
2355 | "(sector %llu on %s).\n", | |
2356 | mdname(conf->mddev), | |
05616be5 | 2357 | (unsigned long long)s, |
8bda470e | 2358 | bdn); |
2e8ac303 | 2359 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2360 | /* Oh, no!!! */ |
2e8ac303 | 2361 | set_bad = 1; |
8bda470e CD |
2362 | printk_ratelimited( |
2363 | KERN_WARNING | |
2364 | "md/raid:%s: read error NOT corrected!! " | |
2365 | "(sector %llu on %s).\n", | |
2366 | mdname(conf->mddev), | |
05616be5 | 2367 | (unsigned long long)s, |
8bda470e | 2368 | bdn); |
2e8ac303 | 2369 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2370 | > conf->max_nr_stripes) |
14f8d26b | 2371 | printk(KERN_WARNING |
0c55e022 | 2372 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2373 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2374 | else |
2375 | retry = 1; | |
edfa1f65 BY |
2376 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2377 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2378 | retry = 1; | |
ba22dcbf | 2379 | if (retry) |
3f9e7c14 | 2380 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2381 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2382 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2383 | } else | |
2384 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2385 | else { |
4e5314b5 N |
2386 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2387 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2388 | if (!(set_bad |
2389 | && test_bit(In_sync, &rdev->flags) | |
2390 | && rdev_set_badblocks( | |
2391 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2392 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2393 | } |
1da177e4 | 2394 | } |
14a75d3e | 2395 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
2396 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2397 | set_bit(STRIPE_HANDLE, &sh->state); | |
2398 | release_stripe(sh); | |
1da177e4 LT |
2399 | } |
2400 | ||
4246a0b6 | 2401 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2402 | { |
99c0fb5f | 2403 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2404 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2405 | int disks = sh->disks, i; |
977df362 | 2406 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2407 | sector_t first_bad; |
2408 | int bad_sectors; | |
977df362 | 2409 | int replacement = 0; |
1da177e4 | 2410 | |
977df362 N |
2411 | for (i = 0 ; i < disks; i++) { |
2412 | if (bi == &sh->dev[i].req) { | |
2413 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2414 | break; |
977df362 N |
2415 | } |
2416 | if (bi == &sh->dev[i].rreq) { | |
2417 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2418 | if (rdev) |
2419 | replacement = 1; | |
2420 | else | |
2421 | /* rdev was removed and 'replacement' | |
2422 | * replaced it. rdev is not removed | |
2423 | * until all requests are finished. | |
2424 | */ | |
2425 | rdev = conf->disks[i].rdev; | |
977df362 N |
2426 | break; |
2427 | } | |
2428 | } | |
4246a0b6 | 2429 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2430 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2431 | bi->bi_error); |
1da177e4 LT |
2432 | if (i == disks) { |
2433 | BUG(); | |
6712ecf8 | 2434 | return; |
1da177e4 LT |
2435 | } |
2436 | ||
977df362 | 2437 | if (replacement) { |
4246a0b6 | 2438 | if (bi->bi_error) |
977df362 N |
2439 | md_error(conf->mddev, rdev); |
2440 | else if (is_badblock(rdev, sh->sector, | |
2441 | STRIPE_SECTORS, | |
2442 | &first_bad, &bad_sectors)) | |
2443 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2444 | } else { | |
4246a0b6 | 2445 | if (bi->bi_error) { |
9f97e4b1 | 2446 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2447 | set_bit(WriteErrorSeen, &rdev->flags); |
2448 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2449 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2450 | set_bit(MD_RECOVERY_NEEDED, | |
2451 | &rdev->mddev->recovery); | |
977df362 N |
2452 | } else if (is_badblock(rdev, sh->sector, |
2453 | STRIPE_SECTORS, | |
c0b32972 | 2454 | &first_bad, &bad_sectors)) { |
977df362 | 2455 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2456 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2457 | /* That was a successful write so make | |
2458 | * sure it looks like we already did | |
2459 | * a re-write. | |
2460 | */ | |
2461 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2462 | } | |
977df362 N |
2463 | } |
2464 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2465 | |
4246a0b6 | 2466 | if (sh->batch_head && bi->bi_error && !replacement) |
72ac7330 | 2467 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2468 | ||
977df362 N |
2469 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2470 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2471 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 2472 | release_stripe(sh); |
59fc630b | 2473 | |
2474 | if (sh->batch_head && sh != sh->batch_head) | |
2475 | release_stripe(sh->batch_head); | |
1da177e4 LT |
2476 | } |
2477 | ||
784052ec | 2478 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
d592a996 | 2479 | |
784052ec | 2480 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2481 | { |
2482 | struct r5dev *dev = &sh->dev[i]; | |
2483 | ||
2484 | bio_init(&dev->req); | |
2485 | dev->req.bi_io_vec = &dev->vec; | |
d592a996 | 2486 | dev->req.bi_max_vecs = 1; |
1da177e4 LT |
2487 | dev->req.bi_private = sh; |
2488 | ||
977df362 N |
2489 | bio_init(&dev->rreq); |
2490 | dev->rreq.bi_io_vec = &dev->rvec; | |
d592a996 | 2491 | dev->rreq.bi_max_vecs = 1; |
977df362 | 2492 | dev->rreq.bi_private = sh; |
977df362 | 2493 | |
1da177e4 | 2494 | dev->flags = 0; |
784052ec | 2495 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
2496 | } |
2497 | ||
fd01b88c | 2498 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2499 | { |
2500 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2501 | struct r5conf *conf = mddev->private; |
908f4fbd | 2502 | unsigned long flags; |
0c55e022 | 2503 | pr_debug("raid456: error called\n"); |
1da177e4 | 2504 | |
908f4fbd N |
2505 | spin_lock_irqsave(&conf->device_lock, flags); |
2506 | clear_bit(In_sync, &rdev->flags); | |
2507 | mddev->degraded = calc_degraded(conf); | |
2508 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2509 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2510 | ||
de393cde | 2511 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
2512 | set_bit(Faulty, &rdev->flags); |
2513 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
c3cce6cd | 2514 | set_bit(MD_CHANGE_PENDING, &mddev->flags); |
6f8d0c77 N |
2515 | printk(KERN_ALERT |
2516 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
2517 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2518 | mdname(mddev), | |
2519 | bdevname(rdev->bdev, b), | |
2520 | mdname(mddev), | |
2521 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 2522 | } |
1da177e4 LT |
2523 | |
2524 | /* | |
2525 | * Input: a 'big' sector number, | |
2526 | * Output: index of the data and parity disk, and the sector # in them. | |
2527 | */ | |
d1688a6d | 2528 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
2529 | int previous, int *dd_idx, |
2530 | struct stripe_head *sh) | |
1da177e4 | 2531 | { |
6e3b96ed | 2532 | sector_t stripe, stripe2; |
35f2a591 | 2533 | sector_t chunk_number; |
1da177e4 | 2534 | unsigned int chunk_offset; |
911d4ee8 | 2535 | int pd_idx, qd_idx; |
67cc2b81 | 2536 | int ddf_layout = 0; |
1da177e4 | 2537 | sector_t new_sector; |
e183eaed N |
2538 | int algorithm = previous ? conf->prev_algo |
2539 | : conf->algorithm; | |
09c9e5fa AN |
2540 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2541 | : conf->chunk_sectors; | |
112bf897 N |
2542 | int raid_disks = previous ? conf->previous_raid_disks |
2543 | : conf->raid_disks; | |
2544 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2545 | |
2546 | /* First compute the information on this sector */ | |
2547 | ||
2548 | /* | |
2549 | * Compute the chunk number and the sector offset inside the chunk | |
2550 | */ | |
2551 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2552 | chunk_number = r_sector; | |
1da177e4 LT |
2553 | |
2554 | /* | |
2555 | * Compute the stripe number | |
2556 | */ | |
35f2a591 N |
2557 | stripe = chunk_number; |
2558 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2559 | stripe2 = stripe; |
1da177e4 LT |
2560 | /* |
2561 | * Select the parity disk based on the user selected algorithm. | |
2562 | */ | |
84789554 | 2563 | pd_idx = qd_idx = -1; |
16a53ecc N |
2564 | switch(conf->level) { |
2565 | case 4: | |
911d4ee8 | 2566 | pd_idx = data_disks; |
16a53ecc N |
2567 | break; |
2568 | case 5: | |
e183eaed | 2569 | switch (algorithm) { |
1da177e4 | 2570 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2571 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2572 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2573 | (*dd_idx)++; |
2574 | break; | |
2575 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2576 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2577 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2578 | (*dd_idx)++; |
2579 | break; | |
2580 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2581 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2582 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2583 | break; |
2584 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2585 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2586 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2587 | break; |
99c0fb5f N |
2588 | case ALGORITHM_PARITY_0: |
2589 | pd_idx = 0; | |
2590 | (*dd_idx)++; | |
2591 | break; | |
2592 | case ALGORITHM_PARITY_N: | |
2593 | pd_idx = data_disks; | |
2594 | break; | |
1da177e4 | 2595 | default: |
99c0fb5f | 2596 | BUG(); |
16a53ecc N |
2597 | } |
2598 | break; | |
2599 | case 6: | |
2600 | ||
e183eaed | 2601 | switch (algorithm) { |
16a53ecc | 2602 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2603 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2604 | qd_idx = pd_idx + 1; |
2605 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2606 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2607 | qd_idx = 0; |
2608 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2609 | (*dd_idx) += 2; /* D D P Q D */ |
2610 | break; | |
2611 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2612 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2613 | qd_idx = pd_idx + 1; |
2614 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2615 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2616 | qd_idx = 0; |
2617 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2618 | (*dd_idx) += 2; /* D D P Q D */ |
2619 | break; | |
2620 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2621 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2622 | qd_idx = (pd_idx + 1) % raid_disks; |
2623 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2624 | break; |
2625 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2626 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2627 | qd_idx = (pd_idx + 1) % raid_disks; |
2628 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2629 | break; |
99c0fb5f N |
2630 | |
2631 | case ALGORITHM_PARITY_0: | |
2632 | pd_idx = 0; | |
2633 | qd_idx = 1; | |
2634 | (*dd_idx) += 2; | |
2635 | break; | |
2636 | case ALGORITHM_PARITY_N: | |
2637 | pd_idx = data_disks; | |
2638 | qd_idx = data_disks + 1; | |
2639 | break; | |
2640 | ||
2641 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2642 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2643 | * of blocks for computing Q is different. | |
2644 | */ | |
6e3b96ed | 2645 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2646 | qd_idx = pd_idx + 1; |
2647 | if (pd_idx == raid_disks-1) { | |
2648 | (*dd_idx)++; /* Q D D D P */ | |
2649 | qd_idx = 0; | |
2650 | } else if (*dd_idx >= pd_idx) | |
2651 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2652 | ddf_layout = 1; |
99c0fb5f N |
2653 | break; |
2654 | ||
2655 | case ALGORITHM_ROTATING_N_RESTART: | |
2656 | /* Same a left_asymmetric, by first stripe is | |
2657 | * D D D P Q rather than | |
2658 | * Q D D D P | |
2659 | */ | |
6e3b96ed N |
2660 | stripe2 += 1; |
2661 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2662 | qd_idx = pd_idx + 1; |
2663 | if (pd_idx == raid_disks-1) { | |
2664 | (*dd_idx)++; /* Q D D D P */ | |
2665 | qd_idx = 0; | |
2666 | } else if (*dd_idx >= pd_idx) | |
2667 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2668 | ddf_layout = 1; |
99c0fb5f N |
2669 | break; |
2670 | ||
2671 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2672 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2673 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2674 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2675 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2676 | ddf_layout = 1; |
99c0fb5f N |
2677 | break; |
2678 | ||
2679 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2680 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2681 | pd_idx = data_disks - 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_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2688 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2689 | if (*dd_idx >= pd_idx) |
2690 | (*dd_idx)++; | |
2691 | qd_idx = raid_disks - 1; | |
2692 | break; | |
2693 | ||
2694 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2695 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2696 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2697 | qd_idx = raid_disks - 1; | |
2698 | break; | |
2699 | ||
2700 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2701 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2702 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2703 | qd_idx = raid_disks - 1; | |
2704 | break; | |
2705 | ||
2706 | case ALGORITHM_PARITY_0_6: | |
2707 | pd_idx = 0; | |
2708 | (*dd_idx)++; | |
2709 | qd_idx = raid_disks - 1; | |
2710 | break; | |
2711 | ||
16a53ecc | 2712 | default: |
99c0fb5f | 2713 | BUG(); |
16a53ecc N |
2714 | } |
2715 | break; | |
1da177e4 LT |
2716 | } |
2717 | ||
911d4ee8 N |
2718 | if (sh) { |
2719 | sh->pd_idx = pd_idx; | |
2720 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2721 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2722 | } |
1da177e4 LT |
2723 | /* |
2724 | * Finally, compute the new sector number | |
2725 | */ | |
2726 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2727 | return new_sector; | |
2728 | } | |
2729 | ||
784052ec | 2730 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2731 | { |
d1688a6d | 2732 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2733 | int raid_disks = sh->disks; |
2734 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2735 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2736 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2737 | : conf->chunk_sectors; | |
e183eaed N |
2738 | int algorithm = previous ? conf->prev_algo |
2739 | : conf->algorithm; | |
1da177e4 LT |
2740 | sector_t stripe; |
2741 | int chunk_offset; | |
35f2a591 N |
2742 | sector_t chunk_number; |
2743 | int dummy1, dd_idx = i; | |
1da177e4 | 2744 | sector_t r_sector; |
911d4ee8 | 2745 | struct stripe_head sh2; |
1da177e4 LT |
2746 | |
2747 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2748 | stripe = new_sector; | |
1da177e4 | 2749 | |
16a53ecc N |
2750 | if (i == sh->pd_idx) |
2751 | return 0; | |
2752 | switch(conf->level) { | |
2753 | case 4: break; | |
2754 | case 5: | |
e183eaed | 2755 | switch (algorithm) { |
1da177e4 LT |
2756 | case ALGORITHM_LEFT_ASYMMETRIC: |
2757 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2758 | if (i > sh->pd_idx) | |
2759 | i--; | |
2760 | break; | |
2761 | case ALGORITHM_LEFT_SYMMETRIC: | |
2762 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2763 | if (i < sh->pd_idx) | |
2764 | i += raid_disks; | |
2765 | i -= (sh->pd_idx + 1); | |
2766 | break; | |
99c0fb5f N |
2767 | case ALGORITHM_PARITY_0: |
2768 | i -= 1; | |
2769 | break; | |
2770 | case ALGORITHM_PARITY_N: | |
2771 | break; | |
1da177e4 | 2772 | default: |
99c0fb5f | 2773 | BUG(); |
16a53ecc N |
2774 | } |
2775 | break; | |
2776 | case 6: | |
d0dabf7e | 2777 | if (i == sh->qd_idx) |
16a53ecc | 2778 | return 0; /* It is the Q disk */ |
e183eaed | 2779 | switch (algorithm) { |
16a53ecc N |
2780 | case ALGORITHM_LEFT_ASYMMETRIC: |
2781 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2782 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2783 | case ALGORITHM_ROTATING_N_RESTART: | |
2784 | if (sh->pd_idx == raid_disks-1) | |
2785 | i--; /* Q D D D P */ | |
16a53ecc N |
2786 | else if (i > sh->pd_idx) |
2787 | i -= 2; /* D D P Q D */ | |
2788 | break; | |
2789 | case ALGORITHM_LEFT_SYMMETRIC: | |
2790 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2791 | if (sh->pd_idx == raid_disks-1) | |
2792 | i--; /* Q D D D P */ | |
2793 | else { | |
2794 | /* D D P Q D */ | |
2795 | if (i < sh->pd_idx) | |
2796 | i += raid_disks; | |
2797 | i -= (sh->pd_idx + 2); | |
2798 | } | |
2799 | break; | |
99c0fb5f N |
2800 | case ALGORITHM_PARITY_0: |
2801 | i -= 2; | |
2802 | break; | |
2803 | case ALGORITHM_PARITY_N: | |
2804 | break; | |
2805 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2806 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2807 | if (sh->pd_idx == 0) |
2808 | i--; /* P D D D Q */ | |
e4424fee N |
2809 | else { |
2810 | /* D D Q P D */ | |
2811 | if (i < sh->pd_idx) | |
2812 | i += raid_disks; | |
2813 | i -= (sh->pd_idx + 1); | |
2814 | } | |
99c0fb5f N |
2815 | break; |
2816 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2817 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2818 | if (i > sh->pd_idx) | |
2819 | i--; | |
2820 | break; | |
2821 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2822 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2823 | if (i < sh->pd_idx) | |
2824 | i += data_disks + 1; | |
2825 | i -= (sh->pd_idx + 1); | |
2826 | break; | |
2827 | case ALGORITHM_PARITY_0_6: | |
2828 | i -= 1; | |
2829 | break; | |
16a53ecc | 2830 | default: |
99c0fb5f | 2831 | BUG(); |
16a53ecc N |
2832 | } |
2833 | break; | |
1da177e4 LT |
2834 | } |
2835 | ||
2836 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2837 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2838 | |
112bf897 | 2839 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2840 | previous, &dummy1, &sh2); |
911d4ee8 N |
2841 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2842 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2843 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2844 | mdname(conf->mddev)); | |
1da177e4 LT |
2845 | return 0; |
2846 | } | |
2847 | return r_sector; | |
2848 | } | |
2849 | ||
600aa109 | 2850 | static void |
c0f7bddb | 2851 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2852 | int rcw, int expand) |
e33129d8 | 2853 | { |
584acdd4 | 2854 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 2855 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2856 | int level = conf->level; |
e33129d8 DW |
2857 | |
2858 | if (rcw) { | |
e33129d8 DW |
2859 | |
2860 | for (i = disks; i--; ) { | |
2861 | struct r5dev *dev = &sh->dev[i]; | |
2862 | ||
2863 | if (dev->towrite) { | |
2864 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2865 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2866 | if (!expand) |
2867 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2868 | s->locked++; |
e33129d8 DW |
2869 | } |
2870 | } | |
ce7d363a N |
2871 | /* if we are not expanding this is a proper write request, and |
2872 | * there will be bios with new data to be drained into the | |
2873 | * stripe cache | |
2874 | */ | |
2875 | if (!expand) { | |
2876 | if (!s->locked) | |
2877 | /* False alarm, nothing to do */ | |
2878 | return; | |
2879 | sh->reconstruct_state = reconstruct_state_drain_run; | |
2880 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2881 | } else | |
2882 | sh->reconstruct_state = reconstruct_state_run; | |
2883 | ||
2884 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
2885 | ||
c0f7bddb | 2886 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2887 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2888 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
2889 | } else { |
2890 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
2891 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
2892 | BUG_ON(level == 6 && |
2893 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
2894 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 2895 | |
e33129d8 DW |
2896 | for (i = disks; i--; ) { |
2897 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 2898 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
2899 | continue; |
2900 | ||
e33129d8 DW |
2901 | if (dev->towrite && |
2902 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2903 | test_bit(R5_Wantcompute, &dev->flags))) { |
2904 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2905 | set_bit(R5_LOCKED, &dev->flags); |
2906 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2907 | s->locked++; |
e33129d8 DW |
2908 | } |
2909 | } | |
ce7d363a N |
2910 | if (!s->locked) |
2911 | /* False alarm - nothing to do */ | |
2912 | return; | |
2913 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
2914 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
2915 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2916 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
2917 | } |
2918 | ||
c0f7bddb | 2919 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2920 | * are in flight |
2921 | */ | |
2922 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2923 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2924 | s->locked++; |
e33129d8 | 2925 | |
c0f7bddb YT |
2926 | if (level == 6) { |
2927 | int qd_idx = sh->qd_idx; | |
2928 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2929 | ||
2930 | set_bit(R5_LOCKED, &dev->flags); | |
2931 | clear_bit(R5_UPTODATE, &dev->flags); | |
2932 | s->locked++; | |
2933 | } | |
2934 | ||
600aa109 | 2935 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2936 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2937 | s->locked, s->ops_request); |
e33129d8 | 2938 | } |
16a53ecc | 2939 | |
1da177e4 LT |
2940 | /* |
2941 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2942 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2943 | * The bi_next chain must be in order. |
2944 | */ | |
da41ba65 | 2945 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
2946 | int forwrite, int previous) | |
1da177e4 LT |
2947 | { |
2948 | struct bio **bip; | |
d1688a6d | 2949 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2950 | int firstwrite=0; |
1da177e4 | 2951 | |
cbe47ec5 | 2952 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 2953 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
2954 | (unsigned long long)sh->sector); |
2955 | ||
b17459c0 SL |
2956 | /* |
2957 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2958 | * reference count to avoid race. The reference count should already be | |
2959 | * increased before this function is called (for example, in | |
2960 | * make_request()), so other bio sharing this stripe will not free the | |
2961 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2962 | * protect it. | |
2963 | */ | |
2964 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 2965 | /* Don't allow new IO added to stripes in batch list */ |
2966 | if (sh->batch_head) | |
2967 | goto overlap; | |
72626685 | 2968 | if (forwrite) { |
1da177e4 | 2969 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2970 | if (*bip == NULL) |
72626685 N |
2971 | firstwrite = 1; |
2972 | } else | |
1da177e4 | 2973 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
2974 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
2975 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
2976 | goto overlap; |
2977 | bip = & (*bip)->bi_next; | |
2978 | } | |
4f024f37 | 2979 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
2980 | goto overlap; |
2981 | ||
da41ba65 | 2982 | if (!forwrite || previous) |
2983 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
2984 | ||
78bafebd | 2985 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2986 | if (*bip) |
2987 | bi->bi_next = *bip; | |
2988 | *bip = bi; | |
e7836bd6 | 2989 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2990 | |
1da177e4 LT |
2991 | if (forwrite) { |
2992 | /* check if page is covered */ | |
2993 | sector_t sector = sh->dev[dd_idx].sector; | |
2994 | for (bi=sh->dev[dd_idx].towrite; | |
2995 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 2996 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 2997 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
2998 | if (bio_end_sector(bi) >= sector) |
2999 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3000 | } |
3001 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3002 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3003 | sh->overwrite_disks++; | |
1da177e4 | 3004 | } |
cbe47ec5 N |
3005 | |
3006 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3007 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3008 | (unsigned long long)sh->sector, dd_idx); |
3009 | ||
3010 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3011 | /* Cannot hold spinlock over bitmap_startwrite, |
3012 | * but must ensure this isn't added to a batch until | |
3013 | * we have added to the bitmap and set bm_seq. | |
3014 | * So set STRIPE_BITMAP_PENDING to prevent | |
3015 | * batching. | |
3016 | * If multiple add_stripe_bio() calls race here they | |
3017 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3018 | * to complete "bitmap_startwrite" gets to set | |
3019 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3020 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3021 | * any more. | |
3022 | */ | |
3023 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3024 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3025 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3026 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3027 | spin_lock_irq(&sh->stripe_lock); |
3028 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3029 | if (!sh->batch_head) { | |
3030 | sh->bm_seq = conf->seq_flush+1; | |
3031 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3032 | } | |
cbe47ec5 | 3033 | } |
d0852df5 | 3034 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3035 | |
3036 | if (stripe_can_batch(sh)) | |
3037 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3038 | return 1; |
3039 | ||
3040 | overlap: | |
3041 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3042 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3043 | return 0; |
3044 | } | |
3045 | ||
d1688a6d | 3046 | static void end_reshape(struct r5conf *conf); |
29269553 | 3047 | |
d1688a6d | 3048 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3049 | struct stripe_head *sh) |
ccfcc3c1 | 3050 | { |
784052ec | 3051 | int sectors_per_chunk = |
09c9e5fa | 3052 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3053 | int dd_idx; |
2d2063ce | 3054 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3055 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3056 | |
112bf897 N |
3057 | raid5_compute_sector(conf, |
3058 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3059 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3060 | previous, |
911d4ee8 | 3061 | &dd_idx, sh); |
ccfcc3c1 N |
3062 | } |
3063 | ||
a4456856 | 3064 | static void |
d1688a6d | 3065 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 | 3066 | struct stripe_head_state *s, int disks, |
34a6f80e | 3067 | struct bio_list *return_bi) |
a4456856 DW |
3068 | { |
3069 | int i; | |
59fc630b | 3070 | BUG_ON(sh->batch_head); |
a4456856 DW |
3071 | for (i = disks; i--; ) { |
3072 | struct bio *bi; | |
3073 | int bitmap_end = 0; | |
3074 | ||
3075 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3076 | struct md_rdev *rdev; |
a4456856 DW |
3077 | rcu_read_lock(); |
3078 | rdev = rcu_dereference(conf->disks[i].rdev); | |
3079 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
3080 | atomic_inc(&rdev->nr_pending); |
3081 | else | |
3082 | rdev = NULL; | |
a4456856 | 3083 | rcu_read_unlock(); |
7f0da59b N |
3084 | if (rdev) { |
3085 | if (!rdev_set_badblocks( | |
3086 | rdev, | |
3087 | sh->sector, | |
3088 | STRIPE_SECTORS, 0)) | |
3089 | md_error(conf->mddev, rdev); | |
3090 | rdev_dec_pending(rdev, conf->mddev); | |
3091 | } | |
a4456856 | 3092 | } |
b17459c0 | 3093 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3094 | /* fail all writes first */ |
3095 | bi = sh->dev[i].towrite; | |
3096 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3097 | sh->overwrite_disks = 0; |
b17459c0 | 3098 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3099 | if (bi) |
a4456856 | 3100 | bitmap_end = 1; |
a4456856 DW |
3101 | |
3102 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
3103 | wake_up(&conf->wait_for_overlap); | |
3104 | ||
4f024f37 | 3105 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3106 | sh->dev[i].sector + STRIPE_SECTORS) { |
3107 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3108 | |
3109 | bi->bi_error = -EIO; | |
e7836bd6 | 3110 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3111 | md_write_end(conf->mddev); |
34a6f80e | 3112 | bio_list_add(return_bi, bi); |
a4456856 DW |
3113 | } |
3114 | bi = nextbi; | |
3115 | } | |
7eaf7e8e SL |
3116 | if (bitmap_end) |
3117 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3118 | STRIPE_SECTORS, 0, 0); | |
3119 | bitmap_end = 0; | |
a4456856 DW |
3120 | /* and fail all 'written' */ |
3121 | bi = sh->dev[i].written; | |
3122 | sh->dev[i].written = NULL; | |
d592a996 SL |
3123 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3124 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3125 | sh->dev[i].page = sh->dev[i].orig_page; | |
3126 | } | |
3127 | ||
a4456856 | 3128 | if (bi) bitmap_end = 1; |
4f024f37 | 3129 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3130 | sh->dev[i].sector + STRIPE_SECTORS) { |
3131 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3132 | |
3133 | bi->bi_error = -EIO; | |
e7836bd6 | 3134 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3135 | md_write_end(conf->mddev); |
34a6f80e | 3136 | bio_list_add(return_bi, bi); |
a4456856 DW |
3137 | } |
3138 | bi = bi2; | |
3139 | } | |
3140 | ||
b5e98d65 DW |
3141 | /* fail any reads if this device is non-operational and |
3142 | * the data has not reached the cache yet. | |
3143 | */ | |
3144 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
3145 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
3146 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3147 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3148 | bi = sh->dev[i].toread; |
3149 | sh->dev[i].toread = NULL; | |
143c4d05 | 3150 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3151 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3152 | wake_up(&conf->wait_for_overlap); | |
4f024f37 | 3153 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3154 | sh->dev[i].sector + STRIPE_SECTORS) { |
3155 | struct bio *nextbi = | |
3156 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3157 | |
3158 | bi->bi_error = -EIO; | |
34a6f80e N |
3159 | if (!raid5_dec_bi_active_stripes(bi)) |
3160 | bio_list_add(return_bi, bi); | |
a4456856 DW |
3161 | bi = nextbi; |
3162 | } | |
3163 | } | |
a4456856 DW |
3164 | if (bitmap_end) |
3165 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3166 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3167 | /* If we were in the middle of a write the parity block might |
3168 | * still be locked - so just clear all R5_LOCKED flags | |
3169 | */ | |
3170 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
3171 | } |
3172 | ||
8b3e6cdc DW |
3173 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3174 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3175 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3176 | } |
3177 | ||
7f0da59b | 3178 | static void |
d1688a6d | 3179 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3180 | struct stripe_head_state *s) |
3181 | { | |
3182 | int abort = 0; | |
3183 | int i; | |
3184 | ||
59fc630b | 3185 | BUG_ON(sh->batch_head); |
7f0da59b | 3186 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3187 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3188 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3189 | s->syncing = 0; |
9a3e1101 | 3190 | s->replacing = 0; |
7f0da59b | 3191 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3192 | * Don't even need to abort as that is handled elsewhere |
3193 | * if needed, and not always wanted e.g. if there is a known | |
3194 | * bad block here. | |
9a3e1101 | 3195 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3196 | * non-sync devices, or abort the recovery |
3197 | */ | |
18b9837e N |
3198 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3199 | /* During recovery devices cannot be removed, so | |
3200 | * locking and refcounting of rdevs is not needed | |
3201 | */ | |
3202 | for (i = 0; i < conf->raid_disks; i++) { | |
3203 | struct md_rdev *rdev = conf->disks[i].rdev; | |
3204 | if (rdev | |
3205 | && !test_bit(Faulty, &rdev->flags) | |
3206 | && !test_bit(In_sync, &rdev->flags) | |
3207 | && !rdev_set_badblocks(rdev, sh->sector, | |
3208 | STRIPE_SECTORS, 0)) | |
3209 | abort = 1; | |
3210 | rdev = conf->disks[i].replacement; | |
3211 | if (rdev | |
3212 | && !test_bit(Faulty, &rdev->flags) | |
3213 | && !test_bit(In_sync, &rdev->flags) | |
3214 | && !rdev_set_badblocks(rdev, sh->sector, | |
3215 | STRIPE_SECTORS, 0)) | |
3216 | abort = 1; | |
3217 | } | |
3218 | if (abort) | |
3219 | conf->recovery_disabled = | |
3220 | conf->mddev->recovery_disabled; | |
7f0da59b | 3221 | } |
18b9837e | 3222 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3223 | } |
3224 | ||
9a3e1101 N |
3225 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3226 | { | |
3227 | struct md_rdev *rdev; | |
3228 | int rv = 0; | |
3229 | /* Doing recovery so rcu locking not required */ | |
3230 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
3231 | if (rdev | |
3232 | && !test_bit(Faulty, &rdev->flags) | |
3233 | && !test_bit(In_sync, &rdev->flags) | |
3234 | && (rdev->recovery_offset <= sh->sector | |
3235 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3236 | rv = 1; | |
3237 | ||
3238 | return rv; | |
3239 | } | |
3240 | ||
93b3dbce | 3241 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
3242 | * to be read or computed to satisfy a request. |
3243 | * | |
3244 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 3245 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 3246 | */ |
2c58f06e N |
3247 | |
3248 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3249 | int disk_idx, int disks) | |
a4456856 | 3250 | { |
5599becc | 3251 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3252 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3253 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3254 | int i; |
5599becc | 3255 | |
a79cfe12 N |
3256 | |
3257 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3258 | test_bit(R5_UPTODATE, &dev->flags)) | |
3259 | /* No point reading this as we already have it or have | |
3260 | * decided to get it. | |
3261 | */ | |
3262 | return 0; | |
3263 | ||
3264 | if (dev->toread || | |
3265 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3266 | /* We need this block to directly satisfy a request */ | |
3267 | return 1; | |
3268 | ||
3269 | if (s->syncing || s->expanding || | |
3270 | (s->replacing && want_replace(sh, disk_idx))) | |
3271 | /* When syncing, or expanding we read everything. | |
3272 | * When replacing, we need the replaced block. | |
3273 | */ | |
3274 | return 1; | |
3275 | ||
3276 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3277 | (s->failed >= 2 && fdev[1]->toread)) | |
3278 | /* If we want to read from a failed device, then | |
3279 | * we need to actually read every other device. | |
3280 | */ | |
3281 | return 1; | |
3282 | ||
a9d56950 N |
3283 | /* Sometimes neither read-modify-write nor reconstruct-write |
3284 | * cycles can work. In those cases we read every block we | |
3285 | * can. Then the parity-update is certain to have enough to | |
3286 | * work with. | |
3287 | * This can only be a problem when we need to write something, | |
3288 | * and some device has failed. If either of those tests | |
3289 | * fail we need look no further. | |
3290 | */ | |
3291 | if (!s->failed || !s->to_write) | |
3292 | return 0; | |
3293 | ||
3294 | if (test_bit(R5_Insync, &dev->flags) && | |
3295 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3296 | /* Pre-reads at not permitted until after short delay | |
3297 | * to gather multiple requests. However if this | |
3298 | * device is no Insync, the block could only be be computed | |
3299 | * and there is no need to delay that. | |
3300 | */ | |
3301 | return 0; | |
ea664c82 N |
3302 | |
3303 | for (i = 0; i < s->failed; i++) { | |
3304 | if (fdev[i]->towrite && | |
3305 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3306 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3307 | /* If we have a partial write to a failed | |
3308 | * device, then we will need to reconstruct | |
3309 | * the content of that device, so all other | |
3310 | * devices must be read. | |
3311 | */ | |
3312 | return 1; | |
3313 | } | |
3314 | ||
3315 | /* If we are forced to do a reconstruct-write, either because | |
3316 | * the current RAID6 implementation only supports that, or | |
3317 | * or because parity cannot be trusted and we are currently | |
3318 | * recovering it, there is extra need to be careful. | |
3319 | * If one of the devices that we would need to read, because | |
3320 | * it is not being overwritten (and maybe not written at all) | |
3321 | * is missing/faulty, then we need to read everything we can. | |
3322 | */ | |
3323 | if (sh->raid_conf->level != 6 && | |
3324 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3325 | /* reconstruct-write isn't being forced */ | |
3326 | return 0; | |
3327 | for (i = 0; i < s->failed; i++) { | |
10d82c5f N |
3328 | if (s->failed_num[i] != sh->pd_idx && |
3329 | s->failed_num[i] != sh->qd_idx && | |
3330 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3331 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3332 | return 1; | |
3333 | } | |
3334 | ||
2c58f06e N |
3335 | return 0; |
3336 | } | |
3337 | ||
3338 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3339 | int disk_idx, int disks) | |
3340 | { | |
3341 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3342 | ||
3343 | /* is the data in this block needed, and can we get it? */ | |
3344 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3345 | /* we would like to get this block, possibly by computing it, |
3346 | * otherwise read it if the backing disk is insync | |
3347 | */ | |
3348 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3349 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3350 | BUG_ON(sh->batch_head); |
5599becc | 3351 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3352 | (s->failed && (disk_idx == s->failed_num[0] || |
3353 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3354 | /* have disk failed, and we're requested to fetch it; |
3355 | * do compute it | |
a4456856 | 3356 | */ |
5599becc YT |
3357 | pr_debug("Computing stripe %llu block %d\n", |
3358 | (unsigned long long)sh->sector, disk_idx); | |
3359 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3360 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3361 | set_bit(R5_Wantcompute, &dev->flags); | |
3362 | sh->ops.target = disk_idx; | |
3363 | sh->ops.target2 = -1; /* no 2nd target */ | |
3364 | s->req_compute = 1; | |
93b3dbce N |
3365 | /* Careful: from this point on 'uptodate' is in the eye |
3366 | * of raid_run_ops which services 'compute' operations | |
3367 | * before writes. R5_Wantcompute flags a block that will | |
3368 | * be R5_UPTODATE by the time it is needed for a | |
3369 | * subsequent operation. | |
3370 | */ | |
5599becc YT |
3371 | s->uptodate++; |
3372 | return 1; | |
3373 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3374 | /* Computing 2-failure is *very* expensive; only | |
3375 | * do it if failed >= 2 | |
3376 | */ | |
3377 | int other; | |
3378 | for (other = disks; other--; ) { | |
3379 | if (other == disk_idx) | |
3380 | continue; | |
3381 | if (!test_bit(R5_UPTODATE, | |
3382 | &sh->dev[other].flags)) | |
3383 | break; | |
a4456856 | 3384 | } |
5599becc YT |
3385 | BUG_ON(other < 0); |
3386 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3387 | (unsigned long long)sh->sector, | |
3388 | disk_idx, other); | |
3389 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3390 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3391 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3392 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3393 | sh->ops.target = disk_idx; | |
3394 | sh->ops.target2 = other; | |
3395 | s->uptodate += 2; | |
3396 | s->req_compute = 1; | |
3397 | return 1; | |
3398 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3399 | set_bit(R5_LOCKED, &dev->flags); | |
3400 | set_bit(R5_Wantread, &dev->flags); | |
3401 | s->locked++; | |
3402 | pr_debug("Reading block %d (sync=%d)\n", | |
3403 | disk_idx, s->syncing); | |
a4456856 DW |
3404 | } |
3405 | } | |
5599becc YT |
3406 | |
3407 | return 0; | |
3408 | } | |
3409 | ||
3410 | /** | |
93b3dbce | 3411 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3412 | */ |
93b3dbce N |
3413 | static void handle_stripe_fill(struct stripe_head *sh, |
3414 | struct stripe_head_state *s, | |
3415 | int disks) | |
5599becc YT |
3416 | { |
3417 | int i; | |
3418 | ||
3419 | /* look for blocks to read/compute, skip this if a compute | |
3420 | * is already in flight, or if the stripe contents are in the | |
3421 | * midst of changing due to a write | |
3422 | */ | |
3423 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
3424 | !sh->reconstruct_state) | |
3425 | for (i = disks; i--; ) | |
93b3dbce | 3426 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3427 | break; |
a4456856 DW |
3428 | set_bit(STRIPE_HANDLE, &sh->state); |
3429 | } | |
3430 | ||
787b76fa N |
3431 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3432 | unsigned long handle_flags); | |
1fe797e6 | 3433 | /* handle_stripe_clean_event |
a4456856 DW |
3434 | * any written block on an uptodate or failed drive can be returned. |
3435 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3436 | * never LOCKED, so we don't need to test 'failed' directly. | |
3437 | */ | |
d1688a6d | 3438 | static void handle_stripe_clean_event(struct r5conf *conf, |
34a6f80e | 3439 | struct stripe_head *sh, int disks, struct bio_list *return_bi) |
a4456856 DW |
3440 | { |
3441 | int i; | |
3442 | struct r5dev *dev; | |
f8dfcffd | 3443 | int discard_pending = 0; |
59fc630b | 3444 | struct stripe_head *head_sh = sh; |
3445 | bool do_endio = false; | |
a4456856 DW |
3446 | |
3447 | for (i = disks; i--; ) | |
3448 | if (sh->dev[i].written) { | |
3449 | dev = &sh->dev[i]; | |
3450 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3451 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3452 | test_bit(R5_Discard, &dev->flags) || |
3453 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3454 | /* We can return any write requests */ |
3455 | struct bio *wbi, *wbi2; | |
45b4233c | 3456 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3457 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3458 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3459 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3460 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3461 | } |
59fc630b | 3462 | do_endio = true; |
3463 | ||
3464 | returnbi: | |
3465 | dev->page = dev->orig_page; | |
a4456856 DW |
3466 | wbi = dev->written; |
3467 | dev->written = NULL; | |
4f024f37 | 3468 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3469 | dev->sector + STRIPE_SECTORS) { |
3470 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3471 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 | 3472 | md_write_end(conf->mddev); |
34a6f80e | 3473 | bio_list_add(return_bi, wbi); |
a4456856 DW |
3474 | } |
3475 | wbi = wbi2; | |
3476 | } | |
7eaf7e8e SL |
3477 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3478 | STRIPE_SECTORS, | |
a4456856 | 3479 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3480 | 0); |
59fc630b | 3481 | if (head_sh->batch_head) { |
3482 | sh = list_first_entry(&sh->batch_list, | |
3483 | struct stripe_head, | |
3484 | batch_list); | |
3485 | if (sh != head_sh) { | |
3486 | dev = &sh->dev[i]; | |
3487 | goto returnbi; | |
3488 | } | |
3489 | } | |
3490 | sh = head_sh; | |
3491 | dev = &sh->dev[i]; | |
f8dfcffd N |
3492 | } else if (test_bit(R5_Discard, &dev->flags)) |
3493 | discard_pending = 1; | |
d592a996 SL |
3494 | WARN_ON(test_bit(R5_SkipCopy, &dev->flags)); |
3495 | WARN_ON(dev->page != dev->orig_page); | |
f8dfcffd N |
3496 | } |
3497 | if (!discard_pending && | |
3498 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
3499 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); | |
3500 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3501 | if (sh->qd_idx >= 0) { | |
3502 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3503 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3504 | } | |
3505 | /* now that discard is done we can proceed with any sync */ | |
3506 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3507 | /* |
3508 | * SCSI discard will change some bio fields and the stripe has | |
3509 | * no updated data, so remove it from hash list and the stripe | |
3510 | * will be reinitialized | |
3511 | */ | |
3512 | spin_lock_irq(&conf->device_lock); | |
59fc630b | 3513 | unhash: |
d47648fc | 3514 | remove_hash(sh); |
59fc630b | 3515 | if (head_sh->batch_head) { |
3516 | sh = list_first_entry(&sh->batch_list, | |
3517 | struct stripe_head, batch_list); | |
3518 | if (sh != head_sh) | |
3519 | goto unhash; | |
3520 | } | |
d47648fc | 3521 | spin_unlock_irq(&conf->device_lock); |
59fc630b | 3522 | sh = head_sh; |
3523 | ||
f8dfcffd N |
3524 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3525 | set_bit(STRIPE_HANDLE, &sh->state); | |
3526 | ||
3527 | } | |
8b3e6cdc DW |
3528 | |
3529 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3530 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3531 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3532 | |
787b76fa N |
3533 | if (head_sh->batch_head && do_endio) |
3534 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3535 | } |
3536 | ||
d1688a6d | 3537 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
3538 | struct stripe_head *sh, |
3539 | struct stripe_head_state *s, | |
3540 | int disks) | |
a4456856 DW |
3541 | { |
3542 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3543 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3544 | ||
584acdd4 | 3545 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3546 | * If yes, then the array is dirty (after unclean shutdown or |
3547 | * initial creation), so parity in some stripes might be inconsistent. | |
3548 | * In this case, we need to always do reconstruct-write, to ensure | |
3549 | * that in case of drive failure or read-error correction, we | |
3550 | * generate correct data from the parity. | |
3551 | */ | |
584acdd4 | 3552 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3553 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3554 | s->failed == 0)) { | |
a7854487 | 3555 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3556 | * look like rcw is cheaper |
3557 | */ | |
3558 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3559 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3560 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3561 | (unsigned long long)sh->sector); |
c8ac1803 | 3562 | } else for (i = disks; i--; ) { |
a4456856 DW |
3563 | /* would I have to read this buffer for read_modify_write */ |
3564 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3565 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3566 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3567 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3568 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
3569 | if (test_bit(R5_Insync, &dev->flags)) |
3570 | rmw++; | |
3571 | else | |
3572 | rmw += 2*disks; /* cannot read it */ | |
3573 | } | |
3574 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3575 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3576 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3577 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3578 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3579 | test_bit(R5_Wantcompute, &dev->flags))) { | |
67f45548 N |
3580 | if (test_bit(R5_Insync, &dev->flags)) |
3581 | rcw++; | |
a4456856 DW |
3582 | else |
3583 | rcw += 2*disks; | |
3584 | } | |
3585 | } | |
45b4233c | 3586 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
3587 | (unsigned long long)sh->sector, rmw, rcw); |
3588 | set_bit(STRIPE_HANDLE, &sh->state); | |
584acdd4 | 3589 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) { |
a4456856 | 3590 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3591 | if (conf->mddev->queue) |
3592 | blk_add_trace_msg(conf->mddev->queue, | |
3593 | "raid5 rmw %llu %d", | |
3594 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3595 | for (i = disks; i--; ) { |
3596 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3597 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3598 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3599 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3600 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 | 3601 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3602 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3603 | &sh->state)) { | |
3604 | pr_debug("Read_old block %d for r-m-w\n", | |
3605 | i); | |
a4456856 DW |
3606 | set_bit(R5_LOCKED, &dev->flags); |
3607 | set_bit(R5_Wantread, &dev->flags); | |
3608 | s->locked++; | |
3609 | } else { | |
3610 | set_bit(STRIPE_DELAYED, &sh->state); | |
3611 | set_bit(STRIPE_HANDLE, &sh->state); | |
3612 | } | |
3613 | } | |
3614 | } | |
a9add5d9 | 3615 | } |
584acdd4 | 3616 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) { |
a4456856 | 3617 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3618 | int qread =0; |
c8ac1803 | 3619 | rcw = 0; |
a4456856 DW |
3620 | for (i = disks; i--; ) { |
3621 | struct r5dev *dev = &sh->dev[i]; | |
3622 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3623 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3624 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3625 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3626 | test_bit(R5_Wantcompute, &dev->flags))) { |
3627 | rcw++; | |
67f45548 N |
3628 | if (test_bit(R5_Insync, &dev->flags) && |
3629 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3630 | &sh->state)) { | |
45b4233c | 3631 | pr_debug("Read_old block " |
a4456856 DW |
3632 | "%d for Reconstruct\n", i); |
3633 | set_bit(R5_LOCKED, &dev->flags); | |
3634 | set_bit(R5_Wantread, &dev->flags); | |
3635 | s->locked++; | |
a9add5d9 | 3636 | qread++; |
a4456856 DW |
3637 | } else { |
3638 | set_bit(STRIPE_DELAYED, &sh->state); | |
3639 | set_bit(STRIPE_HANDLE, &sh->state); | |
3640 | } | |
3641 | } | |
3642 | } | |
e3620a3a | 3643 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3644 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3645 | (unsigned long long)sh->sector, | |
3646 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3647 | } |
b1b02fe9 N |
3648 | |
3649 | if (rcw > disks && rmw > disks && | |
3650 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3651 | set_bit(STRIPE_DELAYED, &sh->state); | |
3652 | ||
a4456856 DW |
3653 | /* now if nothing is locked, and if we have enough data, |
3654 | * we can start a write request | |
3655 | */ | |
f38e1219 DW |
3656 | /* since handle_stripe can be called at any time we need to handle the |
3657 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3658 | * subsequent call wants to start a write request. raid_run_ops only |
3659 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3660 | * simultaneously. If this is not the case then new writes need to be |
3661 | * held off until the compute completes. | |
3662 | */ | |
976ea8d4 DW |
3663 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3664 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
3665 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 3666 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
3667 | } |
3668 | ||
d1688a6d | 3669 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3670 | struct stripe_head_state *s, int disks) |
3671 | { | |
ecc65c9b | 3672 | struct r5dev *dev = NULL; |
bd2ab670 | 3673 | |
59fc630b | 3674 | BUG_ON(sh->batch_head); |
a4456856 | 3675 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3676 | |
ecc65c9b DW |
3677 | switch (sh->check_state) { |
3678 | case check_state_idle: | |
3679 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3680 | if (s->failed == 0) { |
bd2ab670 | 3681 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3682 | sh->check_state = check_state_run; |
3683 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3684 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3685 | s->uptodate--; |
ecc65c9b | 3686 | break; |
bd2ab670 | 3687 | } |
f2b3b44d | 3688 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3689 | /* fall through */ |
3690 | case check_state_compute_result: | |
3691 | sh->check_state = check_state_idle; | |
3692 | if (!dev) | |
3693 | dev = &sh->dev[sh->pd_idx]; | |
3694 | ||
3695 | /* check that a write has not made the stripe insync */ | |
3696 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3697 | break; | |
c8894419 | 3698 | |
a4456856 | 3699 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3700 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3701 | BUG_ON(s->uptodate != disks); | |
3702 | ||
3703 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3704 | s->locked++; |
a4456856 | 3705 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3706 | |
a4456856 | 3707 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3708 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3709 | break; |
3710 | case check_state_run: | |
3711 | break; /* we will be called again upon completion */ | |
3712 | case check_state_check_result: | |
3713 | sh->check_state = check_state_idle; | |
3714 | ||
3715 | /* if a failure occurred during the check operation, leave | |
3716 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3717 | */ | |
3718 | if (s->failed) | |
3719 | break; | |
3720 | ||
3721 | /* handle a successful check operation, if parity is correct | |
3722 | * we are done. Otherwise update the mismatch count and repair | |
3723 | * parity if !MD_RECOVERY_CHECK | |
3724 | */ | |
ad283ea4 | 3725 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3726 | /* parity is correct (on disc, |
3727 | * not in buffer any more) | |
3728 | */ | |
3729 | set_bit(STRIPE_INSYNC, &sh->state); | |
3730 | else { | |
7f7583d4 | 3731 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3732 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3733 | /* don't try to repair!! */ | |
3734 | set_bit(STRIPE_INSYNC, &sh->state); | |
3735 | else { | |
3736 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3737 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3738 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3739 | set_bit(R5_Wantcompute, | |
3740 | &sh->dev[sh->pd_idx].flags); | |
3741 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3742 | sh->ops.target2 = -1; |
ecc65c9b DW |
3743 | s->uptodate++; |
3744 | } | |
3745 | } | |
3746 | break; | |
3747 | case check_state_compute_run: | |
3748 | break; | |
3749 | default: | |
3750 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3751 | __func__, sh->check_state, | |
3752 | (unsigned long long) sh->sector); | |
3753 | BUG(); | |
a4456856 DW |
3754 | } |
3755 | } | |
3756 | ||
d1688a6d | 3757 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3758 | struct stripe_head_state *s, |
f2b3b44d | 3759 | int disks) |
a4456856 | 3760 | { |
a4456856 | 3761 | int pd_idx = sh->pd_idx; |
34e04e87 | 3762 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3763 | struct r5dev *dev; |
a4456856 | 3764 | |
59fc630b | 3765 | BUG_ON(sh->batch_head); |
a4456856 DW |
3766 | set_bit(STRIPE_HANDLE, &sh->state); |
3767 | ||
3768 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3769 | |
a4456856 DW |
3770 | /* Want to check and possibly repair P and Q. |
3771 | * However there could be one 'failed' device, in which | |
3772 | * case we can only check one of them, possibly using the | |
3773 | * other to generate missing data | |
3774 | */ | |
3775 | ||
d82dfee0 DW |
3776 | switch (sh->check_state) { |
3777 | case check_state_idle: | |
3778 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 3779 | if (s->failed == s->q_failed) { |
d82dfee0 | 3780 | /* The only possible failed device holds Q, so it |
a4456856 DW |
3781 | * makes sense to check P (If anything else were failed, |
3782 | * we would have used P to recreate it). | |
3783 | */ | |
d82dfee0 | 3784 | sh->check_state = check_state_run; |
a4456856 | 3785 | } |
f2b3b44d | 3786 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 3787 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
3788 | * anything, so it makes sense to check it |
3789 | */ | |
d82dfee0 DW |
3790 | if (sh->check_state == check_state_run) |
3791 | sh->check_state = check_state_run_pq; | |
3792 | else | |
3793 | sh->check_state = check_state_run_q; | |
a4456856 | 3794 | } |
a4456856 | 3795 | |
d82dfee0 DW |
3796 | /* discard potentially stale zero_sum_result */ |
3797 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3798 | |
d82dfee0 DW |
3799 | if (sh->check_state == check_state_run) { |
3800 | /* async_xor_zero_sum destroys the contents of P */ | |
3801 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3802 | s->uptodate--; | |
a4456856 | 3803 | } |
d82dfee0 DW |
3804 | if (sh->check_state >= check_state_run && |
3805 | sh->check_state <= check_state_run_pq) { | |
3806 | /* async_syndrome_zero_sum preserves P and Q, so | |
3807 | * no need to mark them !uptodate here | |
3808 | */ | |
3809 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3810 | break; | |
a4456856 DW |
3811 | } |
3812 | ||
d82dfee0 DW |
3813 | /* we have 2-disk failure */ |
3814 | BUG_ON(s->failed != 2); | |
3815 | /* fall through */ | |
3816 | case check_state_compute_result: | |
3817 | sh->check_state = check_state_idle; | |
a4456856 | 3818 | |
d82dfee0 DW |
3819 | /* check that a write has not made the stripe insync */ |
3820 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3821 | break; | |
a4456856 DW |
3822 | |
3823 | /* now write out any block on a failed drive, | |
d82dfee0 | 3824 | * or P or Q if they were recomputed |
a4456856 | 3825 | */ |
d82dfee0 | 3826 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3827 | if (s->failed == 2) { |
f2b3b44d | 3828 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3829 | s->locked++; |
3830 | set_bit(R5_LOCKED, &dev->flags); | |
3831 | set_bit(R5_Wantwrite, &dev->flags); | |
3832 | } | |
3833 | if (s->failed >= 1) { | |
f2b3b44d | 3834 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3835 | s->locked++; |
3836 | set_bit(R5_LOCKED, &dev->flags); | |
3837 | set_bit(R5_Wantwrite, &dev->flags); | |
3838 | } | |
d82dfee0 | 3839 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3840 | dev = &sh->dev[pd_idx]; |
3841 | s->locked++; | |
3842 | set_bit(R5_LOCKED, &dev->flags); | |
3843 | set_bit(R5_Wantwrite, &dev->flags); | |
3844 | } | |
d82dfee0 | 3845 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3846 | dev = &sh->dev[qd_idx]; |
3847 | s->locked++; | |
3848 | set_bit(R5_LOCKED, &dev->flags); | |
3849 | set_bit(R5_Wantwrite, &dev->flags); | |
3850 | } | |
3851 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3852 | ||
3853 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3854 | break; |
3855 | case check_state_run: | |
3856 | case check_state_run_q: | |
3857 | case check_state_run_pq: | |
3858 | break; /* we will be called again upon completion */ | |
3859 | case check_state_check_result: | |
3860 | sh->check_state = check_state_idle; | |
3861 | ||
3862 | /* handle a successful check operation, if parity is correct | |
3863 | * we are done. Otherwise update the mismatch count and repair | |
3864 | * parity if !MD_RECOVERY_CHECK | |
3865 | */ | |
3866 | if (sh->ops.zero_sum_result == 0) { | |
3867 | /* both parities are correct */ | |
3868 | if (!s->failed) | |
3869 | set_bit(STRIPE_INSYNC, &sh->state); | |
3870 | else { | |
3871 | /* in contrast to the raid5 case we can validate | |
3872 | * parity, but still have a failure to write | |
3873 | * back | |
3874 | */ | |
3875 | sh->check_state = check_state_compute_result; | |
3876 | /* Returning at this point means that we may go | |
3877 | * off and bring p and/or q uptodate again so | |
3878 | * we make sure to check zero_sum_result again | |
3879 | * to verify if p or q need writeback | |
3880 | */ | |
3881 | } | |
3882 | } else { | |
7f7583d4 | 3883 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
3884 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3885 | /* don't try to repair!! */ | |
3886 | set_bit(STRIPE_INSYNC, &sh->state); | |
3887 | else { | |
3888 | int *target = &sh->ops.target; | |
3889 | ||
3890 | sh->ops.target = -1; | |
3891 | sh->ops.target2 = -1; | |
3892 | sh->check_state = check_state_compute_run; | |
3893 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3894 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3895 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3896 | set_bit(R5_Wantcompute, | |
3897 | &sh->dev[pd_idx].flags); | |
3898 | *target = pd_idx; | |
3899 | target = &sh->ops.target2; | |
3900 | s->uptodate++; | |
3901 | } | |
3902 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3903 | set_bit(R5_Wantcompute, | |
3904 | &sh->dev[qd_idx].flags); | |
3905 | *target = qd_idx; | |
3906 | s->uptodate++; | |
3907 | } | |
3908 | } | |
3909 | } | |
3910 | break; | |
3911 | case check_state_compute_run: | |
3912 | break; | |
3913 | default: | |
3914 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3915 | __func__, sh->check_state, | |
3916 | (unsigned long long) sh->sector); | |
3917 | BUG(); | |
a4456856 DW |
3918 | } |
3919 | } | |
3920 | ||
d1688a6d | 3921 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3922 | { |
3923 | int i; | |
3924 | ||
3925 | /* We have read all the blocks in this stripe and now we need to | |
3926 | * copy some of them into a target stripe for expand. | |
3927 | */ | |
f0a50d37 | 3928 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 3929 | BUG_ON(sh->batch_head); |
a4456856 DW |
3930 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3931 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3932 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3933 | int dd_idx, j; |
a4456856 | 3934 | struct stripe_head *sh2; |
a08abd8c | 3935 | struct async_submit_ctl submit; |
a4456856 | 3936 | |
784052ec | 3937 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3938 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3939 | &dd_idx, NULL); | |
a8c906ca | 3940 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3941 | if (sh2 == NULL) |
3942 | /* so far only the early blocks of this stripe | |
3943 | * have been requested. When later blocks | |
3944 | * get requested, we will try again | |
3945 | */ | |
3946 | continue; | |
3947 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3948 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3949 | /* must have already done this block */ | |
3950 | release_stripe(sh2); | |
3951 | continue; | |
3952 | } | |
f0a50d37 DW |
3953 | |
3954 | /* place all the copies on one channel */ | |
a08abd8c | 3955 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3956 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3957 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3958 | &submit); |
f0a50d37 | 3959 | |
a4456856 DW |
3960 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3961 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3962 | for (j = 0; j < conf->raid_disks; j++) | |
3963 | if (j != sh2->pd_idx && | |
86c374ba | 3964 | j != sh2->qd_idx && |
a4456856 DW |
3965 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3966 | break; | |
3967 | if (j == conf->raid_disks) { | |
3968 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3969 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3970 | } | |
3971 | release_stripe(sh2); | |
f0a50d37 | 3972 | |
a4456856 | 3973 | } |
a2e08551 | 3974 | /* done submitting copies, wait for them to complete */ |
749586b7 | 3975 | async_tx_quiesce(&tx); |
a4456856 | 3976 | } |
1da177e4 LT |
3977 | |
3978 | /* | |
3979 | * handle_stripe - do things to a stripe. | |
3980 | * | |
9a3e1101 N |
3981 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3982 | * state of various bits to see what needs to be done. | |
1da177e4 | 3983 | * Possible results: |
9a3e1101 N |
3984 | * return some read requests which now have data |
3985 | * return some write requests which are safely on storage | |
1da177e4 LT |
3986 | * schedule a read on some buffers |
3987 | * schedule a write of some buffers | |
3988 | * return confirmation of parity correctness | |
3989 | * | |
1da177e4 | 3990 | */ |
a4456856 | 3991 | |
acfe726b | 3992 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 3993 | { |
d1688a6d | 3994 | struct r5conf *conf = sh->raid_conf; |
f416885e | 3995 | int disks = sh->disks; |
474af965 N |
3996 | struct r5dev *dev; |
3997 | int i; | |
9a3e1101 | 3998 | int do_recovery = 0; |
1da177e4 | 3999 | |
acfe726b N |
4000 | memset(s, 0, sizeof(*s)); |
4001 | ||
dabc4ec6 | 4002 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4003 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4004 | s->failed_num[0] = -1; |
4005 | s->failed_num[1] = -1; | |
1da177e4 | 4006 | |
acfe726b | 4007 | /* Now to look around and see what can be done */ |
1da177e4 | 4008 | rcu_read_lock(); |
16a53ecc | 4009 | for (i=disks; i--; ) { |
3cb03002 | 4010 | struct md_rdev *rdev; |
31c176ec N |
4011 | sector_t first_bad; |
4012 | int bad_sectors; | |
4013 | int is_bad = 0; | |
acfe726b | 4014 | |
16a53ecc | 4015 | dev = &sh->dev[i]; |
1da177e4 | 4016 | |
45b4233c | 4017 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4018 | i, dev->flags, |
4019 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4020 | /* maybe we can reply to a read |
4021 | * | |
4022 | * new wantfill requests are only permitted while | |
4023 | * ops_complete_biofill is guaranteed to be inactive | |
4024 | */ | |
4025 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4026 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4027 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4028 | |
16a53ecc | 4029 | /* now count some things */ |
cc94015a N |
4030 | if (test_bit(R5_LOCKED, &dev->flags)) |
4031 | s->locked++; | |
4032 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4033 | s->uptodate++; | |
2d6e4ecc | 4034 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4035 | s->compute++; |
4036 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4037 | } |
1da177e4 | 4038 | |
acfe726b | 4039 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4040 | s->to_fill++; |
acfe726b | 4041 | else if (dev->toread) |
cc94015a | 4042 | s->to_read++; |
16a53ecc | 4043 | if (dev->towrite) { |
cc94015a | 4044 | s->to_write++; |
16a53ecc | 4045 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4046 | s->non_overwrite++; |
16a53ecc | 4047 | } |
a4456856 | 4048 | if (dev->written) |
cc94015a | 4049 | s->written++; |
14a75d3e N |
4050 | /* Prefer to use the replacement for reads, but only |
4051 | * if it is recovered enough and has no bad blocks. | |
4052 | */ | |
4053 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4054 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4055 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4056 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4057 | &first_bad, &bad_sectors)) | |
4058 | set_bit(R5_ReadRepl, &dev->flags); | |
4059 | else { | |
e6030cb0 | 4060 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4061 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4062 | else |
4063 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4064 | rdev = rcu_dereference(conf->disks[i].rdev); |
4065 | clear_bit(R5_ReadRepl, &dev->flags); | |
4066 | } | |
9283d8c5 N |
4067 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4068 | rdev = NULL; | |
31c176ec N |
4069 | if (rdev) { |
4070 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4071 | &first_bad, &bad_sectors); | |
4072 | if (s->blocked_rdev == NULL | |
4073 | && (test_bit(Blocked, &rdev->flags) | |
4074 | || is_bad < 0)) { | |
4075 | if (is_bad < 0) | |
4076 | set_bit(BlockedBadBlocks, | |
4077 | &rdev->flags); | |
4078 | s->blocked_rdev = rdev; | |
4079 | atomic_inc(&rdev->nr_pending); | |
4080 | } | |
6bfe0b49 | 4081 | } |
415e72d0 N |
4082 | clear_bit(R5_Insync, &dev->flags); |
4083 | if (!rdev) | |
4084 | /* Not in-sync */; | |
31c176ec N |
4085 | else if (is_bad) { |
4086 | /* also not in-sync */ | |
18b9837e N |
4087 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4088 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4089 | /* treat as in-sync, but with a read error |
4090 | * which we can now try to correct | |
4091 | */ | |
4092 | set_bit(R5_Insync, &dev->flags); | |
4093 | set_bit(R5_ReadError, &dev->flags); | |
4094 | } | |
4095 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4096 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4097 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4098 | /* in sync if before recovery_offset */ |
30d7a483 N |
4099 | set_bit(R5_Insync, &dev->flags); |
4100 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4101 | test_bit(R5_Expanded, &dev->flags)) | |
4102 | /* If we've reshaped into here, we assume it is Insync. | |
4103 | * We will shortly update recovery_offset to make | |
4104 | * it official. | |
4105 | */ | |
4106 | set_bit(R5_Insync, &dev->flags); | |
4107 | ||
1cc03eb9 | 4108 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4109 | /* This flag does not apply to '.replacement' |
4110 | * only to .rdev, so make sure to check that*/ | |
4111 | struct md_rdev *rdev2 = rcu_dereference( | |
4112 | conf->disks[i].rdev); | |
4113 | if (rdev2 == rdev) | |
4114 | clear_bit(R5_Insync, &dev->flags); | |
4115 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4116 | s->handle_bad_blocks = 1; |
14a75d3e | 4117 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4118 | } else |
4119 | clear_bit(R5_WriteError, &dev->flags); | |
4120 | } | |
1cc03eb9 | 4121 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4122 | /* This flag does not apply to '.replacement' |
4123 | * only to .rdev, so make sure to check that*/ | |
4124 | struct md_rdev *rdev2 = rcu_dereference( | |
4125 | conf->disks[i].rdev); | |
4126 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4127 | s->handle_bad_blocks = 1; |
14a75d3e | 4128 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4129 | } else |
4130 | clear_bit(R5_MadeGood, &dev->flags); | |
4131 | } | |
977df362 N |
4132 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4133 | struct md_rdev *rdev2 = rcu_dereference( | |
4134 | conf->disks[i].replacement); | |
4135 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4136 | s->handle_bad_blocks = 1; | |
4137 | atomic_inc(&rdev2->nr_pending); | |
4138 | } else | |
4139 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4140 | } | |
415e72d0 | 4141 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4142 | /* The ReadError flag will just be confusing now */ |
4143 | clear_bit(R5_ReadError, &dev->flags); | |
4144 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4145 | } |
415e72d0 N |
4146 | if (test_bit(R5_ReadError, &dev->flags)) |
4147 | clear_bit(R5_Insync, &dev->flags); | |
4148 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4149 | if (s->failed < 2) |
4150 | s->failed_num[s->failed] = i; | |
4151 | s->failed++; | |
9a3e1101 N |
4152 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4153 | do_recovery = 1; | |
415e72d0 | 4154 | } |
1da177e4 | 4155 | } |
9a3e1101 N |
4156 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4157 | /* If there is a failed device being replaced, | |
4158 | * we must be recovering. | |
4159 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4160 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4161 | * else we can only be replacing |
4162 | * sync and recovery both need to read all devices, and so | |
4163 | * use the same flag. | |
4164 | */ | |
4165 | if (do_recovery || | |
c6d2e084 | 4166 | sh->sector >= conf->mddev->recovery_cp || |
4167 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4168 | s->syncing = 1; |
4169 | else | |
4170 | s->replacing = 1; | |
4171 | } | |
1da177e4 | 4172 | rcu_read_unlock(); |
cc94015a N |
4173 | } |
4174 | ||
59fc630b | 4175 | static int clear_batch_ready(struct stripe_head *sh) |
4176 | { | |
b15a9dbd N |
4177 | /* Return '1' if this is a member of batch, or |
4178 | * '0' if it is a lone stripe or a head which can now be | |
4179 | * handled. | |
4180 | */ | |
59fc630b | 4181 | struct stripe_head *tmp; |
4182 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4183 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4184 | spin_lock(&sh->stripe_lock); |
4185 | if (!sh->batch_head) { | |
4186 | spin_unlock(&sh->stripe_lock); | |
4187 | return 0; | |
4188 | } | |
4189 | ||
4190 | /* | |
4191 | * this stripe could be added to a batch list before we check | |
4192 | * BATCH_READY, skips it | |
4193 | */ | |
4194 | if (sh->batch_head != sh) { | |
4195 | spin_unlock(&sh->stripe_lock); | |
4196 | return 1; | |
4197 | } | |
4198 | spin_lock(&sh->batch_lock); | |
4199 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4200 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4201 | spin_unlock(&sh->batch_lock); | |
4202 | spin_unlock(&sh->stripe_lock); | |
4203 | ||
4204 | /* | |
4205 | * BATCH_READY is cleared, no new stripes can be added. | |
4206 | * batch_list can be accessed without lock | |
4207 | */ | |
4208 | return 0; | |
4209 | } | |
4210 | ||
3960ce79 N |
4211 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4212 | unsigned long handle_flags) | |
72ac7330 | 4213 | { |
4e3d62ff | 4214 | struct stripe_head *sh, *next; |
72ac7330 | 4215 | int i; |
fb642b92 | 4216 | int do_wakeup = 0; |
72ac7330 | 4217 | |
bb27051f N |
4218 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4219 | ||
72ac7330 | 4220 | list_del_init(&sh->batch_list); |
4221 | ||
1b956f7a N |
4222 | WARN_ON_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
4223 | (1 << STRIPE_SYNCING) | | |
4224 | (1 << STRIPE_REPLACED) | | |
4225 | (1 << STRIPE_PREREAD_ACTIVE) | | |
4226 | (1 << STRIPE_DELAYED) | | |
4227 | (1 << STRIPE_BIT_DELAY) | | |
4228 | (1 << STRIPE_FULL_WRITE) | | |
4229 | (1 << STRIPE_BIOFILL_RUN) | | |
4230 | (1 << STRIPE_COMPUTE_RUN) | | |
4231 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4232 | (1 << STRIPE_DISCARD) | | |
4233 | (1 << STRIPE_BATCH_READY) | | |
4234 | (1 << STRIPE_BATCH_ERR) | | |
4235 | (1 << STRIPE_BITMAP_PENDING))); | |
4236 | WARN_ON_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4237 | (1 << STRIPE_REPLACED))); | |
4238 | ||
4239 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
4240 | (1 << STRIPE_DEGRADED)), | |
4241 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4242 | ||
72ac7330 | 4243 | sh->check_state = head_sh->check_state; |
4244 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4245 | for (i = 0; i < sh->disks; i++) { |
4246 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4247 | do_wakeup = 1; | |
72ac7330 | 4248 | sh->dev[i].flags = head_sh->dev[i].flags & |
4249 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4250 | } |
72ac7330 | 4251 | spin_lock_irq(&sh->stripe_lock); |
4252 | sh->batch_head = NULL; | |
4253 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4254 | if (handle_flags == 0 || |
4255 | sh->state & handle_flags) | |
4256 | set_bit(STRIPE_HANDLE, &sh->state); | |
72ac7330 | 4257 | release_stripe(sh); |
72ac7330 | 4258 | } |
fb642b92 N |
4259 | spin_lock_irq(&head_sh->stripe_lock); |
4260 | head_sh->batch_head = NULL; | |
4261 | spin_unlock_irq(&head_sh->stripe_lock); | |
4262 | for (i = 0; i < head_sh->disks; i++) | |
4263 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4264 | do_wakeup = 1; | |
3960ce79 N |
4265 | if (head_sh->state & handle_flags) |
4266 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4267 | |
4268 | if (do_wakeup) | |
4269 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4270 | } |
4271 | ||
cc94015a N |
4272 | static void handle_stripe(struct stripe_head *sh) |
4273 | { | |
4274 | struct stripe_head_state s; | |
d1688a6d | 4275 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4276 | int i; |
84789554 N |
4277 | int prexor; |
4278 | int disks = sh->disks; | |
474af965 | 4279 | struct r5dev *pdev, *qdev; |
cc94015a N |
4280 | |
4281 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4282 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4283 | /* already being handled, ensure it gets handled |
4284 | * again when current action finishes */ | |
4285 | set_bit(STRIPE_HANDLE, &sh->state); | |
4286 | return; | |
4287 | } | |
4288 | ||
59fc630b | 4289 | if (clear_batch_ready(sh) ) { |
4290 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4291 | return; | |
4292 | } | |
4293 | ||
4e3d62ff | 4294 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4295 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4296 | |
dabc4ec6 | 4297 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4298 | spin_lock(&sh->stripe_lock); |
4299 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4300 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4301 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4302 | set_bit(STRIPE_SYNCING, &sh->state); | |
4303 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4304 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4305 | } |
4306 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4307 | } |
4308 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4309 | ||
4310 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4311 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4312 | (unsigned long long)sh->sector, sh->state, | |
4313 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4314 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4315 | |
acfe726b | 4316 | analyse_stripe(sh, &s); |
c5a31000 | 4317 | |
bc2607f3 N |
4318 | if (s.handle_bad_blocks) { |
4319 | set_bit(STRIPE_HANDLE, &sh->state); | |
4320 | goto finish; | |
4321 | } | |
4322 | ||
474af965 N |
4323 | if (unlikely(s.blocked_rdev)) { |
4324 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4325 | s.replacing || s.to_write || s.written) { |
474af965 N |
4326 | set_bit(STRIPE_HANDLE, &sh->state); |
4327 | goto finish; | |
4328 | } | |
4329 | /* There is nothing for the blocked_rdev to block */ | |
4330 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4331 | s.blocked_rdev = NULL; | |
4332 | } | |
4333 | ||
4334 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4335 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4336 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4337 | } | |
4338 | ||
4339 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4340 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4341 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4342 | s.failed_num[0], s.failed_num[1]); | |
4343 | /* check if the array has lost more than max_degraded devices and, | |
4344 | * if so, some requests might need to be failed. | |
4345 | */ | |
9a3f530f N |
4346 | if (s.failed > conf->max_degraded) { |
4347 | sh->check_state = 0; | |
4348 | sh->reconstruct_state = 0; | |
626f2092 | 4349 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4350 | if (s.to_read+s.to_write+s.written) |
4351 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4352 | if (s.syncing + s.replacing) |
9a3f530f N |
4353 | handle_failed_sync(conf, sh, &s); |
4354 | } | |
474af965 | 4355 | |
84789554 N |
4356 | /* Now we check to see if any write operations have recently |
4357 | * completed | |
4358 | */ | |
4359 | prexor = 0; | |
4360 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4361 | prexor = 1; | |
4362 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4363 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4364 | sh->reconstruct_state = reconstruct_state_idle; | |
4365 | ||
4366 | /* All the 'written' buffers and the parity block are ready to | |
4367 | * be written back to disk | |
4368 | */ | |
9e444768 SL |
4369 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4370 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4371 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4372 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4373 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4374 | for (i = disks; i--; ) { |
4375 | struct r5dev *dev = &sh->dev[i]; | |
4376 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4377 | (i == sh->pd_idx || i == sh->qd_idx || | |
4378 | dev->written)) { | |
4379 | pr_debug("Writing block %d\n", i); | |
4380 | set_bit(R5_Wantwrite, &dev->flags); | |
4381 | if (prexor) | |
4382 | continue; | |
9c4bdf69 N |
4383 | if (s.failed > 1) |
4384 | continue; | |
84789554 N |
4385 | if (!test_bit(R5_Insync, &dev->flags) || |
4386 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4387 | s.failed == 0)) | |
4388 | set_bit(STRIPE_INSYNC, &sh->state); | |
4389 | } | |
4390 | } | |
4391 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4392 | s.dec_preread_active = 1; | |
4393 | } | |
4394 | ||
ef5b7c69 N |
4395 | /* |
4396 | * might be able to return some write requests if the parity blocks | |
4397 | * are safe, or on a failed drive | |
4398 | */ | |
4399 | pdev = &sh->dev[sh->pd_idx]; | |
4400 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4401 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4402 | qdev = &sh->dev[sh->qd_idx]; | |
4403 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4404 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4405 | || conf->level < 6; | |
4406 | ||
4407 | if (s.written && | |
4408 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4409 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4410 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4411 | test_bit(R5_Discard, &pdev->flags))))) && | |
4412 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4413 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4414 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4415 | test_bit(R5_Discard, &qdev->flags)))))) | |
4416 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4417 | ||
4418 | /* Now we might consider reading some blocks, either to check/generate | |
4419 | * parity, or to satisfy requests | |
4420 | * or to load a block that is being partially written. | |
4421 | */ | |
4422 | if (s.to_read || s.non_overwrite | |
4423 | || (conf->level == 6 && s.to_write && s.failed) | |
4424 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4425 | || s.replacing | |
4426 | || s.expanding) | |
4427 | handle_stripe_fill(sh, &s, disks); | |
4428 | ||
84789554 N |
4429 | /* Now to consider new write requests and what else, if anything |
4430 | * should be read. We do not handle new writes when: | |
4431 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
4432 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4433 | * block. | |
4434 | */ | |
4435 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
4436 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4437 | ||
4438 | /* maybe we need to check and possibly fix the parity for this stripe | |
4439 | * Any reads will already have been scheduled, so we just see if enough | |
4440 | * data is available. The parity check is held off while parity | |
4441 | * dependent operations are in flight. | |
4442 | */ | |
4443 | if (sh->check_state || | |
4444 | (s.syncing && s.locked == 0 && | |
4445 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4446 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4447 | if (conf->level == 6) | |
4448 | handle_parity_checks6(conf, sh, &s, disks); | |
4449 | else | |
4450 | handle_parity_checks5(conf, sh, &s, disks); | |
4451 | } | |
c5a31000 | 4452 | |
f94c0b66 N |
4453 | if ((s.replacing || s.syncing) && s.locked == 0 |
4454 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4455 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4456 | /* Write out to replacement devices where possible */ |
4457 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4458 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4459 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4460 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4461 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4462 | s.locked++; | |
4463 | } | |
f94c0b66 N |
4464 | if (s.replacing) |
4465 | set_bit(STRIPE_INSYNC, &sh->state); | |
4466 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4467 | } |
4468 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4469 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4470 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4471 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4472 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4473 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4474 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4475 | } |
4476 | ||
4477 | /* If the failed drives are just a ReadError, then we might need | |
4478 | * to progress the repair/check process | |
4479 | */ | |
4480 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4481 | for (i = 0; i < s.failed; i++) { | |
4482 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4483 | if (test_bit(R5_ReadError, &dev->flags) | |
4484 | && !test_bit(R5_LOCKED, &dev->flags) | |
4485 | && test_bit(R5_UPTODATE, &dev->flags) | |
4486 | ) { | |
4487 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4488 | set_bit(R5_Wantwrite, &dev->flags); | |
4489 | set_bit(R5_ReWrite, &dev->flags); | |
4490 | set_bit(R5_LOCKED, &dev->flags); | |
4491 | s.locked++; | |
4492 | } else { | |
4493 | /* let's read it back */ | |
4494 | set_bit(R5_Wantread, &dev->flags); | |
4495 | set_bit(R5_LOCKED, &dev->flags); | |
4496 | s.locked++; | |
4497 | } | |
4498 | } | |
4499 | } | |
4500 | ||
3687c061 N |
4501 | /* Finish reconstruct operations initiated by the expansion process */ |
4502 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4503 | struct stripe_head *sh_src | |
4504 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
4505 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
4506 | /* sh cannot be written until sh_src has been read. | |
4507 | * so arrange for sh to be delayed a little | |
4508 | */ | |
4509 | set_bit(STRIPE_DELAYED, &sh->state); | |
4510 | set_bit(STRIPE_HANDLE, &sh->state); | |
4511 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4512 | &sh_src->state)) | |
4513 | atomic_inc(&conf->preread_active_stripes); | |
4514 | release_stripe(sh_src); | |
4515 | goto finish; | |
4516 | } | |
4517 | if (sh_src) | |
4518 | release_stripe(sh_src); | |
4519 | ||
4520 | sh->reconstruct_state = reconstruct_state_idle; | |
4521 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4522 | for (i = conf->raid_disks; i--; ) { | |
4523 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4524 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4525 | s.locked++; | |
4526 | } | |
4527 | } | |
f416885e | 4528 | |
3687c061 N |
4529 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4530 | !sh->reconstruct_state) { | |
4531 | /* Need to write out all blocks after computing parity */ | |
4532 | sh->disks = conf->raid_disks; | |
4533 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4534 | schedule_reconstruction(sh, &s, 1, 1); | |
4535 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4536 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4537 | atomic_dec(&conf->reshape_stripes); | |
4538 | wake_up(&conf->wait_for_overlap); | |
4539 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4540 | } | |
4541 | ||
4542 | if (s.expanding && s.locked == 0 && | |
4543 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4544 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4545 | |
3687c061 | 4546 | finish: |
6bfe0b49 | 4547 | /* wait for this device to become unblocked */ |
5f066c63 N |
4548 | if (unlikely(s.blocked_rdev)) { |
4549 | if (conf->mddev->external) | |
4550 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4551 | conf->mddev); | |
4552 | else | |
4553 | /* Internal metadata will immediately | |
4554 | * be written by raid5d, so we don't | |
4555 | * need to wait here. | |
4556 | */ | |
4557 | rdev_dec_pending(s.blocked_rdev, | |
4558 | conf->mddev); | |
4559 | } | |
6bfe0b49 | 4560 | |
bc2607f3 N |
4561 | if (s.handle_bad_blocks) |
4562 | for (i = disks; i--; ) { | |
3cb03002 | 4563 | struct md_rdev *rdev; |
bc2607f3 N |
4564 | struct r5dev *dev = &sh->dev[i]; |
4565 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4566 | /* We own a safe reference to the rdev */ | |
4567 | rdev = conf->disks[i].rdev; | |
4568 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4569 | STRIPE_SECTORS, 0)) | |
4570 | md_error(conf->mddev, rdev); | |
4571 | rdev_dec_pending(rdev, conf->mddev); | |
4572 | } | |
b84db560 N |
4573 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4574 | rdev = conf->disks[i].rdev; | |
4575 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4576 | STRIPE_SECTORS, 0); |
b84db560 N |
4577 | rdev_dec_pending(rdev, conf->mddev); |
4578 | } | |
977df362 N |
4579 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4580 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4581 | if (!rdev) |
4582 | /* rdev have been moved down */ | |
4583 | rdev = conf->disks[i].rdev; | |
977df362 | 4584 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4585 | STRIPE_SECTORS, 0); |
977df362 N |
4586 | rdev_dec_pending(rdev, conf->mddev); |
4587 | } | |
bc2607f3 N |
4588 | } |
4589 | ||
6c0069c0 YT |
4590 | if (s.ops_request) |
4591 | raid_run_ops(sh, s.ops_request); | |
4592 | ||
f0e43bcd | 4593 | ops_run_io(sh, &s); |
16a53ecc | 4594 | |
c5709ef6 | 4595 | if (s.dec_preread_active) { |
729a1866 | 4596 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4597 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4598 | * have actually been submitted. |
4599 | */ | |
4600 | atomic_dec(&conf->preread_active_stripes); | |
4601 | if (atomic_read(&conf->preread_active_stripes) < | |
4602 | IO_THRESHOLD) | |
4603 | md_wakeup_thread(conf->mddev->thread); | |
4604 | } | |
4605 | ||
c3cce6cd N |
4606 | if (!bio_list_empty(&s.return_bi)) { |
4607 | if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) { | |
4608 | spin_lock_irq(&conf->device_lock); | |
4609 | bio_list_merge(&conf->return_bi, &s.return_bi); | |
4610 | spin_unlock_irq(&conf->device_lock); | |
4611 | md_wakeup_thread(conf->mddev->thread); | |
4612 | } else | |
4613 | return_io(&s.return_bi); | |
4614 | } | |
16a53ecc | 4615 | |
257a4b42 | 4616 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4617 | } |
4618 | ||
d1688a6d | 4619 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4620 | { |
4621 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4622 | while (!list_empty(&conf->delayed_list)) { | |
4623 | struct list_head *l = conf->delayed_list.next; | |
4624 | struct stripe_head *sh; | |
4625 | sh = list_entry(l, struct stripe_head, lru); | |
4626 | list_del_init(l); | |
4627 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4628 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4629 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4630 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4631 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4632 | } |
482c0834 | 4633 | } |
16a53ecc N |
4634 | } |
4635 | ||
566c09c5 SL |
4636 | static void activate_bit_delay(struct r5conf *conf, |
4637 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4638 | { |
4639 | /* device_lock is held */ | |
4640 | struct list_head head; | |
4641 | list_add(&head, &conf->bitmap_list); | |
4642 | list_del_init(&conf->bitmap_list); | |
4643 | while (!list_empty(&head)) { | |
4644 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4645 | int hash; |
16a53ecc N |
4646 | list_del_init(&sh->lru); |
4647 | atomic_inc(&sh->count); | |
566c09c5 SL |
4648 | hash = sh->hash_lock_index; |
4649 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4650 | } |
4651 | } | |
4652 | ||
5c675f83 | 4653 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4654 | { |
d1688a6d | 4655 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4656 | |
4657 | /* No difference between reads and writes. Just check | |
4658 | * how busy the stripe_cache is | |
4659 | */ | |
3fa841d7 | 4660 | |
5423399a | 4661 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd N |
4662 | return 1; |
4663 | if (conf->quiesce) | |
4664 | return 1; | |
4bda556a | 4665 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4666 | return 1; |
4667 | ||
4668 | return 0; | |
4669 | } | |
4670 | ||
fd01b88c | 4671 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4672 | { |
3cb5edf4 | 4673 | struct r5conf *conf = mddev->private; |
4f024f37 | 4674 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 4675 | unsigned int chunk_sectors; |
aa8b57aa | 4676 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4677 | |
3cb5edf4 | 4678 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
4679 | return chunk_sectors >= |
4680 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4681 | } | |
4682 | ||
46031f9a RBJ |
4683 | /* |
4684 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4685 | * later sampled by raid5d. | |
4686 | */ | |
d1688a6d | 4687 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4688 | { |
4689 | unsigned long flags; | |
4690 | ||
4691 | spin_lock_irqsave(&conf->device_lock, flags); | |
4692 | ||
4693 | bi->bi_next = conf->retry_read_aligned_list; | |
4694 | conf->retry_read_aligned_list = bi; | |
4695 | ||
4696 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4697 | md_wakeup_thread(conf->mddev->thread); | |
4698 | } | |
4699 | ||
d1688a6d | 4700 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4701 | { |
4702 | struct bio *bi; | |
4703 | ||
4704 | bi = conf->retry_read_aligned; | |
4705 | if (bi) { | |
4706 | conf->retry_read_aligned = NULL; | |
4707 | return bi; | |
4708 | } | |
4709 | bi = conf->retry_read_aligned_list; | |
4710 | if(bi) { | |
387bb173 | 4711 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4712 | bi->bi_next = NULL; |
960e739d JA |
4713 | /* |
4714 | * this sets the active strip count to 1 and the processed | |
4715 | * strip count to zero (upper 8 bits) | |
4716 | */ | |
e7836bd6 | 4717 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
4718 | } |
4719 | ||
4720 | return bi; | |
4721 | } | |
4722 | ||
f679623f RBJ |
4723 | /* |
4724 | * The "raid5_align_endio" should check if the read succeeded and if it | |
4725 | * did, call bio_endio on the original bio (having bio_put the new bio | |
4726 | * first). | |
4727 | * If the read failed.. | |
4728 | */ | |
4246a0b6 | 4729 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
4730 | { |
4731 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 4732 | struct mddev *mddev; |
d1688a6d | 4733 | struct r5conf *conf; |
3cb03002 | 4734 | struct md_rdev *rdev; |
9b81c842 | 4735 | int error = bi->bi_error; |
46031f9a | 4736 | |
f679623f | 4737 | bio_put(bi); |
46031f9a | 4738 | |
46031f9a RBJ |
4739 | rdev = (void*)raid_bi->bi_next; |
4740 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
4741 | mddev = rdev->mddev; |
4742 | conf = mddev->private; | |
46031f9a RBJ |
4743 | |
4744 | rdev_dec_pending(rdev, conf->mddev); | |
4745 | ||
9b81c842 | 4746 | if (!error) { |
0a82a8d1 LT |
4747 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
4748 | raid_bi, 0); | |
4246a0b6 | 4749 | bio_endio(raid_bi); |
46031f9a | 4750 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 4751 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 4752 | return; |
46031f9a RBJ |
4753 | } |
4754 | ||
45b4233c | 4755 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
4756 | |
4757 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
4758 | } |
4759 | ||
7ef6b12a | 4760 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 4761 | { |
d1688a6d | 4762 | struct r5conf *conf = mddev->private; |
8553fe7e | 4763 | int dd_idx; |
f679623f | 4764 | struct bio* align_bi; |
3cb03002 | 4765 | struct md_rdev *rdev; |
671488cc | 4766 | sector_t end_sector; |
f679623f RBJ |
4767 | |
4768 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 4769 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
4770 | return 0; |
4771 | } | |
4772 | /* | |
a167f663 | 4773 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 4774 | */ |
a167f663 | 4775 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
4776 | if (!align_bi) |
4777 | return 0; | |
4778 | /* | |
4779 | * set bi_end_io to a new function, and set bi_private to the | |
4780 | * original bio. | |
4781 | */ | |
4782 | align_bi->bi_end_io = raid5_align_endio; | |
4783 | align_bi->bi_private = raid_bio; | |
4784 | /* | |
4785 | * compute position | |
4786 | */ | |
4f024f37 KO |
4787 | align_bi->bi_iter.bi_sector = |
4788 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
4789 | 0, &dd_idx, NULL); | |
f679623f | 4790 | |
f73a1c7d | 4791 | end_sector = bio_end_sector(align_bi); |
f679623f | 4792 | rcu_read_lock(); |
671488cc N |
4793 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
4794 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
4795 | rdev->recovery_offset < end_sector) { | |
4796 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
4797 | if (rdev && | |
4798 | (test_bit(Faulty, &rdev->flags) || | |
4799 | !(test_bit(In_sync, &rdev->flags) || | |
4800 | rdev->recovery_offset >= end_sector))) | |
4801 | rdev = NULL; | |
4802 | } | |
4803 | if (rdev) { | |
31c176ec N |
4804 | sector_t first_bad; |
4805 | int bad_sectors; | |
4806 | ||
f679623f RBJ |
4807 | atomic_inc(&rdev->nr_pending); |
4808 | rcu_read_unlock(); | |
46031f9a RBJ |
4809 | raid_bio->bi_next = (void*)rdev; |
4810 | align_bi->bi_bdev = rdev->bdev; | |
b7c44ed9 | 4811 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 4812 | |
7140aafc | 4813 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 4814 | bio_sectors(align_bi), |
31c176ec | 4815 | &first_bad, &bad_sectors)) { |
387bb173 NB |
4816 | bio_put(align_bi); |
4817 | rdev_dec_pending(rdev, mddev); | |
4818 | return 0; | |
4819 | } | |
4820 | ||
6c0544e2 | 4821 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 4822 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 4823 | |
46031f9a | 4824 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 4825 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 4826 | conf->quiesce == 0, |
eed8c02e | 4827 | conf->device_lock); |
46031f9a RBJ |
4828 | atomic_inc(&conf->active_aligned_reads); |
4829 | spin_unlock_irq(&conf->device_lock); | |
4830 | ||
e3620a3a JB |
4831 | if (mddev->gendisk) |
4832 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
4833 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 4834 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
4835 | generic_make_request(align_bi); |
4836 | return 1; | |
4837 | } else { | |
4838 | rcu_read_unlock(); | |
46031f9a | 4839 | bio_put(align_bi); |
f679623f RBJ |
4840 | return 0; |
4841 | } | |
4842 | } | |
4843 | ||
7ef6b12a ML |
4844 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
4845 | { | |
4846 | struct bio *split; | |
4847 | ||
4848 | do { | |
4849 | sector_t sector = raid_bio->bi_iter.bi_sector; | |
4850 | unsigned chunk_sects = mddev->chunk_sectors; | |
4851 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
4852 | ||
4853 | if (sectors < bio_sectors(raid_bio)) { | |
4854 | split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set); | |
4855 | bio_chain(split, raid_bio); | |
4856 | } else | |
4857 | split = raid_bio; | |
4858 | ||
4859 | if (!raid5_read_one_chunk(mddev, split)) { | |
4860 | if (split != raid_bio) | |
4861 | generic_make_request(raid_bio); | |
4862 | return split; | |
4863 | } | |
4864 | } while (split != raid_bio); | |
4865 | ||
4866 | return NULL; | |
4867 | } | |
4868 | ||
8b3e6cdc DW |
4869 | /* __get_priority_stripe - get the next stripe to process |
4870 | * | |
4871 | * Full stripe writes are allowed to pass preread active stripes up until | |
4872 | * the bypass_threshold is exceeded. In general the bypass_count | |
4873 | * increments when the handle_list is handled before the hold_list; however, it | |
4874 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
4875 | * stripe with in flight i/o. The bypass_count will be reset when the | |
4876 | * head of the hold_list has changed, i.e. the head was promoted to the | |
4877 | * handle_list. | |
4878 | */ | |
851c30c9 | 4879 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 4880 | { |
851c30c9 SL |
4881 | struct stripe_head *sh = NULL, *tmp; |
4882 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 4883 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
4884 | |
4885 | if (conf->worker_cnt_per_group == 0) { | |
4886 | handle_list = &conf->handle_list; | |
4887 | } else if (group != ANY_GROUP) { | |
4888 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 4889 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
4890 | } else { |
4891 | int i; | |
4892 | for (i = 0; i < conf->group_cnt; i++) { | |
4893 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 4894 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
4895 | if (!list_empty(handle_list)) |
4896 | break; | |
4897 | } | |
4898 | } | |
8b3e6cdc DW |
4899 | |
4900 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
4901 | __func__, | |
851c30c9 | 4902 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
4903 | list_empty(&conf->hold_list) ? "empty" : "busy", |
4904 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
4905 | ||
851c30c9 SL |
4906 | if (!list_empty(handle_list)) { |
4907 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
4908 | |
4909 | if (list_empty(&conf->hold_list)) | |
4910 | conf->bypass_count = 0; | |
4911 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
4912 | if (conf->hold_list.next == conf->last_hold) | |
4913 | conf->bypass_count++; | |
4914 | else { | |
4915 | conf->last_hold = conf->hold_list.next; | |
4916 | conf->bypass_count -= conf->bypass_threshold; | |
4917 | if (conf->bypass_count < 0) | |
4918 | conf->bypass_count = 0; | |
4919 | } | |
4920 | } | |
4921 | } else if (!list_empty(&conf->hold_list) && | |
4922 | ((conf->bypass_threshold && | |
4923 | conf->bypass_count > conf->bypass_threshold) || | |
4924 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
4925 | |
4926 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
4927 | if (conf->worker_cnt_per_group == 0 || | |
4928 | group == ANY_GROUP || | |
4929 | !cpu_online(tmp->cpu) || | |
4930 | cpu_to_group(tmp->cpu) == group) { | |
4931 | sh = tmp; | |
4932 | break; | |
4933 | } | |
4934 | } | |
4935 | ||
4936 | if (sh) { | |
4937 | conf->bypass_count -= conf->bypass_threshold; | |
4938 | if (conf->bypass_count < 0) | |
4939 | conf->bypass_count = 0; | |
4940 | } | |
bfc90cb0 | 4941 | wg = NULL; |
851c30c9 SL |
4942 | } |
4943 | ||
4944 | if (!sh) | |
8b3e6cdc DW |
4945 | return NULL; |
4946 | ||
bfc90cb0 SL |
4947 | if (wg) { |
4948 | wg->stripes_cnt--; | |
4949 | sh->group = NULL; | |
4950 | } | |
8b3e6cdc | 4951 | list_del_init(&sh->lru); |
c7a6d35e | 4952 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
4953 | return sh; |
4954 | } | |
f679623f | 4955 | |
8811b596 SL |
4956 | struct raid5_plug_cb { |
4957 | struct blk_plug_cb cb; | |
4958 | struct list_head list; | |
566c09c5 | 4959 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
4960 | }; |
4961 | ||
4962 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4963 | { | |
4964 | struct raid5_plug_cb *cb = container_of( | |
4965 | blk_cb, struct raid5_plug_cb, cb); | |
4966 | struct stripe_head *sh; | |
4967 | struct mddev *mddev = cb->cb.data; | |
4968 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 4969 | int cnt = 0; |
566c09c5 | 4970 | int hash; |
8811b596 SL |
4971 | |
4972 | if (cb->list.next && !list_empty(&cb->list)) { | |
4973 | spin_lock_irq(&conf->device_lock); | |
4974 | while (!list_empty(&cb->list)) { | |
4975 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
4976 | list_del_init(&sh->lru); | |
4977 | /* | |
4978 | * avoid race release_stripe_plug() sees | |
4979 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
4980 | * is still in our list | |
4981 | */ | |
4e857c58 | 4982 | smp_mb__before_atomic(); |
8811b596 | 4983 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
4984 | /* |
4985 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
4986 | * case, the count is always > 1 here | |
4987 | */ | |
566c09c5 SL |
4988 | hash = sh->hash_lock_index; |
4989 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 4990 | cnt++; |
8811b596 SL |
4991 | } |
4992 | spin_unlock_irq(&conf->device_lock); | |
4993 | } | |
566c09c5 SL |
4994 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
4995 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
4996 | if (mddev->queue) |
4997 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
4998 | kfree(cb); |
4999 | } | |
5000 | ||
5001 | static void release_stripe_plug(struct mddev *mddev, | |
5002 | struct stripe_head *sh) | |
5003 | { | |
5004 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5005 | raid5_unplug, mddev, | |
5006 | sizeof(struct raid5_plug_cb)); | |
5007 | struct raid5_plug_cb *cb; | |
5008 | ||
5009 | if (!blk_cb) { | |
5010 | release_stripe(sh); | |
5011 | return; | |
5012 | } | |
5013 | ||
5014 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5015 | ||
566c09c5 SL |
5016 | if (cb->list.next == NULL) { |
5017 | int i; | |
8811b596 | 5018 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5019 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5020 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5021 | } | |
8811b596 SL |
5022 | |
5023 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5024 | list_add_tail(&sh->lru, &cb->list); | |
5025 | else | |
5026 | release_stripe(sh); | |
5027 | } | |
5028 | ||
620125f2 SL |
5029 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5030 | { | |
5031 | struct r5conf *conf = mddev->private; | |
5032 | sector_t logical_sector, last_sector; | |
5033 | struct stripe_head *sh; | |
5034 | int remaining; | |
5035 | int stripe_sectors; | |
5036 | ||
5037 | if (mddev->reshape_position != MaxSector) | |
5038 | /* Skip discard while reshape is happening */ | |
5039 | return; | |
5040 | ||
4f024f37 KO |
5041 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5042 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5043 | |
5044 | bi->bi_next = NULL; | |
5045 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5046 | ||
5047 | stripe_sectors = conf->chunk_sectors * | |
5048 | (conf->raid_disks - conf->max_degraded); | |
5049 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5050 | stripe_sectors); | |
5051 | sector_div(last_sector, stripe_sectors); | |
5052 | ||
5053 | logical_sector *= conf->chunk_sectors; | |
5054 | last_sector *= conf->chunk_sectors; | |
5055 | ||
5056 | for (; logical_sector < last_sector; | |
5057 | logical_sector += STRIPE_SECTORS) { | |
5058 | DEFINE_WAIT(w); | |
5059 | int d; | |
5060 | again: | |
5061 | sh = get_active_stripe(conf, logical_sector, 0, 0, 0); | |
5062 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5063 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5064 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5065 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
5066 | release_stripe(sh); | |
5067 | schedule(); | |
5068 | goto again; | |
5069 | } | |
5070 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5071 | spin_lock_irq(&sh->stripe_lock); |
5072 | for (d = 0; d < conf->raid_disks; d++) { | |
5073 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5074 | continue; | |
5075 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5076 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5077 | spin_unlock_irq(&sh->stripe_lock); | |
5078 | release_stripe(sh); | |
5079 | schedule(); | |
5080 | goto again; | |
5081 | } | |
5082 | } | |
f8dfcffd | 5083 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5084 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5085 | sh->overwrite_disks = 0; |
620125f2 SL |
5086 | for (d = 0; d < conf->raid_disks; d++) { |
5087 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5088 | continue; | |
5089 | sh->dev[d].towrite = bi; | |
5090 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5091 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5092 | sh->overwrite_disks++; |
620125f2 SL |
5093 | } |
5094 | spin_unlock_irq(&sh->stripe_lock); | |
5095 | if (conf->mddev->bitmap) { | |
5096 | for (d = 0; | |
5097 | d < conf->raid_disks - conf->max_degraded; | |
5098 | d++) | |
5099 | bitmap_startwrite(mddev->bitmap, | |
5100 | sh->sector, | |
5101 | STRIPE_SECTORS, | |
5102 | 0); | |
5103 | sh->bm_seq = conf->seq_flush + 1; | |
5104 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5105 | } | |
5106 | ||
5107 | set_bit(STRIPE_HANDLE, &sh->state); | |
5108 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5109 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5110 | atomic_inc(&conf->preread_active_stripes); | |
5111 | release_stripe_plug(mddev, sh); | |
5112 | } | |
5113 | ||
5114 | remaining = raid5_dec_bi_active_stripes(bi); | |
5115 | if (remaining == 0) { | |
5116 | md_write_end(mddev); | |
4246a0b6 | 5117 | bio_endio(bi); |
620125f2 SL |
5118 | } |
5119 | } | |
5120 | ||
b4fdcb02 | 5121 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5122 | { |
d1688a6d | 5123 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5124 | int dd_idx; |
1da177e4 LT |
5125 | sector_t new_sector; |
5126 | sector_t logical_sector, last_sector; | |
5127 | struct stripe_head *sh; | |
a362357b | 5128 | const int rw = bio_data_dir(bi); |
49077326 | 5129 | int remaining; |
27c0f68f SL |
5130 | DEFINE_WAIT(w); |
5131 | bool do_prepare; | |
1da177e4 | 5132 | |
e9c7469b TH |
5133 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
5134 | md_flush_request(mddev, bi); | |
5a7bbad2 | 5135 | return; |
e5dcdd80 N |
5136 | } |
5137 | ||
3d310eb7 | 5138 | md_write_start(mddev, bi); |
06d91a5f | 5139 | |
9ffc8f7c EM |
5140 | /* |
5141 | * If array is degraded, better not do chunk aligned read because | |
5142 | * later we might have to read it again in order to reconstruct | |
5143 | * data on failed drives. | |
5144 | */ | |
5145 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5146 | mddev->reshape_position == MaxSector) { |
5147 | bi = chunk_aligned_read(mddev, bi); | |
5148 | if (!bi) | |
5149 | return; | |
5150 | } | |
52488615 | 5151 | |
620125f2 SL |
5152 | if (unlikely(bi->bi_rw & REQ_DISCARD)) { |
5153 | make_discard_request(mddev, bi); | |
5154 | return; | |
5155 | } | |
5156 | ||
4f024f37 | 5157 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5158 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5159 | bi->bi_next = NULL; |
5160 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5161 | |
27c0f68f | 5162 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5163 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5164 | int previous; |
c46501b2 | 5165 | int seq; |
b578d55f | 5166 | |
27c0f68f | 5167 | do_prepare = false; |
7ecaa1e6 | 5168 | retry: |
c46501b2 | 5169 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5170 | previous = 0; |
27c0f68f SL |
5171 | if (do_prepare) |
5172 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5173 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5174 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5175 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5176 | * 64bit on a 32bit platform, and so it might be |
5177 | * possible to see a half-updated value | |
aeb878b0 | 5178 | * Of course reshape_progress could change after |
df8e7f76 N |
5179 | * the lock is dropped, so once we get a reference |
5180 | * to the stripe that we think it is, we will have | |
5181 | * to check again. | |
5182 | */ | |
7ecaa1e6 | 5183 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5184 | if (mddev->reshape_backwards |
fef9c61f N |
5185 | ? logical_sector < conf->reshape_progress |
5186 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5187 | previous = 1; |
5188 | } else { | |
2c810cdd | 5189 | if (mddev->reshape_backwards |
fef9c61f N |
5190 | ? logical_sector < conf->reshape_safe |
5191 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5192 | spin_unlock_irq(&conf->device_lock); |
5193 | schedule(); | |
27c0f68f | 5194 | do_prepare = true; |
b578d55f N |
5195 | goto retry; |
5196 | } | |
5197 | } | |
7ecaa1e6 N |
5198 | spin_unlock_irq(&conf->device_lock); |
5199 | } | |
16a53ecc | 5200 | |
112bf897 N |
5201 | new_sector = raid5_compute_sector(conf, logical_sector, |
5202 | previous, | |
911d4ee8 | 5203 | &dd_idx, NULL); |
0c55e022 | 5204 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
c46501b2 | 5205 | (unsigned long long)new_sector, |
1da177e4 LT |
5206 | (unsigned long long)logical_sector); |
5207 | ||
b5663ba4 | 5208 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 5209 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 5210 | if (sh) { |
b0f9ec04 | 5211 | if (unlikely(previous)) { |
7ecaa1e6 | 5212 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5213 | * stripe, so we must do the range check again. |
5214 | * Expansion could still move past after this | |
5215 | * test, but as we are holding a reference to | |
5216 | * 'sh', we know that if that happens, | |
5217 | * STRIPE_EXPANDING will get set and the expansion | |
5218 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5219 | */ |
5220 | int must_retry = 0; | |
5221 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5222 | if (mddev->reshape_backwards |
b0f9ec04 N |
5223 | ? logical_sector >= conf->reshape_progress |
5224 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5225 | /* mismatch, need to try again */ |
5226 | must_retry = 1; | |
5227 | spin_unlock_irq(&conf->device_lock); | |
5228 | if (must_retry) { | |
5229 | release_stripe(sh); | |
7a3ab908 | 5230 | schedule(); |
27c0f68f | 5231 | do_prepare = true; |
7ecaa1e6 N |
5232 | goto retry; |
5233 | } | |
5234 | } | |
c46501b2 N |
5235 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5236 | /* Might have got the wrong stripe_head | |
5237 | * by accident | |
5238 | */ | |
5239 | release_stripe(sh); | |
5240 | goto retry; | |
5241 | } | |
e62e58a5 | 5242 | |
ffd96e35 | 5243 | if (rw == WRITE && |
a5c308d4 | 5244 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
5245 | logical_sector < mddev->suspend_hi) { |
5246 | release_stripe(sh); | |
e62e58a5 N |
5247 | /* As the suspend_* range is controlled by |
5248 | * userspace, we want an interruptible | |
5249 | * wait. | |
5250 | */ | |
5251 | flush_signals(current); | |
5252 | prepare_to_wait(&conf->wait_for_overlap, | |
5253 | &w, TASK_INTERRUPTIBLE); | |
5254 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5255 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5256 | schedule(); |
27c0f68f SL |
5257 | do_prepare = true; |
5258 | } | |
e464eafd N |
5259 | goto retry; |
5260 | } | |
7ecaa1e6 N |
5261 | |
5262 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5263 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5264 | /* Stripe is busy expanding or |
5265 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5266 | * and wait a while |
5267 | */ | |
482c0834 | 5268 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5269 | release_stripe(sh); |
5270 | schedule(); | |
27c0f68f | 5271 | do_prepare = true; |
1da177e4 LT |
5272 | goto retry; |
5273 | } | |
6ed3003c N |
5274 | set_bit(STRIPE_HANDLE, &sh->state); |
5275 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5276 | if ((!sh->batch_head || sh == sh->batch_head) && |
5277 | (bi->bi_rw & REQ_SYNC) && | |
729a1866 N |
5278 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5279 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5280 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5281 | } else { |
5282 | /* cannot get stripe for read-ahead, just give-up */ | |
4246a0b6 | 5283 | bi->bi_error = -EIO; |
1da177e4 LT |
5284 | break; |
5285 | } | |
1da177e4 | 5286 | } |
27c0f68f | 5287 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5288 | |
e7836bd6 | 5289 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5290 | if (remaining == 0) { |
1da177e4 | 5291 | |
16a53ecc | 5292 | if ( rw == WRITE ) |
1da177e4 | 5293 | md_write_end(mddev); |
6712ecf8 | 5294 | |
0a82a8d1 LT |
5295 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5296 | bi, 0); | |
4246a0b6 | 5297 | bio_endio(bi); |
1da177e4 | 5298 | } |
1da177e4 LT |
5299 | } |
5300 | ||
fd01b88c | 5301 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5302 | |
fd01b88c | 5303 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5304 | { |
52c03291 N |
5305 | /* reshaping is quite different to recovery/resync so it is |
5306 | * handled quite separately ... here. | |
5307 | * | |
5308 | * On each call to sync_request, we gather one chunk worth of | |
5309 | * destination stripes and flag them as expanding. | |
5310 | * Then we find all the source stripes and request reads. | |
5311 | * As the reads complete, handle_stripe will copy the data | |
5312 | * into the destination stripe and release that stripe. | |
5313 | */ | |
d1688a6d | 5314 | struct r5conf *conf = mddev->private; |
1da177e4 | 5315 | struct stripe_head *sh; |
ccfcc3c1 | 5316 | sector_t first_sector, last_sector; |
f416885e N |
5317 | int raid_disks = conf->previous_raid_disks; |
5318 | int data_disks = raid_disks - conf->max_degraded; | |
5319 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5320 | int i; |
5321 | int dd_idx; | |
c8f517c4 | 5322 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5323 | sector_t stripe_addr; |
7a661381 | 5324 | int reshape_sectors; |
ab69ae12 | 5325 | struct list_head stripes; |
92140480 | 5326 | sector_t retn; |
52c03291 | 5327 | |
fef9c61f N |
5328 | if (sector_nr == 0) { |
5329 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5330 | if (mddev->reshape_backwards && |
fef9c61f N |
5331 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5332 | sector_nr = raid5_size(mddev, 0, 0) | |
5333 | - conf->reshape_progress; | |
6cbd8148 N |
5334 | } else if (mddev->reshape_backwards && |
5335 | conf->reshape_progress == MaxSector) { | |
5336 | /* shouldn't happen, but just in case, finish up.*/ | |
5337 | sector_nr = MaxSector; | |
2c810cdd | 5338 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5339 | conf->reshape_progress > 0) |
5340 | sector_nr = conf->reshape_progress; | |
f416885e | 5341 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5342 | if (sector_nr) { |
8dee7211 N |
5343 | mddev->curr_resync_completed = sector_nr; |
5344 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5345 | *skipped = 1; |
92140480 N |
5346 | retn = sector_nr; |
5347 | goto finish; | |
fef9c61f | 5348 | } |
52c03291 N |
5349 | } |
5350 | ||
7a661381 N |
5351 | /* We need to process a full chunk at a time. |
5352 | * If old and new chunk sizes differ, we need to process the | |
5353 | * largest of these | |
5354 | */ | |
3cb5edf4 N |
5355 | |
5356 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5357 | |
b5254dd5 N |
5358 | /* We update the metadata at least every 10 seconds, or when |
5359 | * the data about to be copied would over-write the source of | |
5360 | * the data at the front of the range. i.e. one new_stripe | |
5361 | * along from reshape_progress new_maps to after where | |
5362 | * reshape_safe old_maps to | |
52c03291 | 5363 | */ |
fef9c61f | 5364 | writepos = conf->reshape_progress; |
f416885e | 5365 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5366 | readpos = conf->reshape_progress; |
5367 | sector_div(readpos, data_disks); | |
fef9c61f | 5368 | safepos = conf->reshape_safe; |
f416885e | 5369 | sector_div(safepos, data_disks); |
2c810cdd | 5370 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5371 | BUG_ON(writepos < reshape_sectors); |
5372 | writepos -= reshape_sectors; | |
c8f517c4 | 5373 | readpos += reshape_sectors; |
7a661381 | 5374 | safepos += reshape_sectors; |
fef9c61f | 5375 | } else { |
7a661381 | 5376 | writepos += reshape_sectors; |
c74c0d76 N |
5377 | /* readpos and safepos are worst-case calculations. |
5378 | * A negative number is overly pessimistic, and causes | |
5379 | * obvious problems for unsigned storage. So clip to 0. | |
5380 | */ | |
ed37d83e N |
5381 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5382 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5383 | } |
52c03291 | 5384 | |
b5254dd5 N |
5385 | /* Having calculated the 'writepos' possibly use it |
5386 | * to set 'stripe_addr' which is where we will write to. | |
5387 | */ | |
5388 | if (mddev->reshape_backwards) { | |
5389 | BUG_ON(conf->reshape_progress == 0); | |
5390 | stripe_addr = writepos; | |
5391 | BUG_ON((mddev->dev_sectors & | |
5392 | ~((sector_t)reshape_sectors - 1)) | |
5393 | - reshape_sectors - stripe_addr | |
5394 | != sector_nr); | |
5395 | } else { | |
5396 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5397 | stripe_addr = sector_nr; | |
5398 | } | |
5399 | ||
c8f517c4 N |
5400 | /* 'writepos' is the most advanced device address we might write. |
5401 | * 'readpos' is the least advanced device address we might read. | |
5402 | * 'safepos' is the least address recorded in the metadata as having | |
5403 | * been reshaped. | |
b5254dd5 N |
5404 | * If there is a min_offset_diff, these are adjusted either by |
5405 | * increasing the safepos/readpos if diff is negative, or | |
5406 | * increasing writepos if diff is positive. | |
5407 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5408 | * ensure safety in the face of a crash - that must be done by userspace |
5409 | * making a backup of the data. So in that case there is no particular | |
5410 | * rush to update metadata. | |
5411 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5412 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5413 | * we can be safe in the event of a crash. | |
5414 | * So we insist on updating metadata if safepos is behind writepos and | |
5415 | * readpos is beyond writepos. | |
5416 | * In any case, update the metadata every 10 seconds. | |
5417 | * Maybe that number should be configurable, but I'm not sure it is | |
5418 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5419 | */ | |
b5254dd5 N |
5420 | if (conf->min_offset_diff < 0) { |
5421 | safepos += -conf->min_offset_diff; | |
5422 | readpos += -conf->min_offset_diff; | |
5423 | } else | |
5424 | writepos += conf->min_offset_diff; | |
5425 | ||
2c810cdd | 5426 | if ((mddev->reshape_backwards |
c8f517c4 N |
5427 | ? (safepos > writepos && readpos < writepos) |
5428 | : (safepos < writepos && readpos > writepos)) || | |
5429 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5430 | /* Cannot proceed until we've updated the superblock... */ |
5431 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5432 | atomic_read(&conf->reshape_stripes)==0 |
5433 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5434 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5435 | return 0; | |
fef9c61f | 5436 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5437 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5438 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 5439 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 5440 | md_wakeup_thread(mddev->thread); |
850b2b42 | 5441 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
c91abf5a N |
5442 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5443 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5444 | return 0; | |
52c03291 | 5445 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5446 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5447 | spin_unlock_irq(&conf->device_lock); |
5448 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5449 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5450 | } |
5451 | ||
ab69ae12 | 5452 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5453 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5454 | int j; |
a9f326eb | 5455 | int skipped_disk = 0; |
a8c906ca | 5456 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5457 | set_bit(STRIPE_EXPANDING, &sh->state); |
5458 | atomic_inc(&conf->reshape_stripes); | |
5459 | /* If any of this stripe is beyond the end of the old | |
5460 | * array, then we need to zero those blocks | |
5461 | */ | |
5462 | for (j=sh->disks; j--;) { | |
5463 | sector_t s; | |
5464 | if (j == sh->pd_idx) | |
5465 | continue; | |
f416885e | 5466 | if (conf->level == 6 && |
d0dabf7e | 5467 | j == sh->qd_idx) |
f416885e | 5468 | continue; |
784052ec | 5469 | s = compute_blocknr(sh, j, 0); |
b522adcd | 5470 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5471 | skipped_disk = 1; |
52c03291 N |
5472 | continue; |
5473 | } | |
5474 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5475 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5476 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5477 | } | |
a9f326eb | 5478 | if (!skipped_disk) { |
52c03291 N |
5479 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5480 | set_bit(STRIPE_HANDLE, &sh->state); | |
5481 | } | |
ab69ae12 | 5482 | list_add(&sh->lru, &stripes); |
52c03291 N |
5483 | } |
5484 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5485 | if (mddev->reshape_backwards) |
7a661381 | 5486 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5487 | else |
7a661381 | 5488 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5489 | spin_unlock_irq(&conf->device_lock); |
5490 | /* Ok, those stripe are ready. We can start scheduling | |
5491 | * reads on the source stripes. | |
5492 | * The source stripes are determined by mapping the first and last | |
5493 | * block on the destination stripes. | |
5494 | */ | |
52c03291 | 5495 | first_sector = |
ec32a2bd | 5496 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5497 | 1, &dd_idx, NULL); |
52c03291 | 5498 | last_sector = |
0e6e0271 | 5499 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5500 | * new_data_disks - 1), |
911d4ee8 | 5501 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5502 | if (last_sector >= mddev->dev_sectors) |
5503 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5504 | while (first_sector <= last_sector) { |
a8c906ca | 5505 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5506 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5507 | set_bit(STRIPE_HANDLE, &sh->state); | |
5508 | release_stripe(sh); | |
5509 | first_sector += STRIPE_SECTORS; | |
5510 | } | |
ab69ae12 N |
5511 | /* Now that the sources are clearly marked, we can release |
5512 | * the destination stripes | |
5513 | */ | |
5514 | while (!list_empty(&stripes)) { | |
5515 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5516 | list_del_init(&sh->lru); | |
5517 | release_stripe(sh); | |
5518 | } | |
c6207277 N |
5519 | /* If this takes us to the resync_max point where we have to pause, |
5520 | * then we need to write out the superblock. | |
5521 | */ | |
7a661381 | 5522 | sector_nr += reshape_sectors; |
92140480 N |
5523 | retn = reshape_sectors; |
5524 | finish: | |
c5e19d90 N |
5525 | if (mddev->curr_resync_completed > mddev->resync_max || |
5526 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5527 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5528 | /* Cannot proceed until we've updated the superblock... */ |
5529 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5530 | atomic_read(&conf->reshape_stripes) == 0 |
5531 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5532 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5533 | goto ret; | |
fef9c61f | 5534 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5535 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5536 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
5537 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
5538 | md_wakeup_thread(mddev->thread); | |
5539 | wait_event(mddev->sb_wait, | |
5540 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
c91abf5a N |
5541 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5542 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5543 | goto ret; | |
c6207277 | 5544 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5545 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5546 | spin_unlock_irq(&conf->device_lock); |
5547 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5548 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5549 | } |
c91abf5a | 5550 | ret: |
92140480 | 5551 | return retn; |
52c03291 N |
5552 | } |
5553 | ||
09314799 | 5554 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
52c03291 | 5555 | { |
d1688a6d | 5556 | struct r5conf *conf = mddev->private; |
52c03291 | 5557 | struct stripe_head *sh; |
58c0fed4 | 5558 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5559 | sector_t sync_blocks; |
16a53ecc N |
5560 | int still_degraded = 0; |
5561 | int i; | |
1da177e4 | 5562 | |
72626685 | 5563 | if (sector_nr >= max_sector) { |
1da177e4 | 5564 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5565 | |
29269553 N |
5566 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5567 | end_reshape(conf); | |
5568 | return 0; | |
5569 | } | |
72626685 N |
5570 | |
5571 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5572 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5573 | &sync_blocks, 1); | |
16a53ecc | 5574 | else /* completed sync */ |
72626685 N |
5575 | conf->fullsync = 0; |
5576 | bitmap_close_sync(mddev->bitmap); | |
5577 | ||
1da177e4 LT |
5578 | return 0; |
5579 | } | |
ccfcc3c1 | 5580 | |
64bd660b N |
5581 | /* Allow raid5_quiesce to complete */ |
5582 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5583 | ||
52c03291 N |
5584 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5585 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5586 | |
c6207277 N |
5587 | /* No need to check resync_max as we never do more than one |
5588 | * stripe, and as resync_max will always be on a chunk boundary, | |
5589 | * if the check in md_do_sync didn't fire, there is no chance | |
5590 | * of overstepping resync_max here | |
5591 | */ | |
5592 | ||
16a53ecc | 5593 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5594 | * to resync, then assert that we are finished, because there is |
5595 | * nothing we can do. | |
5596 | */ | |
3285edf1 | 5597 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5598 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5599 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5600 | *skipped = 1; |
1da177e4 LT |
5601 | return rv; |
5602 | } | |
6f608040 | 5603 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5604 | !conf->fullsync && | |
5605 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5606 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5607 | /* we can skip this block, and probably more */ |
5608 | sync_blocks /= STRIPE_SECTORS; | |
5609 | *skipped = 1; | |
5610 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5611 | } | |
1da177e4 | 5612 | |
b47490c9 N |
5613 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
5614 | ||
a8c906ca | 5615 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5616 | if (sh == NULL) { |
a8c906ca | 5617 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5618 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5619 | * is trying to get access |
1da177e4 | 5620 | */ |
66c006a5 | 5621 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5622 | } |
16a53ecc | 5623 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5624 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5625 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5626 | */ |
16d9cfab EM |
5627 | rcu_read_lock(); |
5628 | for (i = 0; i < conf->raid_disks; i++) { | |
5629 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5630 | ||
5631 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5632 | still_degraded = 1; |
16d9cfab EM |
5633 | } |
5634 | rcu_read_unlock(); | |
16a53ecc N |
5635 | |
5636 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5637 | ||
83206d66 | 5638 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5639 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5640 | |
1da177e4 LT |
5641 | release_stripe(sh); |
5642 | ||
5643 | return STRIPE_SECTORS; | |
5644 | } | |
5645 | ||
d1688a6d | 5646 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5647 | { |
5648 | /* We may not be able to submit a whole bio at once as there | |
5649 | * may not be enough stripe_heads available. | |
5650 | * We cannot pre-allocate enough stripe_heads as we may need | |
5651 | * more than exist in the cache (if we allow ever large chunks). | |
5652 | * So we do one stripe head at a time and record in | |
5653 | * ->bi_hw_segments how many have been done. | |
5654 | * | |
5655 | * We *know* that this entire raid_bio is in one chunk, so | |
5656 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5657 | */ | |
5658 | struct stripe_head *sh; | |
911d4ee8 | 5659 | int dd_idx; |
46031f9a RBJ |
5660 | sector_t sector, logical_sector, last_sector; |
5661 | int scnt = 0; | |
5662 | int remaining; | |
5663 | int handled = 0; | |
5664 | ||
4f024f37 KO |
5665 | logical_sector = raid_bio->bi_iter.bi_sector & |
5666 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5667 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5668 | 0, &dd_idx, NULL); |
f73a1c7d | 5669 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5670 | |
5671 | for (; logical_sector < last_sector; | |
387bb173 NB |
5672 | logical_sector += STRIPE_SECTORS, |
5673 | sector += STRIPE_SECTORS, | |
5674 | scnt++) { | |
46031f9a | 5675 | |
e7836bd6 | 5676 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5677 | /* already done this stripe */ |
5678 | continue; | |
5679 | ||
2844dc32 | 5680 | sh = get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5681 | |
5682 | if (!sh) { | |
5683 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5684 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5685 | conf->retry_read_aligned = raid_bio; |
5686 | return handled; | |
5687 | } | |
5688 | ||
da41ba65 | 5689 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
387bb173 | 5690 | release_stripe(sh); |
e7836bd6 | 5691 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
5692 | conf->retry_read_aligned = raid_bio; |
5693 | return handled; | |
5694 | } | |
5695 | ||
3f9e7c14 | 5696 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 5697 | handle_stripe(sh); |
46031f9a RBJ |
5698 | release_stripe(sh); |
5699 | handled++; | |
5700 | } | |
e7836bd6 | 5701 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
5702 | if (remaining == 0) { |
5703 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
5704 | raid_bio, 0); | |
4246a0b6 | 5705 | bio_endio(raid_bio); |
0a82a8d1 | 5706 | } |
46031f9a | 5707 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5708 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
5709 | return handled; |
5710 | } | |
5711 | ||
bfc90cb0 | 5712 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
5713 | struct r5worker *worker, |
5714 | struct list_head *temp_inactive_list) | |
46a06401 SL |
5715 | { |
5716 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
5717 | int i, batch_size = 0, hash; |
5718 | bool release_inactive = false; | |
46a06401 SL |
5719 | |
5720 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 5721 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
5722 | batch[batch_size++] = sh; |
5723 | ||
566c09c5 SL |
5724 | if (batch_size == 0) { |
5725 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5726 | if (!list_empty(temp_inactive_list + i)) | |
5727 | break; | |
5728 | if (i == NR_STRIPE_HASH_LOCKS) | |
5729 | return batch_size; | |
5730 | release_inactive = true; | |
5731 | } | |
46a06401 SL |
5732 | spin_unlock_irq(&conf->device_lock); |
5733 | ||
566c09c5 SL |
5734 | release_inactive_stripe_list(conf, temp_inactive_list, |
5735 | NR_STRIPE_HASH_LOCKS); | |
5736 | ||
5737 | if (release_inactive) { | |
5738 | spin_lock_irq(&conf->device_lock); | |
5739 | return 0; | |
5740 | } | |
5741 | ||
46a06401 SL |
5742 | for (i = 0; i < batch_size; i++) |
5743 | handle_stripe(batch[i]); | |
5744 | ||
5745 | cond_resched(); | |
5746 | ||
5747 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
5748 | for (i = 0; i < batch_size; i++) { |
5749 | hash = batch[i]->hash_lock_index; | |
5750 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
5751 | } | |
46a06401 SL |
5752 | return batch_size; |
5753 | } | |
46031f9a | 5754 | |
851c30c9 SL |
5755 | static void raid5_do_work(struct work_struct *work) |
5756 | { | |
5757 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
5758 | struct r5worker_group *group = worker->group; | |
5759 | struct r5conf *conf = group->conf; | |
5760 | int group_id = group - conf->worker_groups; | |
5761 | int handled; | |
5762 | struct blk_plug plug; | |
5763 | ||
5764 | pr_debug("+++ raid5worker active\n"); | |
5765 | ||
5766 | blk_start_plug(&plug); | |
5767 | handled = 0; | |
5768 | spin_lock_irq(&conf->device_lock); | |
5769 | while (1) { | |
5770 | int batch_size, released; | |
5771 | ||
566c09c5 | 5772 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 5773 | |
566c09c5 SL |
5774 | batch_size = handle_active_stripes(conf, group_id, worker, |
5775 | worker->temp_inactive_list); | |
bfc90cb0 | 5776 | worker->working = false; |
851c30c9 SL |
5777 | if (!batch_size && !released) |
5778 | break; | |
5779 | handled += batch_size; | |
5780 | } | |
5781 | pr_debug("%d stripes handled\n", handled); | |
5782 | ||
5783 | spin_unlock_irq(&conf->device_lock); | |
5784 | blk_finish_plug(&plug); | |
5785 | ||
5786 | pr_debug("--- raid5worker inactive\n"); | |
5787 | } | |
5788 | ||
1da177e4 LT |
5789 | /* |
5790 | * This is our raid5 kernel thread. | |
5791 | * | |
5792 | * We scan the hash table for stripes which can be handled now. | |
5793 | * During the scan, completed stripes are saved for us by the interrupt | |
5794 | * handler, so that they will not have to wait for our next wakeup. | |
5795 | */ | |
4ed8731d | 5796 | static void raid5d(struct md_thread *thread) |
1da177e4 | 5797 | { |
4ed8731d | 5798 | struct mddev *mddev = thread->mddev; |
d1688a6d | 5799 | struct r5conf *conf = mddev->private; |
1da177e4 | 5800 | int handled; |
e1dfa0a2 | 5801 | struct blk_plug plug; |
1da177e4 | 5802 | |
45b4233c | 5803 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
5804 | |
5805 | md_check_recovery(mddev); | |
1da177e4 | 5806 | |
c3cce6cd N |
5807 | if (!bio_list_empty(&conf->return_bi) && |
5808 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
5809 | struct bio_list tmp = BIO_EMPTY_LIST; | |
5810 | spin_lock_irq(&conf->device_lock); | |
5811 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
5812 | bio_list_merge(&tmp, &conf->return_bi); | |
5813 | bio_list_init(&conf->return_bi); | |
5814 | } | |
5815 | spin_unlock_irq(&conf->device_lock); | |
5816 | return_io(&tmp); | |
5817 | } | |
5818 | ||
e1dfa0a2 | 5819 | blk_start_plug(&plug); |
1da177e4 LT |
5820 | handled = 0; |
5821 | spin_lock_irq(&conf->device_lock); | |
5822 | while (1) { | |
46031f9a | 5823 | struct bio *bio; |
773ca82f SL |
5824 | int batch_size, released; |
5825 | ||
566c09c5 | 5826 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
5827 | if (released) |
5828 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 5829 | |
0021b7bc | 5830 | if ( |
7c13edc8 N |
5831 | !list_empty(&conf->bitmap_list)) { |
5832 | /* Now is a good time to flush some bitmap updates */ | |
5833 | conf->seq_flush++; | |
700e432d | 5834 | spin_unlock_irq(&conf->device_lock); |
72626685 | 5835 | bitmap_unplug(mddev->bitmap); |
700e432d | 5836 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 5837 | conf->seq_write = conf->seq_flush; |
566c09c5 | 5838 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 5839 | } |
0021b7bc | 5840 | raid5_activate_delayed(conf); |
72626685 | 5841 | |
46031f9a RBJ |
5842 | while ((bio = remove_bio_from_retry(conf))) { |
5843 | int ok; | |
5844 | spin_unlock_irq(&conf->device_lock); | |
5845 | ok = retry_aligned_read(conf, bio); | |
5846 | spin_lock_irq(&conf->device_lock); | |
5847 | if (!ok) | |
5848 | break; | |
5849 | handled++; | |
5850 | } | |
5851 | ||
566c09c5 SL |
5852 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
5853 | conf->temp_inactive_list); | |
773ca82f | 5854 | if (!batch_size && !released) |
1da177e4 | 5855 | break; |
46a06401 | 5856 | handled += batch_size; |
1da177e4 | 5857 | |
46a06401 SL |
5858 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
5859 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 5860 | md_check_recovery(mddev); |
46a06401 SL |
5861 | spin_lock_irq(&conf->device_lock); |
5862 | } | |
1da177e4 | 5863 | } |
45b4233c | 5864 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
5865 | |
5866 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
5867 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
5868 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
5869 | grow_one_stripe(conf, __GFP_NOWARN); |
5870 | /* Set flag even if allocation failed. This helps | |
5871 | * slow down allocation requests when mem is short | |
5872 | */ | |
5873 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 5874 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 5875 | } |
1da177e4 | 5876 | |
c9f21aaf | 5877 | async_tx_issue_pending_all(); |
e1dfa0a2 | 5878 | blk_finish_plug(&plug); |
1da177e4 | 5879 | |
45b4233c | 5880 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
5881 | } |
5882 | ||
3f294f4f | 5883 | static ssize_t |
fd01b88c | 5884 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 5885 | { |
7b1485ba N |
5886 | struct r5conf *conf; |
5887 | int ret = 0; | |
5888 | spin_lock(&mddev->lock); | |
5889 | conf = mddev->private; | |
96de1e66 | 5890 | if (conf) |
edbe83ab | 5891 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
5892 | spin_unlock(&mddev->lock); |
5893 | return ret; | |
3f294f4f N |
5894 | } |
5895 | ||
c41d4ac4 | 5896 | int |
fd01b88c | 5897 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 5898 | { |
d1688a6d | 5899 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
5900 | int err; |
5901 | ||
c41d4ac4 | 5902 | if (size <= 16 || size > 32768) |
3f294f4f | 5903 | return -EINVAL; |
486f0644 | 5904 | |
edbe83ab | 5905 | conf->min_nr_stripes = size; |
2d5b569b | 5906 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5907 | while (size < conf->max_nr_stripes && |
5908 | drop_one_stripe(conf)) | |
5909 | ; | |
2d5b569b | 5910 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5911 | |
edbe83ab | 5912 | |
b5470dc5 DW |
5913 | err = md_allow_write(mddev); |
5914 | if (err) | |
5915 | return err; | |
486f0644 | 5916 | |
2d5b569b | 5917 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5918 | while (size > conf->max_nr_stripes) |
5919 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
5920 | break; | |
2d5b569b | 5921 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5922 | |
c41d4ac4 N |
5923 | return 0; |
5924 | } | |
5925 | EXPORT_SYMBOL(raid5_set_cache_size); | |
5926 | ||
5927 | static ssize_t | |
fd01b88c | 5928 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 5929 | { |
6791875e | 5930 | struct r5conf *conf; |
c41d4ac4 N |
5931 | unsigned long new; |
5932 | int err; | |
5933 | ||
5934 | if (len >= PAGE_SIZE) | |
5935 | return -EINVAL; | |
b29bebd6 | 5936 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 5937 | return -EINVAL; |
6791875e | 5938 | err = mddev_lock(mddev); |
c41d4ac4 N |
5939 | if (err) |
5940 | return err; | |
6791875e N |
5941 | conf = mddev->private; |
5942 | if (!conf) | |
5943 | err = -ENODEV; | |
5944 | else | |
5945 | err = raid5_set_cache_size(mddev, new); | |
5946 | mddev_unlock(mddev); | |
5947 | ||
5948 | return err ?: len; | |
3f294f4f | 5949 | } |
007583c9 | 5950 | |
96de1e66 N |
5951 | static struct md_sysfs_entry |
5952 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
5953 | raid5_show_stripe_cache_size, | |
5954 | raid5_store_stripe_cache_size); | |
3f294f4f | 5955 | |
d06f191f MS |
5956 | static ssize_t |
5957 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
5958 | { | |
5959 | struct r5conf *conf = mddev->private; | |
5960 | if (conf) | |
5961 | return sprintf(page, "%d\n", conf->rmw_level); | |
5962 | else | |
5963 | return 0; | |
5964 | } | |
5965 | ||
5966 | static ssize_t | |
5967 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
5968 | { | |
5969 | struct r5conf *conf = mddev->private; | |
5970 | unsigned long new; | |
5971 | ||
5972 | if (!conf) | |
5973 | return -ENODEV; | |
5974 | ||
5975 | if (len >= PAGE_SIZE) | |
5976 | return -EINVAL; | |
5977 | ||
5978 | if (kstrtoul(page, 10, &new)) | |
5979 | return -EINVAL; | |
5980 | ||
5981 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
5982 | return -EINVAL; | |
5983 | ||
5984 | if (new != PARITY_DISABLE_RMW && | |
5985 | new != PARITY_ENABLE_RMW && | |
5986 | new != PARITY_PREFER_RMW) | |
5987 | return -EINVAL; | |
5988 | ||
5989 | conf->rmw_level = new; | |
5990 | return len; | |
5991 | } | |
5992 | ||
5993 | static struct md_sysfs_entry | |
5994 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
5995 | raid5_show_rmw_level, | |
5996 | raid5_store_rmw_level); | |
5997 | ||
5998 | ||
8b3e6cdc | 5999 | static ssize_t |
fd01b88c | 6000 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6001 | { |
7b1485ba N |
6002 | struct r5conf *conf; |
6003 | int ret = 0; | |
6004 | spin_lock(&mddev->lock); | |
6005 | conf = mddev->private; | |
8b3e6cdc | 6006 | if (conf) |
7b1485ba N |
6007 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6008 | spin_unlock(&mddev->lock); | |
6009 | return ret; | |
8b3e6cdc DW |
6010 | } |
6011 | ||
6012 | static ssize_t | |
fd01b88c | 6013 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6014 | { |
6791875e | 6015 | struct r5conf *conf; |
4ef197d8 | 6016 | unsigned long new; |
6791875e N |
6017 | int err; |
6018 | ||
8b3e6cdc DW |
6019 | if (len >= PAGE_SIZE) |
6020 | return -EINVAL; | |
b29bebd6 | 6021 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6022 | return -EINVAL; |
6791875e N |
6023 | |
6024 | err = mddev_lock(mddev); | |
6025 | if (err) | |
6026 | return err; | |
6027 | conf = mddev->private; | |
6028 | if (!conf) | |
6029 | err = -ENODEV; | |
edbe83ab | 6030 | else if (new > conf->min_nr_stripes) |
6791875e N |
6031 | err = -EINVAL; |
6032 | else | |
6033 | conf->bypass_threshold = new; | |
6034 | mddev_unlock(mddev); | |
6035 | return err ?: len; | |
8b3e6cdc DW |
6036 | } |
6037 | ||
6038 | static struct md_sysfs_entry | |
6039 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6040 | S_IRUGO | S_IWUSR, | |
6041 | raid5_show_preread_threshold, | |
6042 | raid5_store_preread_threshold); | |
6043 | ||
d592a996 SL |
6044 | static ssize_t |
6045 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6046 | { | |
7b1485ba N |
6047 | struct r5conf *conf; |
6048 | int ret = 0; | |
6049 | spin_lock(&mddev->lock); | |
6050 | conf = mddev->private; | |
d592a996 | 6051 | if (conf) |
7b1485ba N |
6052 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6053 | spin_unlock(&mddev->lock); | |
6054 | return ret; | |
d592a996 SL |
6055 | } |
6056 | ||
6057 | static ssize_t | |
6058 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6059 | { | |
6791875e | 6060 | struct r5conf *conf; |
d592a996 | 6061 | unsigned long new; |
6791875e N |
6062 | int err; |
6063 | ||
d592a996 SL |
6064 | if (len >= PAGE_SIZE) |
6065 | return -EINVAL; | |
d592a996 SL |
6066 | if (kstrtoul(page, 10, &new)) |
6067 | return -EINVAL; | |
6068 | new = !!new; | |
6791875e N |
6069 | |
6070 | err = mddev_lock(mddev); | |
6071 | if (err) | |
6072 | return err; | |
6073 | conf = mddev->private; | |
6074 | if (!conf) | |
6075 | err = -ENODEV; | |
6076 | else if (new != conf->skip_copy) { | |
6077 | mddev_suspend(mddev); | |
6078 | conf->skip_copy = new; | |
6079 | if (new) | |
6080 | mddev->queue->backing_dev_info.capabilities |= | |
6081 | BDI_CAP_STABLE_WRITES; | |
6082 | else | |
6083 | mddev->queue->backing_dev_info.capabilities &= | |
6084 | ~BDI_CAP_STABLE_WRITES; | |
6085 | mddev_resume(mddev); | |
6086 | } | |
6087 | mddev_unlock(mddev); | |
6088 | return err ?: len; | |
d592a996 SL |
6089 | } |
6090 | ||
6091 | static struct md_sysfs_entry | |
6092 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6093 | raid5_show_skip_copy, | |
6094 | raid5_store_skip_copy); | |
6095 | ||
3f294f4f | 6096 | static ssize_t |
fd01b88c | 6097 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6098 | { |
d1688a6d | 6099 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6100 | if (conf) |
6101 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6102 | else | |
6103 | return 0; | |
3f294f4f N |
6104 | } |
6105 | ||
96de1e66 N |
6106 | static struct md_sysfs_entry |
6107 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6108 | |
b721420e SL |
6109 | static ssize_t |
6110 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6111 | { | |
7b1485ba N |
6112 | struct r5conf *conf; |
6113 | int ret = 0; | |
6114 | spin_lock(&mddev->lock); | |
6115 | conf = mddev->private; | |
b721420e | 6116 | if (conf) |
7b1485ba N |
6117 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6118 | spin_unlock(&mddev->lock); | |
6119 | return ret; | |
b721420e SL |
6120 | } |
6121 | ||
60aaf933 | 6122 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6123 | int *group_cnt, | |
6124 | int *worker_cnt_per_group, | |
6125 | struct r5worker_group **worker_groups); | |
b721420e SL |
6126 | static ssize_t |
6127 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6128 | { | |
6791875e | 6129 | struct r5conf *conf; |
b721420e SL |
6130 | unsigned long new; |
6131 | int err; | |
60aaf933 | 6132 | struct r5worker_group *new_groups, *old_groups; |
6133 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6134 | |
6135 | if (len >= PAGE_SIZE) | |
6136 | return -EINVAL; | |
b721420e SL |
6137 | if (kstrtoul(page, 10, &new)) |
6138 | return -EINVAL; | |
6139 | ||
6791875e N |
6140 | err = mddev_lock(mddev); |
6141 | if (err) | |
6142 | return err; | |
6143 | conf = mddev->private; | |
6144 | if (!conf) | |
6145 | err = -ENODEV; | |
6146 | else if (new != conf->worker_cnt_per_group) { | |
6147 | mddev_suspend(mddev); | |
b721420e | 6148 | |
6791875e N |
6149 | old_groups = conf->worker_groups; |
6150 | if (old_groups) | |
6151 | flush_workqueue(raid5_wq); | |
d206dcfa | 6152 | |
6791875e N |
6153 | err = alloc_thread_groups(conf, new, |
6154 | &group_cnt, &worker_cnt_per_group, | |
6155 | &new_groups); | |
6156 | if (!err) { | |
6157 | spin_lock_irq(&conf->device_lock); | |
6158 | conf->group_cnt = group_cnt; | |
6159 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6160 | conf->worker_groups = new_groups; | |
6161 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6162 | |
6791875e N |
6163 | if (old_groups) |
6164 | kfree(old_groups[0].workers); | |
6165 | kfree(old_groups); | |
6166 | } | |
6167 | mddev_resume(mddev); | |
b721420e | 6168 | } |
6791875e | 6169 | mddev_unlock(mddev); |
b721420e | 6170 | |
6791875e | 6171 | return err ?: len; |
b721420e SL |
6172 | } |
6173 | ||
6174 | static struct md_sysfs_entry | |
6175 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6176 | raid5_show_group_thread_cnt, | |
6177 | raid5_store_group_thread_cnt); | |
6178 | ||
007583c9 | 6179 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6180 | &raid5_stripecache_size.attr, |
6181 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6182 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6183 | &raid5_group_thread_cnt.attr, |
d592a996 | 6184 | &raid5_skip_copy.attr, |
d06f191f | 6185 | &raid5_rmw_level.attr, |
3f294f4f N |
6186 | NULL, |
6187 | }; | |
007583c9 N |
6188 | static struct attribute_group raid5_attrs_group = { |
6189 | .name = NULL, | |
6190 | .attrs = raid5_attrs, | |
3f294f4f N |
6191 | }; |
6192 | ||
60aaf933 | 6193 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6194 | int *group_cnt, | |
6195 | int *worker_cnt_per_group, | |
6196 | struct r5worker_group **worker_groups) | |
851c30c9 | 6197 | { |
566c09c5 | 6198 | int i, j, k; |
851c30c9 SL |
6199 | ssize_t size; |
6200 | struct r5worker *workers; | |
6201 | ||
60aaf933 | 6202 | *worker_cnt_per_group = cnt; |
851c30c9 | 6203 | if (cnt == 0) { |
60aaf933 | 6204 | *group_cnt = 0; |
6205 | *worker_groups = NULL; | |
851c30c9 SL |
6206 | return 0; |
6207 | } | |
60aaf933 | 6208 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6209 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6210 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6211 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6212 | *group_cnt, GFP_NOIO); | |
6213 | if (!*worker_groups || !workers) { | |
851c30c9 | 6214 | kfree(workers); |
60aaf933 | 6215 | kfree(*worker_groups); |
851c30c9 SL |
6216 | return -ENOMEM; |
6217 | } | |
6218 | ||
60aaf933 | 6219 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6220 | struct r5worker_group *group; |
6221 | ||
0c775d52 | 6222 | group = &(*worker_groups)[i]; |
851c30c9 SL |
6223 | INIT_LIST_HEAD(&group->handle_list); |
6224 | group->conf = conf; | |
6225 | group->workers = workers + i * cnt; | |
6226 | ||
6227 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6228 | struct r5worker *worker = group->workers + j; |
6229 | worker->group = group; | |
6230 | INIT_WORK(&worker->work, raid5_do_work); | |
6231 | ||
6232 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6233 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6234 | } |
6235 | } | |
6236 | ||
6237 | return 0; | |
6238 | } | |
6239 | ||
6240 | static void free_thread_groups(struct r5conf *conf) | |
6241 | { | |
6242 | if (conf->worker_groups) | |
6243 | kfree(conf->worker_groups[0].workers); | |
6244 | kfree(conf->worker_groups); | |
6245 | conf->worker_groups = NULL; | |
6246 | } | |
6247 | ||
80c3a6ce | 6248 | static sector_t |
fd01b88c | 6249 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6250 | { |
d1688a6d | 6251 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6252 | |
6253 | if (!sectors) | |
6254 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6255 | if (!raid_disks) |
7ec05478 | 6256 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6257 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6258 | |
3cb5edf4 N |
6259 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6260 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6261 | return sectors * (raid_disks - conf->max_degraded); |
6262 | } | |
6263 | ||
789b5e03 ON |
6264 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6265 | { | |
6266 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6267 | if (percpu->scribble) |
6268 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6269 | percpu->spare_page = NULL; |
6270 | percpu->scribble = NULL; | |
6271 | } | |
6272 | ||
6273 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6274 | { | |
6275 | if (conf->level == 6 && !percpu->spare_page) | |
6276 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6277 | if (!percpu->scribble) | |
46d5b785 | 6278 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6279 | conf->previous_raid_disks), |
6280 | max(conf->chunk_sectors, | |
6281 | conf->prev_chunk_sectors) | |
6282 | / STRIPE_SECTORS, | |
6283 | GFP_KERNEL); | |
789b5e03 ON |
6284 | |
6285 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6286 | free_scratch_buffer(conf, percpu); | |
6287 | return -ENOMEM; | |
6288 | } | |
6289 | ||
6290 | return 0; | |
6291 | } | |
6292 | ||
d1688a6d | 6293 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 | 6294 | { |
36d1c647 DW |
6295 | unsigned long cpu; |
6296 | ||
6297 | if (!conf->percpu) | |
6298 | return; | |
6299 | ||
36d1c647 DW |
6300 | #ifdef CONFIG_HOTPLUG_CPU |
6301 | unregister_cpu_notifier(&conf->cpu_notify); | |
6302 | #endif | |
789b5e03 ON |
6303 | |
6304 | get_online_cpus(); | |
6305 | for_each_possible_cpu(cpu) | |
6306 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
36d1c647 DW |
6307 | put_online_cpus(); |
6308 | ||
6309 | free_percpu(conf->percpu); | |
6310 | } | |
6311 | ||
d1688a6d | 6312 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6313 | { |
edbe83ab N |
6314 | if (conf->shrinker.seeks) |
6315 | unregister_shrinker(&conf->shrinker); | |
851c30c9 | 6316 | free_thread_groups(conf); |
95fc17aa | 6317 | shrink_stripes(conf); |
36d1c647 | 6318 | raid5_free_percpu(conf); |
95fc17aa DW |
6319 | kfree(conf->disks); |
6320 | kfree(conf->stripe_hashtbl); | |
6321 | kfree(conf); | |
6322 | } | |
6323 | ||
36d1c647 DW |
6324 | #ifdef CONFIG_HOTPLUG_CPU |
6325 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
6326 | void *hcpu) | |
6327 | { | |
d1688a6d | 6328 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
6329 | long cpu = (long)hcpu; |
6330 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
6331 | ||
6332 | switch (action) { | |
6333 | case CPU_UP_PREPARE: | |
6334 | case CPU_UP_PREPARE_FROZEN: | |
789b5e03 | 6335 | if (alloc_scratch_buffer(conf, percpu)) { |
36d1c647 DW |
6336 | pr_err("%s: failed memory allocation for cpu%ld\n", |
6337 | __func__, cpu); | |
55af6bb5 | 6338 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
6339 | } |
6340 | break; | |
6341 | case CPU_DEAD: | |
6342 | case CPU_DEAD_FROZEN: | |
789b5e03 | 6343 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
36d1c647 DW |
6344 | break; |
6345 | default: | |
6346 | break; | |
6347 | } | |
6348 | return NOTIFY_OK; | |
6349 | } | |
6350 | #endif | |
6351 | ||
d1688a6d | 6352 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
6353 | { |
6354 | unsigned long cpu; | |
789b5e03 | 6355 | int err = 0; |
36d1c647 | 6356 | |
789b5e03 ON |
6357 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6358 | if (!conf->percpu) | |
36d1c647 | 6359 | return -ENOMEM; |
789b5e03 ON |
6360 | |
6361 | #ifdef CONFIG_HOTPLUG_CPU | |
6362 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
6363 | conf->cpu_notify.priority = 0; | |
6364 | err = register_cpu_notifier(&conf->cpu_notify); | |
6365 | if (err) | |
6366 | return err; | |
6367 | #endif | |
36d1c647 DW |
6368 | |
6369 | get_online_cpus(); | |
36d1c647 | 6370 | for_each_present_cpu(cpu) { |
789b5e03 ON |
6371 | err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
6372 | if (err) { | |
6373 | pr_err("%s: failed memory allocation for cpu%ld\n", | |
6374 | __func__, cpu); | |
36d1c647 DW |
6375 | break; |
6376 | } | |
36d1c647 | 6377 | } |
36d1c647 DW |
6378 | put_online_cpus(); |
6379 | ||
6380 | return err; | |
6381 | } | |
6382 | ||
edbe83ab N |
6383 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6384 | struct shrink_control *sc) | |
6385 | { | |
6386 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6387 | unsigned long ret = SHRINK_STOP; |
6388 | ||
6389 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6390 | ret= 0; | |
49895bcc N |
6391 | while (ret < sc->nr_to_scan && |
6392 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6393 | if (drop_one_stripe(conf) == 0) { |
6394 | ret = SHRINK_STOP; | |
6395 | break; | |
6396 | } | |
6397 | ret++; | |
6398 | } | |
6399 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6400 | } |
6401 | return ret; | |
6402 | } | |
6403 | ||
6404 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6405 | struct shrink_control *sc) | |
6406 | { | |
6407 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6408 | ||
6409 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6410 | /* unlikely, but not impossible */ | |
6411 | return 0; | |
6412 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6413 | } | |
6414 | ||
d1688a6d | 6415 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6416 | { |
d1688a6d | 6417 | struct r5conf *conf; |
5e5e3e78 | 6418 | int raid_disk, memory, max_disks; |
3cb03002 | 6419 | struct md_rdev *rdev; |
1da177e4 | 6420 | struct disk_info *disk; |
0232605d | 6421 | char pers_name[6]; |
566c09c5 | 6422 | int i; |
60aaf933 | 6423 | int group_cnt, worker_cnt_per_group; |
6424 | struct r5worker_group *new_group; | |
1da177e4 | 6425 | |
91adb564 N |
6426 | if (mddev->new_level != 5 |
6427 | && mddev->new_level != 4 | |
6428 | && mddev->new_level != 6) { | |
0c55e022 | 6429 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
6430 | mdname(mddev), mddev->new_level); |
6431 | return ERR_PTR(-EIO); | |
1da177e4 | 6432 | } |
91adb564 N |
6433 | if ((mddev->new_level == 5 |
6434 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6435 | (mddev->new_level == 6 | |
6436 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 6437 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
6438 | mdname(mddev), mddev->new_layout); |
6439 | return ERR_PTR(-EIO); | |
99c0fb5f | 6440 | } |
91adb564 | 6441 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 6442 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
6443 | mdname(mddev), mddev->raid_disks); |
6444 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
6445 | } |
6446 | ||
664e7c41 AN |
6447 | if (!mddev->new_chunk_sectors || |
6448 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6449 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
6450 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
6451 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6452 | return ERR_PTR(-EINVAL); |
f6705578 N |
6453 | } |
6454 | ||
d1688a6d | 6455 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6456 | if (conf == NULL) |
1da177e4 | 6457 | goto abort; |
851c30c9 | 6458 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6459 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6460 | &new_group)) { | |
6461 | conf->group_cnt = group_cnt; | |
6462 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6463 | conf->worker_groups = new_group; | |
6464 | } else | |
851c30c9 | 6465 | goto abort; |
f5efd45a | 6466 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6467 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6468 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6469 | init_waitqueue_head(&conf->wait_for_quiescent); |
e9e4c377 YL |
6470 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) { |
6471 | init_waitqueue_head(&conf->wait_for_stripe[i]); | |
6472 | } | |
f5efd45a DW |
6473 | init_waitqueue_head(&conf->wait_for_overlap); |
6474 | INIT_LIST_HEAD(&conf->handle_list); | |
6475 | INIT_LIST_HEAD(&conf->hold_list); | |
6476 | INIT_LIST_HEAD(&conf->delayed_list); | |
6477 | INIT_LIST_HEAD(&conf->bitmap_list); | |
c3cce6cd | 6478 | bio_list_init(&conf->return_bi); |
773ca82f | 6479 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6480 | atomic_set(&conf->active_stripes, 0); |
6481 | atomic_set(&conf->preread_active_stripes, 0); | |
6482 | atomic_set(&conf->active_aligned_reads, 0); | |
6483 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 6484 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6485 | |
6486 | conf->raid_disks = mddev->raid_disks; | |
6487 | if (mddev->reshape_position == MaxSector) | |
6488 | conf->previous_raid_disks = mddev->raid_disks; | |
6489 | else | |
f6705578 | 6490 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6491 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6492 | |
5e5e3e78 | 6493 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
6494 | GFP_KERNEL); |
6495 | if (!conf->disks) | |
6496 | goto abort; | |
9ffae0cf | 6497 | |
1da177e4 LT |
6498 | conf->mddev = mddev; |
6499 | ||
fccddba0 | 6500 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6501 | goto abort; |
1da177e4 | 6502 | |
566c09c5 SL |
6503 | /* We init hash_locks[0] separately to that it can be used |
6504 | * as the reference lock in the spin_lock_nest_lock() call | |
6505 | * in lock_all_device_hash_locks_irq in order to convince | |
6506 | * lockdep that we know what we are doing. | |
6507 | */ | |
6508 | spin_lock_init(conf->hash_locks); | |
6509 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6510 | spin_lock_init(conf->hash_locks + i); | |
6511 | ||
6512 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6513 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6514 | ||
6515 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6516 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6517 | ||
36d1c647 | 6518 | conf->level = mddev->new_level; |
46d5b785 | 6519 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6520 | if (raid5_alloc_percpu(conf) != 0) |
6521 | goto abort; | |
6522 | ||
0c55e022 | 6523 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6524 | |
dafb20fa | 6525 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6526 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6527 | if (raid_disk >= max_disks |
1da177e4 LT |
6528 | || raid_disk < 0) |
6529 | continue; | |
6530 | disk = conf->disks + raid_disk; | |
6531 | ||
17045f52 N |
6532 | if (test_bit(Replacement, &rdev->flags)) { |
6533 | if (disk->replacement) | |
6534 | goto abort; | |
6535 | disk->replacement = rdev; | |
6536 | } else { | |
6537 | if (disk->rdev) | |
6538 | goto abort; | |
6539 | disk->rdev = rdev; | |
6540 | } | |
1da177e4 | 6541 | |
b2d444d7 | 6542 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6543 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
6544 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
6545 | " disk %d\n", | |
6546 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6547 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6548 | /* Cannot rely on bitmap to complete recovery */ |
6549 | conf->fullsync = 1; | |
1da177e4 LT |
6550 | } |
6551 | ||
91adb564 | 6552 | conf->level = mddev->new_level; |
584acdd4 | 6553 | if (conf->level == 6) { |
16a53ecc | 6554 | conf->max_degraded = 2; |
584acdd4 MS |
6555 | if (raid6_call.xor_syndrome) |
6556 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6557 | else | |
6558 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6559 | } else { | |
16a53ecc | 6560 | conf->max_degraded = 1; |
584acdd4 MS |
6561 | conf->rmw_level = PARITY_ENABLE_RMW; |
6562 | } | |
91adb564 | 6563 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6564 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6565 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6566 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 6567 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
6568 | } else { |
6569 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
6570 | conf->prev_algo = conf->algorithm; | |
e183eaed | 6571 | } |
1da177e4 | 6572 | |
edbe83ab N |
6573 | conf->min_nr_stripes = NR_STRIPES; |
6574 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + | |
5e5e3e78 | 6575 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6576 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6577 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
91adb564 | 6578 | printk(KERN_ERR |
0c55e022 N |
6579 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
6580 | mdname(mddev), memory); | |
91adb564 N |
6581 | goto abort; |
6582 | } else | |
0c55e022 N |
6583 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
6584 | mdname(mddev), memory); | |
edbe83ab N |
6585 | /* |
6586 | * Losing a stripe head costs more than the time to refill it, | |
6587 | * it reduces the queue depth and so can hurt throughput. | |
6588 | * So set it rather large, scaled by number of devices. | |
6589 | */ | |
6590 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6591 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6592 | conf->shrinker.count_objects = raid5_cache_count; | |
6593 | conf->shrinker.batch = 128; | |
6594 | conf->shrinker.flags = 0; | |
6595 | register_shrinker(&conf->shrinker); | |
1da177e4 | 6596 | |
0232605d N |
6597 | sprintf(pers_name, "raid%d", mddev->new_level); |
6598 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
6599 | if (!conf->thread) { |
6600 | printk(KERN_ERR | |
0c55e022 | 6601 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 6602 | mdname(mddev)); |
16a53ecc N |
6603 | goto abort; |
6604 | } | |
91adb564 N |
6605 | |
6606 | return conf; | |
6607 | ||
6608 | abort: | |
6609 | if (conf) { | |
95fc17aa | 6610 | free_conf(conf); |
91adb564 N |
6611 | return ERR_PTR(-EIO); |
6612 | } else | |
6613 | return ERR_PTR(-ENOMEM); | |
6614 | } | |
6615 | ||
c148ffdc N |
6616 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6617 | { | |
6618 | switch (algo) { | |
6619 | case ALGORITHM_PARITY_0: | |
6620 | if (raid_disk < max_degraded) | |
6621 | return 1; | |
6622 | break; | |
6623 | case ALGORITHM_PARITY_N: | |
6624 | if (raid_disk >= raid_disks - max_degraded) | |
6625 | return 1; | |
6626 | break; | |
6627 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6628 | if (raid_disk == 0 || |
c148ffdc N |
6629 | raid_disk == raid_disks - 1) |
6630 | return 1; | |
6631 | break; | |
6632 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6633 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6634 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6635 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6636 | if (raid_disk == raid_disks - 1) | |
6637 | return 1; | |
6638 | } | |
6639 | return 0; | |
6640 | } | |
6641 | ||
fd01b88c | 6642 | static int run(struct mddev *mddev) |
91adb564 | 6643 | { |
d1688a6d | 6644 | struct r5conf *conf; |
9f7c2220 | 6645 | int working_disks = 0; |
c148ffdc | 6646 | int dirty_parity_disks = 0; |
3cb03002 | 6647 | struct md_rdev *rdev; |
c148ffdc | 6648 | sector_t reshape_offset = 0; |
17045f52 | 6649 | int i; |
b5254dd5 N |
6650 | long long min_offset_diff = 0; |
6651 | int first = 1; | |
91adb564 | 6652 | |
8c6ac868 | 6653 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 6654 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
6655 | " -- starting background reconstruction\n", |
6656 | mdname(mddev)); | |
b5254dd5 N |
6657 | |
6658 | rdev_for_each(rdev, mddev) { | |
6659 | long long diff; | |
6660 | if (rdev->raid_disk < 0) | |
6661 | continue; | |
6662 | diff = (rdev->new_data_offset - rdev->data_offset); | |
6663 | if (first) { | |
6664 | min_offset_diff = diff; | |
6665 | first = 0; | |
6666 | } else if (mddev->reshape_backwards && | |
6667 | diff < min_offset_diff) | |
6668 | min_offset_diff = diff; | |
6669 | else if (!mddev->reshape_backwards && | |
6670 | diff > min_offset_diff) | |
6671 | min_offset_diff = diff; | |
6672 | } | |
6673 | ||
91adb564 N |
6674 | if (mddev->reshape_position != MaxSector) { |
6675 | /* Check that we can continue the reshape. | |
b5254dd5 N |
6676 | * Difficulties arise if the stripe we would write to |
6677 | * next is at or after the stripe we would read from next. | |
6678 | * For a reshape that changes the number of devices, this | |
6679 | * is only possible for a very short time, and mdadm makes | |
6680 | * sure that time appears to have past before assembling | |
6681 | * the array. So we fail if that time hasn't passed. | |
6682 | * For a reshape that keeps the number of devices the same | |
6683 | * mdadm must be monitoring the reshape can keeping the | |
6684 | * critical areas read-only and backed up. It will start | |
6685 | * the array in read-only mode, so we check for that. | |
91adb564 N |
6686 | */ |
6687 | sector_t here_new, here_old; | |
6688 | int old_disks; | |
18b00334 | 6689 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
6690 | int chunk_sectors; |
6691 | int new_data_disks; | |
91adb564 | 6692 | |
88ce4930 | 6693 | if (mddev->new_level != mddev->level) { |
0c55e022 | 6694 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
6695 | "required - aborting.\n", |
6696 | mdname(mddev)); | |
6697 | return -EINVAL; | |
6698 | } | |
91adb564 N |
6699 | old_disks = mddev->raid_disks - mddev->delta_disks; |
6700 | /* reshape_position must be on a new-stripe boundary, and one | |
6701 | * further up in new geometry must map after here in old | |
6702 | * geometry. | |
05256d98 N |
6703 | * If the chunk sizes are different, then as we perform reshape |
6704 | * in units of the largest of the two, reshape_position needs | |
6705 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
6706 | */ |
6707 | here_new = mddev->reshape_position; | |
05256d98 N |
6708 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
6709 | new_data_disks = mddev->raid_disks - max_degraded; | |
6710 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
0c55e022 N |
6711 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
6712 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
6713 | return -EINVAL; |
6714 | } | |
05256d98 | 6715 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
6716 | /* here_new is the stripe we will write to */ |
6717 | here_old = mddev->reshape_position; | |
05256d98 | 6718 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
6719 | /* here_old is the first stripe that we might need to read |
6720 | * from */ | |
67ac6011 N |
6721 | if (mddev->delta_disks == 0) { |
6722 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 6723 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
6724 | * and taking constant backups. |
6725 | * mdadm always starts a situation like this in | |
6726 | * readonly mode so it can take control before | |
6727 | * allowing any writes. So just check for that. | |
6728 | */ | |
b5254dd5 N |
6729 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
6730 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
6731 | /* not really in-place - so OK */; | |
6732 | else if (mddev->ro == 0) { | |
6733 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
6734 | "must be started in read-only mode " | |
6735 | "- aborting\n", | |
0c55e022 | 6736 | mdname(mddev)); |
67ac6011 N |
6737 | return -EINVAL; |
6738 | } | |
2c810cdd | 6739 | } else if (mddev->reshape_backwards |
05256d98 N |
6740 | ? (here_new * chunk_sectors + min_offset_diff <= |
6741 | here_old * chunk_sectors) | |
6742 | : (here_new * chunk_sectors >= | |
6743 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 6744 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
6745 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
6746 | "auto-recovery - aborting.\n", | |
6747 | mdname(mddev)); | |
91adb564 N |
6748 | return -EINVAL; |
6749 | } | |
0c55e022 N |
6750 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
6751 | mdname(mddev)); | |
91adb564 N |
6752 | /* OK, we should be able to continue; */ |
6753 | } else { | |
6754 | BUG_ON(mddev->level != mddev->new_level); | |
6755 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 6756 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 6757 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 6758 | } |
91adb564 | 6759 | |
245f46c2 N |
6760 | if (mddev->private == NULL) |
6761 | conf = setup_conf(mddev); | |
6762 | else | |
6763 | conf = mddev->private; | |
6764 | ||
91adb564 N |
6765 | if (IS_ERR(conf)) |
6766 | return PTR_ERR(conf); | |
6767 | ||
b5254dd5 | 6768 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
6769 | mddev->thread = conf->thread; |
6770 | conf->thread = NULL; | |
6771 | mddev->private = conf; | |
6772 | ||
17045f52 N |
6773 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
6774 | i++) { | |
6775 | rdev = conf->disks[i].rdev; | |
6776 | if (!rdev && conf->disks[i].replacement) { | |
6777 | /* The replacement is all we have yet */ | |
6778 | rdev = conf->disks[i].replacement; | |
6779 | conf->disks[i].replacement = NULL; | |
6780 | clear_bit(Replacement, &rdev->flags); | |
6781 | conf->disks[i].rdev = rdev; | |
6782 | } | |
6783 | if (!rdev) | |
c148ffdc | 6784 | continue; |
17045f52 N |
6785 | if (conf->disks[i].replacement && |
6786 | conf->reshape_progress != MaxSector) { | |
6787 | /* replacements and reshape simply do not mix. */ | |
6788 | printk(KERN_ERR "md: cannot handle concurrent " | |
6789 | "replacement and reshape.\n"); | |
6790 | goto abort; | |
6791 | } | |
2f115882 | 6792 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 6793 | working_disks++; |
2f115882 N |
6794 | continue; |
6795 | } | |
c148ffdc N |
6796 | /* This disc is not fully in-sync. However if it |
6797 | * just stored parity (beyond the recovery_offset), | |
6798 | * when we don't need to be concerned about the | |
6799 | * array being dirty. | |
6800 | * When reshape goes 'backwards', we never have | |
6801 | * partially completed devices, so we only need | |
6802 | * to worry about reshape going forwards. | |
6803 | */ | |
6804 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
6805 | if (mddev->major_version == 0 && | |
6806 | mddev->minor_version > 90) | |
6807 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 6808 | |
c148ffdc N |
6809 | if (rdev->recovery_offset < reshape_offset) { |
6810 | /* We need to check old and new layout */ | |
6811 | if (!only_parity(rdev->raid_disk, | |
6812 | conf->algorithm, | |
6813 | conf->raid_disks, | |
6814 | conf->max_degraded)) | |
6815 | continue; | |
6816 | } | |
6817 | if (!only_parity(rdev->raid_disk, | |
6818 | conf->prev_algo, | |
6819 | conf->previous_raid_disks, | |
6820 | conf->max_degraded)) | |
6821 | continue; | |
6822 | dirty_parity_disks++; | |
6823 | } | |
91adb564 | 6824 | |
17045f52 N |
6825 | /* |
6826 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
6827 | */ | |
908f4fbd | 6828 | mddev->degraded = calc_degraded(conf); |
91adb564 | 6829 | |
674806d6 | 6830 | if (has_failed(conf)) { |
0c55e022 | 6831 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 6832 | " (%d/%d failed)\n", |
02c2de8c | 6833 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
6834 | goto abort; |
6835 | } | |
6836 | ||
91adb564 | 6837 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 6838 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
6839 | mddev->resync_max_sectors = mddev->dev_sectors; |
6840 | ||
c148ffdc | 6841 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 6842 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
6843 | if (mddev->ok_start_degraded) |
6844 | printk(KERN_WARNING | |
0c55e022 N |
6845 | "md/raid:%s: starting dirty degraded array" |
6846 | " - data corruption possible.\n", | |
6ff8d8ec N |
6847 | mdname(mddev)); |
6848 | else { | |
6849 | printk(KERN_ERR | |
0c55e022 | 6850 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
6851 | mdname(mddev)); |
6852 | goto abort; | |
6853 | } | |
1da177e4 LT |
6854 | } |
6855 | ||
1da177e4 | 6856 | if (mddev->degraded == 0) |
0c55e022 N |
6857 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
6858 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
6859 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
6860 | mddev->new_layout); | |
1da177e4 | 6861 | else |
0c55e022 N |
6862 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
6863 | " out of %d devices, algorithm %d\n", | |
6864 | mdname(mddev), conf->level, | |
6865 | mddev->raid_disks - mddev->degraded, | |
6866 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
6867 | |
6868 | print_raid5_conf(conf); | |
6869 | ||
fef9c61f | 6870 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 6871 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
6872 | atomic_set(&conf->reshape_stripes, 0); |
6873 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
6874 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6875 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6876 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6877 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6878 | "reshape"); |
f6705578 N |
6879 | } |
6880 | ||
1da177e4 | 6881 | /* Ok, everything is just fine now */ |
a64c876f N |
6882 | if (mddev->to_remove == &raid5_attrs_group) |
6883 | mddev->to_remove = NULL; | |
00bcb4ac N |
6884 | else if (mddev->kobj.sd && |
6885 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 6886 | printk(KERN_WARNING |
4a5add49 | 6887 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 6888 | mdname(mddev)); |
4a5add49 | 6889 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 6890 | |
4a5add49 | 6891 | if (mddev->queue) { |
9f7c2220 | 6892 | int chunk_size; |
620125f2 | 6893 | bool discard_supported = true; |
4a5add49 N |
6894 | /* read-ahead size must cover two whole stripes, which |
6895 | * is 2 * (datadisks) * chunksize where 'n' is the | |
6896 | * number of raid devices | |
6897 | */ | |
6898 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
6899 | int stripe = data_disks * | |
6900 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
6901 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
6902 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 6903 | |
9f7c2220 N |
6904 | chunk_size = mddev->chunk_sectors << 9; |
6905 | blk_queue_io_min(mddev->queue, chunk_size); | |
6906 | blk_queue_io_opt(mddev->queue, chunk_size * | |
6907 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 6908 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
6909 | /* |
6910 | * We can only discard a whole stripe. It doesn't make sense to | |
6911 | * discard data disk but write parity disk | |
6912 | */ | |
6913 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
6914 | /* Round up to power of 2, as discard handling |
6915 | * currently assumes that */ | |
6916 | while ((stripe-1) & stripe) | |
6917 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
6918 | mddev->queue->limits.discard_alignment = stripe; |
6919 | mddev->queue->limits.discard_granularity = stripe; | |
6920 | /* | |
6921 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 6922 | * guarantee discard_zeroes_data |
620125f2 SL |
6923 | */ |
6924 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 6925 | |
5026d7a9 PA |
6926 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
6927 | ||
05616be5 | 6928 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
6929 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6930 | rdev->data_offset << 9); | |
05616be5 N |
6931 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6932 | rdev->new_data_offset << 9); | |
620125f2 SL |
6933 | /* |
6934 | * discard_zeroes_data is required, otherwise data | |
6935 | * could be lost. Consider a scenario: discard a stripe | |
6936 | * (the stripe could be inconsistent if | |
6937 | * discard_zeroes_data is 0); write one disk of the | |
6938 | * stripe (the stripe could be inconsistent again | |
6939 | * depending on which disks are used to calculate | |
6940 | * parity); the disk is broken; The stripe data of this | |
6941 | * disk is lost. | |
6942 | */ | |
6943 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
6944 | !bdev_get_queue(rdev->bdev)-> | |
6945 | limits.discard_zeroes_data) | |
6946 | discard_supported = false; | |
8e0e99ba N |
6947 | /* Unfortunately, discard_zeroes_data is not currently |
6948 | * a guarantee - just a hint. So we only allow DISCARD | |
6949 | * if the sysadmin has confirmed that only safe devices | |
6950 | * are in use by setting a module parameter. | |
6951 | */ | |
6952 | if (!devices_handle_discard_safely) { | |
6953 | if (discard_supported) { | |
6954 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
6955 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
6956 | } | |
6957 | discard_supported = false; | |
6958 | } | |
05616be5 | 6959 | } |
620125f2 SL |
6960 | |
6961 | if (discard_supported && | |
6962 | mddev->queue->limits.max_discard_sectors >= stripe && | |
6963 | mddev->queue->limits.discard_granularity >= stripe) | |
6964 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
6965 | mddev->queue); | |
6966 | else | |
6967 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
6968 | mddev->queue); | |
9f7c2220 | 6969 | } |
23032a0e | 6970 | |
1da177e4 LT |
6971 | return 0; |
6972 | abort: | |
01f96c0a | 6973 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
6974 | print_raid5_conf(conf); |
6975 | free_conf(conf); | |
1da177e4 | 6976 | mddev->private = NULL; |
0c55e022 | 6977 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
6978 | return -EIO; |
6979 | } | |
6980 | ||
afa0f557 | 6981 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 6982 | { |
afa0f557 | 6983 | struct r5conf *conf = priv; |
1da177e4 | 6984 | |
95fc17aa | 6985 | free_conf(conf); |
a64c876f | 6986 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
6987 | } |
6988 | ||
fd01b88c | 6989 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6990 | { |
d1688a6d | 6991 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
6992 | int i; |
6993 | ||
9d8f0363 | 6994 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 6995 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 6996 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
6997 | for (i = 0; i < conf->raid_disks; i++) |
6998 | seq_printf (seq, "%s", | |
6999 | conf->disks[i].rdev && | |
b2d444d7 | 7000 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 7001 | seq_printf (seq, "]"); |
1da177e4 LT |
7002 | } |
7003 | ||
d1688a6d | 7004 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7005 | { |
7006 | int i; | |
7007 | struct disk_info *tmp; | |
7008 | ||
0c55e022 | 7009 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
7010 | if (!conf) { |
7011 | printk("(conf==NULL)\n"); | |
7012 | return; | |
7013 | } | |
0c55e022 N |
7014 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
7015 | conf->raid_disks, | |
7016 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7017 | |
7018 | for (i = 0; i < conf->raid_disks; i++) { | |
7019 | char b[BDEVNAME_SIZE]; | |
7020 | tmp = conf->disks + i; | |
7021 | if (tmp->rdev) | |
0c55e022 N |
7022 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
7023 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
7024 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7025 | } |
7026 | } | |
7027 | ||
fd01b88c | 7028 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7029 | { |
7030 | int i; | |
d1688a6d | 7031 | struct r5conf *conf = mddev->private; |
1da177e4 | 7032 | struct disk_info *tmp; |
6b965620 N |
7033 | int count = 0; |
7034 | unsigned long flags; | |
1da177e4 LT |
7035 | |
7036 | for (i = 0; i < conf->raid_disks; i++) { | |
7037 | tmp = conf->disks + i; | |
dd054fce N |
7038 | if (tmp->replacement |
7039 | && tmp->replacement->recovery_offset == MaxSector | |
7040 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7041 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7042 | /* Replacement has just become active. */ | |
7043 | if (!tmp->rdev | |
7044 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7045 | count++; | |
7046 | if (tmp->rdev) { | |
7047 | /* Replaced device not technically faulty, | |
7048 | * but we need to be sure it gets removed | |
7049 | * and never re-added. | |
7050 | */ | |
7051 | set_bit(Faulty, &tmp->rdev->flags); | |
7052 | sysfs_notify_dirent_safe( | |
7053 | tmp->rdev->sysfs_state); | |
7054 | } | |
7055 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7056 | } else if (tmp->rdev | |
70fffd0b | 7057 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7058 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7059 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7060 | count++; |
43c73ca4 | 7061 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7062 | } |
7063 | } | |
6b965620 | 7064 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7065 | mddev->degraded = calc_degraded(conf); |
6b965620 | 7066 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7067 | print_raid5_conf(conf); |
6b965620 | 7068 | return count; |
1da177e4 LT |
7069 | } |
7070 | ||
b8321b68 | 7071 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7072 | { |
d1688a6d | 7073 | struct r5conf *conf = mddev->private; |
1da177e4 | 7074 | int err = 0; |
b8321b68 | 7075 | int number = rdev->raid_disk; |
657e3e4d | 7076 | struct md_rdev **rdevp; |
1da177e4 LT |
7077 | struct disk_info *p = conf->disks + number; |
7078 | ||
7079 | print_raid5_conf(conf); | |
657e3e4d N |
7080 | if (rdev == p->rdev) |
7081 | rdevp = &p->rdev; | |
7082 | else if (rdev == p->replacement) | |
7083 | rdevp = &p->replacement; | |
7084 | else | |
7085 | return 0; | |
7086 | ||
7087 | if (number >= conf->raid_disks && | |
7088 | conf->reshape_progress == MaxSector) | |
7089 | clear_bit(In_sync, &rdev->flags); | |
7090 | ||
7091 | if (test_bit(In_sync, &rdev->flags) || | |
7092 | atomic_read(&rdev->nr_pending)) { | |
7093 | err = -EBUSY; | |
7094 | goto abort; | |
7095 | } | |
7096 | /* Only remove non-faulty devices if recovery | |
7097 | * isn't possible. | |
7098 | */ | |
7099 | if (!test_bit(Faulty, &rdev->flags) && | |
7100 | mddev->recovery_disabled != conf->recovery_disabled && | |
7101 | !has_failed(conf) && | |
dd054fce | 7102 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7103 | number < conf->raid_disks) { |
7104 | err = -EBUSY; | |
7105 | goto abort; | |
7106 | } | |
7107 | *rdevp = NULL; | |
7108 | synchronize_rcu(); | |
7109 | if (atomic_read(&rdev->nr_pending)) { | |
7110 | /* lost the race, try later */ | |
7111 | err = -EBUSY; | |
7112 | *rdevp = rdev; | |
dd054fce N |
7113 | } else if (p->replacement) { |
7114 | /* We must have just cleared 'rdev' */ | |
7115 | p->rdev = p->replacement; | |
7116 | clear_bit(Replacement, &p->replacement->flags); | |
7117 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7118 | * but will never see neither - if they are careful | |
7119 | */ | |
7120 | p->replacement = NULL; | |
7121 | clear_bit(WantReplacement, &rdev->flags); | |
7122 | } else | |
7123 | /* We might have just removed the Replacement as faulty- | |
7124 | * clear the bit just in case | |
7125 | */ | |
7126 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7127 | abort: |
7128 | ||
7129 | print_raid5_conf(conf); | |
7130 | return err; | |
7131 | } | |
7132 | ||
fd01b88c | 7133 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7134 | { |
d1688a6d | 7135 | struct r5conf *conf = mddev->private; |
199050ea | 7136 | int err = -EEXIST; |
1da177e4 LT |
7137 | int disk; |
7138 | struct disk_info *p; | |
6c2fce2e NB |
7139 | int first = 0; |
7140 | int last = conf->raid_disks - 1; | |
1da177e4 | 7141 | |
7f0da59b N |
7142 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7143 | return -EBUSY; | |
7144 | ||
dc10c643 | 7145 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7146 | /* no point adding a device */ |
199050ea | 7147 | return -EINVAL; |
1da177e4 | 7148 | |
6c2fce2e NB |
7149 | if (rdev->raid_disk >= 0) |
7150 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7151 | |
7152 | /* | |
16a53ecc N |
7153 | * find the disk ... but prefer rdev->saved_raid_disk |
7154 | * if possible. | |
1da177e4 | 7155 | */ |
16a53ecc | 7156 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7157 | rdev->saved_raid_disk >= first && |
16a53ecc | 7158 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7159 | first = rdev->saved_raid_disk; |
7160 | ||
7161 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7162 | p = conf->disks + disk; |
7163 | if (p->rdev == NULL) { | |
b2d444d7 | 7164 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7165 | rdev->raid_disk = disk; |
199050ea | 7166 | err = 0; |
72626685 N |
7167 | if (rdev->saved_raid_disk != disk) |
7168 | conf->fullsync = 1; | |
d6065f7b | 7169 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7170 | goto out; |
1da177e4 | 7171 | } |
5cfb22a1 N |
7172 | } |
7173 | for (disk = first; disk <= last; disk++) { | |
7174 | p = conf->disks + disk; | |
7bfec5f3 N |
7175 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7176 | p->replacement == NULL) { | |
7177 | clear_bit(In_sync, &rdev->flags); | |
7178 | set_bit(Replacement, &rdev->flags); | |
7179 | rdev->raid_disk = disk; | |
7180 | err = 0; | |
7181 | conf->fullsync = 1; | |
7182 | rcu_assign_pointer(p->replacement, rdev); | |
7183 | break; | |
7184 | } | |
7185 | } | |
5cfb22a1 | 7186 | out: |
1da177e4 | 7187 | print_raid5_conf(conf); |
199050ea | 7188 | return err; |
1da177e4 LT |
7189 | } |
7190 | ||
fd01b88c | 7191 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7192 | { |
7193 | /* no resync is happening, and there is enough space | |
7194 | * on all devices, so we can resize. | |
7195 | * We need to make sure resync covers any new space. | |
7196 | * If the array is shrinking we should possibly wait until | |
7197 | * any io in the removed space completes, but it hardly seems | |
7198 | * worth it. | |
7199 | */ | |
a4a6125a | 7200 | sector_t newsize; |
3cb5edf4 N |
7201 | struct r5conf *conf = mddev->private; |
7202 | ||
7203 | sectors &= ~((sector_t)conf->chunk_sectors - 1); | |
a4a6125a N |
7204 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7205 | if (mddev->external_size && | |
7206 | mddev->array_sectors > newsize) | |
b522adcd | 7207 | return -EINVAL; |
a4a6125a N |
7208 | if (mddev->bitmap) { |
7209 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7210 | if (ret) | |
7211 | return ret; | |
7212 | } | |
7213 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 7214 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 7215 | revalidate_disk(mddev->gendisk); |
b098636c N |
7216 | if (sectors > mddev->dev_sectors && |
7217 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7218 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7219 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7220 | } | |
58c0fed4 | 7221 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7222 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7223 | return 0; |
7224 | } | |
7225 | ||
fd01b88c | 7226 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7227 | { |
7228 | /* Can only proceed if there are plenty of stripe_heads. | |
7229 | * We need a minimum of one full stripe,, and for sensible progress | |
7230 | * it is best to have about 4 times that. | |
7231 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7232 | * allow for chunk sizes up to 256K, which is probably OK. | |
7233 | * If the chunk size is greater, user-space should request more | |
7234 | * stripe_heads first. | |
7235 | */ | |
d1688a6d | 7236 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7237 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7238 | > conf->min_nr_stripes || |
01ee22b4 | 7239 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7240 | > conf->min_nr_stripes) { |
0c55e022 N |
7241 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7242 | mdname(mddev), | |
01ee22b4 N |
7243 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
7244 | / STRIPE_SIZE)*4); | |
7245 | return 0; | |
7246 | } | |
7247 | return 1; | |
7248 | } | |
7249 | ||
fd01b88c | 7250 | static int check_reshape(struct mddev *mddev) |
29269553 | 7251 | { |
d1688a6d | 7252 | struct r5conf *conf = mddev->private; |
29269553 | 7253 | |
88ce4930 N |
7254 | if (mddev->delta_disks == 0 && |
7255 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7256 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7257 | return 0; /* nothing to do */ |
674806d6 | 7258 | if (has_failed(conf)) |
ec32a2bd | 7259 | return -EINVAL; |
fdcfbbb6 | 7260 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7261 | /* We might be able to shrink, but the devices must |
7262 | * be made bigger first. | |
7263 | * For raid6, 4 is the minimum size. | |
7264 | * Otherwise 2 is the minimum | |
7265 | */ | |
7266 | int min = 2; | |
7267 | if (mddev->level == 6) | |
7268 | min = 4; | |
7269 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7270 | return -EINVAL; | |
7271 | } | |
29269553 | 7272 | |
01ee22b4 | 7273 | if (!check_stripe_cache(mddev)) |
29269553 | 7274 | return -ENOSPC; |
29269553 | 7275 | |
738a2738 N |
7276 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7277 | mddev->delta_disks > 0) | |
7278 | if (resize_chunks(conf, | |
7279 | conf->previous_raid_disks | |
7280 | + max(0, mddev->delta_disks), | |
7281 | max(mddev->new_chunk_sectors, | |
7282 | mddev->chunk_sectors) | |
7283 | ) < 0) | |
7284 | return -ENOMEM; | |
e56108d6 N |
7285 | return resize_stripes(conf, (conf->previous_raid_disks |
7286 | + mddev->delta_disks)); | |
63c70c4f N |
7287 | } |
7288 | ||
fd01b88c | 7289 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7290 | { |
d1688a6d | 7291 | struct r5conf *conf = mddev->private; |
3cb03002 | 7292 | struct md_rdev *rdev; |
63c70c4f | 7293 | int spares = 0; |
c04be0aa | 7294 | unsigned long flags; |
63c70c4f | 7295 | |
f416885e | 7296 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7297 | return -EBUSY; |
7298 | ||
01ee22b4 N |
7299 | if (!check_stripe_cache(mddev)) |
7300 | return -ENOSPC; | |
7301 | ||
30b67645 N |
7302 | if (has_failed(conf)) |
7303 | return -EINVAL; | |
7304 | ||
c6563a8c | 7305 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7306 | if (!test_bit(In_sync, &rdev->flags) |
7307 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7308 | spares++; |
c6563a8c | 7309 | } |
63c70c4f | 7310 | |
f416885e | 7311 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7312 | /* Not enough devices even to make a degraded array |
7313 | * of that size | |
7314 | */ | |
7315 | return -EINVAL; | |
7316 | ||
ec32a2bd N |
7317 | /* Refuse to reduce size of the array. Any reductions in |
7318 | * array size must be through explicit setting of array_size | |
7319 | * attribute. | |
7320 | */ | |
7321 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7322 | < mddev->array_sectors) { | |
0c55e022 | 7323 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
7324 | "before number of disks\n", mdname(mddev)); |
7325 | return -EINVAL; | |
7326 | } | |
7327 | ||
f6705578 | 7328 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7329 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7330 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7331 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7332 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7333 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7334 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7335 | conf->prev_algo = conf->algorithm; |
7336 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7337 | conf->generation++; |
7338 | /* Code that selects data_offset needs to see the generation update | |
7339 | * if reshape_progress has been set - so a memory barrier needed. | |
7340 | */ | |
7341 | smp_mb(); | |
2c810cdd | 7342 | if (mddev->reshape_backwards) |
fef9c61f N |
7343 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7344 | else | |
7345 | conf->reshape_progress = 0; | |
7346 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7347 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7348 | spin_unlock_irq(&conf->device_lock); |
7349 | ||
4d77e3ba N |
7350 | /* Now make sure any requests that proceeded on the assumption |
7351 | * the reshape wasn't running - like Discard or Read - have | |
7352 | * completed. | |
7353 | */ | |
7354 | mddev_suspend(mddev); | |
7355 | mddev_resume(mddev); | |
7356 | ||
29269553 N |
7357 | /* Add some new drives, as many as will fit. |
7358 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7359 | * Don't add devices if we are reducing the number of |
7360 | * devices in the array. This is because it is not possible | |
7361 | * to correctly record the "partially reconstructed" state of | |
7362 | * such devices during the reshape and confusion could result. | |
29269553 | 7363 | */ |
87a8dec9 | 7364 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7365 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7366 | if (rdev->raid_disk < 0 && |
7367 | !test_bit(Faulty, &rdev->flags)) { | |
7368 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7369 | if (rdev->raid_disk |
9d4c7d87 | 7370 | >= conf->previous_raid_disks) |
87a8dec9 | 7371 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7372 | else |
87a8dec9 | 7373 | rdev->recovery_offset = 0; |
36fad858 NK |
7374 | |
7375 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7376 | /* Failure here is OK */; |
50da0840 | 7377 | } |
87a8dec9 N |
7378 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7379 | && !test_bit(Faulty, &rdev->flags)) { | |
7380 | /* This is a spare that was manually added */ | |
7381 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7382 | } |
29269553 | 7383 | |
87a8dec9 N |
7384 | /* When a reshape changes the number of devices, |
7385 | * ->degraded is measured against the larger of the | |
7386 | * pre and post number of devices. | |
7387 | */ | |
ec32a2bd | 7388 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7389 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
7390 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7391 | } | |
63c70c4f | 7392 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7393 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 7394 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 7395 | |
29269553 N |
7396 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7397 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7398 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7399 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7400 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7401 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7402 | "reshape"); |
29269553 N |
7403 | if (!mddev->sync_thread) { |
7404 | mddev->recovery = 0; | |
7405 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7406 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7407 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7408 | mddev->new_chunk_sectors = |
7409 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7410 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7411 | rdev_for_each(rdev, mddev) |
7412 | rdev->new_data_offset = rdev->data_offset; | |
7413 | smp_wmb(); | |
ba8805b9 | 7414 | conf->generation --; |
fef9c61f | 7415 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7416 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7417 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7418 | spin_unlock_irq(&conf->device_lock); |
7419 | return -EAGAIN; | |
7420 | } | |
c8f517c4 | 7421 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7422 | md_wakeup_thread(mddev->sync_thread); |
7423 | md_new_event(mddev); | |
7424 | return 0; | |
7425 | } | |
29269553 | 7426 | |
ec32a2bd N |
7427 | /* This is called from the reshape thread and should make any |
7428 | * changes needed in 'conf' | |
7429 | */ | |
d1688a6d | 7430 | static void end_reshape(struct r5conf *conf) |
29269553 | 7431 | { |
29269553 | 7432 | |
f6705578 | 7433 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7434 | struct md_rdev *rdev; |
f6705578 | 7435 | |
f6705578 | 7436 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7437 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7438 | rdev_for_each(rdev, conf->mddev) |
7439 | rdev->data_offset = rdev->new_data_offset; | |
7440 | smp_wmb(); | |
fef9c61f | 7441 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7442 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 7443 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7444 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7445 | |
7446 | /* read-ahead size must cover two whole stripes, which is | |
7447 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7448 | */ | |
4a5add49 | 7449 | if (conf->mddev->queue) { |
cea9c228 | 7450 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7451 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7452 | / PAGE_SIZE); |
16a53ecc N |
7453 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
7454 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
7455 | } | |
29269553 | 7456 | } |
29269553 N |
7457 | } |
7458 | ||
ec32a2bd N |
7459 | /* This is called from the raid5d thread with mddev_lock held. |
7460 | * It makes config changes to the device. | |
7461 | */ | |
fd01b88c | 7462 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7463 | { |
d1688a6d | 7464 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7465 | |
7466 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7467 | ||
ec32a2bd N |
7468 | if (mddev->delta_disks > 0) { |
7469 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
7470 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 7471 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
7472 | } else { |
7473 | int d; | |
908f4fbd N |
7474 | spin_lock_irq(&conf->device_lock); |
7475 | mddev->degraded = calc_degraded(conf); | |
7476 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
7477 | for (d = conf->raid_disks ; |
7478 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7479 | d++) { |
3cb03002 | 7480 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7481 | if (rdev) |
7482 | clear_bit(In_sync, &rdev->flags); | |
7483 | rdev = conf->disks[d].replacement; | |
7484 | if (rdev) | |
7485 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7486 | } |
cea9c228 | 7487 | } |
88ce4930 | 7488 | mddev->layout = conf->algorithm; |
09c9e5fa | 7489 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7490 | mddev->reshape_position = MaxSector; |
7491 | mddev->delta_disks = 0; | |
2c810cdd | 7492 | mddev->reshape_backwards = 0; |
cea9c228 N |
7493 | } |
7494 | } | |
7495 | ||
fd01b88c | 7496 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7497 | { |
d1688a6d | 7498 | struct r5conf *conf = mddev->private; |
72626685 N |
7499 | |
7500 | switch(state) { | |
e464eafd N |
7501 | case 2: /* resume for a suspend */ |
7502 | wake_up(&conf->wait_for_overlap); | |
7503 | break; | |
7504 | ||
72626685 | 7505 | case 1: /* stop all writes */ |
566c09c5 | 7506 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7507 | /* '2' tells resync/reshape to pause so that all |
7508 | * active stripes can drain | |
7509 | */ | |
7510 | conf->quiesce = 2; | |
b1b46486 | 7511 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7512 | atomic_read(&conf->active_stripes) == 0 && |
7513 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7514 | unlock_all_device_hash_locks_irq(conf), |
7515 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7516 | conf->quiesce = 1; |
566c09c5 | 7517 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7518 | /* allow reshape to continue */ |
7519 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7520 | break; |
7521 | ||
7522 | case 0: /* re-enable writes */ | |
566c09c5 | 7523 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7524 | conf->quiesce = 0; |
b1b46486 | 7525 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7526 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7527 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7528 | break; |
7529 | } | |
72626685 | 7530 | } |
b15c2e57 | 7531 | |
fd01b88c | 7532 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7533 | { |
e373ab10 | 7534 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7535 | sector_t sectors; |
54071b38 | 7536 | |
f1b29bca | 7537 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7538 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
7539 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7540 | mdname(mddev)); | |
f1b29bca DW |
7541 | return ERR_PTR(-EINVAL); |
7542 | } | |
7543 | ||
e373ab10 N |
7544 | sectors = raid0_conf->strip_zone[0].zone_end; |
7545 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7546 | mddev->dev_sectors = sectors; |
f1b29bca | 7547 | mddev->new_level = level; |
54071b38 TM |
7548 | mddev->new_layout = ALGORITHM_PARITY_N; |
7549 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7550 | mddev->raid_disks += 1; | |
7551 | mddev->delta_disks = 1; | |
7552 | /* make sure it will be not marked as dirty */ | |
7553 | mddev->recovery_cp = MaxSector; | |
7554 | ||
7555 | return setup_conf(mddev); | |
7556 | } | |
7557 | ||
fd01b88c | 7558 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7559 | { |
7560 | int chunksect; | |
7561 | ||
7562 | if (mddev->raid_disks != 2 || | |
7563 | mddev->degraded > 1) | |
7564 | return ERR_PTR(-EINVAL); | |
7565 | ||
7566 | /* Should check if there are write-behind devices? */ | |
7567 | ||
7568 | chunksect = 64*2; /* 64K by default */ | |
7569 | ||
7570 | /* The array must be an exact multiple of chunksize */ | |
7571 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
7572 | chunksect >>= 1; | |
7573 | ||
7574 | if ((chunksect<<9) < STRIPE_SIZE) | |
7575 | /* array size does not allow a suitable chunk size */ | |
7576 | return ERR_PTR(-EINVAL); | |
7577 | ||
7578 | mddev->new_level = 5; | |
7579 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 7580 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
7581 | |
7582 | return setup_conf(mddev); | |
7583 | } | |
7584 | ||
fd01b88c | 7585 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
7586 | { |
7587 | int new_layout; | |
7588 | ||
7589 | switch (mddev->layout) { | |
7590 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7591 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7592 | break; | |
7593 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7594 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
7595 | break; | |
7596 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7597 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
7598 | break; | |
7599 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7600 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
7601 | break; | |
7602 | case ALGORITHM_PARITY_0_6: | |
7603 | new_layout = ALGORITHM_PARITY_0; | |
7604 | break; | |
7605 | case ALGORITHM_PARITY_N: | |
7606 | new_layout = ALGORITHM_PARITY_N; | |
7607 | break; | |
7608 | default: | |
7609 | return ERR_PTR(-EINVAL); | |
7610 | } | |
7611 | mddev->new_level = 5; | |
7612 | mddev->new_layout = new_layout; | |
7613 | mddev->delta_disks = -1; | |
7614 | mddev->raid_disks -= 1; | |
7615 | return setup_conf(mddev); | |
7616 | } | |
7617 | ||
fd01b88c | 7618 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 7619 | { |
88ce4930 N |
7620 | /* For a 2-drive array, the layout and chunk size can be changed |
7621 | * immediately as not restriping is needed. | |
7622 | * For larger arrays we record the new value - after validation | |
7623 | * to be used by a reshape pass. | |
b3546035 | 7624 | */ |
d1688a6d | 7625 | struct r5conf *conf = mddev->private; |
597a711b | 7626 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 7627 | |
597a711b | 7628 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
7629 | return -EINVAL; |
7630 | if (new_chunk > 0) { | |
0ba459d2 | 7631 | if (!is_power_of_2(new_chunk)) |
b3546035 | 7632 | return -EINVAL; |
597a711b | 7633 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 7634 | return -EINVAL; |
597a711b | 7635 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
7636 | /* not factor of array size */ |
7637 | return -EINVAL; | |
7638 | } | |
7639 | ||
7640 | /* They look valid */ | |
7641 | ||
88ce4930 | 7642 | if (mddev->raid_disks == 2) { |
597a711b N |
7643 | /* can make the change immediately */ |
7644 | if (mddev->new_layout >= 0) { | |
7645 | conf->algorithm = mddev->new_layout; | |
7646 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
7647 | } |
7648 | if (new_chunk > 0) { | |
597a711b N |
7649 | conf->chunk_sectors = new_chunk ; |
7650 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
7651 | } |
7652 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7653 | md_wakeup_thread(mddev->thread); | |
b3546035 | 7654 | } |
50ac168a | 7655 | return check_reshape(mddev); |
88ce4930 N |
7656 | } |
7657 | ||
fd01b88c | 7658 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 7659 | { |
597a711b | 7660 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 7661 | |
597a711b | 7662 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 7663 | return -EINVAL; |
b3546035 | 7664 | if (new_chunk > 0) { |
0ba459d2 | 7665 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 7666 | return -EINVAL; |
597a711b | 7667 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 7668 | return -EINVAL; |
597a711b | 7669 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
7670 | /* not factor of array size */ |
7671 | return -EINVAL; | |
b3546035 | 7672 | } |
88ce4930 N |
7673 | |
7674 | /* They look valid */ | |
50ac168a | 7675 | return check_reshape(mddev); |
b3546035 N |
7676 | } |
7677 | ||
fd01b88c | 7678 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
7679 | { |
7680 | /* raid5 can take over: | |
f1b29bca | 7681 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
7682 | * raid1 - if there are two drives. We need to know the chunk size |
7683 | * raid4 - trivial - just use a raid4 layout. | |
7684 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 7685 | */ |
f1b29bca DW |
7686 | if (mddev->level == 0) |
7687 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
7688 | if (mddev->level == 1) |
7689 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
7690 | if (mddev->level == 4) { |
7691 | mddev->new_layout = ALGORITHM_PARITY_N; | |
7692 | mddev->new_level = 5; | |
7693 | return setup_conf(mddev); | |
7694 | } | |
fc9739c6 N |
7695 | if (mddev->level == 6) |
7696 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
7697 | |
7698 | return ERR_PTR(-EINVAL); | |
7699 | } | |
7700 | ||
fd01b88c | 7701 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 7702 | { |
f1b29bca DW |
7703 | /* raid4 can take over: |
7704 | * raid0 - if there is only one strip zone | |
7705 | * raid5 - if layout is right | |
a78d38a1 | 7706 | */ |
f1b29bca DW |
7707 | if (mddev->level == 0) |
7708 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
7709 | if (mddev->level == 5 && |
7710 | mddev->layout == ALGORITHM_PARITY_N) { | |
7711 | mddev->new_layout = 0; | |
7712 | mddev->new_level = 4; | |
7713 | return setup_conf(mddev); | |
7714 | } | |
7715 | return ERR_PTR(-EINVAL); | |
7716 | } | |
d562b0c4 | 7717 | |
84fc4b56 | 7718 | static struct md_personality raid5_personality; |
245f46c2 | 7719 | |
fd01b88c | 7720 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
7721 | { |
7722 | /* Currently can only take over a raid5. We map the | |
7723 | * personality to an equivalent raid6 personality | |
7724 | * with the Q block at the end. | |
7725 | */ | |
7726 | int new_layout; | |
7727 | ||
7728 | if (mddev->pers != &raid5_personality) | |
7729 | return ERR_PTR(-EINVAL); | |
7730 | if (mddev->degraded > 1) | |
7731 | return ERR_PTR(-EINVAL); | |
7732 | if (mddev->raid_disks > 253) | |
7733 | return ERR_PTR(-EINVAL); | |
7734 | if (mddev->raid_disks < 3) | |
7735 | return ERR_PTR(-EINVAL); | |
7736 | ||
7737 | switch (mddev->layout) { | |
7738 | case ALGORITHM_LEFT_ASYMMETRIC: | |
7739 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
7740 | break; | |
7741 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
7742 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
7743 | break; | |
7744 | case ALGORITHM_LEFT_SYMMETRIC: | |
7745 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
7746 | break; | |
7747 | case ALGORITHM_RIGHT_SYMMETRIC: | |
7748 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
7749 | break; | |
7750 | case ALGORITHM_PARITY_0: | |
7751 | new_layout = ALGORITHM_PARITY_0_6; | |
7752 | break; | |
7753 | case ALGORITHM_PARITY_N: | |
7754 | new_layout = ALGORITHM_PARITY_N; | |
7755 | break; | |
7756 | default: | |
7757 | return ERR_PTR(-EINVAL); | |
7758 | } | |
7759 | mddev->new_level = 6; | |
7760 | mddev->new_layout = new_layout; | |
7761 | mddev->delta_disks = 1; | |
7762 | mddev->raid_disks += 1; | |
7763 | return setup_conf(mddev); | |
7764 | } | |
7765 | ||
84fc4b56 | 7766 | static struct md_personality raid6_personality = |
16a53ecc N |
7767 | { |
7768 | .name = "raid6", | |
7769 | .level = 6, | |
7770 | .owner = THIS_MODULE, | |
7771 | .make_request = make_request, | |
7772 | .run = run, | |
afa0f557 | 7773 | .free = raid5_free, |
16a53ecc N |
7774 | .status = status, |
7775 | .error_handler = error, | |
7776 | .hot_add_disk = raid5_add_disk, | |
7777 | .hot_remove_disk= raid5_remove_disk, | |
7778 | .spare_active = raid5_spare_active, | |
7779 | .sync_request = sync_request, | |
7780 | .resize = raid5_resize, | |
80c3a6ce | 7781 | .size = raid5_size, |
50ac168a | 7782 | .check_reshape = raid6_check_reshape, |
f416885e | 7783 | .start_reshape = raid5_start_reshape, |
cea9c228 | 7784 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 7785 | .quiesce = raid5_quiesce, |
245f46c2 | 7786 | .takeover = raid6_takeover, |
5c675f83 | 7787 | .congested = raid5_congested, |
16a53ecc | 7788 | }; |
84fc4b56 | 7789 | static struct md_personality raid5_personality = |
1da177e4 LT |
7790 | { |
7791 | .name = "raid5", | |
2604b703 | 7792 | .level = 5, |
1da177e4 LT |
7793 | .owner = THIS_MODULE, |
7794 | .make_request = make_request, | |
7795 | .run = run, | |
afa0f557 | 7796 | .free = raid5_free, |
1da177e4 LT |
7797 | .status = status, |
7798 | .error_handler = error, | |
7799 | .hot_add_disk = raid5_add_disk, | |
7800 | .hot_remove_disk= raid5_remove_disk, | |
7801 | .spare_active = raid5_spare_active, | |
7802 | .sync_request = sync_request, | |
7803 | .resize = raid5_resize, | |
80c3a6ce | 7804 | .size = raid5_size, |
63c70c4f N |
7805 | .check_reshape = raid5_check_reshape, |
7806 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7807 | .finish_reshape = raid5_finish_reshape, |
72626685 | 7808 | .quiesce = raid5_quiesce, |
d562b0c4 | 7809 | .takeover = raid5_takeover, |
5c675f83 | 7810 | .congested = raid5_congested, |
1da177e4 LT |
7811 | }; |
7812 | ||
84fc4b56 | 7813 | static struct md_personality raid4_personality = |
1da177e4 | 7814 | { |
2604b703 N |
7815 | .name = "raid4", |
7816 | .level = 4, | |
7817 | .owner = THIS_MODULE, | |
7818 | .make_request = make_request, | |
7819 | .run = run, | |
afa0f557 | 7820 | .free = raid5_free, |
2604b703 N |
7821 | .status = status, |
7822 | .error_handler = error, | |
7823 | .hot_add_disk = raid5_add_disk, | |
7824 | .hot_remove_disk= raid5_remove_disk, | |
7825 | .spare_active = raid5_spare_active, | |
7826 | .sync_request = sync_request, | |
7827 | .resize = raid5_resize, | |
80c3a6ce | 7828 | .size = raid5_size, |
3d37890b N |
7829 | .check_reshape = raid5_check_reshape, |
7830 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7831 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 7832 | .quiesce = raid5_quiesce, |
a78d38a1 | 7833 | .takeover = raid4_takeover, |
5c675f83 | 7834 | .congested = raid5_congested, |
2604b703 N |
7835 | }; |
7836 | ||
7837 | static int __init raid5_init(void) | |
7838 | { | |
851c30c9 SL |
7839 | raid5_wq = alloc_workqueue("raid5wq", |
7840 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
7841 | if (!raid5_wq) | |
7842 | return -ENOMEM; | |
16a53ecc | 7843 | register_md_personality(&raid6_personality); |
2604b703 N |
7844 | register_md_personality(&raid5_personality); |
7845 | register_md_personality(&raid4_personality); | |
7846 | return 0; | |
1da177e4 LT |
7847 | } |
7848 | ||
2604b703 | 7849 | static void raid5_exit(void) |
1da177e4 | 7850 | { |
16a53ecc | 7851 | unregister_md_personality(&raid6_personality); |
2604b703 N |
7852 | unregister_md_personality(&raid5_personality); |
7853 | unregister_md_personality(&raid4_personality); | |
851c30c9 | 7854 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
7855 | } |
7856 | ||
7857 | module_init(raid5_init); | |
7858 | module_exit(raid5_exit); | |
7859 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7860 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 7861 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
7862 | MODULE_ALIAS("md-raid5"); |
7863 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
7864 | MODULE_ALIAS("md-level-5"); |
7865 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
7866 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
7867 | MODULE_ALIAS("md-raid6"); | |
7868 | MODULE_ALIAS("md-level-6"); | |
7869 | ||
7870 | /* This used to be two separate modules, they were: */ | |
7871 | MODULE_ALIAS("raid5"); | |
7872 | MODULE_ALIAS("raid6"); |