]>
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; | |
6ab2a4b8 | 343 | bool do_wakeup = false; |
566c09c5 SL |
344 | unsigned long flags; |
345 | ||
346 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
347 | size = NR_STRIPE_HASH_LOCKS; | |
348 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
349 | } else | |
350 | size = 1; | |
351 | while (size) { | |
352 | struct list_head *list = &temp_inactive_list[size - 1]; | |
353 | ||
354 | /* | |
6d036f7d | 355 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
356 | * remove stripes from the list |
357 | */ | |
358 | if (!list_empty_careful(list)) { | |
359 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
360 | if (list_empty(conf->inactive_list + hash) && |
361 | !list_empty(list)) | |
362 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 363 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 364 | do_wakeup = true; |
566c09c5 SL |
365 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
366 | } | |
367 | size--; | |
368 | hash--; | |
369 | } | |
370 | ||
371 | if (do_wakeup) { | |
6ab2a4b8 | 372 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
373 | if (atomic_read(&conf->active_stripes) == 0) |
374 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
375 | if (conf->retry_read_aligned) |
376 | md_wakeup_thread(conf->mddev->thread); | |
377 | } | |
4eb788df SL |
378 | } |
379 | ||
773ca82f | 380 | /* should hold conf->device_lock already */ |
566c09c5 SL |
381 | static int release_stripe_list(struct r5conf *conf, |
382 | struct list_head *temp_inactive_list) | |
773ca82f SL |
383 | { |
384 | struct stripe_head *sh; | |
385 | int count = 0; | |
386 | struct llist_node *head; | |
387 | ||
388 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 389 | head = llist_reverse_order(head); |
773ca82f | 390 | while (head) { |
566c09c5 SL |
391 | int hash; |
392 | ||
773ca82f SL |
393 | sh = llist_entry(head, struct stripe_head, release_list); |
394 | head = llist_next(head); | |
395 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ | |
396 | smp_mb(); | |
397 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
398 | /* | |
399 | * Don't worry the bit is set here, because if the bit is set | |
400 | * again, the count is always > 1. This is true for | |
401 | * STRIPE_ON_UNPLUG_LIST bit too. | |
402 | */ | |
566c09c5 SL |
403 | hash = sh->hash_lock_index; |
404 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
405 | count++; |
406 | } | |
407 | ||
408 | return count; | |
409 | } | |
410 | ||
6d036f7d | 411 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 412 | { |
d1688a6d | 413 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 414 | unsigned long flags; |
566c09c5 SL |
415 | struct list_head list; |
416 | int hash; | |
773ca82f | 417 | bool wakeup; |
16a53ecc | 418 | |
cf170f3f ES |
419 | /* Avoid release_list until the last reference. |
420 | */ | |
421 | if (atomic_add_unless(&sh->count, -1, 1)) | |
422 | return; | |
423 | ||
ad4068de | 424 | if (unlikely(!conf->mddev->thread) || |
425 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
426 | goto slow_path; |
427 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
428 | if (wakeup) | |
429 | md_wakeup_thread(conf->mddev->thread); | |
430 | return; | |
431 | slow_path: | |
4eb788df | 432 | local_irq_save(flags); |
773ca82f | 433 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 434 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
435 | INIT_LIST_HEAD(&list); |
436 | hash = sh->hash_lock_index; | |
437 | do_release_stripe(conf, sh, &list); | |
4eb788df | 438 | spin_unlock(&conf->device_lock); |
566c09c5 | 439 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
440 | } |
441 | local_irq_restore(flags); | |
1da177e4 LT |
442 | } |
443 | ||
fccddba0 | 444 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 445 | { |
45b4233c DW |
446 | pr_debug("remove_hash(), stripe %llu\n", |
447 | (unsigned long long)sh->sector); | |
1da177e4 | 448 | |
fccddba0 | 449 | hlist_del_init(&sh->hash); |
1da177e4 LT |
450 | } |
451 | ||
d1688a6d | 452 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 453 | { |
fccddba0 | 454 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 455 | |
45b4233c DW |
456 | pr_debug("insert_hash(), stripe %llu\n", |
457 | (unsigned long long)sh->sector); | |
1da177e4 | 458 | |
fccddba0 | 459 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
460 | } |
461 | ||
1da177e4 | 462 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 463 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
464 | { |
465 | struct stripe_head *sh = NULL; | |
466 | struct list_head *first; | |
467 | ||
566c09c5 | 468 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 469 | goto out; |
566c09c5 | 470 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
471 | sh = list_entry(first, struct stripe_head, lru); |
472 | list_del_init(first); | |
473 | remove_hash(sh); | |
474 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 475 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
476 | if (list_empty(conf->inactive_list + hash)) |
477 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
478 | out: |
479 | return sh; | |
480 | } | |
481 | ||
e4e11e38 | 482 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
483 | { |
484 | struct page *p; | |
485 | int i; | |
e4e11e38 | 486 | int num = sh->raid_conf->pool_size; |
1da177e4 | 487 | |
e4e11e38 | 488 | for (i = 0; i < num ; i++) { |
d592a996 | 489 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
490 | p = sh->dev[i].page; |
491 | if (!p) | |
492 | continue; | |
493 | sh->dev[i].page = NULL; | |
2d1f3b5d | 494 | put_page(p); |
1da177e4 LT |
495 | } |
496 | } | |
497 | ||
a9683a79 | 498 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
499 | { |
500 | int i; | |
e4e11e38 | 501 | int num = sh->raid_conf->pool_size; |
1da177e4 | 502 | |
e4e11e38 | 503 | for (i = 0; i < num; i++) { |
1da177e4 LT |
504 | struct page *page; |
505 | ||
a9683a79 | 506 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
507 | return 1; |
508 | } | |
509 | sh->dev[i].page = page; | |
d592a996 | 510 | sh->dev[i].orig_page = page; |
1da177e4 LT |
511 | } |
512 | return 0; | |
513 | } | |
514 | ||
784052ec | 515 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 516 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 517 | struct stripe_head *sh); |
1da177e4 | 518 | |
b5663ba4 | 519 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 520 | { |
d1688a6d | 521 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 522 | int i, seq; |
1da177e4 | 523 | |
78bafebd ES |
524 | BUG_ON(atomic_read(&sh->count) != 0); |
525 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 526 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 527 | BUG_ON(sh->batch_head); |
d84e0f10 | 528 | |
45b4233c | 529 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 530 | (unsigned long long)sector); |
566c09c5 SL |
531 | retry: |
532 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 533 | sh->generation = conf->generation - previous; |
b5663ba4 | 534 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 535 | sh->sector = sector; |
911d4ee8 | 536 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
537 | sh->state = 0; |
538 | ||
7ecaa1e6 | 539 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
540 | struct r5dev *dev = &sh->dev[i]; |
541 | ||
d84e0f10 | 542 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 543 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 544 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 545 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 546 | dev->read, dev->towrite, dev->written, |
1da177e4 | 547 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 548 | WARN_ON(1); |
1da177e4 LT |
549 | } |
550 | dev->flags = 0; | |
784052ec | 551 | raid5_build_block(sh, i, previous); |
1da177e4 | 552 | } |
566c09c5 SL |
553 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
554 | goto retry; | |
7a87f434 | 555 | sh->overwrite_disks = 0; |
1da177e4 | 556 | insert_hash(conf, sh); |
851c30c9 | 557 | sh->cpu = smp_processor_id(); |
da41ba65 | 558 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
559 | } |
560 | ||
d1688a6d | 561 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 562 | short generation) |
1da177e4 LT |
563 | { |
564 | struct stripe_head *sh; | |
565 | ||
45b4233c | 566 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 567 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 568 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 569 | return sh; |
45b4233c | 570 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
571 | return NULL; |
572 | } | |
573 | ||
674806d6 N |
574 | /* |
575 | * Need to check if array has failed when deciding whether to: | |
576 | * - start an array | |
577 | * - remove non-faulty devices | |
578 | * - add a spare | |
579 | * - allow a reshape | |
580 | * This determination is simple when no reshape is happening. | |
581 | * However if there is a reshape, we need to carefully check | |
582 | * both the before and after sections. | |
583 | * This is because some failed devices may only affect one | |
584 | * of the two sections, and some non-in_sync devices may | |
585 | * be insync in the section most affected by failed devices. | |
586 | */ | |
908f4fbd | 587 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 588 | { |
908f4fbd | 589 | int degraded, degraded2; |
674806d6 | 590 | int i; |
674806d6 N |
591 | |
592 | rcu_read_lock(); | |
593 | degraded = 0; | |
594 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 595 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
596 | if (rdev && test_bit(Faulty, &rdev->flags)) |
597 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
598 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
599 | degraded++; | |
600 | else if (test_bit(In_sync, &rdev->flags)) | |
601 | ; | |
602 | else | |
603 | /* not in-sync or faulty. | |
604 | * If the reshape increases the number of devices, | |
605 | * this is being recovered by the reshape, so | |
606 | * this 'previous' section is not in_sync. | |
607 | * If the number of devices is being reduced however, | |
608 | * the device can only be part of the array if | |
609 | * we are reverting a reshape, so this section will | |
610 | * be in-sync. | |
611 | */ | |
612 | if (conf->raid_disks >= conf->previous_raid_disks) | |
613 | degraded++; | |
614 | } | |
615 | rcu_read_unlock(); | |
908f4fbd N |
616 | if (conf->raid_disks == conf->previous_raid_disks) |
617 | return degraded; | |
674806d6 | 618 | rcu_read_lock(); |
908f4fbd | 619 | degraded2 = 0; |
674806d6 | 620 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 621 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
622 | if (rdev && test_bit(Faulty, &rdev->flags)) |
623 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 624 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 625 | degraded2++; |
674806d6 N |
626 | else if (test_bit(In_sync, &rdev->flags)) |
627 | ; | |
628 | else | |
629 | /* not in-sync or faulty. | |
630 | * If reshape increases the number of devices, this | |
631 | * section has already been recovered, else it | |
632 | * almost certainly hasn't. | |
633 | */ | |
634 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 635 | degraded2++; |
674806d6 N |
636 | } |
637 | rcu_read_unlock(); | |
908f4fbd N |
638 | if (degraded2 > degraded) |
639 | return degraded2; | |
640 | return degraded; | |
641 | } | |
642 | ||
643 | static int has_failed(struct r5conf *conf) | |
644 | { | |
645 | int degraded; | |
646 | ||
647 | if (conf->mddev->reshape_position == MaxSector) | |
648 | return conf->mddev->degraded > conf->max_degraded; | |
649 | ||
650 | degraded = calc_degraded(conf); | |
674806d6 N |
651 | if (degraded > conf->max_degraded) |
652 | return 1; | |
653 | return 0; | |
654 | } | |
655 | ||
6d036f7d SL |
656 | struct stripe_head * |
657 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
658 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
659 | { |
660 | struct stripe_head *sh; | |
566c09c5 | 661 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 662 | int inc_empty_inactive_list_flag; |
1da177e4 | 663 | |
45b4233c | 664 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 665 | |
566c09c5 | 666 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
667 | |
668 | do { | |
b1b46486 | 669 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 670 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 671 | *(conf->hash_locks + hash)); |
86b42c71 | 672 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 673 | if (!sh) { |
edbe83ab | 674 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 675 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
676 | if (!sh && !test_bit(R5_DID_ALLOC, |
677 | &conf->cache_state)) | |
edbe83ab N |
678 | set_bit(R5_ALLOC_MORE, |
679 | &conf->cache_state); | |
680 | } | |
1da177e4 LT |
681 | if (noblock && sh == NULL) |
682 | break; | |
683 | if (!sh) { | |
5423399a N |
684 | set_bit(R5_INACTIVE_BLOCKED, |
685 | &conf->cache_state); | |
6ab2a4b8 SL |
686 | wait_event_lock_irq( |
687 | conf->wait_for_stripe, | |
566c09c5 SL |
688 | !list_empty(conf->inactive_list + hash) && |
689 | (atomic_read(&conf->active_stripes) | |
690 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
691 | || !test_bit(R5_INACTIVE_BLOCKED, |
692 | &conf->cache_state)), | |
6ab2a4b8 | 693 | *(conf->hash_locks + hash)); |
5423399a N |
694 | clear_bit(R5_INACTIVE_BLOCKED, |
695 | &conf->cache_state); | |
7da9d450 | 696 | } else { |
b5663ba4 | 697 | init_stripe(sh, sector, previous); |
7da9d450 N |
698 | atomic_inc(&sh->count); |
699 | } | |
e240c183 | 700 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 701 | spin_lock(&conf->device_lock); |
e240c183 | 702 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
703 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
704 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
705 | BUG_ON(list_empty(&sh->lru) && |
706 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
707 | inc_empty_inactive_list_flag = 0; |
708 | if (!list_empty(conf->inactive_list + hash)) | |
709 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 710 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
711 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
712 | atomic_inc(&conf->empty_inactive_list_nr); | |
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 | ||
566c09c5 | 723 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
724 | return sh; |
725 | } | |
726 | ||
7a87f434 | 727 | static bool is_full_stripe_write(struct stripe_head *sh) |
728 | { | |
729 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
730 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
731 | } | |
732 | ||
59fc630b | 733 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
734 | { | |
735 | local_irq_disable(); | |
736 | if (sh1 > sh2) { | |
737 | spin_lock(&sh2->stripe_lock); | |
738 | spin_lock_nested(&sh1->stripe_lock, 1); | |
739 | } else { | |
740 | spin_lock(&sh1->stripe_lock); | |
741 | spin_lock_nested(&sh2->stripe_lock, 1); | |
742 | } | |
743 | } | |
744 | ||
745 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
746 | { | |
747 | spin_unlock(&sh1->stripe_lock); | |
748 | spin_unlock(&sh2->stripe_lock); | |
749 | local_irq_enable(); | |
750 | } | |
751 | ||
752 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
753 | static bool stripe_can_batch(struct stripe_head *sh) | |
754 | { | |
9c3e333d SL |
755 | struct r5conf *conf = sh->raid_conf; |
756 | ||
757 | if (conf->log) | |
758 | return false; | |
59fc630b | 759 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 760 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 761 | is_full_stripe_write(sh); |
762 | } | |
763 | ||
764 | /* we only do back search */ | |
765 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
766 | { | |
767 | struct stripe_head *head; | |
768 | sector_t head_sector, tmp_sec; | |
769 | int hash; | |
770 | int dd_idx; | |
ff00d3b4 | 771 | int inc_empty_inactive_list_flag; |
59fc630b | 772 | |
59fc630b | 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)); | |
ff00d3b4 ZL |
789 | inc_empty_inactive_list_flag = 0; |
790 | if (!list_empty(conf->inactive_list + hash)) | |
791 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 792 | list_del_init(&head->lru); |
ff00d3b4 ZL |
793 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
794 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 795 | if (head->group) { |
796 | head->group->stripes_cnt--; | |
797 | head->group = NULL; | |
798 | } | |
799 | } | |
800 | atomic_inc(&head->count); | |
801 | spin_unlock(&conf->device_lock); | |
802 | } | |
803 | spin_unlock_irq(conf->hash_locks + hash); | |
804 | ||
805 | if (!head) | |
806 | return; | |
807 | if (!stripe_can_batch(head)) | |
808 | goto out; | |
809 | ||
810 | lock_two_stripes(head, sh); | |
811 | /* clear_batch_ready clear the flag */ | |
812 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
813 | goto unlock_out; | |
814 | ||
815 | if (sh->batch_head) | |
816 | goto unlock_out; | |
817 | ||
818 | dd_idx = 0; | |
819 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
820 | dd_idx++; | |
1eff9d32 | 821 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 822 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 823 | goto unlock_out; |
824 | ||
825 | if (head->batch_head) { | |
826 | spin_lock(&head->batch_head->batch_lock); | |
827 | /* This batch list is already running */ | |
828 | if (!stripe_can_batch(head)) { | |
829 | spin_unlock(&head->batch_head->batch_lock); | |
830 | goto unlock_out; | |
831 | } | |
832 | ||
833 | /* | |
834 | * at this point, head's BATCH_READY could be cleared, but we | |
835 | * can still add the stripe to batch list | |
836 | */ | |
837 | list_add(&sh->batch_list, &head->batch_list); | |
838 | spin_unlock(&head->batch_head->batch_lock); | |
839 | ||
840 | sh->batch_head = head->batch_head; | |
841 | } else { | |
842 | head->batch_head = head; | |
843 | sh->batch_head = head->batch_head; | |
844 | spin_lock(&head->batch_lock); | |
845 | list_add_tail(&sh->batch_list, &head->batch_list); | |
846 | spin_unlock(&head->batch_lock); | |
847 | } | |
848 | ||
849 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
850 | if (atomic_dec_return(&conf->preread_active_stripes) | |
851 | < IO_THRESHOLD) | |
852 | md_wakeup_thread(conf->mddev->thread); | |
853 | ||
2b6b2457 N |
854 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
855 | int seq = sh->bm_seq; | |
856 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
857 | sh->batch_head->bm_seq > seq) | |
858 | seq = sh->batch_head->bm_seq; | |
859 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
860 | sh->batch_head->bm_seq = seq; | |
861 | } | |
862 | ||
59fc630b | 863 | atomic_inc(&sh->count); |
864 | unlock_out: | |
865 | unlock_two_stripes(head, sh); | |
866 | out: | |
6d036f7d | 867 | raid5_release_stripe(head); |
59fc630b | 868 | } |
869 | ||
05616be5 N |
870 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
871 | * in this stripe_head. | |
872 | */ | |
873 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
874 | { | |
875 | sector_t progress = conf->reshape_progress; | |
876 | /* Need a memory barrier to make sure we see the value | |
877 | * of conf->generation, or ->data_offset that was set before | |
878 | * reshape_progress was updated. | |
879 | */ | |
880 | smp_rmb(); | |
881 | if (progress == MaxSector) | |
882 | return 0; | |
883 | if (sh->generation == conf->generation - 1) | |
884 | return 0; | |
885 | /* We are in a reshape, and this is a new-generation stripe, | |
886 | * so use new_data_offset. | |
887 | */ | |
888 | return 1; | |
889 | } | |
890 | ||
6712ecf8 | 891 | static void |
4246a0b6 | 892 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 893 | static void |
4246a0b6 | 894 | raid5_end_write_request(struct bio *bi); |
91c00924 | 895 | |
c4e5ac0a | 896 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 897 | { |
d1688a6d | 898 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 899 | int i, disks = sh->disks; |
59fc630b | 900 | struct stripe_head *head_sh = sh; |
91c00924 DW |
901 | |
902 | might_sleep(); | |
903 | ||
f6bed0ef SL |
904 | if (r5l_write_stripe(conf->log, sh) == 0) |
905 | return; | |
91c00924 | 906 | for (i = disks; i--; ) { |
796a5cf0 | 907 | int op, op_flags = 0; |
9a3e1101 | 908 | int replace_only = 0; |
977df362 N |
909 | struct bio *bi, *rbi; |
910 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 911 | |
912 | sh = head_sh; | |
e9c7469b | 913 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 914 | op = REQ_OP_WRITE; |
e9c7469b | 915 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
796a5cf0 | 916 | op_flags = WRITE_FUA; |
9e444768 | 917 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 918 | op = REQ_OP_DISCARD; |
e9c7469b | 919 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 920 | op = REQ_OP_READ; |
9a3e1101 N |
921 | else if (test_and_clear_bit(R5_WantReplace, |
922 | &sh->dev[i].flags)) { | |
796a5cf0 | 923 | op = REQ_OP_WRITE; |
9a3e1101 N |
924 | replace_only = 1; |
925 | } else | |
91c00924 | 926 | continue; |
bc0934f0 | 927 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 928 | op_flags |= REQ_SYNC; |
91c00924 | 929 | |
59fc630b | 930 | again: |
91c00924 | 931 | bi = &sh->dev[i].req; |
977df362 | 932 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 933 | |
91c00924 | 934 | rcu_read_lock(); |
9a3e1101 | 935 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
936 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
937 | rdev = rcu_dereference(conf->disks[i].rdev); | |
938 | if (!rdev) { | |
939 | rdev = rrdev; | |
940 | rrdev = NULL; | |
941 | } | |
796a5cf0 | 942 | if (op_is_write(op)) { |
9a3e1101 N |
943 | if (replace_only) |
944 | rdev = NULL; | |
dd054fce N |
945 | if (rdev == rrdev) |
946 | /* We raced and saw duplicates */ | |
947 | rrdev = NULL; | |
9a3e1101 | 948 | } else { |
59fc630b | 949 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
950 | rdev = rrdev; |
951 | rrdev = NULL; | |
952 | } | |
977df362 | 953 | |
91c00924 DW |
954 | if (rdev && test_bit(Faulty, &rdev->flags)) |
955 | rdev = NULL; | |
956 | if (rdev) | |
957 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
958 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
959 | rrdev = NULL; | |
960 | if (rrdev) | |
961 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
962 | rcu_read_unlock(); |
963 | ||
73e92e51 | 964 | /* We have already checked bad blocks for reads. Now |
977df362 N |
965 | * need to check for writes. We never accept write errors |
966 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 967 | */ |
796a5cf0 | 968 | while (op_is_write(op) && rdev && |
73e92e51 N |
969 | test_bit(WriteErrorSeen, &rdev->flags)) { |
970 | sector_t first_bad; | |
971 | int bad_sectors; | |
972 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
973 | &first_bad, &bad_sectors); | |
974 | if (!bad) | |
975 | break; | |
976 | ||
977 | if (bad < 0) { | |
978 | set_bit(BlockedBadBlocks, &rdev->flags); | |
979 | if (!conf->mddev->external && | |
980 | conf->mddev->flags) { | |
981 | /* It is very unlikely, but we might | |
982 | * still need to write out the | |
983 | * bad block log - better give it | |
984 | * a chance*/ | |
985 | md_check_recovery(conf->mddev); | |
986 | } | |
1850753d | 987 | /* |
988 | * Because md_wait_for_blocked_rdev | |
989 | * will dec nr_pending, we must | |
990 | * increment it first. | |
991 | */ | |
992 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
993 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
994 | } else { | |
995 | /* Acknowledged bad block - skip the write */ | |
996 | rdev_dec_pending(rdev, conf->mddev); | |
997 | rdev = NULL; | |
998 | } | |
999 | } | |
1000 | ||
91c00924 | 1001 | if (rdev) { |
9a3e1101 N |
1002 | if (s->syncing || s->expanding || s->expanded |
1003 | || s->replacing) | |
91c00924 DW |
1004 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1005 | ||
2b7497f0 DW |
1006 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1007 | ||
91c00924 | 1008 | bi->bi_bdev = rdev->bdev; |
796a5cf0 MC |
1009 | bio_set_op_attrs(bi, op, op_flags); |
1010 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1011 | ? raid5_end_write_request |
1012 | : raid5_end_read_request; | |
1013 | bi->bi_private = sh; | |
1014 | ||
6296b960 | 1015 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1016 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1017 | bi->bi_opf, i); |
91c00924 | 1018 | atomic_inc(&sh->count); |
59fc630b | 1019 | if (sh != head_sh) |
1020 | atomic_inc(&head_sh->count); | |
05616be5 | 1021 | if (use_new_offset(conf, sh)) |
4f024f37 | 1022 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1023 | + rdev->new_data_offset); |
1024 | else | |
4f024f37 | 1025 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1026 | + rdev->data_offset); |
59fc630b | 1027 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1028 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1029 | |
d592a996 SL |
1030 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1031 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1032 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1033 | bi->bi_vcnt = 1; |
91c00924 DW |
1034 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1035 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1036 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1037 | /* |
1038 | * If this is discard request, set bi_vcnt 0. We don't | |
1039 | * want to confuse SCSI because SCSI will replace payload | |
1040 | */ | |
796a5cf0 | 1041 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1042 | bi->bi_vcnt = 0; |
977df362 N |
1043 | if (rrdev) |
1044 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1045 | |
1046 | if (conf->mddev->gendisk) | |
1047 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1048 | bi, disk_devt(conf->mddev->gendisk), | |
1049 | sh->dev[i].sector); | |
91c00924 | 1050 | generic_make_request(bi); |
977df362 N |
1051 | } |
1052 | if (rrdev) { | |
9a3e1101 N |
1053 | if (s->syncing || s->expanding || s->expanded |
1054 | || s->replacing) | |
977df362 N |
1055 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1056 | ||
1057 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1058 | ||
1059 | rbi->bi_bdev = rrdev->bdev; | |
796a5cf0 MC |
1060 | bio_set_op_attrs(rbi, op, op_flags); |
1061 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1062 | rbi->bi_end_io = raid5_end_write_request; |
1063 | rbi->bi_private = sh; | |
1064 | ||
6296b960 | 1065 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1066 | "replacement disc %d\n", |
1067 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1068 | rbi->bi_opf, i); |
977df362 | 1069 | atomic_inc(&sh->count); |
59fc630b | 1070 | if (sh != head_sh) |
1071 | atomic_inc(&head_sh->count); | |
05616be5 | 1072 | if (use_new_offset(conf, sh)) |
4f024f37 | 1073 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1074 | + rrdev->new_data_offset); |
1075 | else | |
4f024f37 | 1076 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1077 | + rrdev->data_offset); |
d592a996 SL |
1078 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1079 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1080 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1081 | rbi->bi_vcnt = 1; |
977df362 N |
1082 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1083 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1084 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1085 | /* |
1086 | * If this is discard request, set bi_vcnt 0. We don't | |
1087 | * want to confuse SCSI because SCSI will replace payload | |
1088 | */ | |
796a5cf0 | 1089 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1090 | rbi->bi_vcnt = 0; |
e3620a3a JB |
1091 | if (conf->mddev->gendisk) |
1092 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1093 | rbi, disk_devt(conf->mddev->gendisk), | |
1094 | sh->dev[i].sector); | |
977df362 N |
1095 | generic_make_request(rbi); |
1096 | } | |
1097 | if (!rdev && !rrdev) { | |
796a5cf0 | 1098 | if (op_is_write(op)) |
91c00924 | 1099 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1100 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1101 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1102 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1103 | set_bit(STRIPE_HANDLE, &sh->state); | |
1104 | } | |
59fc630b | 1105 | |
1106 | if (!head_sh->batch_head) | |
1107 | continue; | |
1108 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1109 | batch_list); | |
1110 | if (sh != head_sh) | |
1111 | goto again; | |
91c00924 DW |
1112 | } |
1113 | } | |
1114 | ||
1115 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1116 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1117 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1118 | struct stripe_head *sh) | |
91c00924 | 1119 | { |
7988613b KO |
1120 | struct bio_vec bvl; |
1121 | struct bvec_iter iter; | |
91c00924 | 1122 | struct page *bio_page; |
91c00924 | 1123 | int page_offset; |
a08abd8c | 1124 | struct async_submit_ctl submit; |
0403e382 | 1125 | enum async_tx_flags flags = 0; |
91c00924 | 1126 | |
4f024f37 KO |
1127 | if (bio->bi_iter.bi_sector >= sector) |
1128 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1129 | else |
4f024f37 | 1130 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1131 | |
0403e382 DW |
1132 | if (frombio) |
1133 | flags |= ASYNC_TX_FENCE; | |
1134 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1135 | ||
7988613b KO |
1136 | bio_for_each_segment(bvl, bio, iter) { |
1137 | int len = bvl.bv_len; | |
91c00924 DW |
1138 | int clen; |
1139 | int b_offset = 0; | |
1140 | ||
1141 | if (page_offset < 0) { | |
1142 | b_offset = -page_offset; | |
1143 | page_offset += b_offset; | |
1144 | len -= b_offset; | |
1145 | } | |
1146 | ||
1147 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1148 | clen = STRIPE_SIZE - page_offset; | |
1149 | else | |
1150 | clen = len; | |
1151 | ||
1152 | if (clen > 0) { | |
7988613b KO |
1153 | b_offset += bvl.bv_offset; |
1154 | bio_page = bvl.bv_page; | |
d592a996 SL |
1155 | if (frombio) { |
1156 | if (sh->raid_conf->skip_copy && | |
1157 | b_offset == 0 && page_offset == 0 && | |
1158 | clen == STRIPE_SIZE) | |
1159 | *page = bio_page; | |
1160 | else | |
1161 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1162 | b_offset, clen, &submit); |
d592a996 SL |
1163 | } else |
1164 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1165 | page_offset, clen, &submit); |
91c00924 | 1166 | } |
a08abd8c DW |
1167 | /* chain the operations */ |
1168 | submit.depend_tx = tx; | |
1169 | ||
91c00924 DW |
1170 | if (clen < len) /* hit end of page */ |
1171 | break; | |
1172 | page_offset += len; | |
1173 | } | |
1174 | ||
1175 | return tx; | |
1176 | } | |
1177 | ||
1178 | static void ops_complete_biofill(void *stripe_head_ref) | |
1179 | { | |
1180 | struct stripe_head *sh = stripe_head_ref; | |
34a6f80e | 1181 | struct bio_list return_bi = BIO_EMPTY_LIST; |
e4d84909 | 1182 | int i; |
91c00924 | 1183 | |
e46b272b | 1184 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1185 | (unsigned long long)sh->sector); |
1186 | ||
1187 | /* clear completed biofills */ | |
1188 | for (i = sh->disks; i--; ) { | |
1189 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1190 | |
1191 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1192 | /* and check if we need to reply to a read request, |
1193 | * new R5_Wantfill requests are held off until | |
83de75cc | 1194 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1195 | */ |
1196 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1197 | struct bio *rbi, *rbi2; |
91c00924 | 1198 | |
91c00924 DW |
1199 | BUG_ON(!dev->read); |
1200 | rbi = dev->read; | |
1201 | dev->read = NULL; | |
4f024f37 | 1202 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1203 | dev->sector + STRIPE_SECTORS) { |
1204 | rbi2 = r5_next_bio(rbi, dev->sector); | |
34a6f80e N |
1205 | if (!raid5_dec_bi_active_stripes(rbi)) |
1206 | bio_list_add(&return_bi, rbi); | |
91c00924 DW |
1207 | rbi = rbi2; |
1208 | } | |
1209 | } | |
1210 | } | |
83de75cc | 1211 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1212 | |
34a6f80e | 1213 | return_io(&return_bi); |
91c00924 | 1214 | |
e4d84909 | 1215 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1216 | raid5_release_stripe(sh); |
91c00924 DW |
1217 | } |
1218 | ||
1219 | static void ops_run_biofill(struct stripe_head *sh) | |
1220 | { | |
1221 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1222 | struct async_submit_ctl submit; |
91c00924 DW |
1223 | int i; |
1224 | ||
59fc630b | 1225 | BUG_ON(sh->batch_head); |
e46b272b | 1226 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1227 | (unsigned long long)sh->sector); |
1228 | ||
1229 | for (i = sh->disks; i--; ) { | |
1230 | struct r5dev *dev = &sh->dev[i]; | |
1231 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1232 | struct bio *rbi; | |
b17459c0 | 1233 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1234 | dev->read = rbi = dev->toread; |
1235 | dev->toread = NULL; | |
b17459c0 | 1236 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1237 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1238 | dev->sector + STRIPE_SECTORS) { |
d592a996 SL |
1239 | tx = async_copy_data(0, rbi, &dev->page, |
1240 | dev->sector, tx, sh); | |
91c00924 DW |
1241 | rbi = r5_next_bio(rbi, dev->sector); |
1242 | } | |
1243 | } | |
1244 | } | |
1245 | ||
1246 | atomic_inc(&sh->count); | |
a08abd8c DW |
1247 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1248 | async_trigger_callback(&submit); | |
91c00924 DW |
1249 | } |
1250 | ||
4e7d2c0a | 1251 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1252 | { |
4e7d2c0a | 1253 | struct r5dev *tgt; |
91c00924 | 1254 | |
4e7d2c0a DW |
1255 | if (target < 0) |
1256 | return; | |
91c00924 | 1257 | |
4e7d2c0a | 1258 | tgt = &sh->dev[target]; |
91c00924 DW |
1259 | set_bit(R5_UPTODATE, &tgt->flags); |
1260 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1261 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1262 | } |
1263 | ||
ac6b53b6 | 1264 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1265 | { |
1266 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1267 | |
e46b272b | 1268 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1269 | (unsigned long long)sh->sector); |
1270 | ||
ac6b53b6 | 1271 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1272 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1273 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1274 | |
ecc65c9b DW |
1275 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1276 | if (sh->check_state == check_state_compute_run) | |
1277 | sh->check_state = check_state_compute_result; | |
91c00924 | 1278 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1279 | raid5_release_stripe(sh); |
91c00924 DW |
1280 | } |
1281 | ||
d6f38f31 DW |
1282 | /* return a pointer to the address conversion region of the scribble buffer */ |
1283 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1284 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1285 | { |
46d5b785 | 1286 | void *addr; |
1287 | ||
1288 | addr = flex_array_get(percpu->scribble, i); | |
1289 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1290 | } | |
1291 | ||
1292 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1293 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1294 | { | |
1295 | void *addr; | |
1296 | ||
1297 | addr = flex_array_get(percpu->scribble, i); | |
1298 | return addr; | |
d6f38f31 DW |
1299 | } |
1300 | ||
1301 | static struct dma_async_tx_descriptor * | |
1302 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1303 | { |
91c00924 | 1304 | int disks = sh->disks; |
46d5b785 | 1305 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1306 | int target = sh->ops.target; |
1307 | struct r5dev *tgt = &sh->dev[target]; | |
1308 | struct page *xor_dest = tgt->page; | |
1309 | int count = 0; | |
1310 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1311 | struct async_submit_ctl submit; |
91c00924 DW |
1312 | int i; |
1313 | ||
59fc630b | 1314 | BUG_ON(sh->batch_head); |
1315 | ||
91c00924 | 1316 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1317 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1318 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1319 | ||
1320 | for (i = disks; i--; ) | |
1321 | if (i != target) | |
1322 | xor_srcs[count++] = sh->dev[i].page; | |
1323 | ||
1324 | atomic_inc(&sh->count); | |
1325 | ||
0403e382 | 1326 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1327 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1328 | if (unlikely(count == 1)) |
a08abd8c | 1329 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1330 | else |
a08abd8c | 1331 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1332 | |
91c00924 DW |
1333 | return tx; |
1334 | } | |
1335 | ||
ac6b53b6 DW |
1336 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1337 | * @srcs - (struct page *) array of size sh->disks | |
1338 | * @sh - stripe_head to parse | |
1339 | * | |
1340 | * Populates srcs in proper layout order for the stripe and returns the | |
1341 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1342 | * destination buffer is recorded in srcs[count] and the Q destination | |
1343 | * is recorded in srcs[count+1]]. | |
1344 | */ | |
584acdd4 MS |
1345 | static int set_syndrome_sources(struct page **srcs, |
1346 | struct stripe_head *sh, | |
1347 | int srctype) | |
ac6b53b6 DW |
1348 | { |
1349 | int disks = sh->disks; | |
1350 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1351 | int d0_idx = raid6_d0(sh); | |
1352 | int count; | |
1353 | int i; | |
1354 | ||
1355 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1356 | srcs[i] = NULL; |
ac6b53b6 DW |
1357 | |
1358 | count = 0; | |
1359 | i = d0_idx; | |
1360 | do { | |
1361 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1362 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1363 | |
584acdd4 MS |
1364 | if (i == sh->qd_idx || i == sh->pd_idx || |
1365 | (srctype == SYNDROME_SRC_ALL) || | |
1366 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1367 | test_bit(R5_Wantdrain, &dev->flags)) || | |
1368 | (srctype == SYNDROME_SRC_WRITTEN && | |
1369 | dev->written)) | |
1370 | srcs[slot] = sh->dev[i].page; | |
ac6b53b6 DW |
1371 | i = raid6_next_disk(i, disks); |
1372 | } while (i != d0_idx); | |
ac6b53b6 | 1373 | |
e4424fee | 1374 | return syndrome_disks; |
ac6b53b6 DW |
1375 | } |
1376 | ||
1377 | static struct dma_async_tx_descriptor * | |
1378 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1379 | { | |
1380 | int disks = sh->disks; | |
46d5b785 | 1381 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1382 | int target; |
1383 | int qd_idx = sh->qd_idx; | |
1384 | struct dma_async_tx_descriptor *tx; | |
1385 | struct async_submit_ctl submit; | |
1386 | struct r5dev *tgt; | |
1387 | struct page *dest; | |
1388 | int i; | |
1389 | int count; | |
1390 | ||
59fc630b | 1391 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1392 | if (sh->ops.target < 0) |
1393 | target = sh->ops.target2; | |
1394 | else if (sh->ops.target2 < 0) | |
1395 | target = sh->ops.target; | |
91c00924 | 1396 | else |
ac6b53b6 DW |
1397 | /* we should only have one valid target */ |
1398 | BUG(); | |
1399 | BUG_ON(target < 0); | |
1400 | pr_debug("%s: stripe %llu block: %d\n", | |
1401 | __func__, (unsigned long long)sh->sector, target); | |
1402 | ||
1403 | tgt = &sh->dev[target]; | |
1404 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1405 | dest = tgt->page; | |
1406 | ||
1407 | atomic_inc(&sh->count); | |
1408 | ||
1409 | if (target == qd_idx) { | |
584acdd4 | 1410 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1411 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1412 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1413 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1414 | ops_complete_compute, sh, | |
46d5b785 | 1415 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1416 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1417 | } else { | |
1418 | /* Compute any data- or p-drive using XOR */ | |
1419 | count = 0; | |
1420 | for (i = disks; i-- ; ) { | |
1421 | if (i == target || i == qd_idx) | |
1422 | continue; | |
1423 | blocks[count++] = sh->dev[i].page; | |
1424 | } | |
1425 | ||
0403e382 DW |
1426 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1427 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1428 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1429 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1430 | } | |
91c00924 | 1431 | |
91c00924 DW |
1432 | return tx; |
1433 | } | |
1434 | ||
ac6b53b6 DW |
1435 | static struct dma_async_tx_descriptor * |
1436 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1437 | { | |
1438 | int i, count, disks = sh->disks; | |
1439 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1440 | int d0_idx = raid6_d0(sh); | |
1441 | int faila = -1, failb = -1; | |
1442 | int target = sh->ops.target; | |
1443 | int target2 = sh->ops.target2; | |
1444 | struct r5dev *tgt = &sh->dev[target]; | |
1445 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1446 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1447 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1448 | struct async_submit_ctl submit; |
1449 | ||
59fc630b | 1450 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1451 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1452 | __func__, (unsigned long long)sh->sector, target, target2); | |
1453 | BUG_ON(target < 0 || target2 < 0); | |
1454 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1455 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1456 | ||
6c910a78 | 1457 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1458 | * slot number conversion for 'faila' and 'failb' |
1459 | */ | |
1460 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1461 | blocks[i] = NULL; |
ac6b53b6 DW |
1462 | count = 0; |
1463 | i = d0_idx; | |
1464 | do { | |
1465 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1466 | ||
1467 | blocks[slot] = sh->dev[i].page; | |
1468 | ||
1469 | if (i == target) | |
1470 | faila = slot; | |
1471 | if (i == target2) | |
1472 | failb = slot; | |
1473 | i = raid6_next_disk(i, disks); | |
1474 | } while (i != d0_idx); | |
ac6b53b6 DW |
1475 | |
1476 | BUG_ON(faila == failb); | |
1477 | if (failb < faila) | |
1478 | swap(faila, failb); | |
1479 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1480 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1481 | ||
1482 | atomic_inc(&sh->count); | |
1483 | ||
1484 | if (failb == syndrome_disks+1) { | |
1485 | /* Q disk is one of the missing disks */ | |
1486 | if (faila == syndrome_disks) { | |
1487 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1488 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1489 | ops_complete_compute, sh, | |
46d5b785 | 1490 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1491 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1492 | STRIPE_SIZE, &submit); |
1493 | } else { | |
1494 | struct page *dest; | |
1495 | int data_target; | |
1496 | int qd_idx = sh->qd_idx; | |
1497 | ||
1498 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1499 | if (target == qd_idx) | |
1500 | data_target = target2; | |
1501 | else | |
1502 | data_target = target; | |
1503 | ||
1504 | count = 0; | |
1505 | for (i = disks; i-- ; ) { | |
1506 | if (i == data_target || i == qd_idx) | |
1507 | continue; | |
1508 | blocks[count++] = sh->dev[i].page; | |
1509 | } | |
1510 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1511 | init_async_submit(&submit, |
1512 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1513 | NULL, NULL, NULL, | |
46d5b785 | 1514 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1515 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1516 | &submit); | |
1517 | ||
584acdd4 | 1518 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1519 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1520 | ops_complete_compute, sh, | |
46d5b785 | 1521 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1522 | return async_gen_syndrome(blocks, 0, count+2, |
1523 | STRIPE_SIZE, &submit); | |
1524 | } | |
ac6b53b6 | 1525 | } else { |
6c910a78 DW |
1526 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1527 | ops_complete_compute, sh, | |
46d5b785 | 1528 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1529 | if (failb == syndrome_disks) { |
1530 | /* We're missing D+P. */ | |
1531 | return async_raid6_datap_recov(syndrome_disks+2, | |
1532 | STRIPE_SIZE, faila, | |
1533 | blocks, &submit); | |
1534 | } else { | |
1535 | /* We're missing D+D. */ | |
1536 | return async_raid6_2data_recov(syndrome_disks+2, | |
1537 | STRIPE_SIZE, faila, failb, | |
1538 | blocks, &submit); | |
1539 | } | |
ac6b53b6 DW |
1540 | } |
1541 | } | |
1542 | ||
91c00924 DW |
1543 | static void ops_complete_prexor(void *stripe_head_ref) |
1544 | { | |
1545 | struct stripe_head *sh = stripe_head_ref; | |
1546 | ||
e46b272b | 1547 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1548 | (unsigned long long)sh->sector); |
91c00924 DW |
1549 | } |
1550 | ||
1551 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1552 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1553 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1554 | { |
91c00924 | 1555 | int disks = sh->disks; |
46d5b785 | 1556 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1557 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1558 | struct async_submit_ctl submit; |
91c00924 DW |
1559 | |
1560 | /* existing parity data subtracted */ | |
1561 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1562 | ||
59fc630b | 1563 | BUG_ON(sh->batch_head); |
e46b272b | 1564 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1565 | (unsigned long long)sh->sector); |
1566 | ||
1567 | for (i = disks; i--; ) { | |
1568 | struct r5dev *dev = &sh->dev[i]; | |
1569 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1570 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1571 | xor_srcs[count++] = dev->page; |
1572 | } | |
1573 | ||
0403e382 | 1574 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1575 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1576 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1577 | |
1578 | return tx; | |
1579 | } | |
1580 | ||
584acdd4 MS |
1581 | static struct dma_async_tx_descriptor * |
1582 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1583 | struct dma_async_tx_descriptor *tx) | |
1584 | { | |
1585 | struct page **blocks = to_addr_page(percpu, 0); | |
1586 | int count; | |
1587 | struct async_submit_ctl submit; | |
1588 | ||
1589 | pr_debug("%s: stripe %llu\n", __func__, | |
1590 | (unsigned long long)sh->sector); | |
1591 | ||
1592 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1593 | ||
1594 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1595 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1596 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1597 | ||
1598 | return tx; | |
1599 | } | |
1600 | ||
91c00924 | 1601 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1602 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1603 | { |
1604 | int disks = sh->disks; | |
d8ee0728 | 1605 | int i; |
59fc630b | 1606 | struct stripe_head *head_sh = sh; |
91c00924 | 1607 | |
e46b272b | 1608 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1609 | (unsigned long long)sh->sector); |
1610 | ||
1611 | for (i = disks; i--; ) { | |
59fc630b | 1612 | struct r5dev *dev; |
91c00924 | 1613 | struct bio *chosen; |
91c00924 | 1614 | |
59fc630b | 1615 | sh = head_sh; |
1616 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1617 | struct bio *wbi; |
1618 | ||
59fc630b | 1619 | again: |
1620 | dev = &sh->dev[i]; | |
b17459c0 | 1621 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1622 | chosen = dev->towrite; |
1623 | dev->towrite = NULL; | |
7a87f434 | 1624 | sh->overwrite_disks = 0; |
91c00924 DW |
1625 | BUG_ON(dev->written); |
1626 | wbi = dev->written = chosen; | |
b17459c0 | 1627 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1628 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1629 | |
4f024f37 | 1630 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1631 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1632 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1633 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1634 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1635 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1636 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1637 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1638 | else { |
1639 | tx = async_copy_data(1, wbi, &dev->page, | |
1640 | dev->sector, tx, sh); | |
1641 | if (dev->page != dev->orig_page) { | |
1642 | set_bit(R5_SkipCopy, &dev->flags); | |
1643 | clear_bit(R5_UPTODATE, &dev->flags); | |
1644 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1645 | } | |
1646 | } | |
91c00924 DW |
1647 | wbi = r5_next_bio(wbi, dev->sector); |
1648 | } | |
59fc630b | 1649 | |
1650 | if (head_sh->batch_head) { | |
1651 | sh = list_first_entry(&sh->batch_list, | |
1652 | struct stripe_head, | |
1653 | batch_list); | |
1654 | if (sh == head_sh) | |
1655 | continue; | |
1656 | goto again; | |
1657 | } | |
91c00924 DW |
1658 | } |
1659 | } | |
1660 | ||
1661 | return tx; | |
1662 | } | |
1663 | ||
ac6b53b6 | 1664 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1665 | { |
1666 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1667 | int disks = sh->disks; |
1668 | int pd_idx = sh->pd_idx; | |
1669 | int qd_idx = sh->qd_idx; | |
1670 | int i; | |
9e444768 | 1671 | bool fua = false, sync = false, discard = false; |
91c00924 | 1672 | |
e46b272b | 1673 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1674 | (unsigned long long)sh->sector); |
1675 | ||
bc0934f0 | 1676 | for (i = disks; i--; ) { |
e9c7469b | 1677 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1678 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1679 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1680 | } |
e9c7469b | 1681 | |
91c00924 DW |
1682 | for (i = disks; i--; ) { |
1683 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1684 | |
e9c7469b | 1685 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1686 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1687 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1688 | if (fua) |
1689 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1690 | if (sync) |
1691 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1692 | } |
91c00924 DW |
1693 | } |
1694 | ||
d8ee0728 DW |
1695 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1696 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1697 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1698 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1699 | else { | |
1700 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1701 | sh->reconstruct_state = reconstruct_state_result; | |
1702 | } | |
91c00924 DW |
1703 | |
1704 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1705 | raid5_release_stripe(sh); |
91c00924 DW |
1706 | } |
1707 | ||
1708 | static void | |
ac6b53b6 DW |
1709 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1710 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1711 | { |
91c00924 | 1712 | int disks = sh->disks; |
59fc630b | 1713 | struct page **xor_srcs; |
a08abd8c | 1714 | struct async_submit_ctl submit; |
59fc630b | 1715 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1716 | struct page *xor_dest; |
d8ee0728 | 1717 | int prexor = 0; |
91c00924 | 1718 | unsigned long flags; |
59fc630b | 1719 | int j = 0; |
1720 | struct stripe_head *head_sh = sh; | |
1721 | int last_stripe; | |
91c00924 | 1722 | |
e46b272b | 1723 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1724 | (unsigned long long)sh->sector); |
1725 | ||
620125f2 SL |
1726 | for (i = 0; i < sh->disks; i++) { |
1727 | if (pd_idx == i) | |
1728 | continue; | |
1729 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1730 | break; | |
1731 | } | |
1732 | if (i >= sh->disks) { | |
1733 | atomic_inc(&sh->count); | |
620125f2 SL |
1734 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1735 | ops_complete_reconstruct(sh); | |
1736 | return; | |
1737 | } | |
59fc630b | 1738 | again: |
1739 | count = 0; | |
1740 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1741 | /* check if prexor is active which means only process blocks |
1742 | * that are part of a read-modify-write (written) | |
1743 | */ | |
59fc630b | 1744 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1745 | prexor = 1; |
91c00924 DW |
1746 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1747 | for (i = disks; i--; ) { | |
1748 | struct r5dev *dev = &sh->dev[i]; | |
59fc630b | 1749 | if (head_sh->dev[i].written) |
91c00924 DW |
1750 | xor_srcs[count++] = dev->page; |
1751 | } | |
1752 | } else { | |
1753 | xor_dest = sh->dev[pd_idx].page; | |
1754 | for (i = disks; i--; ) { | |
1755 | struct r5dev *dev = &sh->dev[i]; | |
1756 | if (i != pd_idx) | |
1757 | xor_srcs[count++] = dev->page; | |
1758 | } | |
1759 | } | |
1760 | ||
91c00924 DW |
1761 | /* 1/ if we prexor'd then the dest is reused as a source |
1762 | * 2/ if we did not prexor then we are redoing the parity | |
1763 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1764 | * for the synchronous xor case | |
1765 | */ | |
59fc630b | 1766 | last_stripe = !head_sh->batch_head || |
1767 | list_first_entry(&sh->batch_list, | |
1768 | struct stripe_head, batch_list) == head_sh; | |
1769 | if (last_stripe) { | |
1770 | flags = ASYNC_TX_ACK | | |
1771 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1772 | ||
1773 | atomic_inc(&head_sh->count); | |
1774 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1775 | to_addr_conv(sh, percpu, j)); | |
1776 | } else { | |
1777 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1778 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1779 | to_addr_conv(sh, percpu, j)); | |
1780 | } | |
91c00924 | 1781 | |
a08abd8c DW |
1782 | if (unlikely(count == 1)) |
1783 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1784 | else | |
1785 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1786 | if (!last_stripe) { |
1787 | j++; | |
1788 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1789 | batch_list); | |
1790 | goto again; | |
1791 | } | |
91c00924 DW |
1792 | } |
1793 | ||
ac6b53b6 DW |
1794 | static void |
1795 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1796 | struct dma_async_tx_descriptor *tx) | |
1797 | { | |
1798 | struct async_submit_ctl submit; | |
59fc630b | 1799 | struct page **blocks; |
1800 | int count, i, j = 0; | |
1801 | struct stripe_head *head_sh = sh; | |
1802 | int last_stripe; | |
584acdd4 MS |
1803 | int synflags; |
1804 | unsigned long txflags; | |
ac6b53b6 DW |
1805 | |
1806 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1807 | ||
620125f2 SL |
1808 | for (i = 0; i < sh->disks; i++) { |
1809 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1810 | continue; | |
1811 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1812 | break; | |
1813 | } | |
1814 | if (i >= sh->disks) { | |
1815 | atomic_inc(&sh->count); | |
620125f2 SL |
1816 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1817 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1818 | ops_complete_reconstruct(sh); | |
1819 | return; | |
1820 | } | |
1821 | ||
59fc630b | 1822 | again: |
1823 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1824 | |
1825 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1826 | synflags = SYNDROME_SRC_WRITTEN; | |
1827 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1828 | } else { | |
1829 | synflags = SYNDROME_SRC_ALL; | |
1830 | txflags = ASYNC_TX_ACK; | |
1831 | } | |
1832 | ||
1833 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1834 | last_stripe = !head_sh->batch_head || |
1835 | list_first_entry(&sh->batch_list, | |
1836 | struct stripe_head, batch_list) == head_sh; | |
1837 | ||
1838 | if (last_stripe) { | |
1839 | atomic_inc(&head_sh->count); | |
584acdd4 | 1840 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1841 | head_sh, to_addr_conv(sh, percpu, j)); |
1842 | } else | |
1843 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1844 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1845 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1846 | if (!last_stripe) { |
1847 | j++; | |
1848 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1849 | batch_list); | |
1850 | goto again; | |
1851 | } | |
91c00924 DW |
1852 | } |
1853 | ||
1854 | static void ops_complete_check(void *stripe_head_ref) | |
1855 | { | |
1856 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1857 | |
e46b272b | 1858 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1859 | (unsigned long long)sh->sector); |
1860 | ||
ecc65c9b | 1861 | sh->check_state = check_state_check_result; |
91c00924 | 1862 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1863 | raid5_release_stripe(sh); |
91c00924 DW |
1864 | } |
1865 | ||
ac6b53b6 | 1866 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1867 | { |
91c00924 | 1868 | int disks = sh->disks; |
ac6b53b6 DW |
1869 | int pd_idx = sh->pd_idx; |
1870 | int qd_idx = sh->qd_idx; | |
1871 | struct page *xor_dest; | |
46d5b785 | 1872 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1873 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1874 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1875 | int count; |
1876 | int i; | |
91c00924 | 1877 | |
e46b272b | 1878 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1879 | (unsigned long long)sh->sector); |
1880 | ||
59fc630b | 1881 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1882 | count = 0; |
1883 | xor_dest = sh->dev[pd_idx].page; | |
1884 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1885 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1886 | if (i == pd_idx || i == qd_idx) |
1887 | continue; | |
1888 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1889 | } |
1890 | ||
d6f38f31 | 1891 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1892 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1893 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1894 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1895 | |
91c00924 | 1896 | atomic_inc(&sh->count); |
a08abd8c DW |
1897 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1898 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1899 | } |
1900 | ||
ac6b53b6 DW |
1901 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1902 | { | |
46d5b785 | 1903 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1904 | struct async_submit_ctl submit; |
1905 | int count; | |
1906 | ||
1907 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1908 | (unsigned long long)sh->sector, checkp); | |
1909 | ||
59fc630b | 1910 | BUG_ON(sh->batch_head); |
584acdd4 | 1911 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1912 | if (!checkp) |
1913 | srcs[count] = NULL; | |
91c00924 | 1914 | |
91c00924 | 1915 | atomic_inc(&sh->count); |
ac6b53b6 | 1916 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1917 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1918 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1919 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1920 | } |
1921 | ||
51acbcec | 1922 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1923 | { |
1924 | int overlap_clear = 0, i, disks = sh->disks; | |
1925 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1926 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1927 | int level = conf->level; |
d6f38f31 DW |
1928 | struct raid5_percpu *percpu; |
1929 | unsigned long cpu; | |
91c00924 | 1930 | |
d6f38f31 DW |
1931 | cpu = get_cpu(); |
1932 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1933 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1934 | ops_run_biofill(sh); |
1935 | overlap_clear++; | |
1936 | } | |
1937 | ||
7b3a871e | 1938 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1939 | if (level < 6) |
1940 | tx = ops_run_compute5(sh, percpu); | |
1941 | else { | |
1942 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1943 | tx = ops_run_compute6_1(sh, percpu); | |
1944 | else | |
1945 | tx = ops_run_compute6_2(sh, percpu); | |
1946 | } | |
1947 | /* terminate the chain if reconstruct is not set to be run */ | |
1948 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1949 | async_tx_ack(tx); |
1950 | } | |
91c00924 | 1951 | |
584acdd4 MS |
1952 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
1953 | if (level < 6) | |
1954 | tx = ops_run_prexor5(sh, percpu, tx); | |
1955 | else | |
1956 | tx = ops_run_prexor6(sh, percpu, tx); | |
1957 | } | |
91c00924 | 1958 | |
600aa109 | 1959 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1960 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1961 | overlap_clear++; |
1962 | } | |
1963 | ||
ac6b53b6 DW |
1964 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1965 | if (level < 6) | |
1966 | ops_run_reconstruct5(sh, percpu, tx); | |
1967 | else | |
1968 | ops_run_reconstruct6(sh, percpu, tx); | |
1969 | } | |
91c00924 | 1970 | |
ac6b53b6 DW |
1971 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1972 | if (sh->check_state == check_state_run) | |
1973 | ops_run_check_p(sh, percpu); | |
1974 | else if (sh->check_state == check_state_run_q) | |
1975 | ops_run_check_pq(sh, percpu, 0); | |
1976 | else if (sh->check_state == check_state_run_pq) | |
1977 | ops_run_check_pq(sh, percpu, 1); | |
1978 | else | |
1979 | BUG(); | |
1980 | } | |
91c00924 | 1981 | |
59fc630b | 1982 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
1983 | for (i = disks; i--; ) { |
1984 | struct r5dev *dev = &sh->dev[i]; | |
1985 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1986 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1987 | } | |
d6f38f31 | 1988 | put_cpu(); |
91c00924 DW |
1989 | } |
1990 | ||
5f9d1fde SL |
1991 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
1992 | int disks) | |
f18c1a35 N |
1993 | { |
1994 | struct stripe_head *sh; | |
5f9d1fde | 1995 | int i; |
f18c1a35 N |
1996 | |
1997 | sh = kmem_cache_zalloc(sc, gfp); | |
1998 | if (sh) { | |
1999 | spin_lock_init(&sh->stripe_lock); | |
2000 | spin_lock_init(&sh->batch_lock); | |
2001 | INIT_LIST_HEAD(&sh->batch_list); | |
2002 | INIT_LIST_HEAD(&sh->lru); | |
2003 | atomic_set(&sh->count, 1); | |
5f9d1fde SL |
2004 | for (i = 0; i < disks; i++) { |
2005 | struct r5dev *dev = &sh->dev[i]; | |
2006 | ||
2007 | bio_init(&dev->req); | |
45c91d80 SL |
2008 | dev->req.bi_io_vec = &dev->vec; |
2009 | dev->req.bi_max_vecs = 1; | |
2010 | ||
5f9d1fde | 2011 | bio_init(&dev->rreq); |
45c91d80 SL |
2012 | dev->rreq.bi_io_vec = &dev->rvec; |
2013 | dev->rreq.bi_max_vecs = 1; | |
5f9d1fde | 2014 | } |
f18c1a35 N |
2015 | } |
2016 | return sh; | |
2017 | } | |
486f0644 | 2018 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2019 | { |
2020 | struct stripe_head *sh; | |
f18c1a35 | 2021 | |
5f9d1fde | 2022 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size); |
3f294f4f N |
2023 | if (!sh) |
2024 | return 0; | |
6ce32846 | 2025 | |
3f294f4f | 2026 | sh->raid_conf = conf; |
3f294f4f | 2027 | |
a9683a79 | 2028 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2029 | shrink_buffers(sh); |
3f294f4f N |
2030 | kmem_cache_free(conf->slab_cache, sh); |
2031 | return 0; | |
2032 | } | |
486f0644 N |
2033 | sh->hash_lock_index = |
2034 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2035 | /* we just created an active stripe so... */ |
3f294f4f | 2036 | atomic_inc(&conf->active_stripes); |
59fc630b | 2037 | |
6d036f7d | 2038 | raid5_release_stripe(sh); |
486f0644 | 2039 | conf->max_nr_stripes++; |
3f294f4f N |
2040 | return 1; |
2041 | } | |
2042 | ||
d1688a6d | 2043 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2044 | { |
e18b890b | 2045 | struct kmem_cache *sc; |
5e5e3e78 | 2046 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2047 | |
f4be6b43 N |
2048 | if (conf->mddev->gendisk) |
2049 | sprintf(conf->cache_name[0], | |
2050 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2051 | else | |
2052 | sprintf(conf->cache_name[0], | |
2053 | "raid%d-%p", conf->level, conf->mddev); | |
2054 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2055 | ||
ad01c9e3 N |
2056 | conf->active_name = 0; |
2057 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2058 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2059 | 0, 0, NULL); |
1da177e4 LT |
2060 | if (!sc) |
2061 | return 1; | |
2062 | conf->slab_cache = sc; | |
ad01c9e3 | 2063 | conf->pool_size = devs; |
486f0644 N |
2064 | while (num--) |
2065 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2066 | return 1; |
486f0644 | 2067 | |
1da177e4 LT |
2068 | return 0; |
2069 | } | |
29269553 | 2070 | |
d6f38f31 DW |
2071 | /** |
2072 | * scribble_len - return the required size of the scribble region | |
2073 | * @num - total number of disks in the array | |
2074 | * | |
2075 | * The size must be enough to contain: | |
2076 | * 1/ a struct page pointer for each device in the array +2 | |
2077 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2078 | * (dma_map_page()) or page (page_address()) address. | |
2079 | * | |
2080 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2081 | * calculate over all devices (not just the data blocks), using zeros in place | |
2082 | * of the P and Q blocks. | |
2083 | */ | |
46d5b785 | 2084 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2085 | { |
46d5b785 | 2086 | struct flex_array *ret; |
d6f38f31 DW |
2087 | size_t len; |
2088 | ||
2089 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2090 | ret = flex_array_alloc(len, cnt, flags); |
2091 | if (!ret) | |
2092 | return NULL; | |
2093 | /* always prealloc all elements, so no locking is required */ | |
2094 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2095 | flex_array_free(ret); | |
2096 | return NULL; | |
2097 | } | |
2098 | return ret; | |
d6f38f31 DW |
2099 | } |
2100 | ||
738a2738 N |
2101 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2102 | { | |
2103 | unsigned long cpu; | |
2104 | int err = 0; | |
2105 | ||
27a353c0 SL |
2106 | /* |
2107 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2108 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2109 | * should equal to new_disks and new_sectors | |
2110 | */ | |
2111 | if (conf->scribble_disks >= new_disks && | |
2112 | conf->scribble_sectors >= new_sectors) | |
2113 | return 0; | |
738a2738 N |
2114 | mddev_suspend(conf->mddev); |
2115 | get_online_cpus(); | |
2116 | for_each_present_cpu(cpu) { | |
2117 | struct raid5_percpu *percpu; | |
2118 | struct flex_array *scribble; | |
2119 | ||
2120 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2121 | scribble = scribble_alloc(new_disks, | |
2122 | new_sectors / STRIPE_SECTORS, | |
2123 | GFP_NOIO); | |
2124 | ||
2125 | if (scribble) { | |
2126 | flex_array_free(percpu->scribble); | |
2127 | percpu->scribble = scribble; | |
2128 | } else { | |
2129 | err = -ENOMEM; | |
2130 | break; | |
2131 | } | |
2132 | } | |
2133 | put_online_cpus(); | |
2134 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2135 | if (!err) { |
2136 | conf->scribble_disks = new_disks; | |
2137 | conf->scribble_sectors = new_sectors; | |
2138 | } | |
738a2738 N |
2139 | return err; |
2140 | } | |
2141 | ||
d1688a6d | 2142 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2143 | { |
2144 | /* Make all the stripes able to hold 'newsize' devices. | |
2145 | * New slots in each stripe get 'page' set to a new page. | |
2146 | * | |
2147 | * This happens in stages: | |
2148 | * 1/ create a new kmem_cache and allocate the required number of | |
2149 | * stripe_heads. | |
83f0d77a | 2150 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2151 | * to the new stripe_heads. This will have the side effect of |
2152 | * freezing the array as once all stripe_heads have been collected, | |
2153 | * no IO will be possible. Old stripe heads are freed once their | |
2154 | * pages have been transferred over, and the old kmem_cache is | |
2155 | * freed when all stripes are done. | |
2156 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2157 | * we simple return a failre status - no need to clean anything up. | |
2158 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2159 | * If this fails, we don't bother trying the shrink the | |
2160 | * stripe_heads down again, we just leave them as they are. | |
2161 | * As each stripe_head is processed the new one is released into | |
2162 | * active service. | |
2163 | * | |
2164 | * Once step2 is started, we cannot afford to wait for a write, | |
2165 | * so we use GFP_NOIO allocations. | |
2166 | */ | |
2167 | struct stripe_head *osh, *nsh; | |
2168 | LIST_HEAD(newstripes); | |
2169 | struct disk_info *ndisks; | |
b5470dc5 | 2170 | int err; |
e18b890b | 2171 | struct kmem_cache *sc; |
ad01c9e3 | 2172 | int i; |
566c09c5 | 2173 | int hash, cnt; |
ad01c9e3 N |
2174 | |
2175 | if (newsize <= conf->pool_size) | |
2176 | return 0; /* never bother to shrink */ | |
2177 | ||
b5470dc5 DW |
2178 | err = md_allow_write(conf->mddev); |
2179 | if (err) | |
2180 | return err; | |
2a2275d6 | 2181 | |
ad01c9e3 N |
2182 | /* Step 1 */ |
2183 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2184 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2185 | 0, 0, NULL); |
ad01c9e3 N |
2186 | if (!sc) |
2187 | return -ENOMEM; | |
2188 | ||
2d5b569b N |
2189 | /* Need to ensure auto-resizing doesn't interfere */ |
2190 | mutex_lock(&conf->cache_size_mutex); | |
2191 | ||
ad01c9e3 | 2192 | for (i = conf->max_nr_stripes; i; i--) { |
5f9d1fde | 2193 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize); |
ad01c9e3 N |
2194 | if (!nsh) |
2195 | break; | |
2196 | ||
ad01c9e3 | 2197 | nsh->raid_conf = conf; |
ad01c9e3 N |
2198 | list_add(&nsh->lru, &newstripes); |
2199 | } | |
2200 | if (i) { | |
2201 | /* didn't get enough, give up */ | |
2202 | while (!list_empty(&newstripes)) { | |
2203 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2204 | list_del(&nsh->lru); | |
2205 | kmem_cache_free(sc, nsh); | |
2206 | } | |
2207 | kmem_cache_destroy(sc); | |
2d5b569b | 2208 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2209 | return -ENOMEM; |
2210 | } | |
2211 | /* Step 2 - Must use GFP_NOIO now. | |
2212 | * OK, we have enough stripes, start collecting inactive | |
2213 | * stripes and copying them over | |
2214 | */ | |
566c09c5 SL |
2215 | hash = 0; |
2216 | cnt = 0; | |
ad01c9e3 | 2217 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2218 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2219 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2220 | !list_empty(conf->inactive_list + hash), |
2221 | unlock_device_hash_lock(conf, hash), | |
2222 | lock_device_hash_lock(conf, hash)); | |
2223 | osh = get_free_stripe(conf, hash); | |
2224 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2225 | |
d592a996 | 2226 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2227 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2228 | nsh->dev[i].orig_page = osh->dev[i].page; |
2229 | } | |
566c09c5 | 2230 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2231 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2232 | cnt++; |
2233 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2234 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2235 | hash++; | |
2236 | cnt = 0; | |
2237 | } | |
ad01c9e3 N |
2238 | } |
2239 | kmem_cache_destroy(conf->slab_cache); | |
2240 | ||
2241 | /* Step 3. | |
2242 | * At this point, we are holding all the stripes so the array | |
2243 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2244 | * conf->disks and the scribble region |
ad01c9e3 N |
2245 | */ |
2246 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2247 | if (ndisks) { | |
2248 | for (i=0; i<conf->raid_disks; i++) | |
2249 | ndisks[i] = conf->disks[i]; | |
2250 | kfree(conf->disks); | |
2251 | conf->disks = ndisks; | |
2252 | } else | |
2253 | err = -ENOMEM; | |
2254 | ||
2d5b569b | 2255 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2256 | /* Step 4, return new stripes to service */ |
2257 | while(!list_empty(&newstripes)) { | |
2258 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2259 | list_del_init(&nsh->lru); | |
d6f38f31 | 2260 | |
ad01c9e3 N |
2261 | for (i=conf->raid_disks; i < newsize; i++) |
2262 | if (nsh->dev[i].page == NULL) { | |
2263 | struct page *p = alloc_page(GFP_NOIO); | |
2264 | nsh->dev[i].page = p; | |
d592a996 | 2265 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2266 | if (!p) |
2267 | err = -ENOMEM; | |
2268 | } | |
6d036f7d | 2269 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2270 | } |
2271 | /* critical section pass, GFP_NOIO no longer needed */ | |
2272 | ||
2273 | conf->slab_cache = sc; | |
2274 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2275 | if (!err) |
2276 | conf->pool_size = newsize; | |
ad01c9e3 N |
2277 | return err; |
2278 | } | |
1da177e4 | 2279 | |
486f0644 | 2280 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2281 | { |
2282 | struct stripe_head *sh; | |
49895bcc | 2283 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2284 | |
566c09c5 SL |
2285 | spin_lock_irq(conf->hash_locks + hash); |
2286 | sh = get_free_stripe(conf, hash); | |
2287 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2288 | if (!sh) |
2289 | return 0; | |
78bafebd | 2290 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2291 | shrink_buffers(sh); |
3f294f4f N |
2292 | kmem_cache_free(conf->slab_cache, sh); |
2293 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2294 | conf->max_nr_stripes--; |
3f294f4f N |
2295 | return 1; |
2296 | } | |
2297 | ||
d1688a6d | 2298 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2299 | { |
486f0644 N |
2300 | while (conf->max_nr_stripes && |
2301 | drop_one_stripe(conf)) | |
2302 | ; | |
3f294f4f | 2303 | |
644df1a8 | 2304 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2305 | conf->slab_cache = NULL; |
2306 | } | |
2307 | ||
4246a0b6 | 2308 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2309 | { |
99c0fb5f | 2310 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2311 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2312 | int disks = sh->disks, i; |
d6950432 | 2313 | char b[BDEVNAME_SIZE]; |
dd054fce | 2314 | struct md_rdev *rdev = NULL; |
05616be5 | 2315 | sector_t s; |
1da177e4 LT |
2316 | |
2317 | for (i=0 ; i<disks; i++) | |
2318 | if (bi == &sh->dev[i].req) | |
2319 | break; | |
2320 | ||
4246a0b6 | 2321 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2322 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2323 | bi->bi_error); |
1da177e4 | 2324 | if (i == disks) { |
5f9d1fde | 2325 | bio_reset(bi); |
1da177e4 | 2326 | BUG(); |
6712ecf8 | 2327 | return; |
1da177e4 | 2328 | } |
14a75d3e | 2329 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2330 | /* If replacement finished while this request was outstanding, |
2331 | * 'replacement' might be NULL already. | |
2332 | * In that case it moved down to 'rdev'. | |
2333 | * rdev is not removed until all requests are finished. | |
2334 | */ | |
14a75d3e | 2335 | rdev = conf->disks[i].replacement; |
dd054fce | 2336 | if (!rdev) |
14a75d3e | 2337 | rdev = conf->disks[i].rdev; |
1da177e4 | 2338 | |
05616be5 N |
2339 | if (use_new_offset(conf, sh)) |
2340 | s = sh->sector + rdev->new_data_offset; | |
2341 | else | |
2342 | s = sh->sector + rdev->data_offset; | |
4246a0b6 | 2343 | if (!bi->bi_error) { |
1da177e4 | 2344 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2345 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2346 | /* Note that this cannot happen on a |
2347 | * replacement device. We just fail those on | |
2348 | * any error | |
2349 | */ | |
8bda470e CD |
2350 | printk_ratelimited( |
2351 | KERN_INFO | |
2352 | "md/raid:%s: read error corrected" | |
2353 | " (%lu sectors at %llu on %s)\n", | |
2354 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 2355 | (unsigned long long)s, |
8bda470e | 2356 | bdevname(rdev->bdev, b)); |
ddd5115f | 2357 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2358 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2359 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2360 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2361 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2362 | ||
14a75d3e N |
2363 | if (atomic_read(&rdev->read_errors)) |
2364 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2365 | } else { |
14a75d3e | 2366 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2367 | int retry = 0; |
2e8ac303 | 2368 | int set_bad = 0; |
d6950432 | 2369 | |
1da177e4 | 2370 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2371 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
2372 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
2373 | printk_ratelimited( | |
2374 | KERN_WARNING | |
2375 | "md/raid:%s: read error on replacement device " | |
2376 | "(sector %llu on %s).\n", | |
2377 | mdname(conf->mddev), | |
05616be5 | 2378 | (unsigned long long)s, |
14a75d3e | 2379 | bdn); |
2e8ac303 | 2380 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2381 | set_bad = 1; | |
8bda470e CD |
2382 | printk_ratelimited( |
2383 | KERN_WARNING | |
2384 | "md/raid:%s: read error not correctable " | |
2385 | "(sector %llu on %s).\n", | |
2386 | mdname(conf->mddev), | |
05616be5 | 2387 | (unsigned long long)s, |
8bda470e | 2388 | bdn); |
2e8ac303 | 2389 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2390 | /* Oh, no!!! */ |
2e8ac303 | 2391 | set_bad = 1; |
8bda470e CD |
2392 | printk_ratelimited( |
2393 | KERN_WARNING | |
2394 | "md/raid:%s: read error NOT corrected!! " | |
2395 | "(sector %llu on %s).\n", | |
2396 | mdname(conf->mddev), | |
05616be5 | 2397 | (unsigned long long)s, |
8bda470e | 2398 | bdn); |
2e8ac303 | 2399 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2400 | > conf->max_nr_stripes) |
14f8d26b | 2401 | printk(KERN_WARNING |
0c55e022 | 2402 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2403 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2404 | else |
2405 | retry = 1; | |
edfa1f65 BY |
2406 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2407 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2408 | retry = 1; | |
ba22dcbf | 2409 | if (retry) |
3f9e7c14 | 2410 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2411 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2412 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2413 | } else | |
2414 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2415 | else { |
4e5314b5 N |
2416 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2417 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2418 | if (!(set_bad |
2419 | && test_bit(In_sync, &rdev->flags) | |
2420 | && rdev_set_badblocks( | |
2421 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2422 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2423 | } |
1da177e4 | 2424 | } |
14a75d3e | 2425 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2426 | bio_reset(bi); |
1da177e4 LT |
2427 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2428 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2429 | raid5_release_stripe(sh); |
1da177e4 LT |
2430 | } |
2431 | ||
4246a0b6 | 2432 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2433 | { |
99c0fb5f | 2434 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2435 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2436 | int disks = sh->disks, i; |
977df362 | 2437 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2438 | sector_t first_bad; |
2439 | int bad_sectors; | |
977df362 | 2440 | int replacement = 0; |
1da177e4 | 2441 | |
977df362 N |
2442 | for (i = 0 ; i < disks; i++) { |
2443 | if (bi == &sh->dev[i].req) { | |
2444 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2445 | break; |
977df362 N |
2446 | } |
2447 | if (bi == &sh->dev[i].rreq) { | |
2448 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2449 | if (rdev) |
2450 | replacement = 1; | |
2451 | else | |
2452 | /* rdev was removed and 'replacement' | |
2453 | * replaced it. rdev is not removed | |
2454 | * until all requests are finished. | |
2455 | */ | |
2456 | rdev = conf->disks[i].rdev; | |
977df362 N |
2457 | break; |
2458 | } | |
2459 | } | |
4246a0b6 | 2460 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2461 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2462 | bi->bi_error); |
1da177e4 | 2463 | if (i == disks) { |
5f9d1fde | 2464 | bio_reset(bi); |
1da177e4 | 2465 | BUG(); |
6712ecf8 | 2466 | return; |
1da177e4 LT |
2467 | } |
2468 | ||
977df362 | 2469 | if (replacement) { |
4246a0b6 | 2470 | if (bi->bi_error) |
977df362 N |
2471 | md_error(conf->mddev, rdev); |
2472 | else if (is_badblock(rdev, sh->sector, | |
2473 | STRIPE_SECTORS, | |
2474 | &first_bad, &bad_sectors)) | |
2475 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2476 | } else { | |
4246a0b6 | 2477 | if (bi->bi_error) { |
9f97e4b1 | 2478 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2479 | set_bit(WriteErrorSeen, &rdev->flags); |
2480 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2481 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2482 | set_bit(MD_RECOVERY_NEEDED, | |
2483 | &rdev->mddev->recovery); | |
977df362 N |
2484 | } else if (is_badblock(rdev, sh->sector, |
2485 | STRIPE_SECTORS, | |
c0b32972 | 2486 | &first_bad, &bad_sectors)) { |
977df362 | 2487 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2488 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2489 | /* That was a successful write so make | |
2490 | * sure it looks like we already did | |
2491 | * a re-write. | |
2492 | */ | |
2493 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2494 | } | |
977df362 N |
2495 | } |
2496 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2497 | |
4246a0b6 | 2498 | if (sh->batch_head && bi->bi_error && !replacement) |
72ac7330 | 2499 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2500 | ||
c9445555 | 2501 | bio_reset(bi); |
977df362 N |
2502 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2503 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2504 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2505 | raid5_release_stripe(sh); |
59fc630b | 2506 | |
2507 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2508 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2509 | } |
2510 | ||
784052ec | 2511 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2512 | { |
2513 | struct r5dev *dev = &sh->dev[i]; | |
2514 | ||
1da177e4 | 2515 | dev->flags = 0; |
6d036f7d | 2516 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 LT |
2517 | } |
2518 | ||
849674e4 | 2519 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2520 | { |
2521 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2522 | struct r5conf *conf = mddev->private; |
908f4fbd | 2523 | unsigned long flags; |
0c55e022 | 2524 | pr_debug("raid456: error called\n"); |
1da177e4 | 2525 | |
908f4fbd N |
2526 | spin_lock_irqsave(&conf->device_lock, flags); |
2527 | clear_bit(In_sync, &rdev->flags); | |
2528 | mddev->degraded = calc_degraded(conf); | |
2529 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2530 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2531 | ||
de393cde | 2532 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 | 2533 | set_bit(Faulty, &rdev->flags); |
85ad1d13 GJ |
2534 | set_mask_bits(&mddev->flags, 0, |
2535 | BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING)); | |
6f8d0c77 N |
2536 | printk(KERN_ALERT |
2537 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
2538 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2539 | mdname(mddev), | |
2540 | bdevname(rdev->bdev, b), | |
2541 | mdname(mddev), | |
2542 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 2543 | } |
1da177e4 LT |
2544 | |
2545 | /* | |
2546 | * Input: a 'big' sector number, | |
2547 | * Output: index of the data and parity disk, and the sector # in them. | |
2548 | */ | |
6d036f7d SL |
2549 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2550 | int previous, int *dd_idx, | |
2551 | struct stripe_head *sh) | |
1da177e4 | 2552 | { |
6e3b96ed | 2553 | sector_t stripe, stripe2; |
35f2a591 | 2554 | sector_t chunk_number; |
1da177e4 | 2555 | unsigned int chunk_offset; |
911d4ee8 | 2556 | int pd_idx, qd_idx; |
67cc2b81 | 2557 | int ddf_layout = 0; |
1da177e4 | 2558 | sector_t new_sector; |
e183eaed N |
2559 | int algorithm = previous ? conf->prev_algo |
2560 | : conf->algorithm; | |
09c9e5fa AN |
2561 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2562 | : conf->chunk_sectors; | |
112bf897 N |
2563 | int raid_disks = previous ? conf->previous_raid_disks |
2564 | : conf->raid_disks; | |
2565 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2566 | |
2567 | /* First compute the information on this sector */ | |
2568 | ||
2569 | /* | |
2570 | * Compute the chunk number and the sector offset inside the chunk | |
2571 | */ | |
2572 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2573 | chunk_number = r_sector; | |
1da177e4 LT |
2574 | |
2575 | /* | |
2576 | * Compute the stripe number | |
2577 | */ | |
35f2a591 N |
2578 | stripe = chunk_number; |
2579 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2580 | stripe2 = stripe; |
1da177e4 LT |
2581 | /* |
2582 | * Select the parity disk based on the user selected algorithm. | |
2583 | */ | |
84789554 | 2584 | pd_idx = qd_idx = -1; |
16a53ecc N |
2585 | switch(conf->level) { |
2586 | case 4: | |
911d4ee8 | 2587 | pd_idx = data_disks; |
16a53ecc N |
2588 | break; |
2589 | case 5: | |
e183eaed | 2590 | switch (algorithm) { |
1da177e4 | 2591 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2592 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2593 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2594 | (*dd_idx)++; |
2595 | break; | |
2596 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2597 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2598 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2599 | (*dd_idx)++; |
2600 | break; | |
2601 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2602 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2603 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2604 | break; |
2605 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2606 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2607 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2608 | break; |
99c0fb5f N |
2609 | case ALGORITHM_PARITY_0: |
2610 | pd_idx = 0; | |
2611 | (*dd_idx)++; | |
2612 | break; | |
2613 | case ALGORITHM_PARITY_N: | |
2614 | pd_idx = data_disks; | |
2615 | break; | |
1da177e4 | 2616 | default: |
99c0fb5f | 2617 | BUG(); |
16a53ecc N |
2618 | } |
2619 | break; | |
2620 | case 6: | |
2621 | ||
e183eaed | 2622 | switch (algorithm) { |
16a53ecc | 2623 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2624 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2625 | qd_idx = pd_idx + 1; |
2626 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2627 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2628 | qd_idx = 0; |
2629 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2630 | (*dd_idx) += 2; /* D D P Q D */ |
2631 | break; | |
2632 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2633 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2634 | qd_idx = pd_idx + 1; |
2635 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2636 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2637 | qd_idx = 0; |
2638 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2639 | (*dd_idx) += 2; /* D D P Q D */ |
2640 | break; | |
2641 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2642 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2643 | qd_idx = (pd_idx + 1) % raid_disks; |
2644 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2645 | break; |
2646 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2647 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2648 | qd_idx = (pd_idx + 1) % raid_disks; |
2649 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2650 | break; |
99c0fb5f N |
2651 | |
2652 | case ALGORITHM_PARITY_0: | |
2653 | pd_idx = 0; | |
2654 | qd_idx = 1; | |
2655 | (*dd_idx) += 2; | |
2656 | break; | |
2657 | case ALGORITHM_PARITY_N: | |
2658 | pd_idx = data_disks; | |
2659 | qd_idx = data_disks + 1; | |
2660 | break; | |
2661 | ||
2662 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2663 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2664 | * of blocks for computing Q is different. | |
2665 | */ | |
6e3b96ed | 2666 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2667 | qd_idx = pd_idx + 1; |
2668 | if (pd_idx == raid_disks-1) { | |
2669 | (*dd_idx)++; /* Q D D D P */ | |
2670 | qd_idx = 0; | |
2671 | } else if (*dd_idx >= pd_idx) | |
2672 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2673 | ddf_layout = 1; |
99c0fb5f N |
2674 | break; |
2675 | ||
2676 | case ALGORITHM_ROTATING_N_RESTART: | |
2677 | /* Same a left_asymmetric, by first stripe is | |
2678 | * D D D P Q rather than | |
2679 | * Q D D D P | |
2680 | */ | |
6e3b96ed N |
2681 | stripe2 += 1; |
2682 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2683 | qd_idx = pd_idx + 1; |
2684 | if (pd_idx == raid_disks-1) { | |
2685 | (*dd_idx)++; /* Q D D D P */ | |
2686 | qd_idx = 0; | |
2687 | } else if (*dd_idx >= pd_idx) | |
2688 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2689 | ddf_layout = 1; |
99c0fb5f N |
2690 | break; |
2691 | ||
2692 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2693 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2694 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2695 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2696 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2697 | ddf_layout = 1; |
99c0fb5f N |
2698 | break; |
2699 | ||
2700 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2701 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2702 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2703 | if (*dd_idx >= pd_idx) |
2704 | (*dd_idx)++; | |
2705 | qd_idx = raid_disks - 1; | |
2706 | break; | |
2707 | ||
2708 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2709 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2710 | if (*dd_idx >= pd_idx) |
2711 | (*dd_idx)++; | |
2712 | qd_idx = raid_disks - 1; | |
2713 | break; | |
2714 | ||
2715 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2716 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2717 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2718 | qd_idx = raid_disks - 1; | |
2719 | break; | |
2720 | ||
2721 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2722 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2723 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2724 | qd_idx = raid_disks - 1; | |
2725 | break; | |
2726 | ||
2727 | case ALGORITHM_PARITY_0_6: | |
2728 | pd_idx = 0; | |
2729 | (*dd_idx)++; | |
2730 | qd_idx = raid_disks - 1; | |
2731 | break; | |
2732 | ||
16a53ecc | 2733 | default: |
99c0fb5f | 2734 | BUG(); |
16a53ecc N |
2735 | } |
2736 | break; | |
1da177e4 LT |
2737 | } |
2738 | ||
911d4ee8 N |
2739 | if (sh) { |
2740 | sh->pd_idx = pd_idx; | |
2741 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2742 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2743 | } |
1da177e4 LT |
2744 | /* |
2745 | * Finally, compute the new sector number | |
2746 | */ | |
2747 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2748 | return new_sector; | |
2749 | } | |
2750 | ||
6d036f7d | 2751 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2752 | { |
d1688a6d | 2753 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2754 | int raid_disks = sh->disks; |
2755 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2756 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2757 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2758 | : conf->chunk_sectors; | |
e183eaed N |
2759 | int algorithm = previous ? conf->prev_algo |
2760 | : conf->algorithm; | |
1da177e4 LT |
2761 | sector_t stripe; |
2762 | int chunk_offset; | |
35f2a591 N |
2763 | sector_t chunk_number; |
2764 | int dummy1, dd_idx = i; | |
1da177e4 | 2765 | sector_t r_sector; |
911d4ee8 | 2766 | struct stripe_head sh2; |
1da177e4 LT |
2767 | |
2768 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2769 | stripe = new_sector; | |
1da177e4 | 2770 | |
16a53ecc N |
2771 | if (i == sh->pd_idx) |
2772 | return 0; | |
2773 | switch(conf->level) { | |
2774 | case 4: break; | |
2775 | case 5: | |
e183eaed | 2776 | switch (algorithm) { |
1da177e4 LT |
2777 | case ALGORITHM_LEFT_ASYMMETRIC: |
2778 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2779 | if (i > sh->pd_idx) | |
2780 | i--; | |
2781 | break; | |
2782 | case ALGORITHM_LEFT_SYMMETRIC: | |
2783 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2784 | if (i < sh->pd_idx) | |
2785 | i += raid_disks; | |
2786 | i -= (sh->pd_idx + 1); | |
2787 | break; | |
99c0fb5f N |
2788 | case ALGORITHM_PARITY_0: |
2789 | i -= 1; | |
2790 | break; | |
2791 | case ALGORITHM_PARITY_N: | |
2792 | break; | |
1da177e4 | 2793 | default: |
99c0fb5f | 2794 | BUG(); |
16a53ecc N |
2795 | } |
2796 | break; | |
2797 | case 6: | |
d0dabf7e | 2798 | if (i == sh->qd_idx) |
16a53ecc | 2799 | return 0; /* It is the Q disk */ |
e183eaed | 2800 | switch (algorithm) { |
16a53ecc N |
2801 | case ALGORITHM_LEFT_ASYMMETRIC: |
2802 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2803 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2804 | case ALGORITHM_ROTATING_N_RESTART: | |
2805 | if (sh->pd_idx == raid_disks-1) | |
2806 | i--; /* Q D D D P */ | |
16a53ecc N |
2807 | else if (i > sh->pd_idx) |
2808 | i -= 2; /* D D P Q D */ | |
2809 | break; | |
2810 | case ALGORITHM_LEFT_SYMMETRIC: | |
2811 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2812 | if (sh->pd_idx == raid_disks-1) | |
2813 | i--; /* Q D D D P */ | |
2814 | else { | |
2815 | /* D D P Q D */ | |
2816 | if (i < sh->pd_idx) | |
2817 | i += raid_disks; | |
2818 | i -= (sh->pd_idx + 2); | |
2819 | } | |
2820 | break; | |
99c0fb5f N |
2821 | case ALGORITHM_PARITY_0: |
2822 | i -= 2; | |
2823 | break; | |
2824 | case ALGORITHM_PARITY_N: | |
2825 | break; | |
2826 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2827 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2828 | if (sh->pd_idx == 0) |
2829 | i--; /* P D D D Q */ | |
e4424fee N |
2830 | else { |
2831 | /* D D Q P D */ | |
2832 | if (i < sh->pd_idx) | |
2833 | i += raid_disks; | |
2834 | i -= (sh->pd_idx + 1); | |
2835 | } | |
99c0fb5f N |
2836 | break; |
2837 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2838 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2839 | if (i > sh->pd_idx) | |
2840 | i--; | |
2841 | break; | |
2842 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2843 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2844 | if (i < sh->pd_idx) | |
2845 | i += data_disks + 1; | |
2846 | i -= (sh->pd_idx + 1); | |
2847 | break; | |
2848 | case ALGORITHM_PARITY_0_6: | |
2849 | i -= 1; | |
2850 | break; | |
16a53ecc | 2851 | default: |
99c0fb5f | 2852 | BUG(); |
16a53ecc N |
2853 | } |
2854 | break; | |
1da177e4 LT |
2855 | } |
2856 | ||
2857 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2858 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2859 | |
112bf897 | 2860 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2861 | previous, &dummy1, &sh2); |
911d4ee8 N |
2862 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2863 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2864 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2865 | mdname(conf->mddev)); | |
1da177e4 LT |
2866 | return 0; |
2867 | } | |
2868 | return r_sector; | |
2869 | } | |
2870 | ||
600aa109 | 2871 | static void |
c0f7bddb | 2872 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2873 | int rcw, int expand) |
e33129d8 | 2874 | { |
584acdd4 | 2875 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 2876 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2877 | int level = conf->level; |
e33129d8 DW |
2878 | |
2879 | if (rcw) { | |
e33129d8 DW |
2880 | |
2881 | for (i = disks; i--; ) { | |
2882 | struct r5dev *dev = &sh->dev[i]; | |
2883 | ||
2884 | if (dev->towrite) { | |
2885 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2886 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2887 | if (!expand) |
2888 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2889 | s->locked++; |
e33129d8 DW |
2890 | } |
2891 | } | |
ce7d363a N |
2892 | /* if we are not expanding this is a proper write request, and |
2893 | * there will be bios with new data to be drained into the | |
2894 | * stripe cache | |
2895 | */ | |
2896 | if (!expand) { | |
2897 | if (!s->locked) | |
2898 | /* False alarm, nothing to do */ | |
2899 | return; | |
2900 | sh->reconstruct_state = reconstruct_state_drain_run; | |
2901 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2902 | } else | |
2903 | sh->reconstruct_state = reconstruct_state_run; | |
2904 | ||
2905 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
2906 | ||
c0f7bddb | 2907 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2908 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2909 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
2910 | } else { |
2911 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
2912 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
2913 | BUG_ON(level == 6 && |
2914 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
2915 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 2916 | |
e33129d8 DW |
2917 | for (i = disks; i--; ) { |
2918 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 2919 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
2920 | continue; |
2921 | ||
e33129d8 DW |
2922 | if (dev->towrite && |
2923 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2924 | test_bit(R5_Wantcompute, &dev->flags))) { |
2925 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2926 | set_bit(R5_LOCKED, &dev->flags); |
2927 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2928 | s->locked++; |
e33129d8 DW |
2929 | } |
2930 | } | |
ce7d363a N |
2931 | if (!s->locked) |
2932 | /* False alarm - nothing to do */ | |
2933 | return; | |
2934 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
2935 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
2936 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2937 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
2938 | } |
2939 | ||
c0f7bddb | 2940 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2941 | * are in flight |
2942 | */ | |
2943 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2944 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2945 | s->locked++; |
e33129d8 | 2946 | |
c0f7bddb YT |
2947 | if (level == 6) { |
2948 | int qd_idx = sh->qd_idx; | |
2949 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2950 | ||
2951 | set_bit(R5_LOCKED, &dev->flags); | |
2952 | clear_bit(R5_UPTODATE, &dev->flags); | |
2953 | s->locked++; | |
2954 | } | |
2955 | ||
600aa109 | 2956 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2957 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2958 | s->locked, s->ops_request); |
e33129d8 | 2959 | } |
16a53ecc | 2960 | |
1da177e4 LT |
2961 | /* |
2962 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2963 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2964 | * The bi_next chain must be in order. |
2965 | */ | |
da41ba65 | 2966 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
2967 | int forwrite, int previous) | |
1da177e4 LT |
2968 | { |
2969 | struct bio **bip; | |
d1688a6d | 2970 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2971 | int firstwrite=0; |
1da177e4 | 2972 | |
cbe47ec5 | 2973 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 2974 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
2975 | (unsigned long long)sh->sector); |
2976 | ||
b17459c0 SL |
2977 | /* |
2978 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2979 | * reference count to avoid race. The reference count should already be | |
2980 | * increased before this function is called (for example, in | |
849674e4 | 2981 | * raid5_make_request()), so other bio sharing this stripe will not free the |
b17459c0 SL |
2982 | * stripe. If a stripe is owned by one stripe, the stripe lock will |
2983 | * protect it. | |
2984 | */ | |
2985 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 2986 | /* Don't allow new IO added to stripes in batch list */ |
2987 | if (sh->batch_head) | |
2988 | goto overlap; | |
72626685 | 2989 | if (forwrite) { |
1da177e4 | 2990 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2991 | if (*bip == NULL) |
72626685 N |
2992 | firstwrite = 1; |
2993 | } else | |
1da177e4 | 2994 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
2995 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
2996 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
2997 | goto overlap; |
2998 | bip = & (*bip)->bi_next; | |
2999 | } | |
4f024f37 | 3000 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3001 | goto overlap; |
3002 | ||
da41ba65 | 3003 | if (!forwrite || previous) |
3004 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3005 | ||
78bafebd | 3006 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3007 | if (*bip) |
3008 | bi->bi_next = *bip; | |
3009 | *bip = bi; | |
e7836bd6 | 3010 | raid5_inc_bi_active_stripes(bi); |
72626685 | 3011 | |
1da177e4 LT |
3012 | if (forwrite) { |
3013 | /* check if page is covered */ | |
3014 | sector_t sector = sh->dev[dd_idx].sector; | |
3015 | for (bi=sh->dev[dd_idx].towrite; | |
3016 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3017 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3018 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3019 | if (bio_end_sector(bi) >= sector) |
3020 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3021 | } |
3022 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3023 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3024 | sh->overwrite_disks++; | |
1da177e4 | 3025 | } |
cbe47ec5 N |
3026 | |
3027 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3028 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3029 | (unsigned long long)sh->sector, dd_idx); |
3030 | ||
3031 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3032 | /* Cannot hold spinlock over bitmap_startwrite, |
3033 | * but must ensure this isn't added to a batch until | |
3034 | * we have added to the bitmap and set bm_seq. | |
3035 | * So set STRIPE_BITMAP_PENDING to prevent | |
3036 | * batching. | |
3037 | * If multiple add_stripe_bio() calls race here they | |
3038 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3039 | * to complete "bitmap_startwrite" gets to set | |
3040 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3041 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3042 | * any more. | |
3043 | */ | |
3044 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3045 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3046 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3047 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3048 | spin_lock_irq(&sh->stripe_lock); |
3049 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3050 | if (!sh->batch_head) { | |
3051 | sh->bm_seq = conf->seq_flush+1; | |
3052 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3053 | } | |
cbe47ec5 | 3054 | } |
d0852df5 | 3055 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3056 | |
3057 | if (stripe_can_batch(sh)) | |
3058 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3059 | return 1; |
3060 | ||
3061 | overlap: | |
3062 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3063 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3064 | return 0; |
3065 | } | |
3066 | ||
d1688a6d | 3067 | static void end_reshape(struct r5conf *conf); |
29269553 | 3068 | |
d1688a6d | 3069 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3070 | struct stripe_head *sh) |
ccfcc3c1 | 3071 | { |
784052ec | 3072 | int sectors_per_chunk = |
09c9e5fa | 3073 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3074 | int dd_idx; |
2d2063ce | 3075 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3076 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3077 | |
112bf897 N |
3078 | raid5_compute_sector(conf, |
3079 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3080 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3081 | previous, |
911d4ee8 | 3082 | &dd_idx, sh); |
ccfcc3c1 N |
3083 | } |
3084 | ||
a4456856 | 3085 | static void |
d1688a6d | 3086 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 | 3087 | struct stripe_head_state *s, int disks, |
34a6f80e | 3088 | struct bio_list *return_bi) |
a4456856 DW |
3089 | { |
3090 | int i; | |
59fc630b | 3091 | BUG_ON(sh->batch_head); |
a4456856 DW |
3092 | for (i = disks; i--; ) { |
3093 | struct bio *bi; | |
3094 | int bitmap_end = 0; | |
3095 | ||
3096 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3097 | struct md_rdev *rdev; |
a4456856 DW |
3098 | rcu_read_lock(); |
3099 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3100 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3101 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3102 | atomic_inc(&rdev->nr_pending); |
3103 | else | |
3104 | rdev = NULL; | |
a4456856 | 3105 | rcu_read_unlock(); |
7f0da59b N |
3106 | if (rdev) { |
3107 | if (!rdev_set_badblocks( | |
3108 | rdev, | |
3109 | sh->sector, | |
3110 | STRIPE_SECTORS, 0)) | |
3111 | md_error(conf->mddev, rdev); | |
3112 | rdev_dec_pending(rdev, conf->mddev); | |
3113 | } | |
a4456856 | 3114 | } |
b17459c0 | 3115 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3116 | /* fail all writes first */ |
3117 | bi = sh->dev[i].towrite; | |
3118 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3119 | sh->overwrite_disks = 0; |
b17459c0 | 3120 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3121 | if (bi) |
a4456856 | 3122 | bitmap_end = 1; |
a4456856 | 3123 | |
0576b1c6 SL |
3124 | r5l_stripe_write_finished(sh); |
3125 | ||
a4456856 DW |
3126 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3127 | wake_up(&conf->wait_for_overlap); | |
3128 | ||
4f024f37 | 3129 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3130 | sh->dev[i].sector + STRIPE_SECTORS) { |
3131 | struct bio *nextbi = 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 = nextbi; | |
3139 | } | |
7eaf7e8e SL |
3140 | if (bitmap_end) |
3141 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3142 | STRIPE_SECTORS, 0, 0); | |
3143 | bitmap_end = 0; | |
a4456856 DW |
3144 | /* and fail all 'written' */ |
3145 | bi = sh->dev[i].written; | |
3146 | sh->dev[i].written = NULL; | |
d592a996 SL |
3147 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3148 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3149 | sh->dev[i].page = sh->dev[i].orig_page; | |
3150 | } | |
3151 | ||
a4456856 | 3152 | if (bi) bitmap_end = 1; |
4f024f37 | 3153 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3154 | sh->dev[i].sector + STRIPE_SECTORS) { |
3155 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3156 | |
3157 | bi->bi_error = -EIO; | |
e7836bd6 | 3158 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3159 | md_write_end(conf->mddev); |
34a6f80e | 3160 | bio_list_add(return_bi, bi); |
a4456856 DW |
3161 | } |
3162 | bi = bi2; | |
3163 | } | |
3164 | ||
b5e98d65 DW |
3165 | /* fail any reads if this device is non-operational and |
3166 | * the data has not reached the cache yet. | |
3167 | */ | |
3168 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3169 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3170 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3171 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3172 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3173 | bi = sh->dev[i].toread; |
3174 | sh->dev[i].toread = NULL; | |
143c4d05 | 3175 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3176 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3177 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3178 | if (bi) |
3179 | s->to_read--; | |
4f024f37 | 3180 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3181 | sh->dev[i].sector + STRIPE_SECTORS) { |
3182 | struct bio *nextbi = | |
3183 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3184 | |
3185 | bi->bi_error = -EIO; | |
34a6f80e N |
3186 | if (!raid5_dec_bi_active_stripes(bi)) |
3187 | bio_list_add(return_bi, bi); | |
a4456856 DW |
3188 | bi = nextbi; |
3189 | } | |
3190 | } | |
a4456856 DW |
3191 | if (bitmap_end) |
3192 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3193 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3194 | /* If we were in the middle of a write the parity block might |
3195 | * still be locked - so just clear all R5_LOCKED flags | |
3196 | */ | |
3197 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3198 | } |
ebda780b SL |
3199 | s->to_write = 0; |
3200 | s->written = 0; | |
a4456856 | 3201 | |
8b3e6cdc DW |
3202 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3203 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3204 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3205 | } |
3206 | ||
7f0da59b | 3207 | static void |
d1688a6d | 3208 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3209 | struct stripe_head_state *s) |
3210 | { | |
3211 | int abort = 0; | |
3212 | int i; | |
3213 | ||
59fc630b | 3214 | BUG_ON(sh->batch_head); |
7f0da59b | 3215 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3216 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3217 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3218 | s->syncing = 0; |
9a3e1101 | 3219 | s->replacing = 0; |
7f0da59b | 3220 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3221 | * Don't even need to abort as that is handled elsewhere |
3222 | * if needed, and not always wanted e.g. if there is a known | |
3223 | * bad block here. | |
9a3e1101 | 3224 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3225 | * non-sync devices, or abort the recovery |
3226 | */ | |
18b9837e N |
3227 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3228 | /* During recovery devices cannot be removed, so | |
3229 | * locking and refcounting of rdevs is not needed | |
3230 | */ | |
e50d3992 | 3231 | rcu_read_lock(); |
18b9837e | 3232 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3233 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3234 | if (rdev |
3235 | && !test_bit(Faulty, &rdev->flags) | |
3236 | && !test_bit(In_sync, &rdev->flags) | |
3237 | && !rdev_set_badblocks(rdev, sh->sector, | |
3238 | STRIPE_SECTORS, 0)) | |
3239 | abort = 1; | |
e50d3992 | 3240 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3241 | if (rdev |
3242 | && !test_bit(Faulty, &rdev->flags) | |
3243 | && !test_bit(In_sync, &rdev->flags) | |
3244 | && !rdev_set_badblocks(rdev, sh->sector, | |
3245 | STRIPE_SECTORS, 0)) | |
3246 | abort = 1; | |
3247 | } | |
e50d3992 | 3248 | rcu_read_unlock(); |
18b9837e N |
3249 | if (abort) |
3250 | conf->recovery_disabled = | |
3251 | conf->mddev->recovery_disabled; | |
7f0da59b | 3252 | } |
18b9837e | 3253 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3254 | } |
3255 | ||
9a3e1101 N |
3256 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3257 | { | |
3258 | struct md_rdev *rdev; | |
3259 | int rv = 0; | |
3f232d6a N |
3260 | |
3261 | rcu_read_lock(); | |
3262 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3263 | if (rdev |
3264 | && !test_bit(Faulty, &rdev->flags) | |
3265 | && !test_bit(In_sync, &rdev->flags) | |
3266 | && (rdev->recovery_offset <= sh->sector | |
3267 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3268 | rv = 1; | |
3f232d6a | 3269 | rcu_read_unlock(); |
9a3e1101 N |
3270 | return rv; |
3271 | } | |
3272 | ||
93b3dbce | 3273 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
3274 | * to be read or computed to satisfy a request. |
3275 | * | |
3276 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 3277 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 3278 | */ |
2c58f06e N |
3279 | |
3280 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3281 | int disk_idx, int disks) | |
a4456856 | 3282 | { |
5599becc | 3283 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3284 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3285 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3286 | int i; |
5599becc | 3287 | |
a79cfe12 N |
3288 | |
3289 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3290 | test_bit(R5_UPTODATE, &dev->flags)) | |
3291 | /* No point reading this as we already have it or have | |
3292 | * decided to get it. | |
3293 | */ | |
3294 | return 0; | |
3295 | ||
3296 | if (dev->toread || | |
3297 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3298 | /* We need this block to directly satisfy a request */ | |
3299 | return 1; | |
3300 | ||
3301 | if (s->syncing || s->expanding || | |
3302 | (s->replacing && want_replace(sh, disk_idx))) | |
3303 | /* When syncing, or expanding we read everything. | |
3304 | * When replacing, we need the replaced block. | |
3305 | */ | |
3306 | return 1; | |
3307 | ||
3308 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3309 | (s->failed >= 2 && fdev[1]->toread)) | |
3310 | /* If we want to read from a failed device, then | |
3311 | * we need to actually read every other device. | |
3312 | */ | |
3313 | return 1; | |
3314 | ||
a9d56950 N |
3315 | /* Sometimes neither read-modify-write nor reconstruct-write |
3316 | * cycles can work. In those cases we read every block we | |
3317 | * can. Then the parity-update is certain to have enough to | |
3318 | * work with. | |
3319 | * This can only be a problem when we need to write something, | |
3320 | * and some device has failed. If either of those tests | |
3321 | * fail we need look no further. | |
3322 | */ | |
3323 | if (!s->failed || !s->to_write) | |
3324 | return 0; | |
3325 | ||
3326 | if (test_bit(R5_Insync, &dev->flags) && | |
3327 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3328 | /* Pre-reads at not permitted until after short delay | |
3329 | * to gather multiple requests. However if this | |
3330 | * device is no Insync, the block could only be be computed | |
3331 | * and there is no need to delay that. | |
3332 | */ | |
3333 | return 0; | |
ea664c82 | 3334 | |
36707bb2 | 3335 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3336 | if (fdev[i]->towrite && |
3337 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3338 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3339 | /* If we have a partial write to a failed | |
3340 | * device, then we will need to reconstruct | |
3341 | * the content of that device, so all other | |
3342 | * devices must be read. | |
3343 | */ | |
3344 | return 1; | |
3345 | } | |
3346 | ||
3347 | /* If we are forced to do a reconstruct-write, either because | |
3348 | * the current RAID6 implementation only supports that, or | |
3349 | * or because parity cannot be trusted and we are currently | |
3350 | * recovering it, there is extra need to be careful. | |
3351 | * If one of the devices that we would need to read, because | |
3352 | * it is not being overwritten (and maybe not written at all) | |
3353 | * is missing/faulty, then we need to read everything we can. | |
3354 | */ | |
3355 | if (sh->raid_conf->level != 6 && | |
3356 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3357 | /* reconstruct-write isn't being forced */ | |
3358 | return 0; | |
36707bb2 | 3359 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3360 | if (s->failed_num[i] != sh->pd_idx && |
3361 | s->failed_num[i] != sh->qd_idx && | |
3362 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3363 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3364 | return 1; | |
3365 | } | |
3366 | ||
2c58f06e N |
3367 | return 0; |
3368 | } | |
3369 | ||
3370 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3371 | int disk_idx, int disks) | |
3372 | { | |
3373 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3374 | ||
3375 | /* is the data in this block needed, and can we get it? */ | |
3376 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3377 | /* we would like to get this block, possibly by computing it, |
3378 | * otherwise read it if the backing disk is insync | |
3379 | */ | |
3380 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3381 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3382 | BUG_ON(sh->batch_head); |
5599becc | 3383 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3384 | (s->failed && (disk_idx == s->failed_num[0] || |
3385 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3386 | /* have disk failed, and we're requested to fetch it; |
3387 | * do compute it | |
a4456856 | 3388 | */ |
5599becc YT |
3389 | pr_debug("Computing stripe %llu block %d\n", |
3390 | (unsigned long long)sh->sector, disk_idx); | |
3391 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3392 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3393 | set_bit(R5_Wantcompute, &dev->flags); | |
3394 | sh->ops.target = disk_idx; | |
3395 | sh->ops.target2 = -1; /* no 2nd target */ | |
3396 | s->req_compute = 1; | |
93b3dbce N |
3397 | /* Careful: from this point on 'uptodate' is in the eye |
3398 | * of raid_run_ops which services 'compute' operations | |
3399 | * before writes. R5_Wantcompute flags a block that will | |
3400 | * be R5_UPTODATE by the time it is needed for a | |
3401 | * subsequent operation. | |
3402 | */ | |
5599becc YT |
3403 | s->uptodate++; |
3404 | return 1; | |
3405 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3406 | /* Computing 2-failure is *very* expensive; only | |
3407 | * do it if failed >= 2 | |
3408 | */ | |
3409 | int other; | |
3410 | for (other = disks; other--; ) { | |
3411 | if (other == disk_idx) | |
3412 | continue; | |
3413 | if (!test_bit(R5_UPTODATE, | |
3414 | &sh->dev[other].flags)) | |
3415 | break; | |
a4456856 | 3416 | } |
5599becc YT |
3417 | BUG_ON(other < 0); |
3418 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3419 | (unsigned long long)sh->sector, | |
3420 | disk_idx, other); | |
3421 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3422 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3423 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3424 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3425 | sh->ops.target = disk_idx; | |
3426 | sh->ops.target2 = other; | |
3427 | s->uptodate += 2; | |
3428 | s->req_compute = 1; | |
3429 | return 1; | |
3430 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3431 | set_bit(R5_LOCKED, &dev->flags); | |
3432 | set_bit(R5_Wantread, &dev->flags); | |
3433 | s->locked++; | |
3434 | pr_debug("Reading block %d (sync=%d)\n", | |
3435 | disk_idx, s->syncing); | |
a4456856 DW |
3436 | } |
3437 | } | |
5599becc YT |
3438 | |
3439 | return 0; | |
3440 | } | |
3441 | ||
3442 | /** | |
93b3dbce | 3443 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3444 | */ |
93b3dbce N |
3445 | static void handle_stripe_fill(struct stripe_head *sh, |
3446 | struct stripe_head_state *s, | |
3447 | int disks) | |
5599becc YT |
3448 | { |
3449 | int i; | |
3450 | ||
3451 | /* look for blocks to read/compute, skip this if a compute | |
3452 | * is already in flight, or if the stripe contents are in the | |
3453 | * midst of changing due to a write | |
3454 | */ | |
3455 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
3456 | !sh->reconstruct_state) | |
3457 | for (i = disks; i--; ) | |
93b3dbce | 3458 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3459 | break; |
a4456856 DW |
3460 | set_bit(STRIPE_HANDLE, &sh->state); |
3461 | } | |
3462 | ||
787b76fa N |
3463 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3464 | unsigned long handle_flags); | |
1fe797e6 | 3465 | /* handle_stripe_clean_event |
a4456856 DW |
3466 | * any written block on an uptodate or failed drive can be returned. |
3467 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3468 | * never LOCKED, so we don't need to test 'failed' directly. | |
3469 | */ | |
d1688a6d | 3470 | static void handle_stripe_clean_event(struct r5conf *conf, |
34a6f80e | 3471 | struct stripe_head *sh, int disks, struct bio_list *return_bi) |
a4456856 DW |
3472 | { |
3473 | int i; | |
3474 | struct r5dev *dev; | |
f8dfcffd | 3475 | int discard_pending = 0; |
59fc630b | 3476 | struct stripe_head *head_sh = sh; |
3477 | bool do_endio = false; | |
a4456856 DW |
3478 | |
3479 | for (i = disks; i--; ) | |
3480 | if (sh->dev[i].written) { | |
3481 | dev = &sh->dev[i]; | |
3482 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3483 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3484 | test_bit(R5_Discard, &dev->flags) || |
3485 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3486 | /* We can return any write requests */ |
3487 | struct bio *wbi, *wbi2; | |
45b4233c | 3488 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3489 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3490 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3491 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3492 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3493 | } |
59fc630b | 3494 | do_endio = true; |
3495 | ||
3496 | returnbi: | |
3497 | dev->page = dev->orig_page; | |
a4456856 DW |
3498 | wbi = dev->written; |
3499 | dev->written = NULL; | |
4f024f37 | 3500 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3501 | dev->sector + STRIPE_SECTORS) { |
3502 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3503 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 | 3504 | md_write_end(conf->mddev); |
34a6f80e | 3505 | bio_list_add(return_bi, wbi); |
a4456856 DW |
3506 | } |
3507 | wbi = wbi2; | |
3508 | } | |
7eaf7e8e SL |
3509 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3510 | STRIPE_SECTORS, | |
a4456856 | 3511 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3512 | 0); |
59fc630b | 3513 | if (head_sh->batch_head) { |
3514 | sh = list_first_entry(&sh->batch_list, | |
3515 | struct stripe_head, | |
3516 | batch_list); | |
3517 | if (sh != head_sh) { | |
3518 | dev = &sh->dev[i]; | |
3519 | goto returnbi; | |
3520 | } | |
3521 | } | |
3522 | sh = head_sh; | |
3523 | dev = &sh->dev[i]; | |
f8dfcffd N |
3524 | } else if (test_bit(R5_Discard, &dev->flags)) |
3525 | discard_pending = 1; | |
3526 | } | |
f6bed0ef | 3527 | |
0576b1c6 SL |
3528 | r5l_stripe_write_finished(sh); |
3529 | ||
f8dfcffd N |
3530 | if (!discard_pending && |
3531 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3532 | int hash; |
f8dfcffd N |
3533 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3534 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3535 | if (sh->qd_idx >= 0) { | |
3536 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3537 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3538 | } | |
3539 | /* now that discard is done we can proceed with any sync */ | |
3540 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3541 | /* |
3542 | * SCSI discard will change some bio fields and the stripe has | |
3543 | * no updated data, so remove it from hash list and the stripe | |
3544 | * will be reinitialized | |
3545 | */ | |
59fc630b | 3546 | unhash: |
b8a9d66d RG |
3547 | hash = sh->hash_lock_index; |
3548 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3549 | remove_hash(sh); |
b8a9d66d | 3550 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3551 | if (head_sh->batch_head) { |
3552 | sh = list_first_entry(&sh->batch_list, | |
3553 | struct stripe_head, batch_list); | |
3554 | if (sh != head_sh) | |
3555 | goto unhash; | |
3556 | } | |
59fc630b | 3557 | sh = head_sh; |
3558 | ||
f8dfcffd N |
3559 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3560 | set_bit(STRIPE_HANDLE, &sh->state); | |
3561 | ||
3562 | } | |
8b3e6cdc DW |
3563 | |
3564 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3565 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3566 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3567 | |
787b76fa N |
3568 | if (head_sh->batch_head && do_endio) |
3569 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3570 | } |
3571 | ||
d1688a6d | 3572 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
3573 | struct stripe_head *sh, |
3574 | struct stripe_head_state *s, | |
3575 | int disks) | |
a4456856 DW |
3576 | { |
3577 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3578 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3579 | ||
584acdd4 | 3580 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3581 | * If yes, then the array is dirty (after unclean shutdown or |
3582 | * initial creation), so parity in some stripes might be inconsistent. | |
3583 | * In this case, we need to always do reconstruct-write, to ensure | |
3584 | * that in case of drive failure or read-error correction, we | |
3585 | * generate correct data from the parity. | |
3586 | */ | |
584acdd4 | 3587 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3588 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3589 | s->failed == 0)) { | |
a7854487 | 3590 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3591 | * look like rcw is cheaper |
3592 | */ | |
3593 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3594 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3595 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3596 | (unsigned long long)sh->sector); |
c8ac1803 | 3597 | } else for (i = disks; i--; ) { |
a4456856 DW |
3598 | /* would I have to read this buffer for read_modify_write */ |
3599 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3600 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3601 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3602 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3603 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
3604 | if (test_bit(R5_Insync, &dev->flags)) |
3605 | rmw++; | |
3606 | else | |
3607 | rmw += 2*disks; /* cannot read it */ | |
3608 | } | |
3609 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3610 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3611 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3612 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3613 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3614 | test_bit(R5_Wantcompute, &dev->flags))) { | |
67f45548 N |
3615 | if (test_bit(R5_Insync, &dev->flags)) |
3616 | rcw++; | |
a4456856 DW |
3617 | else |
3618 | rcw += 2*disks; | |
3619 | } | |
3620 | } | |
45b4233c | 3621 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
3622 | (unsigned long long)sh->sector, rmw, rcw); |
3623 | set_bit(STRIPE_HANDLE, &sh->state); | |
41257580 | 3624 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3625 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3626 | if (conf->mddev->queue) |
3627 | blk_add_trace_msg(conf->mddev->queue, | |
3628 | "raid5 rmw %llu %d", | |
3629 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3630 | for (i = disks; i--; ) { |
3631 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3632 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3633 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3634 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3635 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 | 3636 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3637 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3638 | &sh->state)) { | |
3639 | pr_debug("Read_old block %d for r-m-w\n", | |
3640 | i); | |
a4456856 DW |
3641 | set_bit(R5_LOCKED, &dev->flags); |
3642 | set_bit(R5_Wantread, &dev->flags); | |
3643 | s->locked++; | |
3644 | } else { | |
3645 | set_bit(STRIPE_DELAYED, &sh->state); | |
3646 | set_bit(STRIPE_HANDLE, &sh->state); | |
3647 | } | |
3648 | } | |
3649 | } | |
a9add5d9 | 3650 | } |
41257580 | 3651 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3652 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3653 | int qread =0; |
c8ac1803 | 3654 | rcw = 0; |
a4456856 DW |
3655 | for (i = disks; i--; ) { |
3656 | struct r5dev *dev = &sh->dev[i]; | |
3657 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3658 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3659 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3660 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3661 | test_bit(R5_Wantcompute, &dev->flags))) { |
3662 | rcw++; | |
67f45548 N |
3663 | if (test_bit(R5_Insync, &dev->flags) && |
3664 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3665 | &sh->state)) { | |
45b4233c | 3666 | pr_debug("Read_old block " |
a4456856 DW |
3667 | "%d for Reconstruct\n", i); |
3668 | set_bit(R5_LOCKED, &dev->flags); | |
3669 | set_bit(R5_Wantread, &dev->flags); | |
3670 | s->locked++; | |
a9add5d9 | 3671 | qread++; |
a4456856 DW |
3672 | } else { |
3673 | set_bit(STRIPE_DELAYED, &sh->state); | |
3674 | set_bit(STRIPE_HANDLE, &sh->state); | |
3675 | } | |
3676 | } | |
3677 | } | |
e3620a3a | 3678 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3679 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3680 | (unsigned long long)sh->sector, | |
3681 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3682 | } |
b1b02fe9 N |
3683 | |
3684 | if (rcw > disks && rmw > disks && | |
3685 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3686 | set_bit(STRIPE_DELAYED, &sh->state); | |
3687 | ||
a4456856 DW |
3688 | /* now if nothing is locked, and if we have enough data, |
3689 | * we can start a write request | |
3690 | */ | |
f38e1219 DW |
3691 | /* since handle_stripe can be called at any time we need to handle the |
3692 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3693 | * subsequent call wants to start a write request. raid_run_ops only |
3694 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3695 | * simultaneously. If this is not the case then new writes need to be |
3696 | * held off until the compute completes. | |
3697 | */ | |
976ea8d4 DW |
3698 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3699 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
3700 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 3701 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
3702 | } |
3703 | ||
d1688a6d | 3704 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3705 | struct stripe_head_state *s, int disks) |
3706 | { | |
ecc65c9b | 3707 | struct r5dev *dev = NULL; |
bd2ab670 | 3708 | |
59fc630b | 3709 | BUG_ON(sh->batch_head); |
a4456856 | 3710 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3711 | |
ecc65c9b DW |
3712 | switch (sh->check_state) { |
3713 | case check_state_idle: | |
3714 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3715 | if (s->failed == 0) { |
bd2ab670 | 3716 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3717 | sh->check_state = check_state_run; |
3718 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3719 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3720 | s->uptodate--; |
ecc65c9b | 3721 | break; |
bd2ab670 | 3722 | } |
f2b3b44d | 3723 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3724 | /* fall through */ |
3725 | case check_state_compute_result: | |
3726 | sh->check_state = check_state_idle; | |
3727 | if (!dev) | |
3728 | dev = &sh->dev[sh->pd_idx]; | |
3729 | ||
3730 | /* check that a write has not made the stripe insync */ | |
3731 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3732 | break; | |
c8894419 | 3733 | |
a4456856 | 3734 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3735 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3736 | BUG_ON(s->uptodate != disks); | |
3737 | ||
3738 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3739 | s->locked++; |
a4456856 | 3740 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3741 | |
a4456856 | 3742 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3743 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3744 | break; |
3745 | case check_state_run: | |
3746 | break; /* we will be called again upon completion */ | |
3747 | case check_state_check_result: | |
3748 | sh->check_state = check_state_idle; | |
3749 | ||
3750 | /* if a failure occurred during the check operation, leave | |
3751 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3752 | */ | |
3753 | if (s->failed) | |
3754 | break; | |
3755 | ||
3756 | /* handle a successful check operation, if parity is correct | |
3757 | * we are done. Otherwise update the mismatch count and repair | |
3758 | * parity if !MD_RECOVERY_CHECK | |
3759 | */ | |
ad283ea4 | 3760 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3761 | /* parity is correct (on disc, |
3762 | * not in buffer any more) | |
3763 | */ | |
3764 | set_bit(STRIPE_INSYNC, &sh->state); | |
3765 | else { | |
7f7583d4 | 3766 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3767 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3768 | /* don't try to repair!! */ | |
3769 | set_bit(STRIPE_INSYNC, &sh->state); | |
3770 | else { | |
3771 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3772 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3773 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3774 | set_bit(R5_Wantcompute, | |
3775 | &sh->dev[sh->pd_idx].flags); | |
3776 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3777 | sh->ops.target2 = -1; |
ecc65c9b DW |
3778 | s->uptodate++; |
3779 | } | |
3780 | } | |
3781 | break; | |
3782 | case check_state_compute_run: | |
3783 | break; | |
3784 | default: | |
3785 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3786 | __func__, sh->check_state, | |
3787 | (unsigned long long) sh->sector); | |
3788 | BUG(); | |
a4456856 DW |
3789 | } |
3790 | } | |
3791 | ||
d1688a6d | 3792 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3793 | struct stripe_head_state *s, |
f2b3b44d | 3794 | int disks) |
a4456856 | 3795 | { |
a4456856 | 3796 | int pd_idx = sh->pd_idx; |
34e04e87 | 3797 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3798 | struct r5dev *dev; |
a4456856 | 3799 | |
59fc630b | 3800 | BUG_ON(sh->batch_head); |
a4456856 DW |
3801 | set_bit(STRIPE_HANDLE, &sh->state); |
3802 | ||
3803 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3804 | |
a4456856 DW |
3805 | /* Want to check and possibly repair P and Q. |
3806 | * However there could be one 'failed' device, in which | |
3807 | * case we can only check one of them, possibly using the | |
3808 | * other to generate missing data | |
3809 | */ | |
3810 | ||
d82dfee0 DW |
3811 | switch (sh->check_state) { |
3812 | case check_state_idle: | |
3813 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 3814 | if (s->failed == s->q_failed) { |
d82dfee0 | 3815 | /* The only possible failed device holds Q, so it |
a4456856 DW |
3816 | * makes sense to check P (If anything else were failed, |
3817 | * we would have used P to recreate it). | |
3818 | */ | |
d82dfee0 | 3819 | sh->check_state = check_state_run; |
a4456856 | 3820 | } |
f2b3b44d | 3821 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 3822 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
3823 | * anything, so it makes sense to check it |
3824 | */ | |
d82dfee0 DW |
3825 | if (sh->check_state == check_state_run) |
3826 | sh->check_state = check_state_run_pq; | |
3827 | else | |
3828 | sh->check_state = check_state_run_q; | |
a4456856 | 3829 | } |
a4456856 | 3830 | |
d82dfee0 DW |
3831 | /* discard potentially stale zero_sum_result */ |
3832 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3833 | |
d82dfee0 DW |
3834 | if (sh->check_state == check_state_run) { |
3835 | /* async_xor_zero_sum destroys the contents of P */ | |
3836 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3837 | s->uptodate--; | |
a4456856 | 3838 | } |
d82dfee0 DW |
3839 | if (sh->check_state >= check_state_run && |
3840 | sh->check_state <= check_state_run_pq) { | |
3841 | /* async_syndrome_zero_sum preserves P and Q, so | |
3842 | * no need to mark them !uptodate here | |
3843 | */ | |
3844 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3845 | break; | |
a4456856 DW |
3846 | } |
3847 | ||
d82dfee0 DW |
3848 | /* we have 2-disk failure */ |
3849 | BUG_ON(s->failed != 2); | |
3850 | /* fall through */ | |
3851 | case check_state_compute_result: | |
3852 | sh->check_state = check_state_idle; | |
a4456856 | 3853 | |
d82dfee0 DW |
3854 | /* check that a write has not made the stripe insync */ |
3855 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3856 | break; | |
a4456856 DW |
3857 | |
3858 | /* now write out any block on a failed drive, | |
d82dfee0 | 3859 | * or P or Q if they were recomputed |
a4456856 | 3860 | */ |
d82dfee0 | 3861 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3862 | if (s->failed == 2) { |
f2b3b44d | 3863 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3864 | s->locked++; |
3865 | set_bit(R5_LOCKED, &dev->flags); | |
3866 | set_bit(R5_Wantwrite, &dev->flags); | |
3867 | } | |
3868 | if (s->failed >= 1) { | |
f2b3b44d | 3869 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3870 | s->locked++; |
3871 | set_bit(R5_LOCKED, &dev->flags); | |
3872 | set_bit(R5_Wantwrite, &dev->flags); | |
3873 | } | |
d82dfee0 | 3874 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3875 | dev = &sh->dev[pd_idx]; |
3876 | s->locked++; | |
3877 | set_bit(R5_LOCKED, &dev->flags); | |
3878 | set_bit(R5_Wantwrite, &dev->flags); | |
3879 | } | |
d82dfee0 | 3880 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3881 | dev = &sh->dev[qd_idx]; |
3882 | s->locked++; | |
3883 | set_bit(R5_LOCKED, &dev->flags); | |
3884 | set_bit(R5_Wantwrite, &dev->flags); | |
3885 | } | |
3886 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3887 | ||
3888 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3889 | break; |
3890 | case check_state_run: | |
3891 | case check_state_run_q: | |
3892 | case check_state_run_pq: | |
3893 | break; /* we will be called again upon completion */ | |
3894 | case check_state_check_result: | |
3895 | sh->check_state = check_state_idle; | |
3896 | ||
3897 | /* handle a successful check operation, if parity is correct | |
3898 | * we are done. Otherwise update the mismatch count and repair | |
3899 | * parity if !MD_RECOVERY_CHECK | |
3900 | */ | |
3901 | if (sh->ops.zero_sum_result == 0) { | |
3902 | /* both parities are correct */ | |
3903 | if (!s->failed) | |
3904 | set_bit(STRIPE_INSYNC, &sh->state); | |
3905 | else { | |
3906 | /* in contrast to the raid5 case we can validate | |
3907 | * parity, but still have a failure to write | |
3908 | * back | |
3909 | */ | |
3910 | sh->check_state = check_state_compute_result; | |
3911 | /* Returning at this point means that we may go | |
3912 | * off and bring p and/or q uptodate again so | |
3913 | * we make sure to check zero_sum_result again | |
3914 | * to verify if p or q need writeback | |
3915 | */ | |
3916 | } | |
3917 | } else { | |
7f7583d4 | 3918 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
3919 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3920 | /* don't try to repair!! */ | |
3921 | set_bit(STRIPE_INSYNC, &sh->state); | |
3922 | else { | |
3923 | int *target = &sh->ops.target; | |
3924 | ||
3925 | sh->ops.target = -1; | |
3926 | sh->ops.target2 = -1; | |
3927 | sh->check_state = check_state_compute_run; | |
3928 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3929 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3930 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3931 | set_bit(R5_Wantcompute, | |
3932 | &sh->dev[pd_idx].flags); | |
3933 | *target = pd_idx; | |
3934 | target = &sh->ops.target2; | |
3935 | s->uptodate++; | |
3936 | } | |
3937 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3938 | set_bit(R5_Wantcompute, | |
3939 | &sh->dev[qd_idx].flags); | |
3940 | *target = qd_idx; | |
3941 | s->uptodate++; | |
3942 | } | |
3943 | } | |
3944 | } | |
3945 | break; | |
3946 | case check_state_compute_run: | |
3947 | break; | |
3948 | default: | |
3949 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3950 | __func__, sh->check_state, | |
3951 | (unsigned long long) sh->sector); | |
3952 | BUG(); | |
a4456856 DW |
3953 | } |
3954 | } | |
3955 | ||
d1688a6d | 3956 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3957 | { |
3958 | int i; | |
3959 | ||
3960 | /* We have read all the blocks in this stripe and now we need to | |
3961 | * copy some of them into a target stripe for expand. | |
3962 | */ | |
f0a50d37 | 3963 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 3964 | BUG_ON(sh->batch_head); |
a4456856 DW |
3965 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3966 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3967 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3968 | int dd_idx, j; |
a4456856 | 3969 | struct stripe_head *sh2; |
a08abd8c | 3970 | struct async_submit_ctl submit; |
a4456856 | 3971 | |
6d036f7d | 3972 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
3973 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3974 | &dd_idx, NULL); | |
6d036f7d | 3975 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3976 | if (sh2 == NULL) |
3977 | /* so far only the early blocks of this stripe | |
3978 | * have been requested. When later blocks | |
3979 | * get requested, we will try again | |
3980 | */ | |
3981 | continue; | |
3982 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3983 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3984 | /* must have already done this block */ | |
6d036f7d | 3985 | raid5_release_stripe(sh2); |
a4456856 DW |
3986 | continue; |
3987 | } | |
f0a50d37 DW |
3988 | |
3989 | /* place all the copies on one channel */ | |
a08abd8c | 3990 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3991 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3992 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3993 | &submit); |
f0a50d37 | 3994 | |
a4456856 DW |
3995 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3996 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3997 | for (j = 0; j < conf->raid_disks; j++) | |
3998 | if (j != sh2->pd_idx && | |
86c374ba | 3999 | j != sh2->qd_idx && |
a4456856 DW |
4000 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4001 | break; | |
4002 | if (j == conf->raid_disks) { | |
4003 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4004 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4005 | } | |
6d036f7d | 4006 | raid5_release_stripe(sh2); |
f0a50d37 | 4007 | |
a4456856 | 4008 | } |
a2e08551 | 4009 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4010 | async_tx_quiesce(&tx); |
a4456856 | 4011 | } |
1da177e4 LT |
4012 | |
4013 | /* | |
4014 | * handle_stripe - do things to a stripe. | |
4015 | * | |
9a3e1101 N |
4016 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4017 | * state of various bits to see what needs to be done. | |
1da177e4 | 4018 | * Possible results: |
9a3e1101 N |
4019 | * return some read requests which now have data |
4020 | * return some write requests which are safely on storage | |
1da177e4 LT |
4021 | * schedule a read on some buffers |
4022 | * schedule a write of some buffers | |
4023 | * return confirmation of parity correctness | |
4024 | * | |
1da177e4 | 4025 | */ |
a4456856 | 4026 | |
acfe726b | 4027 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4028 | { |
d1688a6d | 4029 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4030 | int disks = sh->disks; |
474af965 N |
4031 | struct r5dev *dev; |
4032 | int i; | |
9a3e1101 | 4033 | int do_recovery = 0; |
1da177e4 | 4034 | |
acfe726b N |
4035 | memset(s, 0, sizeof(*s)); |
4036 | ||
dabc4ec6 | 4037 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4038 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4039 | s->failed_num[0] = -1; |
4040 | s->failed_num[1] = -1; | |
6e74a9cf | 4041 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4042 | |
acfe726b | 4043 | /* Now to look around and see what can be done */ |
1da177e4 | 4044 | rcu_read_lock(); |
16a53ecc | 4045 | for (i=disks; i--; ) { |
3cb03002 | 4046 | struct md_rdev *rdev; |
31c176ec N |
4047 | sector_t first_bad; |
4048 | int bad_sectors; | |
4049 | int is_bad = 0; | |
acfe726b | 4050 | |
16a53ecc | 4051 | dev = &sh->dev[i]; |
1da177e4 | 4052 | |
45b4233c | 4053 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4054 | i, dev->flags, |
4055 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4056 | /* maybe we can reply to a read |
4057 | * | |
4058 | * new wantfill requests are only permitted while | |
4059 | * ops_complete_biofill is guaranteed to be inactive | |
4060 | */ | |
4061 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4062 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4063 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4064 | |
16a53ecc | 4065 | /* now count some things */ |
cc94015a N |
4066 | if (test_bit(R5_LOCKED, &dev->flags)) |
4067 | s->locked++; | |
4068 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4069 | s->uptodate++; | |
2d6e4ecc | 4070 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4071 | s->compute++; |
4072 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4073 | } |
1da177e4 | 4074 | |
acfe726b | 4075 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4076 | s->to_fill++; |
acfe726b | 4077 | else if (dev->toread) |
cc94015a | 4078 | s->to_read++; |
16a53ecc | 4079 | if (dev->towrite) { |
cc94015a | 4080 | s->to_write++; |
16a53ecc | 4081 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4082 | s->non_overwrite++; |
16a53ecc | 4083 | } |
a4456856 | 4084 | if (dev->written) |
cc94015a | 4085 | s->written++; |
14a75d3e N |
4086 | /* Prefer to use the replacement for reads, but only |
4087 | * if it is recovered enough and has no bad blocks. | |
4088 | */ | |
4089 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4090 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4091 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4092 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4093 | &first_bad, &bad_sectors)) | |
4094 | set_bit(R5_ReadRepl, &dev->flags); | |
4095 | else { | |
e6030cb0 | 4096 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4097 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4098 | else |
4099 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4100 | rdev = rcu_dereference(conf->disks[i].rdev); |
4101 | clear_bit(R5_ReadRepl, &dev->flags); | |
4102 | } | |
9283d8c5 N |
4103 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4104 | rdev = NULL; | |
31c176ec N |
4105 | if (rdev) { |
4106 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4107 | &first_bad, &bad_sectors); | |
4108 | if (s->blocked_rdev == NULL | |
4109 | && (test_bit(Blocked, &rdev->flags) | |
4110 | || is_bad < 0)) { | |
4111 | if (is_bad < 0) | |
4112 | set_bit(BlockedBadBlocks, | |
4113 | &rdev->flags); | |
4114 | s->blocked_rdev = rdev; | |
4115 | atomic_inc(&rdev->nr_pending); | |
4116 | } | |
6bfe0b49 | 4117 | } |
415e72d0 N |
4118 | clear_bit(R5_Insync, &dev->flags); |
4119 | if (!rdev) | |
4120 | /* Not in-sync */; | |
31c176ec N |
4121 | else if (is_bad) { |
4122 | /* also not in-sync */ | |
18b9837e N |
4123 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4124 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4125 | /* treat as in-sync, but with a read error |
4126 | * which we can now try to correct | |
4127 | */ | |
4128 | set_bit(R5_Insync, &dev->flags); | |
4129 | set_bit(R5_ReadError, &dev->flags); | |
4130 | } | |
4131 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4132 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4133 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4134 | /* in sync if before recovery_offset */ |
30d7a483 N |
4135 | set_bit(R5_Insync, &dev->flags); |
4136 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4137 | test_bit(R5_Expanded, &dev->flags)) | |
4138 | /* If we've reshaped into here, we assume it is Insync. | |
4139 | * We will shortly update recovery_offset to make | |
4140 | * it official. | |
4141 | */ | |
4142 | set_bit(R5_Insync, &dev->flags); | |
4143 | ||
1cc03eb9 | 4144 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4145 | /* This flag does not apply to '.replacement' |
4146 | * only to .rdev, so make sure to check that*/ | |
4147 | struct md_rdev *rdev2 = rcu_dereference( | |
4148 | conf->disks[i].rdev); | |
4149 | if (rdev2 == rdev) | |
4150 | clear_bit(R5_Insync, &dev->flags); | |
4151 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4152 | s->handle_bad_blocks = 1; |
14a75d3e | 4153 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4154 | } else |
4155 | clear_bit(R5_WriteError, &dev->flags); | |
4156 | } | |
1cc03eb9 | 4157 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4158 | /* This flag does not apply to '.replacement' |
4159 | * only to .rdev, so make sure to check that*/ | |
4160 | struct md_rdev *rdev2 = rcu_dereference( | |
4161 | conf->disks[i].rdev); | |
4162 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4163 | s->handle_bad_blocks = 1; |
14a75d3e | 4164 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4165 | } else |
4166 | clear_bit(R5_MadeGood, &dev->flags); | |
4167 | } | |
977df362 N |
4168 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4169 | struct md_rdev *rdev2 = rcu_dereference( | |
4170 | conf->disks[i].replacement); | |
4171 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4172 | s->handle_bad_blocks = 1; | |
4173 | atomic_inc(&rdev2->nr_pending); | |
4174 | } else | |
4175 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4176 | } | |
415e72d0 | 4177 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4178 | /* The ReadError flag will just be confusing now */ |
4179 | clear_bit(R5_ReadError, &dev->flags); | |
4180 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4181 | } |
415e72d0 N |
4182 | if (test_bit(R5_ReadError, &dev->flags)) |
4183 | clear_bit(R5_Insync, &dev->flags); | |
4184 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4185 | if (s->failed < 2) |
4186 | s->failed_num[s->failed] = i; | |
4187 | s->failed++; | |
9a3e1101 N |
4188 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4189 | do_recovery = 1; | |
415e72d0 | 4190 | } |
1da177e4 | 4191 | } |
9a3e1101 N |
4192 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4193 | /* If there is a failed device being replaced, | |
4194 | * we must be recovering. | |
4195 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4196 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4197 | * else we can only be replacing |
4198 | * sync and recovery both need to read all devices, and so | |
4199 | * use the same flag. | |
4200 | */ | |
4201 | if (do_recovery || | |
c6d2e084 | 4202 | sh->sector >= conf->mddev->recovery_cp || |
4203 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4204 | s->syncing = 1; |
4205 | else | |
4206 | s->replacing = 1; | |
4207 | } | |
1da177e4 | 4208 | rcu_read_unlock(); |
cc94015a N |
4209 | } |
4210 | ||
59fc630b | 4211 | static int clear_batch_ready(struct stripe_head *sh) |
4212 | { | |
b15a9dbd N |
4213 | /* Return '1' if this is a member of batch, or |
4214 | * '0' if it is a lone stripe or a head which can now be | |
4215 | * handled. | |
4216 | */ | |
59fc630b | 4217 | struct stripe_head *tmp; |
4218 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4219 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4220 | spin_lock(&sh->stripe_lock); |
4221 | if (!sh->batch_head) { | |
4222 | spin_unlock(&sh->stripe_lock); | |
4223 | return 0; | |
4224 | } | |
4225 | ||
4226 | /* | |
4227 | * this stripe could be added to a batch list before we check | |
4228 | * BATCH_READY, skips it | |
4229 | */ | |
4230 | if (sh->batch_head != sh) { | |
4231 | spin_unlock(&sh->stripe_lock); | |
4232 | return 1; | |
4233 | } | |
4234 | spin_lock(&sh->batch_lock); | |
4235 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4236 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4237 | spin_unlock(&sh->batch_lock); | |
4238 | spin_unlock(&sh->stripe_lock); | |
4239 | ||
4240 | /* | |
4241 | * BATCH_READY is cleared, no new stripes can be added. | |
4242 | * batch_list can be accessed without lock | |
4243 | */ | |
4244 | return 0; | |
4245 | } | |
4246 | ||
3960ce79 N |
4247 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4248 | unsigned long handle_flags) | |
72ac7330 | 4249 | { |
4e3d62ff | 4250 | struct stripe_head *sh, *next; |
72ac7330 | 4251 | int i; |
fb642b92 | 4252 | int do_wakeup = 0; |
72ac7330 | 4253 | |
bb27051f N |
4254 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4255 | ||
72ac7330 | 4256 | list_del_init(&sh->batch_list); |
4257 | ||
fb3229d5 | 4258 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4259 | (1 << STRIPE_SYNCING) | |
4260 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4261 | (1 << STRIPE_DELAYED) | |
4262 | (1 << STRIPE_BIT_DELAY) | | |
4263 | (1 << STRIPE_FULL_WRITE) | | |
4264 | (1 << STRIPE_BIOFILL_RUN) | | |
4265 | (1 << STRIPE_COMPUTE_RUN) | | |
4266 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4267 | (1 << STRIPE_DISCARD) | | |
4268 | (1 << STRIPE_BATCH_READY) | | |
4269 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4270 | (1 << STRIPE_BITMAP_PENDING)), |
4271 | "stripe state: %lx\n", sh->state); | |
4272 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4273 | (1 << STRIPE_REPLACED)), | |
4274 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4275 | |
4276 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4277 | (1 << STRIPE_PREREAD_ACTIVE) | |
1b956f7a N |
4278 | (1 << STRIPE_DEGRADED)), |
4279 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4280 | ||
72ac7330 | 4281 | sh->check_state = head_sh->check_state; |
4282 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4283 | for (i = 0; i < sh->disks; i++) { |
4284 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4285 | do_wakeup = 1; | |
72ac7330 | 4286 | sh->dev[i].flags = head_sh->dev[i].flags & |
4287 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4288 | } |
72ac7330 | 4289 | spin_lock_irq(&sh->stripe_lock); |
4290 | sh->batch_head = NULL; | |
4291 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4292 | if (handle_flags == 0 || |
4293 | sh->state & handle_flags) | |
4294 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4295 | raid5_release_stripe(sh); |
72ac7330 | 4296 | } |
fb642b92 N |
4297 | spin_lock_irq(&head_sh->stripe_lock); |
4298 | head_sh->batch_head = NULL; | |
4299 | spin_unlock_irq(&head_sh->stripe_lock); | |
4300 | for (i = 0; i < head_sh->disks; i++) | |
4301 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4302 | do_wakeup = 1; | |
3960ce79 N |
4303 | if (head_sh->state & handle_flags) |
4304 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4305 | |
4306 | if (do_wakeup) | |
4307 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4308 | } |
4309 | ||
cc94015a N |
4310 | static void handle_stripe(struct stripe_head *sh) |
4311 | { | |
4312 | struct stripe_head_state s; | |
d1688a6d | 4313 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4314 | int i; |
84789554 N |
4315 | int prexor; |
4316 | int disks = sh->disks; | |
474af965 | 4317 | struct r5dev *pdev, *qdev; |
cc94015a N |
4318 | |
4319 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4320 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4321 | /* already being handled, ensure it gets handled |
4322 | * again when current action finishes */ | |
4323 | set_bit(STRIPE_HANDLE, &sh->state); | |
4324 | return; | |
4325 | } | |
4326 | ||
59fc630b | 4327 | if (clear_batch_ready(sh) ) { |
4328 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4329 | return; | |
4330 | } | |
4331 | ||
4e3d62ff | 4332 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4333 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4334 | |
dabc4ec6 | 4335 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4336 | spin_lock(&sh->stripe_lock); |
4337 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4338 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4339 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4340 | set_bit(STRIPE_SYNCING, &sh->state); | |
4341 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4342 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4343 | } |
4344 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4345 | } |
4346 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4347 | ||
4348 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4349 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4350 | (unsigned long long)sh->sector, sh->state, | |
4351 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4352 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4353 | |
acfe726b | 4354 | analyse_stripe(sh, &s); |
c5a31000 | 4355 | |
b70abcb2 SL |
4356 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4357 | goto finish; | |
4358 | ||
bc2607f3 N |
4359 | if (s.handle_bad_blocks) { |
4360 | set_bit(STRIPE_HANDLE, &sh->state); | |
4361 | goto finish; | |
4362 | } | |
4363 | ||
474af965 N |
4364 | if (unlikely(s.blocked_rdev)) { |
4365 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4366 | s.replacing || s.to_write || s.written) { |
474af965 N |
4367 | set_bit(STRIPE_HANDLE, &sh->state); |
4368 | goto finish; | |
4369 | } | |
4370 | /* There is nothing for the blocked_rdev to block */ | |
4371 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4372 | s.blocked_rdev = NULL; | |
4373 | } | |
4374 | ||
4375 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4376 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4377 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4378 | } | |
4379 | ||
4380 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4381 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4382 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4383 | s.failed_num[0], s.failed_num[1]); | |
4384 | /* check if the array has lost more than max_degraded devices and, | |
4385 | * if so, some requests might need to be failed. | |
4386 | */ | |
6e74a9cf | 4387 | if (s.failed > conf->max_degraded || s.log_failed) { |
9a3f530f N |
4388 | sh->check_state = 0; |
4389 | sh->reconstruct_state = 0; | |
626f2092 | 4390 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4391 | if (s.to_read+s.to_write+s.written) |
4392 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4393 | if (s.syncing + s.replacing) |
9a3f530f N |
4394 | handle_failed_sync(conf, sh, &s); |
4395 | } | |
474af965 | 4396 | |
84789554 N |
4397 | /* Now we check to see if any write operations have recently |
4398 | * completed | |
4399 | */ | |
4400 | prexor = 0; | |
4401 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4402 | prexor = 1; | |
4403 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4404 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4405 | sh->reconstruct_state = reconstruct_state_idle; | |
4406 | ||
4407 | /* All the 'written' buffers and the parity block are ready to | |
4408 | * be written back to disk | |
4409 | */ | |
9e444768 SL |
4410 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4411 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4412 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4413 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4414 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4415 | for (i = disks; i--; ) { |
4416 | struct r5dev *dev = &sh->dev[i]; | |
4417 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4418 | (i == sh->pd_idx || i == sh->qd_idx || | |
4419 | dev->written)) { | |
4420 | pr_debug("Writing block %d\n", i); | |
4421 | set_bit(R5_Wantwrite, &dev->flags); | |
4422 | if (prexor) | |
4423 | continue; | |
9c4bdf69 N |
4424 | if (s.failed > 1) |
4425 | continue; | |
84789554 N |
4426 | if (!test_bit(R5_Insync, &dev->flags) || |
4427 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4428 | s.failed == 0)) | |
4429 | set_bit(STRIPE_INSYNC, &sh->state); | |
4430 | } | |
4431 | } | |
4432 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4433 | s.dec_preread_active = 1; | |
4434 | } | |
4435 | ||
ef5b7c69 N |
4436 | /* |
4437 | * might be able to return some write requests if the parity blocks | |
4438 | * are safe, or on a failed drive | |
4439 | */ | |
4440 | pdev = &sh->dev[sh->pd_idx]; | |
4441 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4442 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4443 | qdev = &sh->dev[sh->qd_idx]; | |
4444 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4445 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4446 | || conf->level < 6; | |
4447 | ||
4448 | if (s.written && | |
4449 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4450 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4451 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4452 | test_bit(R5_Discard, &pdev->flags))))) && | |
4453 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4454 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4455 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4456 | test_bit(R5_Discard, &qdev->flags)))))) | |
4457 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4458 | ||
4459 | /* Now we might consider reading some blocks, either to check/generate | |
4460 | * parity, or to satisfy requests | |
4461 | * or to load a block that is being partially written. | |
4462 | */ | |
4463 | if (s.to_read || s.non_overwrite | |
4464 | || (conf->level == 6 && s.to_write && s.failed) | |
4465 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4466 | || s.replacing | |
4467 | || s.expanding) | |
4468 | handle_stripe_fill(sh, &s, disks); | |
4469 | ||
84789554 N |
4470 | /* Now to consider new write requests and what else, if anything |
4471 | * should be read. We do not handle new writes when: | |
4472 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
4473 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4474 | * block. | |
4475 | */ | |
4476 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
4477 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4478 | ||
4479 | /* maybe we need to check and possibly fix the parity for this stripe | |
4480 | * Any reads will already have been scheduled, so we just see if enough | |
4481 | * data is available. The parity check is held off while parity | |
4482 | * dependent operations are in flight. | |
4483 | */ | |
4484 | if (sh->check_state || | |
4485 | (s.syncing && s.locked == 0 && | |
4486 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4487 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4488 | if (conf->level == 6) | |
4489 | handle_parity_checks6(conf, sh, &s, disks); | |
4490 | else | |
4491 | handle_parity_checks5(conf, sh, &s, disks); | |
4492 | } | |
c5a31000 | 4493 | |
f94c0b66 N |
4494 | if ((s.replacing || s.syncing) && s.locked == 0 |
4495 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4496 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4497 | /* Write out to replacement devices where possible */ |
4498 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4499 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4500 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4501 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4502 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4503 | s.locked++; | |
4504 | } | |
f94c0b66 N |
4505 | if (s.replacing) |
4506 | set_bit(STRIPE_INSYNC, &sh->state); | |
4507 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4508 | } |
4509 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4510 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4511 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4512 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4513 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4514 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4515 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4516 | } |
4517 | ||
4518 | /* If the failed drives are just a ReadError, then we might need | |
4519 | * to progress the repair/check process | |
4520 | */ | |
4521 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4522 | for (i = 0; i < s.failed; i++) { | |
4523 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4524 | if (test_bit(R5_ReadError, &dev->flags) | |
4525 | && !test_bit(R5_LOCKED, &dev->flags) | |
4526 | && test_bit(R5_UPTODATE, &dev->flags) | |
4527 | ) { | |
4528 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4529 | set_bit(R5_Wantwrite, &dev->flags); | |
4530 | set_bit(R5_ReWrite, &dev->flags); | |
4531 | set_bit(R5_LOCKED, &dev->flags); | |
4532 | s.locked++; | |
4533 | } else { | |
4534 | /* let's read it back */ | |
4535 | set_bit(R5_Wantread, &dev->flags); | |
4536 | set_bit(R5_LOCKED, &dev->flags); | |
4537 | s.locked++; | |
4538 | } | |
4539 | } | |
4540 | } | |
4541 | ||
3687c061 N |
4542 | /* Finish reconstruct operations initiated by the expansion process */ |
4543 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4544 | struct stripe_head *sh_src | |
6d036f7d | 4545 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4546 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4547 | /* sh cannot be written until sh_src has been read. | |
4548 | * so arrange for sh to be delayed a little | |
4549 | */ | |
4550 | set_bit(STRIPE_DELAYED, &sh->state); | |
4551 | set_bit(STRIPE_HANDLE, &sh->state); | |
4552 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4553 | &sh_src->state)) | |
4554 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4555 | raid5_release_stripe(sh_src); |
3687c061 N |
4556 | goto finish; |
4557 | } | |
4558 | if (sh_src) | |
6d036f7d | 4559 | raid5_release_stripe(sh_src); |
3687c061 N |
4560 | |
4561 | sh->reconstruct_state = reconstruct_state_idle; | |
4562 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4563 | for (i = conf->raid_disks; i--; ) { | |
4564 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4565 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4566 | s.locked++; | |
4567 | } | |
4568 | } | |
f416885e | 4569 | |
3687c061 N |
4570 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4571 | !sh->reconstruct_state) { | |
4572 | /* Need to write out all blocks after computing parity */ | |
4573 | sh->disks = conf->raid_disks; | |
4574 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4575 | schedule_reconstruction(sh, &s, 1, 1); | |
4576 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4577 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4578 | atomic_dec(&conf->reshape_stripes); | |
4579 | wake_up(&conf->wait_for_overlap); | |
4580 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4581 | } | |
4582 | ||
4583 | if (s.expanding && s.locked == 0 && | |
4584 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4585 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4586 | |
3687c061 | 4587 | finish: |
6bfe0b49 | 4588 | /* wait for this device to become unblocked */ |
5f066c63 N |
4589 | if (unlikely(s.blocked_rdev)) { |
4590 | if (conf->mddev->external) | |
4591 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4592 | conf->mddev); | |
4593 | else | |
4594 | /* Internal metadata will immediately | |
4595 | * be written by raid5d, so we don't | |
4596 | * need to wait here. | |
4597 | */ | |
4598 | rdev_dec_pending(s.blocked_rdev, | |
4599 | conf->mddev); | |
4600 | } | |
6bfe0b49 | 4601 | |
bc2607f3 N |
4602 | if (s.handle_bad_blocks) |
4603 | for (i = disks; i--; ) { | |
3cb03002 | 4604 | struct md_rdev *rdev; |
bc2607f3 N |
4605 | struct r5dev *dev = &sh->dev[i]; |
4606 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4607 | /* We own a safe reference to the rdev */ | |
4608 | rdev = conf->disks[i].rdev; | |
4609 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4610 | STRIPE_SECTORS, 0)) | |
4611 | md_error(conf->mddev, rdev); | |
4612 | rdev_dec_pending(rdev, conf->mddev); | |
4613 | } | |
b84db560 N |
4614 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4615 | rdev = conf->disks[i].rdev; | |
4616 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4617 | STRIPE_SECTORS, 0); |
b84db560 N |
4618 | rdev_dec_pending(rdev, conf->mddev); |
4619 | } | |
977df362 N |
4620 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4621 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4622 | if (!rdev) |
4623 | /* rdev have been moved down */ | |
4624 | rdev = conf->disks[i].rdev; | |
977df362 | 4625 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4626 | STRIPE_SECTORS, 0); |
977df362 N |
4627 | rdev_dec_pending(rdev, conf->mddev); |
4628 | } | |
bc2607f3 N |
4629 | } |
4630 | ||
6c0069c0 YT |
4631 | if (s.ops_request) |
4632 | raid_run_ops(sh, s.ops_request); | |
4633 | ||
f0e43bcd | 4634 | ops_run_io(sh, &s); |
16a53ecc | 4635 | |
c5709ef6 | 4636 | if (s.dec_preread_active) { |
729a1866 | 4637 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4638 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4639 | * have actually been submitted. |
4640 | */ | |
4641 | atomic_dec(&conf->preread_active_stripes); | |
4642 | if (atomic_read(&conf->preread_active_stripes) < | |
4643 | IO_THRESHOLD) | |
4644 | md_wakeup_thread(conf->mddev->thread); | |
4645 | } | |
4646 | ||
c3cce6cd | 4647 | if (!bio_list_empty(&s.return_bi)) { |
7adb072c | 4648 | if (test_bit(MD_CHANGE_PENDING, &conf->mddev->flags)) { |
c3cce6cd N |
4649 | spin_lock_irq(&conf->device_lock); |
4650 | bio_list_merge(&conf->return_bi, &s.return_bi); | |
4651 | spin_unlock_irq(&conf->device_lock); | |
4652 | md_wakeup_thread(conf->mddev->thread); | |
4653 | } else | |
4654 | return_io(&s.return_bi); | |
4655 | } | |
16a53ecc | 4656 | |
257a4b42 | 4657 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4658 | } |
4659 | ||
d1688a6d | 4660 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4661 | { |
4662 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4663 | while (!list_empty(&conf->delayed_list)) { | |
4664 | struct list_head *l = conf->delayed_list.next; | |
4665 | struct stripe_head *sh; | |
4666 | sh = list_entry(l, struct stripe_head, lru); | |
4667 | list_del_init(l); | |
4668 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4669 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4670 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4671 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4672 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4673 | } |
482c0834 | 4674 | } |
16a53ecc N |
4675 | } |
4676 | ||
566c09c5 SL |
4677 | static void activate_bit_delay(struct r5conf *conf, |
4678 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4679 | { |
4680 | /* device_lock is held */ | |
4681 | struct list_head head; | |
4682 | list_add(&head, &conf->bitmap_list); | |
4683 | list_del_init(&conf->bitmap_list); | |
4684 | while (!list_empty(&head)) { | |
4685 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4686 | int hash; |
16a53ecc N |
4687 | list_del_init(&sh->lru); |
4688 | atomic_inc(&sh->count); | |
566c09c5 SL |
4689 | hash = sh->hash_lock_index; |
4690 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4691 | } |
4692 | } | |
4693 | ||
5c675f83 | 4694 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4695 | { |
d1688a6d | 4696 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4697 | |
4698 | /* No difference between reads and writes. Just check | |
4699 | * how busy the stripe_cache is | |
4700 | */ | |
3fa841d7 | 4701 | |
5423399a | 4702 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd N |
4703 | return 1; |
4704 | if (conf->quiesce) | |
4705 | return 1; | |
4bda556a | 4706 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4707 | return 1; |
4708 | ||
4709 | return 0; | |
4710 | } | |
4711 | ||
fd01b88c | 4712 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4713 | { |
3cb5edf4 | 4714 | struct r5conf *conf = mddev->private; |
4f024f37 | 4715 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 4716 | unsigned int chunk_sectors; |
aa8b57aa | 4717 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4718 | |
3cb5edf4 | 4719 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
4720 | return chunk_sectors >= |
4721 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4722 | } | |
4723 | ||
46031f9a RBJ |
4724 | /* |
4725 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4726 | * later sampled by raid5d. | |
4727 | */ | |
d1688a6d | 4728 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4729 | { |
4730 | unsigned long flags; | |
4731 | ||
4732 | spin_lock_irqsave(&conf->device_lock, flags); | |
4733 | ||
4734 | bi->bi_next = conf->retry_read_aligned_list; | |
4735 | conf->retry_read_aligned_list = bi; | |
4736 | ||
4737 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4738 | md_wakeup_thread(conf->mddev->thread); | |
4739 | } | |
4740 | ||
d1688a6d | 4741 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4742 | { |
4743 | struct bio *bi; | |
4744 | ||
4745 | bi = conf->retry_read_aligned; | |
4746 | if (bi) { | |
4747 | conf->retry_read_aligned = NULL; | |
4748 | return bi; | |
4749 | } | |
4750 | bi = conf->retry_read_aligned_list; | |
4751 | if(bi) { | |
387bb173 | 4752 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4753 | bi->bi_next = NULL; |
960e739d JA |
4754 | /* |
4755 | * this sets the active strip count to 1 and the processed | |
4756 | * strip count to zero (upper 8 bits) | |
4757 | */ | |
e7836bd6 | 4758 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
4759 | } |
4760 | ||
4761 | return bi; | |
4762 | } | |
4763 | ||
f679623f RBJ |
4764 | /* |
4765 | * The "raid5_align_endio" should check if the read succeeded and if it | |
4766 | * did, call bio_endio on the original bio (having bio_put the new bio | |
4767 | * first). | |
4768 | * If the read failed.. | |
4769 | */ | |
4246a0b6 | 4770 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
4771 | { |
4772 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 4773 | struct mddev *mddev; |
d1688a6d | 4774 | struct r5conf *conf; |
3cb03002 | 4775 | struct md_rdev *rdev; |
9b81c842 | 4776 | int error = bi->bi_error; |
46031f9a | 4777 | |
f679623f | 4778 | bio_put(bi); |
46031f9a | 4779 | |
46031f9a RBJ |
4780 | rdev = (void*)raid_bi->bi_next; |
4781 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
4782 | mddev = rdev->mddev; |
4783 | conf = mddev->private; | |
46031f9a RBJ |
4784 | |
4785 | rdev_dec_pending(rdev, conf->mddev); | |
4786 | ||
9b81c842 | 4787 | if (!error) { |
0a82a8d1 LT |
4788 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
4789 | raid_bi, 0); | |
4246a0b6 | 4790 | bio_endio(raid_bi); |
46031f9a | 4791 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 4792 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 4793 | return; |
46031f9a RBJ |
4794 | } |
4795 | ||
45b4233c | 4796 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
4797 | |
4798 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
4799 | } |
4800 | ||
7ef6b12a | 4801 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 4802 | { |
d1688a6d | 4803 | struct r5conf *conf = mddev->private; |
8553fe7e | 4804 | int dd_idx; |
f679623f | 4805 | struct bio* align_bi; |
3cb03002 | 4806 | struct md_rdev *rdev; |
671488cc | 4807 | sector_t end_sector; |
f679623f RBJ |
4808 | |
4809 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 4810 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
4811 | return 0; |
4812 | } | |
4813 | /* | |
a167f663 | 4814 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 4815 | */ |
a167f663 | 4816 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
4817 | if (!align_bi) |
4818 | return 0; | |
4819 | /* | |
4820 | * set bi_end_io to a new function, and set bi_private to the | |
4821 | * original bio. | |
4822 | */ | |
4823 | align_bi->bi_end_io = raid5_align_endio; | |
4824 | align_bi->bi_private = raid_bio; | |
4825 | /* | |
4826 | * compute position | |
4827 | */ | |
4f024f37 KO |
4828 | align_bi->bi_iter.bi_sector = |
4829 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
4830 | 0, &dd_idx, NULL); | |
f679623f | 4831 | |
f73a1c7d | 4832 | end_sector = bio_end_sector(align_bi); |
f679623f | 4833 | rcu_read_lock(); |
671488cc N |
4834 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
4835 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
4836 | rdev->recovery_offset < end_sector) { | |
4837 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
4838 | if (rdev && | |
4839 | (test_bit(Faulty, &rdev->flags) || | |
4840 | !(test_bit(In_sync, &rdev->flags) || | |
4841 | rdev->recovery_offset >= end_sector))) | |
4842 | rdev = NULL; | |
4843 | } | |
4844 | if (rdev) { | |
31c176ec N |
4845 | sector_t first_bad; |
4846 | int bad_sectors; | |
4847 | ||
f679623f RBJ |
4848 | atomic_inc(&rdev->nr_pending); |
4849 | rcu_read_unlock(); | |
46031f9a RBJ |
4850 | raid_bio->bi_next = (void*)rdev; |
4851 | align_bi->bi_bdev = rdev->bdev; | |
b7c44ed9 | 4852 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 4853 | |
7140aafc | 4854 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 4855 | bio_sectors(align_bi), |
31c176ec | 4856 | &first_bad, &bad_sectors)) { |
387bb173 NB |
4857 | bio_put(align_bi); |
4858 | rdev_dec_pending(rdev, mddev); | |
4859 | return 0; | |
4860 | } | |
4861 | ||
6c0544e2 | 4862 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 4863 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 4864 | |
46031f9a | 4865 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 4866 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 4867 | conf->quiesce == 0, |
eed8c02e | 4868 | conf->device_lock); |
46031f9a RBJ |
4869 | atomic_inc(&conf->active_aligned_reads); |
4870 | spin_unlock_irq(&conf->device_lock); | |
4871 | ||
e3620a3a JB |
4872 | if (mddev->gendisk) |
4873 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
4874 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 4875 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
4876 | generic_make_request(align_bi); |
4877 | return 1; | |
4878 | } else { | |
4879 | rcu_read_unlock(); | |
46031f9a | 4880 | bio_put(align_bi); |
f679623f RBJ |
4881 | return 0; |
4882 | } | |
4883 | } | |
4884 | ||
7ef6b12a ML |
4885 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
4886 | { | |
4887 | struct bio *split; | |
4888 | ||
4889 | do { | |
4890 | sector_t sector = raid_bio->bi_iter.bi_sector; | |
4891 | unsigned chunk_sects = mddev->chunk_sectors; | |
4892 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
4893 | ||
4894 | if (sectors < bio_sectors(raid_bio)) { | |
4895 | split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set); | |
4896 | bio_chain(split, raid_bio); | |
4897 | } else | |
4898 | split = raid_bio; | |
4899 | ||
4900 | if (!raid5_read_one_chunk(mddev, split)) { | |
4901 | if (split != raid_bio) | |
4902 | generic_make_request(raid_bio); | |
4903 | return split; | |
4904 | } | |
4905 | } while (split != raid_bio); | |
4906 | ||
4907 | return NULL; | |
4908 | } | |
4909 | ||
8b3e6cdc DW |
4910 | /* __get_priority_stripe - get the next stripe to process |
4911 | * | |
4912 | * Full stripe writes are allowed to pass preread active stripes up until | |
4913 | * the bypass_threshold is exceeded. In general the bypass_count | |
4914 | * increments when the handle_list is handled before the hold_list; however, it | |
4915 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
4916 | * stripe with in flight i/o. The bypass_count will be reset when the | |
4917 | * head of the hold_list has changed, i.e. the head was promoted to the | |
4918 | * handle_list. | |
4919 | */ | |
851c30c9 | 4920 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 4921 | { |
851c30c9 SL |
4922 | struct stripe_head *sh = NULL, *tmp; |
4923 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 4924 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
4925 | |
4926 | if (conf->worker_cnt_per_group == 0) { | |
4927 | handle_list = &conf->handle_list; | |
4928 | } else if (group != ANY_GROUP) { | |
4929 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 4930 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
4931 | } else { |
4932 | int i; | |
4933 | for (i = 0; i < conf->group_cnt; i++) { | |
4934 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 4935 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
4936 | if (!list_empty(handle_list)) |
4937 | break; | |
4938 | } | |
4939 | } | |
8b3e6cdc DW |
4940 | |
4941 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
4942 | __func__, | |
851c30c9 | 4943 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
4944 | list_empty(&conf->hold_list) ? "empty" : "busy", |
4945 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
4946 | ||
851c30c9 SL |
4947 | if (!list_empty(handle_list)) { |
4948 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
4949 | |
4950 | if (list_empty(&conf->hold_list)) | |
4951 | conf->bypass_count = 0; | |
4952 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
4953 | if (conf->hold_list.next == conf->last_hold) | |
4954 | conf->bypass_count++; | |
4955 | else { | |
4956 | conf->last_hold = conf->hold_list.next; | |
4957 | conf->bypass_count -= conf->bypass_threshold; | |
4958 | if (conf->bypass_count < 0) | |
4959 | conf->bypass_count = 0; | |
4960 | } | |
4961 | } | |
4962 | } else if (!list_empty(&conf->hold_list) && | |
4963 | ((conf->bypass_threshold && | |
4964 | conf->bypass_count > conf->bypass_threshold) || | |
4965 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
4966 | |
4967 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
4968 | if (conf->worker_cnt_per_group == 0 || | |
4969 | group == ANY_GROUP || | |
4970 | !cpu_online(tmp->cpu) || | |
4971 | cpu_to_group(tmp->cpu) == group) { | |
4972 | sh = tmp; | |
4973 | break; | |
4974 | } | |
4975 | } | |
4976 | ||
4977 | if (sh) { | |
4978 | conf->bypass_count -= conf->bypass_threshold; | |
4979 | if (conf->bypass_count < 0) | |
4980 | conf->bypass_count = 0; | |
4981 | } | |
bfc90cb0 | 4982 | wg = NULL; |
851c30c9 SL |
4983 | } |
4984 | ||
4985 | if (!sh) | |
8b3e6cdc DW |
4986 | return NULL; |
4987 | ||
bfc90cb0 SL |
4988 | if (wg) { |
4989 | wg->stripes_cnt--; | |
4990 | sh->group = NULL; | |
4991 | } | |
8b3e6cdc | 4992 | list_del_init(&sh->lru); |
c7a6d35e | 4993 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
4994 | return sh; |
4995 | } | |
f679623f | 4996 | |
8811b596 SL |
4997 | struct raid5_plug_cb { |
4998 | struct blk_plug_cb cb; | |
4999 | struct list_head list; | |
566c09c5 | 5000 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5001 | }; |
5002 | ||
5003 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5004 | { | |
5005 | struct raid5_plug_cb *cb = container_of( | |
5006 | blk_cb, struct raid5_plug_cb, cb); | |
5007 | struct stripe_head *sh; | |
5008 | struct mddev *mddev = cb->cb.data; | |
5009 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5010 | int cnt = 0; |
566c09c5 | 5011 | int hash; |
8811b596 SL |
5012 | |
5013 | if (cb->list.next && !list_empty(&cb->list)) { | |
5014 | spin_lock_irq(&conf->device_lock); | |
5015 | while (!list_empty(&cb->list)) { | |
5016 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5017 | list_del_init(&sh->lru); | |
5018 | /* | |
5019 | * avoid race release_stripe_plug() sees | |
5020 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5021 | * is still in our list | |
5022 | */ | |
4e857c58 | 5023 | smp_mb__before_atomic(); |
8811b596 | 5024 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5025 | /* |
5026 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5027 | * case, the count is always > 1 here | |
5028 | */ | |
566c09c5 SL |
5029 | hash = sh->hash_lock_index; |
5030 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5031 | cnt++; |
8811b596 SL |
5032 | } |
5033 | spin_unlock_irq(&conf->device_lock); | |
5034 | } | |
566c09c5 SL |
5035 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5036 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5037 | if (mddev->queue) |
5038 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5039 | kfree(cb); |
5040 | } | |
5041 | ||
5042 | static void release_stripe_plug(struct mddev *mddev, | |
5043 | struct stripe_head *sh) | |
5044 | { | |
5045 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5046 | raid5_unplug, mddev, | |
5047 | sizeof(struct raid5_plug_cb)); | |
5048 | struct raid5_plug_cb *cb; | |
5049 | ||
5050 | if (!blk_cb) { | |
6d036f7d | 5051 | raid5_release_stripe(sh); |
8811b596 SL |
5052 | return; |
5053 | } | |
5054 | ||
5055 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5056 | ||
566c09c5 SL |
5057 | if (cb->list.next == NULL) { |
5058 | int i; | |
8811b596 | 5059 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5060 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5061 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5062 | } | |
8811b596 SL |
5063 | |
5064 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5065 | list_add_tail(&sh->lru, &cb->list); | |
5066 | else | |
6d036f7d | 5067 | raid5_release_stripe(sh); |
8811b596 SL |
5068 | } |
5069 | ||
620125f2 SL |
5070 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5071 | { | |
5072 | struct r5conf *conf = mddev->private; | |
5073 | sector_t logical_sector, last_sector; | |
5074 | struct stripe_head *sh; | |
5075 | int remaining; | |
5076 | int stripe_sectors; | |
5077 | ||
5078 | if (mddev->reshape_position != MaxSector) | |
5079 | /* Skip discard while reshape is happening */ | |
5080 | return; | |
5081 | ||
4f024f37 KO |
5082 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5083 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5084 | |
5085 | bi->bi_next = NULL; | |
5086 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5087 | ||
5088 | stripe_sectors = conf->chunk_sectors * | |
5089 | (conf->raid_disks - conf->max_degraded); | |
5090 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5091 | stripe_sectors); | |
5092 | sector_div(last_sector, stripe_sectors); | |
5093 | ||
5094 | logical_sector *= conf->chunk_sectors; | |
5095 | last_sector *= conf->chunk_sectors; | |
5096 | ||
5097 | for (; logical_sector < last_sector; | |
5098 | logical_sector += STRIPE_SECTORS) { | |
5099 | DEFINE_WAIT(w); | |
5100 | int d; | |
5101 | again: | |
6d036f7d | 5102 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5103 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5104 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5105 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5106 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5107 | raid5_release_stripe(sh); |
f8dfcffd N |
5108 | schedule(); |
5109 | goto again; | |
5110 | } | |
5111 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5112 | spin_lock_irq(&sh->stripe_lock); |
5113 | for (d = 0; d < conf->raid_disks; d++) { | |
5114 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5115 | continue; | |
5116 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5117 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5118 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5119 | raid5_release_stripe(sh); |
620125f2 SL |
5120 | schedule(); |
5121 | goto again; | |
5122 | } | |
5123 | } | |
f8dfcffd | 5124 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5125 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5126 | sh->overwrite_disks = 0; |
620125f2 SL |
5127 | for (d = 0; d < conf->raid_disks; d++) { |
5128 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5129 | continue; | |
5130 | sh->dev[d].towrite = bi; | |
5131 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5132 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5133 | sh->overwrite_disks++; |
620125f2 SL |
5134 | } |
5135 | spin_unlock_irq(&sh->stripe_lock); | |
5136 | if (conf->mddev->bitmap) { | |
5137 | for (d = 0; | |
5138 | d < conf->raid_disks - conf->max_degraded; | |
5139 | d++) | |
5140 | bitmap_startwrite(mddev->bitmap, | |
5141 | sh->sector, | |
5142 | STRIPE_SECTORS, | |
5143 | 0); | |
5144 | sh->bm_seq = conf->seq_flush + 1; | |
5145 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5146 | } | |
5147 | ||
5148 | set_bit(STRIPE_HANDLE, &sh->state); | |
5149 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5150 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5151 | atomic_inc(&conf->preread_active_stripes); | |
5152 | release_stripe_plug(mddev, sh); | |
5153 | } | |
5154 | ||
5155 | remaining = raid5_dec_bi_active_stripes(bi); | |
5156 | if (remaining == 0) { | |
5157 | md_write_end(mddev); | |
4246a0b6 | 5158 | bio_endio(bi); |
620125f2 SL |
5159 | } |
5160 | } | |
5161 | ||
849674e4 | 5162 | static void raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5163 | { |
d1688a6d | 5164 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5165 | int dd_idx; |
1da177e4 LT |
5166 | sector_t new_sector; |
5167 | sector_t logical_sector, last_sector; | |
5168 | struct stripe_head *sh; | |
a362357b | 5169 | const int rw = bio_data_dir(bi); |
49077326 | 5170 | int remaining; |
27c0f68f SL |
5171 | DEFINE_WAIT(w); |
5172 | bool do_prepare; | |
1da177e4 | 5173 | |
1eff9d32 | 5174 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
828cbe98 SL |
5175 | int ret = r5l_handle_flush_request(conf->log, bi); |
5176 | ||
5177 | if (ret == 0) | |
5178 | return; | |
5179 | if (ret == -ENODEV) { | |
5180 | md_flush_request(mddev, bi); | |
5181 | return; | |
5182 | } | |
5183 | /* ret == -EAGAIN, fallback */ | |
e5dcdd80 N |
5184 | } |
5185 | ||
3d310eb7 | 5186 | md_write_start(mddev, bi); |
06d91a5f | 5187 | |
9ffc8f7c EM |
5188 | /* |
5189 | * If array is degraded, better not do chunk aligned read because | |
5190 | * later we might have to read it again in order to reconstruct | |
5191 | * data on failed drives. | |
5192 | */ | |
5193 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5194 | mddev->reshape_position == MaxSector) { |
5195 | bi = chunk_aligned_read(mddev, bi); | |
5196 | if (!bi) | |
5197 | return; | |
5198 | } | |
52488615 | 5199 | |
796a5cf0 | 5200 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 SL |
5201 | make_discard_request(mddev, bi); |
5202 | return; | |
5203 | } | |
5204 | ||
4f024f37 | 5205 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5206 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5207 | bi->bi_next = NULL; |
5208 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5209 | |
27c0f68f | 5210 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5211 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5212 | int previous; |
c46501b2 | 5213 | int seq; |
b578d55f | 5214 | |
27c0f68f | 5215 | do_prepare = false; |
7ecaa1e6 | 5216 | retry: |
c46501b2 | 5217 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5218 | previous = 0; |
27c0f68f SL |
5219 | if (do_prepare) |
5220 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5221 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5222 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5223 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5224 | * 64bit on a 32bit platform, and so it might be |
5225 | * possible to see a half-updated value | |
aeb878b0 | 5226 | * Of course reshape_progress could change after |
df8e7f76 N |
5227 | * the lock is dropped, so once we get a reference |
5228 | * to the stripe that we think it is, we will have | |
5229 | * to check again. | |
5230 | */ | |
7ecaa1e6 | 5231 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5232 | if (mddev->reshape_backwards |
fef9c61f N |
5233 | ? logical_sector < conf->reshape_progress |
5234 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5235 | previous = 1; |
5236 | } else { | |
2c810cdd | 5237 | if (mddev->reshape_backwards |
fef9c61f N |
5238 | ? logical_sector < conf->reshape_safe |
5239 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5240 | spin_unlock_irq(&conf->device_lock); |
5241 | schedule(); | |
27c0f68f | 5242 | do_prepare = true; |
b578d55f N |
5243 | goto retry; |
5244 | } | |
5245 | } | |
7ecaa1e6 N |
5246 | spin_unlock_irq(&conf->device_lock); |
5247 | } | |
16a53ecc | 5248 | |
112bf897 N |
5249 | new_sector = raid5_compute_sector(conf, logical_sector, |
5250 | previous, | |
911d4ee8 | 5251 | &dd_idx, NULL); |
849674e4 | 5252 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5253 | (unsigned long long)new_sector, |
1da177e4 LT |
5254 | (unsigned long long)logical_sector); |
5255 | ||
6d036f7d | 5256 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5257 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5258 | if (sh) { |
b0f9ec04 | 5259 | if (unlikely(previous)) { |
7ecaa1e6 | 5260 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5261 | * stripe, so we must do the range check again. |
5262 | * Expansion could still move past after this | |
5263 | * test, but as we are holding a reference to | |
5264 | * 'sh', we know that if that happens, | |
5265 | * STRIPE_EXPANDING will get set and the expansion | |
5266 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5267 | */ |
5268 | int must_retry = 0; | |
5269 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5270 | if (mddev->reshape_backwards |
b0f9ec04 N |
5271 | ? logical_sector >= conf->reshape_progress |
5272 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5273 | /* mismatch, need to try again */ |
5274 | must_retry = 1; | |
5275 | spin_unlock_irq(&conf->device_lock); | |
5276 | if (must_retry) { | |
6d036f7d | 5277 | raid5_release_stripe(sh); |
7a3ab908 | 5278 | schedule(); |
27c0f68f | 5279 | do_prepare = true; |
7ecaa1e6 N |
5280 | goto retry; |
5281 | } | |
5282 | } | |
c46501b2 N |
5283 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5284 | /* Might have got the wrong stripe_head | |
5285 | * by accident | |
5286 | */ | |
6d036f7d | 5287 | raid5_release_stripe(sh); |
c46501b2 N |
5288 | goto retry; |
5289 | } | |
e62e58a5 | 5290 | |
ffd96e35 | 5291 | if (rw == WRITE && |
a5c308d4 | 5292 | logical_sector >= mddev->suspend_lo && |
e464eafd | 5293 | logical_sector < mddev->suspend_hi) { |
6d036f7d | 5294 | raid5_release_stripe(sh); |
e62e58a5 N |
5295 | /* As the suspend_* range is controlled by |
5296 | * userspace, we want an interruptible | |
5297 | * wait. | |
5298 | */ | |
5299 | flush_signals(current); | |
5300 | prepare_to_wait(&conf->wait_for_overlap, | |
5301 | &w, TASK_INTERRUPTIBLE); | |
5302 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5303 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5304 | schedule(); |
27c0f68f SL |
5305 | do_prepare = true; |
5306 | } | |
e464eafd N |
5307 | goto retry; |
5308 | } | |
7ecaa1e6 N |
5309 | |
5310 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5311 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5312 | /* Stripe is busy expanding or |
5313 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5314 | * and wait a while |
5315 | */ | |
482c0834 | 5316 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5317 | raid5_release_stripe(sh); |
1da177e4 | 5318 | schedule(); |
27c0f68f | 5319 | do_prepare = true; |
1da177e4 LT |
5320 | goto retry; |
5321 | } | |
6ed3003c N |
5322 | set_bit(STRIPE_HANDLE, &sh->state); |
5323 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5324 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5325 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5326 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5327 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5328 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5329 | } else { |
5330 | /* cannot get stripe for read-ahead, just give-up */ | |
4246a0b6 | 5331 | bi->bi_error = -EIO; |
1da177e4 LT |
5332 | break; |
5333 | } | |
1da177e4 | 5334 | } |
27c0f68f | 5335 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5336 | |
e7836bd6 | 5337 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5338 | if (remaining == 0) { |
1da177e4 | 5339 | |
16a53ecc | 5340 | if ( rw == WRITE ) |
1da177e4 | 5341 | md_write_end(mddev); |
6712ecf8 | 5342 | |
0a82a8d1 LT |
5343 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5344 | bi, 0); | |
4246a0b6 | 5345 | bio_endio(bi); |
1da177e4 | 5346 | } |
1da177e4 LT |
5347 | } |
5348 | ||
fd01b88c | 5349 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5350 | |
fd01b88c | 5351 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5352 | { |
52c03291 N |
5353 | /* reshaping is quite different to recovery/resync so it is |
5354 | * handled quite separately ... here. | |
5355 | * | |
5356 | * On each call to sync_request, we gather one chunk worth of | |
5357 | * destination stripes and flag them as expanding. | |
5358 | * Then we find all the source stripes and request reads. | |
5359 | * As the reads complete, handle_stripe will copy the data | |
5360 | * into the destination stripe and release that stripe. | |
5361 | */ | |
d1688a6d | 5362 | struct r5conf *conf = mddev->private; |
1da177e4 | 5363 | struct stripe_head *sh; |
ccfcc3c1 | 5364 | sector_t first_sector, last_sector; |
f416885e N |
5365 | int raid_disks = conf->previous_raid_disks; |
5366 | int data_disks = raid_disks - conf->max_degraded; | |
5367 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5368 | int i; |
5369 | int dd_idx; | |
c8f517c4 | 5370 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5371 | sector_t stripe_addr; |
7a661381 | 5372 | int reshape_sectors; |
ab69ae12 | 5373 | struct list_head stripes; |
92140480 | 5374 | sector_t retn; |
52c03291 | 5375 | |
fef9c61f N |
5376 | if (sector_nr == 0) { |
5377 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5378 | if (mddev->reshape_backwards && |
fef9c61f N |
5379 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5380 | sector_nr = raid5_size(mddev, 0, 0) | |
5381 | - conf->reshape_progress; | |
6cbd8148 N |
5382 | } else if (mddev->reshape_backwards && |
5383 | conf->reshape_progress == MaxSector) { | |
5384 | /* shouldn't happen, but just in case, finish up.*/ | |
5385 | sector_nr = MaxSector; | |
2c810cdd | 5386 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5387 | conf->reshape_progress > 0) |
5388 | sector_nr = conf->reshape_progress; | |
f416885e | 5389 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5390 | if (sector_nr) { |
8dee7211 N |
5391 | mddev->curr_resync_completed = sector_nr; |
5392 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5393 | *skipped = 1; |
92140480 N |
5394 | retn = sector_nr; |
5395 | goto finish; | |
fef9c61f | 5396 | } |
52c03291 N |
5397 | } |
5398 | ||
7a661381 N |
5399 | /* We need to process a full chunk at a time. |
5400 | * If old and new chunk sizes differ, we need to process the | |
5401 | * largest of these | |
5402 | */ | |
3cb5edf4 N |
5403 | |
5404 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5405 | |
b5254dd5 N |
5406 | /* We update the metadata at least every 10 seconds, or when |
5407 | * the data about to be copied would over-write the source of | |
5408 | * the data at the front of the range. i.e. one new_stripe | |
5409 | * along from reshape_progress new_maps to after where | |
5410 | * reshape_safe old_maps to | |
52c03291 | 5411 | */ |
fef9c61f | 5412 | writepos = conf->reshape_progress; |
f416885e | 5413 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5414 | readpos = conf->reshape_progress; |
5415 | sector_div(readpos, data_disks); | |
fef9c61f | 5416 | safepos = conf->reshape_safe; |
f416885e | 5417 | sector_div(safepos, data_disks); |
2c810cdd | 5418 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5419 | BUG_ON(writepos < reshape_sectors); |
5420 | writepos -= reshape_sectors; | |
c8f517c4 | 5421 | readpos += reshape_sectors; |
7a661381 | 5422 | safepos += reshape_sectors; |
fef9c61f | 5423 | } else { |
7a661381 | 5424 | writepos += reshape_sectors; |
c74c0d76 N |
5425 | /* readpos and safepos are worst-case calculations. |
5426 | * A negative number is overly pessimistic, and causes | |
5427 | * obvious problems for unsigned storage. So clip to 0. | |
5428 | */ | |
ed37d83e N |
5429 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5430 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5431 | } |
52c03291 | 5432 | |
b5254dd5 N |
5433 | /* Having calculated the 'writepos' possibly use it |
5434 | * to set 'stripe_addr' which is where we will write to. | |
5435 | */ | |
5436 | if (mddev->reshape_backwards) { | |
5437 | BUG_ON(conf->reshape_progress == 0); | |
5438 | stripe_addr = writepos; | |
5439 | BUG_ON((mddev->dev_sectors & | |
5440 | ~((sector_t)reshape_sectors - 1)) | |
5441 | - reshape_sectors - stripe_addr | |
5442 | != sector_nr); | |
5443 | } else { | |
5444 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5445 | stripe_addr = sector_nr; | |
5446 | } | |
5447 | ||
c8f517c4 N |
5448 | /* 'writepos' is the most advanced device address we might write. |
5449 | * 'readpos' is the least advanced device address we might read. | |
5450 | * 'safepos' is the least address recorded in the metadata as having | |
5451 | * been reshaped. | |
b5254dd5 N |
5452 | * If there is a min_offset_diff, these are adjusted either by |
5453 | * increasing the safepos/readpos if diff is negative, or | |
5454 | * increasing writepos if diff is positive. | |
5455 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5456 | * ensure safety in the face of a crash - that must be done by userspace |
5457 | * making a backup of the data. So in that case there is no particular | |
5458 | * rush to update metadata. | |
5459 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5460 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5461 | * we can be safe in the event of a crash. | |
5462 | * So we insist on updating metadata if safepos is behind writepos and | |
5463 | * readpos is beyond writepos. | |
5464 | * In any case, update the metadata every 10 seconds. | |
5465 | * Maybe that number should be configurable, but I'm not sure it is | |
5466 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5467 | */ | |
b5254dd5 N |
5468 | if (conf->min_offset_diff < 0) { |
5469 | safepos += -conf->min_offset_diff; | |
5470 | readpos += -conf->min_offset_diff; | |
5471 | } else | |
5472 | writepos += conf->min_offset_diff; | |
5473 | ||
2c810cdd | 5474 | if ((mddev->reshape_backwards |
c8f517c4 N |
5475 | ? (safepos > writepos && readpos < writepos) |
5476 | : (safepos < writepos && readpos > writepos)) || | |
5477 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5478 | /* Cannot proceed until we've updated the superblock... */ |
5479 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5480 | atomic_read(&conf->reshape_stripes)==0 |
5481 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5482 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5483 | return 0; | |
fef9c61f | 5484 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5485 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5486 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 5487 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 5488 | md_wakeup_thread(mddev->thread); |
850b2b42 | 5489 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
c91abf5a N |
5490 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5491 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5492 | return 0; | |
52c03291 | 5493 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5494 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5495 | spin_unlock_irq(&conf->device_lock); |
5496 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5497 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5498 | } |
5499 | ||
ab69ae12 | 5500 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5501 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5502 | int j; |
a9f326eb | 5503 | int skipped_disk = 0; |
6d036f7d | 5504 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5505 | set_bit(STRIPE_EXPANDING, &sh->state); |
5506 | atomic_inc(&conf->reshape_stripes); | |
5507 | /* If any of this stripe is beyond the end of the old | |
5508 | * array, then we need to zero those blocks | |
5509 | */ | |
5510 | for (j=sh->disks; j--;) { | |
5511 | sector_t s; | |
5512 | if (j == sh->pd_idx) | |
5513 | continue; | |
f416885e | 5514 | if (conf->level == 6 && |
d0dabf7e | 5515 | j == sh->qd_idx) |
f416885e | 5516 | continue; |
6d036f7d | 5517 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5518 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5519 | skipped_disk = 1; |
52c03291 N |
5520 | continue; |
5521 | } | |
5522 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5523 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5524 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5525 | } | |
a9f326eb | 5526 | if (!skipped_disk) { |
52c03291 N |
5527 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5528 | set_bit(STRIPE_HANDLE, &sh->state); | |
5529 | } | |
ab69ae12 | 5530 | list_add(&sh->lru, &stripes); |
52c03291 N |
5531 | } |
5532 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5533 | if (mddev->reshape_backwards) |
7a661381 | 5534 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5535 | else |
7a661381 | 5536 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5537 | spin_unlock_irq(&conf->device_lock); |
5538 | /* Ok, those stripe are ready. We can start scheduling | |
5539 | * reads on the source stripes. | |
5540 | * The source stripes are determined by mapping the first and last | |
5541 | * block on the destination stripes. | |
5542 | */ | |
52c03291 | 5543 | first_sector = |
ec32a2bd | 5544 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5545 | 1, &dd_idx, NULL); |
52c03291 | 5546 | last_sector = |
0e6e0271 | 5547 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5548 | * new_data_disks - 1), |
911d4ee8 | 5549 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5550 | if (last_sector >= mddev->dev_sectors) |
5551 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5552 | while (first_sector <= last_sector) { |
6d036f7d | 5553 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5554 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5555 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5556 | raid5_release_stripe(sh); |
52c03291 N |
5557 | first_sector += STRIPE_SECTORS; |
5558 | } | |
ab69ae12 N |
5559 | /* Now that the sources are clearly marked, we can release |
5560 | * the destination stripes | |
5561 | */ | |
5562 | while (!list_empty(&stripes)) { | |
5563 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5564 | list_del_init(&sh->lru); | |
6d036f7d | 5565 | raid5_release_stripe(sh); |
ab69ae12 | 5566 | } |
c6207277 N |
5567 | /* If this takes us to the resync_max point where we have to pause, |
5568 | * then we need to write out the superblock. | |
5569 | */ | |
7a661381 | 5570 | sector_nr += reshape_sectors; |
92140480 N |
5571 | retn = reshape_sectors; |
5572 | finish: | |
c5e19d90 N |
5573 | if (mddev->curr_resync_completed > mddev->resync_max || |
5574 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5575 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5576 | /* Cannot proceed until we've updated the superblock... */ |
5577 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5578 | atomic_read(&conf->reshape_stripes) == 0 |
5579 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5580 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5581 | goto ret; | |
fef9c61f | 5582 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5583 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5584 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
5585 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
5586 | md_wakeup_thread(mddev->thread); | |
5587 | wait_event(mddev->sb_wait, | |
5588 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
c91abf5a N |
5589 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5590 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5591 | goto ret; | |
c6207277 | 5592 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5593 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5594 | spin_unlock_irq(&conf->device_lock); |
5595 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5596 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5597 | } |
c91abf5a | 5598 | ret: |
92140480 | 5599 | return retn; |
52c03291 N |
5600 | } |
5601 | ||
849674e4 SL |
5602 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
5603 | int *skipped) | |
52c03291 | 5604 | { |
d1688a6d | 5605 | struct r5conf *conf = mddev->private; |
52c03291 | 5606 | struct stripe_head *sh; |
58c0fed4 | 5607 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5608 | sector_t sync_blocks; |
16a53ecc N |
5609 | int still_degraded = 0; |
5610 | int i; | |
1da177e4 | 5611 | |
72626685 | 5612 | if (sector_nr >= max_sector) { |
1da177e4 | 5613 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5614 | |
29269553 N |
5615 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5616 | end_reshape(conf); | |
5617 | return 0; | |
5618 | } | |
72626685 N |
5619 | |
5620 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5621 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5622 | &sync_blocks, 1); | |
16a53ecc | 5623 | else /* completed sync */ |
72626685 N |
5624 | conf->fullsync = 0; |
5625 | bitmap_close_sync(mddev->bitmap); | |
5626 | ||
1da177e4 LT |
5627 | return 0; |
5628 | } | |
ccfcc3c1 | 5629 | |
64bd660b N |
5630 | /* Allow raid5_quiesce to complete */ |
5631 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5632 | ||
52c03291 N |
5633 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5634 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5635 | |
c6207277 N |
5636 | /* No need to check resync_max as we never do more than one |
5637 | * stripe, and as resync_max will always be on a chunk boundary, | |
5638 | * if the check in md_do_sync didn't fire, there is no chance | |
5639 | * of overstepping resync_max here | |
5640 | */ | |
5641 | ||
16a53ecc | 5642 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5643 | * to resync, then assert that we are finished, because there is |
5644 | * nothing we can do. | |
5645 | */ | |
3285edf1 | 5646 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5647 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5648 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5649 | *skipped = 1; |
1da177e4 LT |
5650 | return rv; |
5651 | } | |
6f608040 | 5652 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5653 | !conf->fullsync && | |
5654 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5655 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5656 | /* we can skip this block, and probably more */ |
5657 | sync_blocks /= STRIPE_SECTORS; | |
5658 | *skipped = 1; | |
5659 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5660 | } | |
1da177e4 | 5661 | |
c40f341f | 5662 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 5663 | |
6d036f7d | 5664 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5665 | if (sh == NULL) { |
6d036f7d | 5666 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5667 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5668 | * is trying to get access |
1da177e4 | 5669 | */ |
66c006a5 | 5670 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5671 | } |
16a53ecc | 5672 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5673 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5674 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5675 | */ |
16d9cfab EM |
5676 | rcu_read_lock(); |
5677 | for (i = 0; i < conf->raid_disks; i++) { | |
5678 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5679 | ||
5680 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5681 | still_degraded = 1; |
16d9cfab EM |
5682 | } |
5683 | rcu_read_unlock(); | |
16a53ecc N |
5684 | |
5685 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5686 | ||
83206d66 | 5687 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5688 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5689 | |
6d036f7d | 5690 | raid5_release_stripe(sh); |
1da177e4 LT |
5691 | |
5692 | return STRIPE_SECTORS; | |
5693 | } | |
5694 | ||
d1688a6d | 5695 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5696 | { |
5697 | /* We may not be able to submit a whole bio at once as there | |
5698 | * may not be enough stripe_heads available. | |
5699 | * We cannot pre-allocate enough stripe_heads as we may need | |
5700 | * more than exist in the cache (if we allow ever large chunks). | |
5701 | * So we do one stripe head at a time and record in | |
5702 | * ->bi_hw_segments how many have been done. | |
5703 | * | |
5704 | * We *know* that this entire raid_bio is in one chunk, so | |
5705 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5706 | */ | |
5707 | struct stripe_head *sh; | |
911d4ee8 | 5708 | int dd_idx; |
46031f9a RBJ |
5709 | sector_t sector, logical_sector, last_sector; |
5710 | int scnt = 0; | |
5711 | int remaining; | |
5712 | int handled = 0; | |
5713 | ||
4f024f37 KO |
5714 | logical_sector = raid_bio->bi_iter.bi_sector & |
5715 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5716 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5717 | 0, &dd_idx, NULL); |
f73a1c7d | 5718 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5719 | |
5720 | for (; logical_sector < last_sector; | |
387bb173 NB |
5721 | logical_sector += STRIPE_SECTORS, |
5722 | sector += STRIPE_SECTORS, | |
5723 | scnt++) { | |
46031f9a | 5724 | |
e7836bd6 | 5725 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5726 | /* already done this stripe */ |
5727 | continue; | |
5728 | ||
6d036f7d | 5729 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5730 | |
5731 | if (!sh) { | |
5732 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5733 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5734 | conf->retry_read_aligned = raid_bio; |
5735 | return handled; | |
5736 | } | |
5737 | ||
da41ba65 | 5738 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 5739 | raid5_release_stripe(sh); |
e7836bd6 | 5740 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
5741 | conf->retry_read_aligned = raid_bio; |
5742 | return handled; | |
5743 | } | |
5744 | ||
3f9e7c14 | 5745 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 5746 | handle_stripe(sh); |
6d036f7d | 5747 | raid5_release_stripe(sh); |
46031f9a RBJ |
5748 | handled++; |
5749 | } | |
e7836bd6 | 5750 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
5751 | if (remaining == 0) { |
5752 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
5753 | raid_bio, 0); | |
4246a0b6 | 5754 | bio_endio(raid_bio); |
0a82a8d1 | 5755 | } |
46031f9a | 5756 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5757 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
5758 | return handled; |
5759 | } | |
5760 | ||
bfc90cb0 | 5761 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
5762 | struct r5worker *worker, |
5763 | struct list_head *temp_inactive_list) | |
46a06401 SL |
5764 | { |
5765 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
5766 | int i, batch_size = 0, hash; |
5767 | bool release_inactive = false; | |
46a06401 SL |
5768 | |
5769 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 5770 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
5771 | batch[batch_size++] = sh; |
5772 | ||
566c09c5 SL |
5773 | if (batch_size == 0) { |
5774 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5775 | if (!list_empty(temp_inactive_list + i)) | |
5776 | break; | |
a8c34f91 SL |
5777 | if (i == NR_STRIPE_HASH_LOCKS) { |
5778 | spin_unlock_irq(&conf->device_lock); | |
5779 | r5l_flush_stripe_to_raid(conf->log); | |
5780 | spin_lock_irq(&conf->device_lock); | |
566c09c5 | 5781 | return batch_size; |
a8c34f91 | 5782 | } |
566c09c5 SL |
5783 | release_inactive = true; |
5784 | } | |
46a06401 SL |
5785 | spin_unlock_irq(&conf->device_lock); |
5786 | ||
566c09c5 SL |
5787 | release_inactive_stripe_list(conf, temp_inactive_list, |
5788 | NR_STRIPE_HASH_LOCKS); | |
5789 | ||
a8c34f91 | 5790 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
5791 | if (release_inactive) { |
5792 | spin_lock_irq(&conf->device_lock); | |
5793 | return 0; | |
5794 | } | |
5795 | ||
46a06401 SL |
5796 | for (i = 0; i < batch_size; i++) |
5797 | handle_stripe(batch[i]); | |
f6bed0ef | 5798 | r5l_write_stripe_run(conf->log); |
46a06401 SL |
5799 | |
5800 | cond_resched(); | |
5801 | ||
5802 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
5803 | for (i = 0; i < batch_size; i++) { |
5804 | hash = batch[i]->hash_lock_index; | |
5805 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
5806 | } | |
46a06401 SL |
5807 | return batch_size; |
5808 | } | |
46031f9a | 5809 | |
851c30c9 SL |
5810 | static void raid5_do_work(struct work_struct *work) |
5811 | { | |
5812 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
5813 | struct r5worker_group *group = worker->group; | |
5814 | struct r5conf *conf = group->conf; | |
5815 | int group_id = group - conf->worker_groups; | |
5816 | int handled; | |
5817 | struct blk_plug plug; | |
5818 | ||
5819 | pr_debug("+++ raid5worker active\n"); | |
5820 | ||
5821 | blk_start_plug(&plug); | |
5822 | handled = 0; | |
5823 | spin_lock_irq(&conf->device_lock); | |
5824 | while (1) { | |
5825 | int batch_size, released; | |
5826 | ||
566c09c5 | 5827 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 5828 | |
566c09c5 SL |
5829 | batch_size = handle_active_stripes(conf, group_id, worker, |
5830 | worker->temp_inactive_list); | |
bfc90cb0 | 5831 | worker->working = false; |
851c30c9 SL |
5832 | if (!batch_size && !released) |
5833 | break; | |
5834 | handled += batch_size; | |
5835 | } | |
5836 | pr_debug("%d stripes handled\n", handled); | |
5837 | ||
5838 | spin_unlock_irq(&conf->device_lock); | |
5839 | blk_finish_plug(&plug); | |
5840 | ||
5841 | pr_debug("--- raid5worker inactive\n"); | |
5842 | } | |
5843 | ||
1da177e4 LT |
5844 | /* |
5845 | * This is our raid5 kernel thread. | |
5846 | * | |
5847 | * We scan the hash table for stripes which can be handled now. | |
5848 | * During the scan, completed stripes are saved for us by the interrupt | |
5849 | * handler, so that they will not have to wait for our next wakeup. | |
5850 | */ | |
4ed8731d | 5851 | static void raid5d(struct md_thread *thread) |
1da177e4 | 5852 | { |
4ed8731d | 5853 | struct mddev *mddev = thread->mddev; |
d1688a6d | 5854 | struct r5conf *conf = mddev->private; |
1da177e4 | 5855 | int handled; |
e1dfa0a2 | 5856 | struct blk_plug plug; |
1da177e4 | 5857 | |
45b4233c | 5858 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
5859 | |
5860 | md_check_recovery(mddev); | |
1da177e4 | 5861 | |
c3cce6cd N |
5862 | if (!bio_list_empty(&conf->return_bi) && |
5863 | !test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
5864 | struct bio_list tmp = BIO_EMPTY_LIST; | |
5865 | spin_lock_irq(&conf->device_lock); | |
5866 | if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) { | |
5867 | bio_list_merge(&tmp, &conf->return_bi); | |
5868 | bio_list_init(&conf->return_bi); | |
5869 | } | |
5870 | spin_unlock_irq(&conf->device_lock); | |
5871 | return_io(&tmp); | |
5872 | } | |
5873 | ||
e1dfa0a2 | 5874 | blk_start_plug(&plug); |
1da177e4 LT |
5875 | handled = 0; |
5876 | spin_lock_irq(&conf->device_lock); | |
5877 | while (1) { | |
46031f9a | 5878 | struct bio *bio; |
773ca82f SL |
5879 | int batch_size, released; |
5880 | ||
566c09c5 | 5881 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
5882 | if (released) |
5883 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 5884 | |
0021b7bc | 5885 | if ( |
7c13edc8 N |
5886 | !list_empty(&conf->bitmap_list)) { |
5887 | /* Now is a good time to flush some bitmap updates */ | |
5888 | conf->seq_flush++; | |
700e432d | 5889 | spin_unlock_irq(&conf->device_lock); |
72626685 | 5890 | bitmap_unplug(mddev->bitmap); |
700e432d | 5891 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 5892 | conf->seq_write = conf->seq_flush; |
566c09c5 | 5893 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 5894 | } |
0021b7bc | 5895 | raid5_activate_delayed(conf); |
72626685 | 5896 | |
46031f9a RBJ |
5897 | while ((bio = remove_bio_from_retry(conf))) { |
5898 | int ok; | |
5899 | spin_unlock_irq(&conf->device_lock); | |
5900 | ok = retry_aligned_read(conf, bio); | |
5901 | spin_lock_irq(&conf->device_lock); | |
5902 | if (!ok) | |
5903 | break; | |
5904 | handled++; | |
5905 | } | |
5906 | ||
566c09c5 SL |
5907 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
5908 | conf->temp_inactive_list); | |
773ca82f | 5909 | if (!batch_size && !released) |
1da177e4 | 5910 | break; |
46a06401 | 5911 | handled += batch_size; |
1da177e4 | 5912 | |
46a06401 SL |
5913 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
5914 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 5915 | md_check_recovery(mddev); |
46a06401 SL |
5916 | spin_lock_irq(&conf->device_lock); |
5917 | } | |
1da177e4 | 5918 | } |
45b4233c | 5919 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
5920 | |
5921 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
5922 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
5923 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
5924 | grow_one_stripe(conf, __GFP_NOWARN); |
5925 | /* Set flag even if allocation failed. This helps | |
5926 | * slow down allocation requests when mem is short | |
5927 | */ | |
5928 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 5929 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 5930 | } |
1da177e4 | 5931 | |
0576b1c6 SL |
5932 | r5l_flush_stripe_to_raid(conf->log); |
5933 | ||
c9f21aaf | 5934 | async_tx_issue_pending_all(); |
e1dfa0a2 | 5935 | blk_finish_plug(&plug); |
1da177e4 | 5936 | |
45b4233c | 5937 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
5938 | } |
5939 | ||
3f294f4f | 5940 | static ssize_t |
fd01b88c | 5941 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 5942 | { |
7b1485ba N |
5943 | struct r5conf *conf; |
5944 | int ret = 0; | |
5945 | spin_lock(&mddev->lock); | |
5946 | conf = mddev->private; | |
96de1e66 | 5947 | if (conf) |
edbe83ab | 5948 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
5949 | spin_unlock(&mddev->lock); |
5950 | return ret; | |
3f294f4f N |
5951 | } |
5952 | ||
c41d4ac4 | 5953 | int |
fd01b88c | 5954 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 5955 | { |
d1688a6d | 5956 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
5957 | int err; |
5958 | ||
c41d4ac4 | 5959 | if (size <= 16 || size > 32768) |
3f294f4f | 5960 | return -EINVAL; |
486f0644 | 5961 | |
edbe83ab | 5962 | conf->min_nr_stripes = size; |
2d5b569b | 5963 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5964 | while (size < conf->max_nr_stripes && |
5965 | drop_one_stripe(conf)) | |
5966 | ; | |
2d5b569b | 5967 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5968 | |
edbe83ab | 5969 | |
b5470dc5 DW |
5970 | err = md_allow_write(mddev); |
5971 | if (err) | |
5972 | return err; | |
486f0644 | 5973 | |
2d5b569b | 5974 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5975 | while (size > conf->max_nr_stripes) |
5976 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
5977 | break; | |
2d5b569b | 5978 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5979 | |
c41d4ac4 N |
5980 | return 0; |
5981 | } | |
5982 | EXPORT_SYMBOL(raid5_set_cache_size); | |
5983 | ||
5984 | static ssize_t | |
fd01b88c | 5985 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 5986 | { |
6791875e | 5987 | struct r5conf *conf; |
c41d4ac4 N |
5988 | unsigned long new; |
5989 | int err; | |
5990 | ||
5991 | if (len >= PAGE_SIZE) | |
5992 | return -EINVAL; | |
b29bebd6 | 5993 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 5994 | return -EINVAL; |
6791875e | 5995 | err = mddev_lock(mddev); |
c41d4ac4 N |
5996 | if (err) |
5997 | return err; | |
6791875e N |
5998 | conf = mddev->private; |
5999 | if (!conf) | |
6000 | err = -ENODEV; | |
6001 | else | |
6002 | err = raid5_set_cache_size(mddev, new); | |
6003 | mddev_unlock(mddev); | |
6004 | ||
6005 | return err ?: len; | |
3f294f4f | 6006 | } |
007583c9 | 6007 | |
96de1e66 N |
6008 | static struct md_sysfs_entry |
6009 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6010 | raid5_show_stripe_cache_size, | |
6011 | raid5_store_stripe_cache_size); | |
3f294f4f | 6012 | |
d06f191f MS |
6013 | static ssize_t |
6014 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6015 | { | |
6016 | struct r5conf *conf = mddev->private; | |
6017 | if (conf) | |
6018 | return sprintf(page, "%d\n", conf->rmw_level); | |
6019 | else | |
6020 | return 0; | |
6021 | } | |
6022 | ||
6023 | static ssize_t | |
6024 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6025 | { | |
6026 | struct r5conf *conf = mddev->private; | |
6027 | unsigned long new; | |
6028 | ||
6029 | if (!conf) | |
6030 | return -ENODEV; | |
6031 | ||
6032 | if (len >= PAGE_SIZE) | |
6033 | return -EINVAL; | |
6034 | ||
6035 | if (kstrtoul(page, 10, &new)) | |
6036 | return -EINVAL; | |
6037 | ||
6038 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6039 | return -EINVAL; | |
6040 | ||
6041 | if (new != PARITY_DISABLE_RMW && | |
6042 | new != PARITY_ENABLE_RMW && | |
6043 | new != PARITY_PREFER_RMW) | |
6044 | return -EINVAL; | |
6045 | ||
6046 | conf->rmw_level = new; | |
6047 | return len; | |
6048 | } | |
6049 | ||
6050 | static struct md_sysfs_entry | |
6051 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6052 | raid5_show_rmw_level, | |
6053 | raid5_store_rmw_level); | |
6054 | ||
6055 | ||
8b3e6cdc | 6056 | static ssize_t |
fd01b88c | 6057 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6058 | { |
7b1485ba N |
6059 | struct r5conf *conf; |
6060 | int ret = 0; | |
6061 | spin_lock(&mddev->lock); | |
6062 | conf = mddev->private; | |
8b3e6cdc | 6063 | if (conf) |
7b1485ba N |
6064 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6065 | spin_unlock(&mddev->lock); | |
6066 | return ret; | |
8b3e6cdc DW |
6067 | } |
6068 | ||
6069 | static ssize_t | |
fd01b88c | 6070 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6071 | { |
6791875e | 6072 | struct r5conf *conf; |
4ef197d8 | 6073 | unsigned long new; |
6791875e N |
6074 | int err; |
6075 | ||
8b3e6cdc DW |
6076 | if (len >= PAGE_SIZE) |
6077 | return -EINVAL; | |
b29bebd6 | 6078 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6079 | return -EINVAL; |
6791875e N |
6080 | |
6081 | err = mddev_lock(mddev); | |
6082 | if (err) | |
6083 | return err; | |
6084 | conf = mddev->private; | |
6085 | if (!conf) | |
6086 | err = -ENODEV; | |
edbe83ab | 6087 | else if (new > conf->min_nr_stripes) |
6791875e N |
6088 | err = -EINVAL; |
6089 | else | |
6090 | conf->bypass_threshold = new; | |
6091 | mddev_unlock(mddev); | |
6092 | return err ?: len; | |
8b3e6cdc DW |
6093 | } |
6094 | ||
6095 | static struct md_sysfs_entry | |
6096 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6097 | S_IRUGO | S_IWUSR, | |
6098 | raid5_show_preread_threshold, | |
6099 | raid5_store_preread_threshold); | |
6100 | ||
d592a996 SL |
6101 | static ssize_t |
6102 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6103 | { | |
7b1485ba N |
6104 | struct r5conf *conf; |
6105 | int ret = 0; | |
6106 | spin_lock(&mddev->lock); | |
6107 | conf = mddev->private; | |
d592a996 | 6108 | if (conf) |
7b1485ba N |
6109 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6110 | spin_unlock(&mddev->lock); | |
6111 | return ret; | |
d592a996 SL |
6112 | } |
6113 | ||
6114 | static ssize_t | |
6115 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6116 | { | |
6791875e | 6117 | struct r5conf *conf; |
d592a996 | 6118 | unsigned long new; |
6791875e N |
6119 | int err; |
6120 | ||
d592a996 SL |
6121 | if (len >= PAGE_SIZE) |
6122 | return -EINVAL; | |
d592a996 SL |
6123 | if (kstrtoul(page, 10, &new)) |
6124 | return -EINVAL; | |
6125 | new = !!new; | |
6791875e N |
6126 | |
6127 | err = mddev_lock(mddev); | |
6128 | if (err) | |
6129 | return err; | |
6130 | conf = mddev->private; | |
6131 | if (!conf) | |
6132 | err = -ENODEV; | |
6133 | else if (new != conf->skip_copy) { | |
6134 | mddev_suspend(mddev); | |
6135 | conf->skip_copy = new; | |
6136 | if (new) | |
6137 | mddev->queue->backing_dev_info.capabilities |= | |
6138 | BDI_CAP_STABLE_WRITES; | |
6139 | else | |
6140 | mddev->queue->backing_dev_info.capabilities &= | |
6141 | ~BDI_CAP_STABLE_WRITES; | |
6142 | mddev_resume(mddev); | |
6143 | } | |
6144 | mddev_unlock(mddev); | |
6145 | return err ?: len; | |
d592a996 SL |
6146 | } |
6147 | ||
6148 | static struct md_sysfs_entry | |
6149 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6150 | raid5_show_skip_copy, | |
6151 | raid5_store_skip_copy); | |
6152 | ||
3f294f4f | 6153 | static ssize_t |
fd01b88c | 6154 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6155 | { |
d1688a6d | 6156 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6157 | if (conf) |
6158 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6159 | else | |
6160 | return 0; | |
3f294f4f N |
6161 | } |
6162 | ||
96de1e66 N |
6163 | static struct md_sysfs_entry |
6164 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6165 | |
b721420e SL |
6166 | static ssize_t |
6167 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6168 | { | |
7b1485ba N |
6169 | struct r5conf *conf; |
6170 | int ret = 0; | |
6171 | spin_lock(&mddev->lock); | |
6172 | conf = mddev->private; | |
b721420e | 6173 | if (conf) |
7b1485ba N |
6174 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6175 | spin_unlock(&mddev->lock); | |
6176 | return ret; | |
b721420e SL |
6177 | } |
6178 | ||
60aaf933 | 6179 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6180 | int *group_cnt, | |
6181 | int *worker_cnt_per_group, | |
6182 | struct r5worker_group **worker_groups); | |
b721420e SL |
6183 | static ssize_t |
6184 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6185 | { | |
6791875e | 6186 | struct r5conf *conf; |
b721420e SL |
6187 | unsigned long new; |
6188 | int err; | |
60aaf933 | 6189 | struct r5worker_group *new_groups, *old_groups; |
6190 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6191 | |
6192 | if (len >= PAGE_SIZE) | |
6193 | return -EINVAL; | |
b721420e SL |
6194 | if (kstrtoul(page, 10, &new)) |
6195 | return -EINVAL; | |
6196 | ||
6791875e N |
6197 | err = mddev_lock(mddev); |
6198 | if (err) | |
6199 | return err; | |
6200 | conf = mddev->private; | |
6201 | if (!conf) | |
6202 | err = -ENODEV; | |
6203 | else if (new != conf->worker_cnt_per_group) { | |
6204 | mddev_suspend(mddev); | |
b721420e | 6205 | |
6791875e N |
6206 | old_groups = conf->worker_groups; |
6207 | if (old_groups) | |
6208 | flush_workqueue(raid5_wq); | |
d206dcfa | 6209 | |
6791875e N |
6210 | err = alloc_thread_groups(conf, new, |
6211 | &group_cnt, &worker_cnt_per_group, | |
6212 | &new_groups); | |
6213 | if (!err) { | |
6214 | spin_lock_irq(&conf->device_lock); | |
6215 | conf->group_cnt = group_cnt; | |
6216 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6217 | conf->worker_groups = new_groups; | |
6218 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6219 | |
6791875e N |
6220 | if (old_groups) |
6221 | kfree(old_groups[0].workers); | |
6222 | kfree(old_groups); | |
6223 | } | |
6224 | mddev_resume(mddev); | |
b721420e | 6225 | } |
6791875e | 6226 | mddev_unlock(mddev); |
b721420e | 6227 | |
6791875e | 6228 | return err ?: len; |
b721420e SL |
6229 | } |
6230 | ||
6231 | static struct md_sysfs_entry | |
6232 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6233 | raid5_show_group_thread_cnt, | |
6234 | raid5_store_group_thread_cnt); | |
6235 | ||
007583c9 | 6236 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6237 | &raid5_stripecache_size.attr, |
6238 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6239 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6240 | &raid5_group_thread_cnt.attr, |
d592a996 | 6241 | &raid5_skip_copy.attr, |
d06f191f | 6242 | &raid5_rmw_level.attr, |
3f294f4f N |
6243 | NULL, |
6244 | }; | |
007583c9 N |
6245 | static struct attribute_group raid5_attrs_group = { |
6246 | .name = NULL, | |
6247 | .attrs = raid5_attrs, | |
3f294f4f N |
6248 | }; |
6249 | ||
60aaf933 | 6250 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6251 | int *group_cnt, | |
6252 | int *worker_cnt_per_group, | |
6253 | struct r5worker_group **worker_groups) | |
851c30c9 | 6254 | { |
566c09c5 | 6255 | int i, j, k; |
851c30c9 SL |
6256 | ssize_t size; |
6257 | struct r5worker *workers; | |
6258 | ||
60aaf933 | 6259 | *worker_cnt_per_group = cnt; |
851c30c9 | 6260 | if (cnt == 0) { |
60aaf933 | 6261 | *group_cnt = 0; |
6262 | *worker_groups = NULL; | |
851c30c9 SL |
6263 | return 0; |
6264 | } | |
60aaf933 | 6265 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6266 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6267 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6268 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6269 | *group_cnt, GFP_NOIO); | |
6270 | if (!*worker_groups || !workers) { | |
851c30c9 | 6271 | kfree(workers); |
60aaf933 | 6272 | kfree(*worker_groups); |
851c30c9 SL |
6273 | return -ENOMEM; |
6274 | } | |
6275 | ||
60aaf933 | 6276 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6277 | struct r5worker_group *group; |
6278 | ||
0c775d52 | 6279 | group = &(*worker_groups)[i]; |
851c30c9 SL |
6280 | INIT_LIST_HEAD(&group->handle_list); |
6281 | group->conf = conf; | |
6282 | group->workers = workers + i * cnt; | |
6283 | ||
6284 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6285 | struct r5worker *worker = group->workers + j; |
6286 | worker->group = group; | |
6287 | INIT_WORK(&worker->work, raid5_do_work); | |
6288 | ||
6289 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6290 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6291 | } |
6292 | } | |
6293 | ||
6294 | return 0; | |
6295 | } | |
6296 | ||
6297 | static void free_thread_groups(struct r5conf *conf) | |
6298 | { | |
6299 | if (conf->worker_groups) | |
6300 | kfree(conf->worker_groups[0].workers); | |
6301 | kfree(conf->worker_groups); | |
6302 | conf->worker_groups = NULL; | |
6303 | } | |
6304 | ||
80c3a6ce | 6305 | static sector_t |
fd01b88c | 6306 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6307 | { |
d1688a6d | 6308 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6309 | |
6310 | if (!sectors) | |
6311 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6312 | if (!raid_disks) |
7ec05478 | 6313 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6314 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6315 | |
3cb5edf4 N |
6316 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6317 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6318 | return sectors * (raid_disks - conf->max_degraded); |
6319 | } | |
6320 | ||
789b5e03 ON |
6321 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6322 | { | |
6323 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6324 | if (percpu->scribble) |
6325 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6326 | percpu->spare_page = NULL; |
6327 | percpu->scribble = NULL; | |
6328 | } | |
6329 | ||
6330 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6331 | { | |
6332 | if (conf->level == 6 && !percpu->spare_page) | |
6333 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6334 | if (!percpu->scribble) | |
46d5b785 | 6335 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6336 | conf->previous_raid_disks), |
6337 | max(conf->chunk_sectors, | |
6338 | conf->prev_chunk_sectors) | |
6339 | / STRIPE_SECTORS, | |
6340 | GFP_KERNEL); | |
789b5e03 ON |
6341 | |
6342 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6343 | free_scratch_buffer(conf, percpu); | |
6344 | return -ENOMEM; | |
6345 | } | |
6346 | ||
6347 | return 0; | |
6348 | } | |
6349 | ||
29c6d1bb | 6350 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6351 | { |
29c6d1bb SAS |
6352 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6353 | ||
6354 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6355 | return 0; | |
6356 | } | |
36d1c647 | 6357 | |
29c6d1bb SAS |
6358 | static void raid5_free_percpu(struct r5conf *conf) |
6359 | { | |
36d1c647 DW |
6360 | if (!conf->percpu) |
6361 | return; | |
6362 | ||
29c6d1bb | 6363 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6364 | free_percpu(conf->percpu); |
6365 | } | |
6366 | ||
d1688a6d | 6367 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6368 | { |
5c7e81c3 SL |
6369 | if (conf->log) |
6370 | r5l_exit_log(conf->log); | |
30c89465 | 6371 | if (conf->shrinker.nr_deferred) |
edbe83ab | 6372 | unregister_shrinker(&conf->shrinker); |
5c7e81c3 | 6373 | |
851c30c9 | 6374 | free_thread_groups(conf); |
95fc17aa | 6375 | shrink_stripes(conf); |
36d1c647 | 6376 | raid5_free_percpu(conf); |
95fc17aa DW |
6377 | kfree(conf->disks); |
6378 | kfree(conf->stripe_hashtbl); | |
6379 | kfree(conf); | |
6380 | } | |
6381 | ||
29c6d1bb | 6382 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6383 | { |
29c6d1bb | 6384 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6385 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6386 | ||
29c6d1bb SAS |
6387 | if (alloc_scratch_buffer(conf, percpu)) { |
6388 | pr_err("%s: failed memory allocation for cpu%u\n", | |
6389 | __func__, cpu); | |
6390 | return -ENOMEM; | |
36d1c647 | 6391 | } |
29c6d1bb | 6392 | return 0; |
36d1c647 | 6393 | } |
36d1c647 | 6394 | |
d1688a6d | 6395 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6396 | { |
789b5e03 | 6397 | int err = 0; |
36d1c647 | 6398 | |
789b5e03 ON |
6399 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6400 | if (!conf->percpu) | |
36d1c647 | 6401 | return -ENOMEM; |
789b5e03 | 6402 | |
29c6d1bb | 6403 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6404 | if (!err) { |
6405 | conf->scribble_disks = max(conf->raid_disks, | |
6406 | conf->previous_raid_disks); | |
6407 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6408 | conf->prev_chunk_sectors); | |
6409 | } | |
36d1c647 DW |
6410 | return err; |
6411 | } | |
6412 | ||
edbe83ab N |
6413 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6414 | struct shrink_control *sc) | |
6415 | { | |
6416 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6417 | unsigned long ret = SHRINK_STOP; |
6418 | ||
6419 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6420 | ret= 0; | |
49895bcc N |
6421 | while (ret < sc->nr_to_scan && |
6422 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6423 | if (drop_one_stripe(conf) == 0) { |
6424 | ret = SHRINK_STOP; | |
6425 | break; | |
6426 | } | |
6427 | ret++; | |
6428 | } | |
6429 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6430 | } |
6431 | return ret; | |
6432 | } | |
6433 | ||
6434 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6435 | struct shrink_control *sc) | |
6436 | { | |
6437 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6438 | ||
6439 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6440 | /* unlikely, but not impossible */ | |
6441 | return 0; | |
6442 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6443 | } | |
6444 | ||
d1688a6d | 6445 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6446 | { |
d1688a6d | 6447 | struct r5conf *conf; |
5e5e3e78 | 6448 | int raid_disk, memory, max_disks; |
3cb03002 | 6449 | struct md_rdev *rdev; |
1da177e4 | 6450 | struct disk_info *disk; |
0232605d | 6451 | char pers_name[6]; |
566c09c5 | 6452 | int i; |
60aaf933 | 6453 | int group_cnt, worker_cnt_per_group; |
6454 | struct r5worker_group *new_group; | |
1da177e4 | 6455 | |
91adb564 N |
6456 | if (mddev->new_level != 5 |
6457 | && mddev->new_level != 4 | |
6458 | && mddev->new_level != 6) { | |
0c55e022 | 6459 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
6460 | mdname(mddev), mddev->new_level); |
6461 | return ERR_PTR(-EIO); | |
1da177e4 | 6462 | } |
91adb564 N |
6463 | if ((mddev->new_level == 5 |
6464 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6465 | (mddev->new_level == 6 | |
6466 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 6467 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
6468 | mdname(mddev), mddev->new_layout); |
6469 | return ERR_PTR(-EIO); | |
99c0fb5f | 6470 | } |
91adb564 | 6471 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 6472 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
6473 | mdname(mddev), mddev->raid_disks); |
6474 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
6475 | } |
6476 | ||
664e7c41 AN |
6477 | if (!mddev->new_chunk_sectors || |
6478 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6479 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
6480 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
6481 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6482 | return ERR_PTR(-EINVAL); |
f6705578 N |
6483 | } |
6484 | ||
d1688a6d | 6485 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6486 | if (conf == NULL) |
1da177e4 | 6487 | goto abort; |
851c30c9 | 6488 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6489 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6490 | &new_group)) { | |
6491 | conf->group_cnt = group_cnt; | |
6492 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6493 | conf->worker_groups = new_group; | |
6494 | } else | |
851c30c9 | 6495 | goto abort; |
f5efd45a | 6496 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6497 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6498 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6499 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6500 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6501 | init_waitqueue_head(&conf->wait_for_overlap); |
6502 | INIT_LIST_HEAD(&conf->handle_list); | |
6503 | INIT_LIST_HEAD(&conf->hold_list); | |
6504 | INIT_LIST_HEAD(&conf->delayed_list); | |
6505 | INIT_LIST_HEAD(&conf->bitmap_list); | |
c3cce6cd | 6506 | bio_list_init(&conf->return_bi); |
773ca82f | 6507 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6508 | atomic_set(&conf->active_stripes, 0); |
6509 | atomic_set(&conf->preread_active_stripes, 0); | |
6510 | atomic_set(&conf->active_aligned_reads, 0); | |
6511 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 6512 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6513 | |
6514 | conf->raid_disks = mddev->raid_disks; | |
6515 | if (mddev->reshape_position == MaxSector) | |
6516 | conf->previous_raid_disks = mddev->raid_disks; | |
6517 | else | |
f6705578 | 6518 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6519 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6520 | |
5e5e3e78 | 6521 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
6522 | GFP_KERNEL); |
6523 | if (!conf->disks) | |
6524 | goto abort; | |
9ffae0cf | 6525 | |
1da177e4 LT |
6526 | conf->mddev = mddev; |
6527 | ||
fccddba0 | 6528 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6529 | goto abort; |
1da177e4 | 6530 | |
566c09c5 SL |
6531 | /* We init hash_locks[0] separately to that it can be used |
6532 | * as the reference lock in the spin_lock_nest_lock() call | |
6533 | * in lock_all_device_hash_locks_irq in order to convince | |
6534 | * lockdep that we know what we are doing. | |
6535 | */ | |
6536 | spin_lock_init(conf->hash_locks); | |
6537 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6538 | spin_lock_init(conf->hash_locks + i); | |
6539 | ||
6540 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6541 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6542 | ||
6543 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6544 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6545 | ||
36d1c647 | 6546 | conf->level = mddev->new_level; |
46d5b785 | 6547 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6548 | if (raid5_alloc_percpu(conf) != 0) |
6549 | goto abort; | |
6550 | ||
0c55e022 | 6551 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6552 | |
dafb20fa | 6553 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6554 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6555 | if (raid_disk >= max_disks |
f2076e7d | 6556 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
6557 | continue; |
6558 | disk = conf->disks + raid_disk; | |
6559 | ||
17045f52 N |
6560 | if (test_bit(Replacement, &rdev->flags)) { |
6561 | if (disk->replacement) | |
6562 | goto abort; | |
6563 | disk->replacement = rdev; | |
6564 | } else { | |
6565 | if (disk->rdev) | |
6566 | goto abort; | |
6567 | disk->rdev = rdev; | |
6568 | } | |
1da177e4 | 6569 | |
b2d444d7 | 6570 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6571 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
6572 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
6573 | " disk %d\n", | |
6574 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6575 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6576 | /* Cannot rely on bitmap to complete recovery */ |
6577 | conf->fullsync = 1; | |
1da177e4 LT |
6578 | } |
6579 | ||
91adb564 | 6580 | conf->level = mddev->new_level; |
584acdd4 | 6581 | if (conf->level == 6) { |
16a53ecc | 6582 | conf->max_degraded = 2; |
584acdd4 MS |
6583 | if (raid6_call.xor_syndrome) |
6584 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6585 | else | |
6586 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6587 | } else { | |
16a53ecc | 6588 | conf->max_degraded = 1; |
584acdd4 MS |
6589 | conf->rmw_level = PARITY_ENABLE_RMW; |
6590 | } | |
91adb564 | 6591 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6592 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6593 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6594 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 6595 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
6596 | } else { |
6597 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
6598 | conf->prev_algo = conf->algorithm; | |
e183eaed | 6599 | } |
1da177e4 | 6600 | |
edbe83ab | 6601 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
6602 | if (mddev->reshape_position != MaxSector) { |
6603 | int stripes = max_t(int, | |
6604 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
6605 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
6606 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
6607 | if (conf->min_nr_stripes != NR_STRIPES) | |
6608 | printk(KERN_INFO | |
6609 | "md/raid:%s: force stripe size %d for reshape\n", | |
6610 | mdname(mddev), conf->min_nr_stripes); | |
6611 | } | |
edbe83ab | 6612 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 6613 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6614 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6615 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
91adb564 | 6616 | printk(KERN_ERR |
0c55e022 N |
6617 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
6618 | mdname(mddev), memory); | |
91adb564 N |
6619 | goto abort; |
6620 | } else | |
0c55e022 N |
6621 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
6622 | mdname(mddev), memory); | |
edbe83ab N |
6623 | /* |
6624 | * Losing a stripe head costs more than the time to refill it, | |
6625 | * it reduces the queue depth and so can hurt throughput. | |
6626 | * So set it rather large, scaled by number of devices. | |
6627 | */ | |
6628 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6629 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6630 | conf->shrinker.count_objects = raid5_cache_count; | |
6631 | conf->shrinker.batch = 128; | |
6632 | conf->shrinker.flags = 0; | |
6a0f53ff CY |
6633 | if (register_shrinker(&conf->shrinker)) { |
6634 | printk(KERN_ERR | |
6635 | "md/raid:%s: couldn't register shrinker.\n", | |
6636 | mdname(mddev)); | |
6637 | goto abort; | |
6638 | } | |
1da177e4 | 6639 | |
0232605d N |
6640 | sprintf(pers_name, "raid%d", mddev->new_level); |
6641 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
6642 | if (!conf->thread) { |
6643 | printk(KERN_ERR | |
0c55e022 | 6644 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 6645 | mdname(mddev)); |
16a53ecc N |
6646 | goto abort; |
6647 | } | |
91adb564 N |
6648 | |
6649 | return conf; | |
6650 | ||
6651 | abort: | |
6652 | if (conf) { | |
95fc17aa | 6653 | free_conf(conf); |
91adb564 N |
6654 | return ERR_PTR(-EIO); |
6655 | } else | |
6656 | return ERR_PTR(-ENOMEM); | |
6657 | } | |
6658 | ||
c148ffdc N |
6659 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6660 | { | |
6661 | switch (algo) { | |
6662 | case ALGORITHM_PARITY_0: | |
6663 | if (raid_disk < max_degraded) | |
6664 | return 1; | |
6665 | break; | |
6666 | case ALGORITHM_PARITY_N: | |
6667 | if (raid_disk >= raid_disks - max_degraded) | |
6668 | return 1; | |
6669 | break; | |
6670 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6671 | if (raid_disk == 0 || |
c148ffdc N |
6672 | raid_disk == raid_disks - 1) |
6673 | return 1; | |
6674 | break; | |
6675 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6676 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6677 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6678 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6679 | if (raid_disk == raid_disks - 1) | |
6680 | return 1; | |
6681 | } | |
6682 | return 0; | |
6683 | } | |
6684 | ||
849674e4 | 6685 | static int raid5_run(struct mddev *mddev) |
91adb564 | 6686 | { |
d1688a6d | 6687 | struct r5conf *conf; |
9f7c2220 | 6688 | int working_disks = 0; |
c148ffdc | 6689 | int dirty_parity_disks = 0; |
3cb03002 | 6690 | struct md_rdev *rdev; |
713cf5a6 | 6691 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 6692 | sector_t reshape_offset = 0; |
17045f52 | 6693 | int i; |
b5254dd5 N |
6694 | long long min_offset_diff = 0; |
6695 | int first = 1; | |
91adb564 | 6696 | |
8c6ac868 | 6697 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 6698 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
6699 | " -- starting background reconstruction\n", |
6700 | mdname(mddev)); | |
b5254dd5 N |
6701 | |
6702 | rdev_for_each(rdev, mddev) { | |
6703 | long long diff; | |
713cf5a6 | 6704 | |
f2076e7d | 6705 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 6706 | journal_dev = rdev; |
f2076e7d SL |
6707 | continue; |
6708 | } | |
b5254dd5 N |
6709 | if (rdev->raid_disk < 0) |
6710 | continue; | |
6711 | diff = (rdev->new_data_offset - rdev->data_offset); | |
6712 | if (first) { | |
6713 | min_offset_diff = diff; | |
6714 | first = 0; | |
6715 | } else if (mddev->reshape_backwards && | |
6716 | diff < min_offset_diff) | |
6717 | min_offset_diff = diff; | |
6718 | else if (!mddev->reshape_backwards && | |
6719 | diff > min_offset_diff) | |
6720 | min_offset_diff = diff; | |
6721 | } | |
6722 | ||
91adb564 N |
6723 | if (mddev->reshape_position != MaxSector) { |
6724 | /* Check that we can continue the reshape. | |
b5254dd5 N |
6725 | * Difficulties arise if the stripe we would write to |
6726 | * next is at or after the stripe we would read from next. | |
6727 | * For a reshape that changes the number of devices, this | |
6728 | * is only possible for a very short time, and mdadm makes | |
6729 | * sure that time appears to have past before assembling | |
6730 | * the array. So we fail if that time hasn't passed. | |
6731 | * For a reshape that keeps the number of devices the same | |
6732 | * mdadm must be monitoring the reshape can keeping the | |
6733 | * critical areas read-only and backed up. It will start | |
6734 | * the array in read-only mode, so we check for that. | |
91adb564 N |
6735 | */ |
6736 | sector_t here_new, here_old; | |
6737 | int old_disks; | |
18b00334 | 6738 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
6739 | int chunk_sectors; |
6740 | int new_data_disks; | |
91adb564 | 6741 | |
713cf5a6 SL |
6742 | if (journal_dev) { |
6743 | printk(KERN_ERR "md/raid:%s: don't support reshape with journal - aborting.\n", | |
6744 | mdname(mddev)); | |
6745 | return -EINVAL; | |
6746 | } | |
6747 | ||
88ce4930 | 6748 | if (mddev->new_level != mddev->level) { |
0c55e022 | 6749 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
6750 | "required - aborting.\n", |
6751 | mdname(mddev)); | |
6752 | return -EINVAL; | |
6753 | } | |
91adb564 N |
6754 | old_disks = mddev->raid_disks - mddev->delta_disks; |
6755 | /* reshape_position must be on a new-stripe boundary, and one | |
6756 | * further up in new geometry must map after here in old | |
6757 | * geometry. | |
05256d98 N |
6758 | * If the chunk sizes are different, then as we perform reshape |
6759 | * in units of the largest of the two, reshape_position needs | |
6760 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
6761 | */ |
6762 | here_new = mddev->reshape_position; | |
05256d98 N |
6763 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
6764 | new_data_disks = mddev->raid_disks - max_degraded; | |
6765 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
0c55e022 N |
6766 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
6767 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
6768 | return -EINVAL; |
6769 | } | |
05256d98 | 6770 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
6771 | /* here_new is the stripe we will write to */ |
6772 | here_old = mddev->reshape_position; | |
05256d98 | 6773 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
6774 | /* here_old is the first stripe that we might need to read |
6775 | * from */ | |
67ac6011 N |
6776 | if (mddev->delta_disks == 0) { |
6777 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 6778 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
6779 | * and taking constant backups. |
6780 | * mdadm always starts a situation like this in | |
6781 | * readonly mode so it can take control before | |
6782 | * allowing any writes. So just check for that. | |
6783 | */ | |
b5254dd5 N |
6784 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
6785 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
6786 | /* not really in-place - so OK */; | |
6787 | else if (mddev->ro == 0) { | |
6788 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
6789 | "must be started in read-only mode " | |
6790 | "- aborting\n", | |
0c55e022 | 6791 | mdname(mddev)); |
67ac6011 N |
6792 | return -EINVAL; |
6793 | } | |
2c810cdd | 6794 | } else if (mddev->reshape_backwards |
05256d98 N |
6795 | ? (here_new * chunk_sectors + min_offset_diff <= |
6796 | here_old * chunk_sectors) | |
6797 | : (here_new * chunk_sectors >= | |
6798 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 6799 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
6800 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
6801 | "auto-recovery - aborting.\n", | |
6802 | mdname(mddev)); | |
91adb564 N |
6803 | return -EINVAL; |
6804 | } | |
0c55e022 N |
6805 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
6806 | mdname(mddev)); | |
91adb564 N |
6807 | /* OK, we should be able to continue; */ |
6808 | } else { | |
6809 | BUG_ON(mddev->level != mddev->new_level); | |
6810 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 6811 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 6812 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 6813 | } |
91adb564 | 6814 | |
245f46c2 N |
6815 | if (mddev->private == NULL) |
6816 | conf = setup_conf(mddev); | |
6817 | else | |
6818 | conf = mddev->private; | |
6819 | ||
91adb564 N |
6820 | if (IS_ERR(conf)) |
6821 | return PTR_ERR(conf); | |
6822 | ||
486b0f7b SL |
6823 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
6824 | if (!journal_dev) { | |
6825 | pr_err("md/raid:%s: journal disk is missing, force array readonly\n", | |
6826 | mdname(mddev)); | |
6827 | mddev->ro = 1; | |
6828 | set_disk_ro(mddev->gendisk, 1); | |
6829 | } else if (mddev->recovery_cp == MaxSector) | |
6830 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
6831 | } |
6832 | ||
b5254dd5 | 6833 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
6834 | mddev->thread = conf->thread; |
6835 | conf->thread = NULL; | |
6836 | mddev->private = conf; | |
6837 | ||
17045f52 N |
6838 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
6839 | i++) { | |
6840 | rdev = conf->disks[i].rdev; | |
6841 | if (!rdev && conf->disks[i].replacement) { | |
6842 | /* The replacement is all we have yet */ | |
6843 | rdev = conf->disks[i].replacement; | |
6844 | conf->disks[i].replacement = NULL; | |
6845 | clear_bit(Replacement, &rdev->flags); | |
6846 | conf->disks[i].rdev = rdev; | |
6847 | } | |
6848 | if (!rdev) | |
c148ffdc | 6849 | continue; |
17045f52 N |
6850 | if (conf->disks[i].replacement && |
6851 | conf->reshape_progress != MaxSector) { | |
6852 | /* replacements and reshape simply do not mix. */ | |
6853 | printk(KERN_ERR "md: cannot handle concurrent " | |
6854 | "replacement and reshape.\n"); | |
6855 | goto abort; | |
6856 | } | |
2f115882 | 6857 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 6858 | working_disks++; |
2f115882 N |
6859 | continue; |
6860 | } | |
c148ffdc N |
6861 | /* This disc is not fully in-sync. However if it |
6862 | * just stored parity (beyond the recovery_offset), | |
6863 | * when we don't need to be concerned about the | |
6864 | * array being dirty. | |
6865 | * When reshape goes 'backwards', we never have | |
6866 | * partially completed devices, so we only need | |
6867 | * to worry about reshape going forwards. | |
6868 | */ | |
6869 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
6870 | if (mddev->major_version == 0 && | |
6871 | mddev->minor_version > 90) | |
6872 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 6873 | |
c148ffdc N |
6874 | if (rdev->recovery_offset < reshape_offset) { |
6875 | /* We need to check old and new layout */ | |
6876 | if (!only_parity(rdev->raid_disk, | |
6877 | conf->algorithm, | |
6878 | conf->raid_disks, | |
6879 | conf->max_degraded)) | |
6880 | continue; | |
6881 | } | |
6882 | if (!only_parity(rdev->raid_disk, | |
6883 | conf->prev_algo, | |
6884 | conf->previous_raid_disks, | |
6885 | conf->max_degraded)) | |
6886 | continue; | |
6887 | dirty_parity_disks++; | |
6888 | } | |
91adb564 | 6889 | |
17045f52 N |
6890 | /* |
6891 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
6892 | */ | |
908f4fbd | 6893 | mddev->degraded = calc_degraded(conf); |
91adb564 | 6894 | |
674806d6 | 6895 | if (has_failed(conf)) { |
0c55e022 | 6896 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 6897 | " (%d/%d failed)\n", |
02c2de8c | 6898 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
6899 | goto abort; |
6900 | } | |
6901 | ||
91adb564 | 6902 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 6903 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
6904 | mddev->resync_max_sectors = mddev->dev_sectors; |
6905 | ||
c148ffdc | 6906 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 6907 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
6908 | if (mddev->ok_start_degraded) |
6909 | printk(KERN_WARNING | |
0c55e022 N |
6910 | "md/raid:%s: starting dirty degraded array" |
6911 | " - data corruption possible.\n", | |
6ff8d8ec N |
6912 | mdname(mddev)); |
6913 | else { | |
6914 | printk(KERN_ERR | |
0c55e022 | 6915 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
6916 | mdname(mddev)); |
6917 | goto abort; | |
6918 | } | |
1da177e4 LT |
6919 | } |
6920 | ||
1da177e4 | 6921 | if (mddev->degraded == 0) |
0c55e022 N |
6922 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
6923 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
6924 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
6925 | mddev->new_layout); | |
1da177e4 | 6926 | else |
0c55e022 N |
6927 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
6928 | " out of %d devices, algorithm %d\n", | |
6929 | mdname(mddev), conf->level, | |
6930 | mddev->raid_disks - mddev->degraded, | |
6931 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
6932 | |
6933 | print_raid5_conf(conf); | |
6934 | ||
fef9c61f | 6935 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 6936 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
6937 | atomic_set(&conf->reshape_stripes, 0); |
6938 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
6939 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6940 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6941 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6942 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6943 | "reshape"); |
f6705578 N |
6944 | } |
6945 | ||
1da177e4 | 6946 | /* Ok, everything is just fine now */ |
a64c876f N |
6947 | if (mddev->to_remove == &raid5_attrs_group) |
6948 | mddev->to_remove = NULL; | |
00bcb4ac N |
6949 | else if (mddev->kobj.sd && |
6950 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 6951 | printk(KERN_WARNING |
4a5add49 | 6952 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 6953 | mdname(mddev)); |
4a5add49 | 6954 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 6955 | |
4a5add49 | 6956 | if (mddev->queue) { |
9f7c2220 | 6957 | int chunk_size; |
620125f2 | 6958 | bool discard_supported = true; |
4a5add49 N |
6959 | /* read-ahead size must cover two whole stripes, which |
6960 | * is 2 * (datadisks) * chunksize where 'n' is the | |
6961 | * number of raid devices | |
6962 | */ | |
6963 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
6964 | int stripe = data_disks * | |
6965 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
6966 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
6967 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 6968 | |
9f7c2220 N |
6969 | chunk_size = mddev->chunk_sectors << 9; |
6970 | blk_queue_io_min(mddev->queue, chunk_size); | |
6971 | blk_queue_io_opt(mddev->queue, chunk_size * | |
6972 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 6973 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
6974 | /* |
6975 | * We can only discard a whole stripe. It doesn't make sense to | |
6976 | * discard data disk but write parity disk | |
6977 | */ | |
6978 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
6979 | /* Round up to power of 2, as discard handling |
6980 | * currently assumes that */ | |
6981 | while ((stripe-1) & stripe) | |
6982 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
6983 | mddev->queue->limits.discard_alignment = stripe; |
6984 | mddev->queue->limits.discard_granularity = stripe; | |
6985 | /* | |
6986 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 6987 | * guarantee discard_zeroes_data |
620125f2 SL |
6988 | */ |
6989 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 6990 | |
5026d7a9 PA |
6991 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
6992 | ||
05616be5 | 6993 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
6994 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6995 | rdev->data_offset << 9); | |
05616be5 N |
6996 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6997 | rdev->new_data_offset << 9); | |
620125f2 SL |
6998 | /* |
6999 | * discard_zeroes_data is required, otherwise data | |
7000 | * could be lost. Consider a scenario: discard a stripe | |
7001 | * (the stripe could be inconsistent if | |
7002 | * discard_zeroes_data is 0); write one disk of the | |
7003 | * stripe (the stripe could be inconsistent again | |
7004 | * depending on which disks are used to calculate | |
7005 | * parity); the disk is broken; The stripe data of this | |
7006 | * disk is lost. | |
7007 | */ | |
7008 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
7009 | !bdev_get_queue(rdev->bdev)-> | |
7010 | limits.discard_zeroes_data) | |
7011 | discard_supported = false; | |
8e0e99ba N |
7012 | /* Unfortunately, discard_zeroes_data is not currently |
7013 | * a guarantee - just a hint. So we only allow DISCARD | |
7014 | * if the sysadmin has confirmed that only safe devices | |
7015 | * are in use by setting a module parameter. | |
7016 | */ | |
7017 | if (!devices_handle_discard_safely) { | |
7018 | if (discard_supported) { | |
7019 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
7020 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
7021 | } | |
7022 | discard_supported = false; | |
7023 | } | |
05616be5 | 7024 | } |
620125f2 SL |
7025 | |
7026 | if (discard_supported && | |
e7597e69 JS |
7027 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7028 | mddev->queue->limits.discard_granularity >= stripe) | |
620125f2 SL |
7029 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
7030 | mddev->queue); | |
7031 | else | |
7032 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
7033 | mddev->queue); | |
1dffdddd SL |
7034 | |
7035 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7036 | } |
23032a0e | 7037 | |
5c7e81c3 SL |
7038 | if (journal_dev) { |
7039 | char b[BDEVNAME_SIZE]; | |
7040 | ||
7041 | printk(KERN_INFO"md/raid:%s: using device %s as journal\n", | |
7042 | mdname(mddev), bdevname(journal_dev->bdev, b)); | |
7043 | r5l_init_log(conf, journal_dev); | |
7044 | } | |
7045 | ||
1da177e4 LT |
7046 | return 0; |
7047 | abort: | |
01f96c0a | 7048 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7049 | print_raid5_conf(conf); |
7050 | free_conf(conf); | |
1da177e4 | 7051 | mddev->private = NULL; |
0c55e022 | 7052 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7053 | return -EIO; |
7054 | } | |
7055 | ||
afa0f557 | 7056 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7057 | { |
afa0f557 | 7058 | struct r5conf *conf = priv; |
1da177e4 | 7059 | |
95fc17aa | 7060 | free_conf(conf); |
a64c876f | 7061 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7062 | } |
7063 | ||
849674e4 | 7064 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7065 | { |
d1688a6d | 7066 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7067 | int i; |
7068 | ||
9d8f0363 | 7069 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7070 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7071 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7072 | rcu_read_lock(); |
7073 | for (i = 0; i < conf->raid_disks; i++) { | |
7074 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7075 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7076 | } | |
7077 | rcu_read_unlock(); | |
1da177e4 | 7078 | seq_printf (seq, "]"); |
1da177e4 LT |
7079 | } |
7080 | ||
d1688a6d | 7081 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7082 | { |
7083 | int i; | |
7084 | struct disk_info *tmp; | |
7085 | ||
0c55e022 | 7086 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
7087 | if (!conf) { |
7088 | printk("(conf==NULL)\n"); | |
7089 | return; | |
7090 | } | |
0c55e022 N |
7091 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
7092 | conf->raid_disks, | |
7093 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7094 | |
7095 | for (i = 0; i < conf->raid_disks; i++) { | |
7096 | char b[BDEVNAME_SIZE]; | |
7097 | tmp = conf->disks + i; | |
7098 | if (tmp->rdev) | |
0c55e022 N |
7099 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
7100 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
7101 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7102 | } |
7103 | } | |
7104 | ||
fd01b88c | 7105 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7106 | { |
7107 | int i; | |
d1688a6d | 7108 | struct r5conf *conf = mddev->private; |
1da177e4 | 7109 | struct disk_info *tmp; |
6b965620 N |
7110 | int count = 0; |
7111 | unsigned long flags; | |
1da177e4 LT |
7112 | |
7113 | for (i = 0; i < conf->raid_disks; i++) { | |
7114 | tmp = conf->disks + i; | |
dd054fce N |
7115 | if (tmp->replacement |
7116 | && tmp->replacement->recovery_offset == MaxSector | |
7117 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7118 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7119 | /* Replacement has just become active. */ | |
7120 | if (!tmp->rdev | |
7121 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7122 | count++; | |
7123 | if (tmp->rdev) { | |
7124 | /* Replaced device not technically faulty, | |
7125 | * but we need to be sure it gets removed | |
7126 | * and never re-added. | |
7127 | */ | |
7128 | set_bit(Faulty, &tmp->rdev->flags); | |
7129 | sysfs_notify_dirent_safe( | |
7130 | tmp->rdev->sysfs_state); | |
7131 | } | |
7132 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7133 | } else if (tmp->rdev | |
70fffd0b | 7134 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7135 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7136 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7137 | count++; |
43c73ca4 | 7138 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7139 | } |
7140 | } | |
6b965620 | 7141 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7142 | mddev->degraded = calc_degraded(conf); |
6b965620 | 7143 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7144 | print_raid5_conf(conf); |
6b965620 | 7145 | return count; |
1da177e4 LT |
7146 | } |
7147 | ||
b8321b68 | 7148 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7149 | { |
d1688a6d | 7150 | struct r5conf *conf = mddev->private; |
1da177e4 | 7151 | int err = 0; |
b8321b68 | 7152 | int number = rdev->raid_disk; |
657e3e4d | 7153 | struct md_rdev **rdevp; |
1da177e4 LT |
7154 | struct disk_info *p = conf->disks + number; |
7155 | ||
7156 | print_raid5_conf(conf); | |
f6b6ec5c SL |
7157 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
7158 | struct r5l_log *log; | |
c2bb6242 | 7159 | /* |
f6b6ec5c SL |
7160 | * we can't wait pending write here, as this is called in |
7161 | * raid5d, wait will deadlock. | |
c2bb6242 | 7162 | */ |
f6b6ec5c SL |
7163 | if (atomic_read(&mddev->writes_pending)) |
7164 | return -EBUSY; | |
7165 | log = conf->log; | |
7166 | conf->log = NULL; | |
7167 | synchronize_rcu(); | |
7168 | r5l_exit_log(log); | |
7169 | return 0; | |
c2bb6242 | 7170 | } |
657e3e4d N |
7171 | if (rdev == p->rdev) |
7172 | rdevp = &p->rdev; | |
7173 | else if (rdev == p->replacement) | |
7174 | rdevp = &p->replacement; | |
7175 | else | |
7176 | return 0; | |
7177 | ||
7178 | if (number >= conf->raid_disks && | |
7179 | conf->reshape_progress == MaxSector) | |
7180 | clear_bit(In_sync, &rdev->flags); | |
7181 | ||
7182 | if (test_bit(In_sync, &rdev->flags) || | |
7183 | atomic_read(&rdev->nr_pending)) { | |
7184 | err = -EBUSY; | |
7185 | goto abort; | |
7186 | } | |
7187 | /* Only remove non-faulty devices if recovery | |
7188 | * isn't possible. | |
7189 | */ | |
7190 | if (!test_bit(Faulty, &rdev->flags) && | |
7191 | mddev->recovery_disabled != conf->recovery_disabled && | |
7192 | !has_failed(conf) && | |
dd054fce | 7193 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7194 | number < conf->raid_disks) { |
7195 | err = -EBUSY; | |
7196 | goto abort; | |
7197 | } | |
7198 | *rdevp = NULL; | |
d787be40 N |
7199 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7200 | synchronize_rcu(); | |
7201 | if (atomic_read(&rdev->nr_pending)) { | |
7202 | /* lost the race, try later */ | |
7203 | err = -EBUSY; | |
7204 | *rdevp = rdev; | |
7205 | } | |
7206 | } | |
7207 | if (p->replacement) { | |
dd054fce N |
7208 | /* We must have just cleared 'rdev' */ |
7209 | p->rdev = p->replacement; | |
7210 | clear_bit(Replacement, &p->replacement->flags); | |
7211 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7212 | * but will never see neither - if they are careful | |
7213 | */ | |
7214 | p->replacement = NULL; | |
7215 | clear_bit(WantReplacement, &rdev->flags); | |
7216 | } else | |
7217 | /* We might have just removed the Replacement as faulty- | |
7218 | * clear the bit just in case | |
7219 | */ | |
7220 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7221 | abort: |
7222 | ||
7223 | print_raid5_conf(conf); | |
7224 | return err; | |
7225 | } | |
7226 | ||
fd01b88c | 7227 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7228 | { |
d1688a6d | 7229 | struct r5conf *conf = mddev->private; |
199050ea | 7230 | int err = -EEXIST; |
1da177e4 LT |
7231 | int disk; |
7232 | struct disk_info *p; | |
6c2fce2e NB |
7233 | int first = 0; |
7234 | int last = conf->raid_disks - 1; | |
1da177e4 | 7235 | |
f6b6ec5c SL |
7236 | if (test_bit(Journal, &rdev->flags)) { |
7237 | char b[BDEVNAME_SIZE]; | |
7238 | if (conf->log) | |
7239 | return -EBUSY; | |
7240 | ||
7241 | rdev->raid_disk = 0; | |
7242 | /* | |
7243 | * The array is in readonly mode if journal is missing, so no | |
7244 | * write requests running. We should be safe | |
7245 | */ | |
7246 | r5l_init_log(conf, rdev); | |
7247 | printk(KERN_INFO"md/raid:%s: using device %s as journal\n", | |
7248 | mdname(mddev), bdevname(rdev->bdev, b)); | |
7249 | return 0; | |
7250 | } | |
7f0da59b N |
7251 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7252 | return -EBUSY; | |
7253 | ||
dc10c643 | 7254 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7255 | /* no point adding a device */ |
199050ea | 7256 | return -EINVAL; |
1da177e4 | 7257 | |
6c2fce2e NB |
7258 | if (rdev->raid_disk >= 0) |
7259 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7260 | |
7261 | /* | |
16a53ecc N |
7262 | * find the disk ... but prefer rdev->saved_raid_disk |
7263 | * if possible. | |
1da177e4 | 7264 | */ |
16a53ecc | 7265 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7266 | rdev->saved_raid_disk >= first && |
16a53ecc | 7267 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7268 | first = rdev->saved_raid_disk; |
7269 | ||
7270 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7271 | p = conf->disks + disk; |
7272 | if (p->rdev == NULL) { | |
b2d444d7 | 7273 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7274 | rdev->raid_disk = disk; |
199050ea | 7275 | err = 0; |
72626685 N |
7276 | if (rdev->saved_raid_disk != disk) |
7277 | conf->fullsync = 1; | |
d6065f7b | 7278 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7279 | goto out; |
1da177e4 | 7280 | } |
5cfb22a1 N |
7281 | } |
7282 | for (disk = first; disk <= last; disk++) { | |
7283 | p = conf->disks + disk; | |
7bfec5f3 N |
7284 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7285 | p->replacement == NULL) { | |
7286 | clear_bit(In_sync, &rdev->flags); | |
7287 | set_bit(Replacement, &rdev->flags); | |
7288 | rdev->raid_disk = disk; | |
7289 | err = 0; | |
7290 | conf->fullsync = 1; | |
7291 | rcu_assign_pointer(p->replacement, rdev); | |
7292 | break; | |
7293 | } | |
7294 | } | |
5cfb22a1 | 7295 | out: |
1da177e4 | 7296 | print_raid5_conf(conf); |
199050ea | 7297 | return err; |
1da177e4 LT |
7298 | } |
7299 | ||
fd01b88c | 7300 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7301 | { |
7302 | /* no resync is happening, and there is enough space | |
7303 | * on all devices, so we can resize. | |
7304 | * We need to make sure resync covers any new space. | |
7305 | * If the array is shrinking we should possibly wait until | |
7306 | * any io in the removed space completes, but it hardly seems | |
7307 | * worth it. | |
7308 | */ | |
a4a6125a | 7309 | sector_t newsize; |
3cb5edf4 N |
7310 | struct r5conf *conf = mddev->private; |
7311 | ||
713cf5a6 SL |
7312 | if (conf->log) |
7313 | return -EINVAL; | |
3cb5edf4 | 7314 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7315 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7316 | if (mddev->external_size && | |
7317 | mddev->array_sectors > newsize) | |
b522adcd | 7318 | return -EINVAL; |
a4a6125a N |
7319 | if (mddev->bitmap) { |
7320 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7321 | if (ret) | |
7322 | return ret; | |
7323 | } | |
7324 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 7325 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 7326 | revalidate_disk(mddev->gendisk); |
b098636c N |
7327 | if (sectors > mddev->dev_sectors && |
7328 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7329 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7330 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7331 | } | |
58c0fed4 | 7332 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7333 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7334 | return 0; |
7335 | } | |
7336 | ||
fd01b88c | 7337 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7338 | { |
7339 | /* Can only proceed if there are plenty of stripe_heads. | |
7340 | * We need a minimum of one full stripe,, and for sensible progress | |
7341 | * it is best to have about 4 times that. | |
7342 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7343 | * allow for chunk sizes up to 256K, which is probably OK. | |
7344 | * If the chunk size is greater, user-space should request more | |
7345 | * stripe_heads first. | |
7346 | */ | |
d1688a6d | 7347 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7348 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7349 | > conf->min_nr_stripes || |
01ee22b4 | 7350 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7351 | > conf->min_nr_stripes) { |
0c55e022 N |
7352 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7353 | mdname(mddev), | |
01ee22b4 N |
7354 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
7355 | / STRIPE_SIZE)*4); | |
7356 | return 0; | |
7357 | } | |
7358 | return 1; | |
7359 | } | |
7360 | ||
fd01b88c | 7361 | static int check_reshape(struct mddev *mddev) |
29269553 | 7362 | { |
d1688a6d | 7363 | struct r5conf *conf = mddev->private; |
29269553 | 7364 | |
713cf5a6 SL |
7365 | if (conf->log) |
7366 | return -EINVAL; | |
88ce4930 N |
7367 | if (mddev->delta_disks == 0 && |
7368 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7369 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7370 | return 0; /* nothing to do */ |
674806d6 | 7371 | if (has_failed(conf)) |
ec32a2bd | 7372 | return -EINVAL; |
fdcfbbb6 | 7373 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7374 | /* We might be able to shrink, but the devices must |
7375 | * be made bigger first. | |
7376 | * For raid6, 4 is the minimum size. | |
7377 | * Otherwise 2 is the minimum | |
7378 | */ | |
7379 | int min = 2; | |
7380 | if (mddev->level == 6) | |
7381 | min = 4; | |
7382 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7383 | return -EINVAL; | |
7384 | } | |
29269553 | 7385 | |
01ee22b4 | 7386 | if (!check_stripe_cache(mddev)) |
29269553 | 7387 | return -ENOSPC; |
29269553 | 7388 | |
738a2738 N |
7389 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7390 | mddev->delta_disks > 0) | |
7391 | if (resize_chunks(conf, | |
7392 | conf->previous_raid_disks | |
7393 | + max(0, mddev->delta_disks), | |
7394 | max(mddev->new_chunk_sectors, | |
7395 | mddev->chunk_sectors) | |
7396 | ) < 0) | |
7397 | return -ENOMEM; | |
e56108d6 N |
7398 | return resize_stripes(conf, (conf->previous_raid_disks |
7399 | + mddev->delta_disks)); | |
63c70c4f N |
7400 | } |
7401 | ||
fd01b88c | 7402 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7403 | { |
d1688a6d | 7404 | struct r5conf *conf = mddev->private; |
3cb03002 | 7405 | struct md_rdev *rdev; |
63c70c4f | 7406 | int spares = 0; |
c04be0aa | 7407 | unsigned long flags; |
63c70c4f | 7408 | |
f416885e | 7409 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7410 | return -EBUSY; |
7411 | ||
01ee22b4 N |
7412 | if (!check_stripe_cache(mddev)) |
7413 | return -ENOSPC; | |
7414 | ||
30b67645 N |
7415 | if (has_failed(conf)) |
7416 | return -EINVAL; | |
7417 | ||
c6563a8c | 7418 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7419 | if (!test_bit(In_sync, &rdev->flags) |
7420 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7421 | spares++; |
c6563a8c | 7422 | } |
63c70c4f | 7423 | |
f416885e | 7424 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7425 | /* Not enough devices even to make a degraded array |
7426 | * of that size | |
7427 | */ | |
7428 | return -EINVAL; | |
7429 | ||
ec32a2bd N |
7430 | /* Refuse to reduce size of the array. Any reductions in |
7431 | * array size must be through explicit setting of array_size | |
7432 | * attribute. | |
7433 | */ | |
7434 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7435 | < mddev->array_sectors) { | |
0c55e022 | 7436 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
7437 | "before number of disks\n", mdname(mddev)); |
7438 | return -EINVAL; | |
7439 | } | |
7440 | ||
f6705578 | 7441 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7442 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7443 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7444 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7445 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7446 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7447 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7448 | conf->prev_algo = conf->algorithm; |
7449 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7450 | conf->generation++; |
7451 | /* Code that selects data_offset needs to see the generation update | |
7452 | * if reshape_progress has been set - so a memory barrier needed. | |
7453 | */ | |
7454 | smp_mb(); | |
2c810cdd | 7455 | if (mddev->reshape_backwards) |
fef9c61f N |
7456 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7457 | else | |
7458 | conf->reshape_progress = 0; | |
7459 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7460 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7461 | spin_unlock_irq(&conf->device_lock); |
7462 | ||
4d77e3ba N |
7463 | /* Now make sure any requests that proceeded on the assumption |
7464 | * the reshape wasn't running - like Discard or Read - have | |
7465 | * completed. | |
7466 | */ | |
7467 | mddev_suspend(mddev); | |
7468 | mddev_resume(mddev); | |
7469 | ||
29269553 N |
7470 | /* Add some new drives, as many as will fit. |
7471 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7472 | * Don't add devices if we are reducing the number of |
7473 | * devices in the array. This is because it is not possible | |
7474 | * to correctly record the "partially reconstructed" state of | |
7475 | * such devices during the reshape and confusion could result. | |
29269553 | 7476 | */ |
87a8dec9 | 7477 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7478 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7479 | if (rdev->raid_disk < 0 && |
7480 | !test_bit(Faulty, &rdev->flags)) { | |
7481 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7482 | if (rdev->raid_disk |
9d4c7d87 | 7483 | >= conf->previous_raid_disks) |
87a8dec9 | 7484 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7485 | else |
87a8dec9 | 7486 | rdev->recovery_offset = 0; |
36fad858 NK |
7487 | |
7488 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7489 | /* Failure here is OK */; |
50da0840 | 7490 | } |
87a8dec9 N |
7491 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7492 | && !test_bit(Faulty, &rdev->flags)) { | |
7493 | /* This is a spare that was manually added */ | |
7494 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7495 | } |
29269553 | 7496 | |
87a8dec9 N |
7497 | /* When a reshape changes the number of devices, |
7498 | * ->degraded is measured against the larger of the | |
7499 | * pre and post number of devices. | |
7500 | */ | |
ec32a2bd | 7501 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7502 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
7503 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7504 | } | |
63c70c4f | 7505 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7506 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 7507 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 7508 | |
29269553 N |
7509 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7510 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7511 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7512 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7513 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7514 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7515 | "reshape"); |
29269553 N |
7516 | if (!mddev->sync_thread) { |
7517 | mddev->recovery = 0; | |
7518 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7519 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7520 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7521 | mddev->new_chunk_sectors = |
7522 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7523 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7524 | rdev_for_each(rdev, mddev) |
7525 | rdev->new_data_offset = rdev->data_offset; | |
7526 | smp_wmb(); | |
ba8805b9 | 7527 | conf->generation --; |
fef9c61f | 7528 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7529 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7530 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7531 | spin_unlock_irq(&conf->device_lock); |
7532 | return -EAGAIN; | |
7533 | } | |
c8f517c4 | 7534 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7535 | md_wakeup_thread(mddev->sync_thread); |
7536 | md_new_event(mddev); | |
7537 | return 0; | |
7538 | } | |
29269553 | 7539 | |
ec32a2bd N |
7540 | /* This is called from the reshape thread and should make any |
7541 | * changes needed in 'conf' | |
7542 | */ | |
d1688a6d | 7543 | static void end_reshape(struct r5conf *conf) |
29269553 | 7544 | { |
29269553 | 7545 | |
f6705578 | 7546 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7547 | struct md_rdev *rdev; |
f6705578 | 7548 | |
f6705578 | 7549 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7550 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7551 | rdev_for_each(rdev, conf->mddev) |
7552 | rdev->data_offset = rdev->new_data_offset; | |
7553 | smp_wmb(); | |
fef9c61f | 7554 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7555 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 7556 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7557 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7558 | |
7559 | /* read-ahead size must cover two whole stripes, which is | |
7560 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7561 | */ | |
4a5add49 | 7562 | if (conf->mddev->queue) { |
cea9c228 | 7563 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7564 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7565 | / PAGE_SIZE); |
16a53ecc N |
7566 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
7567 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
7568 | } | |
29269553 | 7569 | } |
29269553 N |
7570 | } |
7571 | ||
ec32a2bd N |
7572 | /* This is called from the raid5d thread with mddev_lock held. |
7573 | * It makes config changes to the device. | |
7574 | */ | |
fd01b88c | 7575 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7576 | { |
d1688a6d | 7577 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7578 | |
7579 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7580 | ||
ec32a2bd N |
7581 | if (mddev->delta_disks > 0) { |
7582 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
fe67d19a HM |
7583 | if (mddev->queue) { |
7584 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
7585 | revalidate_disk(mddev->gendisk); | |
7586 | } | |
ec32a2bd N |
7587 | } else { |
7588 | int d; | |
908f4fbd N |
7589 | spin_lock_irq(&conf->device_lock); |
7590 | mddev->degraded = calc_degraded(conf); | |
7591 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
7592 | for (d = conf->raid_disks ; |
7593 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7594 | d++) { |
3cb03002 | 7595 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7596 | if (rdev) |
7597 | clear_bit(In_sync, &rdev->flags); | |
7598 | rdev = conf->disks[d].replacement; | |
7599 | if (rdev) | |
7600 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7601 | } |
cea9c228 | 7602 | } |
88ce4930 | 7603 | mddev->layout = conf->algorithm; |
09c9e5fa | 7604 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7605 | mddev->reshape_position = MaxSector; |
7606 | mddev->delta_disks = 0; | |
2c810cdd | 7607 | mddev->reshape_backwards = 0; |
cea9c228 N |
7608 | } |
7609 | } | |
7610 | ||
fd01b88c | 7611 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7612 | { |
d1688a6d | 7613 | struct r5conf *conf = mddev->private; |
72626685 N |
7614 | |
7615 | switch(state) { | |
e464eafd N |
7616 | case 2: /* resume for a suspend */ |
7617 | wake_up(&conf->wait_for_overlap); | |
7618 | break; | |
7619 | ||
72626685 | 7620 | case 1: /* stop all writes */ |
566c09c5 | 7621 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7622 | /* '2' tells resync/reshape to pause so that all |
7623 | * active stripes can drain | |
7624 | */ | |
7625 | conf->quiesce = 2; | |
b1b46486 | 7626 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7627 | atomic_read(&conf->active_stripes) == 0 && |
7628 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7629 | unlock_all_device_hash_locks_irq(conf), |
7630 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7631 | conf->quiesce = 1; |
566c09c5 | 7632 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7633 | /* allow reshape to continue */ |
7634 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7635 | break; |
7636 | ||
7637 | case 0: /* re-enable writes */ | |
566c09c5 | 7638 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7639 | conf->quiesce = 0; |
b1b46486 | 7640 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7641 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7642 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7643 | break; |
7644 | } | |
e6c033f7 | 7645 | r5l_quiesce(conf->log, state); |
72626685 | 7646 | } |
b15c2e57 | 7647 | |
fd01b88c | 7648 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7649 | { |
e373ab10 | 7650 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7651 | sector_t sectors; |
54071b38 | 7652 | |
f1b29bca | 7653 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7654 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
7655 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7656 | mdname(mddev)); | |
f1b29bca DW |
7657 | return ERR_PTR(-EINVAL); |
7658 | } | |
7659 | ||
e373ab10 N |
7660 | sectors = raid0_conf->strip_zone[0].zone_end; |
7661 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7662 | mddev->dev_sectors = sectors; |
f1b29bca | 7663 | mddev->new_level = level; |
54071b38 TM |
7664 | mddev->new_layout = ALGORITHM_PARITY_N; |
7665 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7666 | mddev->raid_disks += 1; | |
7667 | mddev->delta_disks = 1; | |
7668 | /* make sure it will be not marked as dirty */ | |
7669 | mddev->recovery_cp = MaxSector; | |
7670 | ||
7671 | return setup_conf(mddev); | |
7672 | } | |
7673 | ||
fd01b88c | 7674 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7675 | { |
7676 | int chunksect; | |
7677 | ||
7678 | if (mddev->raid_disks != 2 || | |
7679 | mddev->degraded > 1) | |
7680 | return ERR_PTR(-EINVAL); | |
7681 | ||
7682 | /* Should check if there are write-behind devices? */ | |
7683 | ||
7684 | chunksect = 64*2; /* 64K by default */ | |
7685 | ||
7686 | /* The array must be an exact multiple of chunksize */ | |
7687 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
7688 | chunksect >>= 1; | |
7689 | ||
7690 | if ((chunksect<<9) < STRIPE_SIZE) | |
7691 | /* array size does not allow a suitable chunk size */ | |
7692 | return ERR_PTR(-EINVAL); | |
7693 | ||
7694 | mddev->new_level = 5; | |
7695 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 7696 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
7697 | |
7698 | return setup_conf(mddev); | |
7699 | } | |
7700 | ||
fd01b88c | 7701 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
7702 | { |
7703 | int new_layout; | |
7704 | ||
7705 | switch (mddev->layout) { | |
7706 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7707 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7708 | break; | |
7709 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7710 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
7711 | break; | |
7712 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7713 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
7714 | break; | |
7715 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7716 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
7717 | break; | |
7718 | case ALGORITHM_PARITY_0_6: | |
7719 | new_layout = ALGORITHM_PARITY_0; | |
7720 | break; | |
7721 | case ALGORITHM_PARITY_N: | |
7722 | new_layout = ALGORITHM_PARITY_N; | |
7723 | break; | |
7724 | default: | |
7725 | return ERR_PTR(-EINVAL); | |
7726 | } | |
7727 | mddev->new_level = 5; | |
7728 | mddev->new_layout = new_layout; | |
7729 | mddev->delta_disks = -1; | |
7730 | mddev->raid_disks -= 1; | |
7731 | return setup_conf(mddev); | |
7732 | } | |
7733 | ||
fd01b88c | 7734 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 7735 | { |
88ce4930 N |
7736 | /* For a 2-drive array, the layout and chunk size can be changed |
7737 | * immediately as not restriping is needed. | |
7738 | * For larger arrays we record the new value - after validation | |
7739 | * to be used by a reshape pass. | |
b3546035 | 7740 | */ |
d1688a6d | 7741 | struct r5conf *conf = mddev->private; |
597a711b | 7742 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 7743 | |
597a711b | 7744 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
7745 | return -EINVAL; |
7746 | if (new_chunk > 0) { | |
0ba459d2 | 7747 | if (!is_power_of_2(new_chunk)) |
b3546035 | 7748 | return -EINVAL; |
597a711b | 7749 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 7750 | return -EINVAL; |
597a711b | 7751 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
7752 | /* not factor of array size */ |
7753 | return -EINVAL; | |
7754 | } | |
7755 | ||
7756 | /* They look valid */ | |
7757 | ||
88ce4930 | 7758 | if (mddev->raid_disks == 2) { |
597a711b N |
7759 | /* can make the change immediately */ |
7760 | if (mddev->new_layout >= 0) { | |
7761 | conf->algorithm = mddev->new_layout; | |
7762 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
7763 | } |
7764 | if (new_chunk > 0) { | |
597a711b N |
7765 | conf->chunk_sectors = new_chunk ; |
7766 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
7767 | } |
7768 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7769 | md_wakeup_thread(mddev->thread); | |
b3546035 | 7770 | } |
50ac168a | 7771 | return check_reshape(mddev); |
88ce4930 N |
7772 | } |
7773 | ||
fd01b88c | 7774 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 7775 | { |
597a711b | 7776 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 7777 | |
597a711b | 7778 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 7779 | return -EINVAL; |
b3546035 | 7780 | if (new_chunk > 0) { |
0ba459d2 | 7781 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 7782 | return -EINVAL; |
597a711b | 7783 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 7784 | return -EINVAL; |
597a711b | 7785 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
7786 | /* not factor of array size */ |
7787 | return -EINVAL; | |
b3546035 | 7788 | } |
88ce4930 N |
7789 | |
7790 | /* They look valid */ | |
50ac168a | 7791 | return check_reshape(mddev); |
b3546035 N |
7792 | } |
7793 | ||
fd01b88c | 7794 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
7795 | { |
7796 | /* raid5 can take over: | |
f1b29bca | 7797 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
7798 | * raid1 - if there are two drives. We need to know the chunk size |
7799 | * raid4 - trivial - just use a raid4 layout. | |
7800 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 7801 | */ |
f1b29bca DW |
7802 | if (mddev->level == 0) |
7803 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
7804 | if (mddev->level == 1) |
7805 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
7806 | if (mddev->level == 4) { |
7807 | mddev->new_layout = ALGORITHM_PARITY_N; | |
7808 | mddev->new_level = 5; | |
7809 | return setup_conf(mddev); | |
7810 | } | |
fc9739c6 N |
7811 | if (mddev->level == 6) |
7812 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
7813 | |
7814 | return ERR_PTR(-EINVAL); | |
7815 | } | |
7816 | ||
fd01b88c | 7817 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 7818 | { |
f1b29bca DW |
7819 | /* raid4 can take over: |
7820 | * raid0 - if there is only one strip zone | |
7821 | * raid5 - if layout is right | |
a78d38a1 | 7822 | */ |
f1b29bca DW |
7823 | if (mddev->level == 0) |
7824 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
7825 | if (mddev->level == 5 && |
7826 | mddev->layout == ALGORITHM_PARITY_N) { | |
7827 | mddev->new_layout = 0; | |
7828 | mddev->new_level = 4; | |
7829 | return setup_conf(mddev); | |
7830 | } | |
7831 | return ERR_PTR(-EINVAL); | |
7832 | } | |
d562b0c4 | 7833 | |
84fc4b56 | 7834 | static struct md_personality raid5_personality; |
245f46c2 | 7835 | |
fd01b88c | 7836 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
7837 | { |
7838 | /* Currently can only take over a raid5. We map the | |
7839 | * personality to an equivalent raid6 personality | |
7840 | * with the Q block at the end. | |
7841 | */ | |
7842 | int new_layout; | |
7843 | ||
7844 | if (mddev->pers != &raid5_personality) | |
7845 | return ERR_PTR(-EINVAL); | |
7846 | if (mddev->degraded > 1) | |
7847 | return ERR_PTR(-EINVAL); | |
7848 | if (mddev->raid_disks > 253) | |
7849 | return ERR_PTR(-EINVAL); | |
7850 | if (mddev->raid_disks < 3) | |
7851 | return ERR_PTR(-EINVAL); | |
7852 | ||
7853 | switch (mddev->layout) { | |
7854 | case ALGORITHM_LEFT_ASYMMETRIC: | |
7855 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
7856 | break; | |
7857 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
7858 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
7859 | break; | |
7860 | case ALGORITHM_LEFT_SYMMETRIC: | |
7861 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
7862 | break; | |
7863 | case ALGORITHM_RIGHT_SYMMETRIC: | |
7864 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
7865 | break; | |
7866 | case ALGORITHM_PARITY_0: | |
7867 | new_layout = ALGORITHM_PARITY_0_6; | |
7868 | break; | |
7869 | case ALGORITHM_PARITY_N: | |
7870 | new_layout = ALGORITHM_PARITY_N; | |
7871 | break; | |
7872 | default: | |
7873 | return ERR_PTR(-EINVAL); | |
7874 | } | |
7875 | mddev->new_level = 6; | |
7876 | mddev->new_layout = new_layout; | |
7877 | mddev->delta_disks = 1; | |
7878 | mddev->raid_disks += 1; | |
7879 | return setup_conf(mddev); | |
7880 | } | |
7881 | ||
84fc4b56 | 7882 | static struct md_personality raid6_personality = |
16a53ecc N |
7883 | { |
7884 | .name = "raid6", | |
7885 | .level = 6, | |
7886 | .owner = THIS_MODULE, | |
849674e4 SL |
7887 | .make_request = raid5_make_request, |
7888 | .run = raid5_run, | |
afa0f557 | 7889 | .free = raid5_free, |
849674e4 SL |
7890 | .status = raid5_status, |
7891 | .error_handler = raid5_error, | |
16a53ecc N |
7892 | .hot_add_disk = raid5_add_disk, |
7893 | .hot_remove_disk= raid5_remove_disk, | |
7894 | .spare_active = raid5_spare_active, | |
849674e4 | 7895 | .sync_request = raid5_sync_request, |
16a53ecc | 7896 | .resize = raid5_resize, |
80c3a6ce | 7897 | .size = raid5_size, |
50ac168a | 7898 | .check_reshape = raid6_check_reshape, |
f416885e | 7899 | .start_reshape = raid5_start_reshape, |
cea9c228 | 7900 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 7901 | .quiesce = raid5_quiesce, |
245f46c2 | 7902 | .takeover = raid6_takeover, |
5c675f83 | 7903 | .congested = raid5_congested, |
16a53ecc | 7904 | }; |
84fc4b56 | 7905 | static struct md_personality raid5_personality = |
1da177e4 LT |
7906 | { |
7907 | .name = "raid5", | |
2604b703 | 7908 | .level = 5, |
1da177e4 | 7909 | .owner = THIS_MODULE, |
849674e4 SL |
7910 | .make_request = raid5_make_request, |
7911 | .run = raid5_run, | |
afa0f557 | 7912 | .free = raid5_free, |
849674e4 SL |
7913 | .status = raid5_status, |
7914 | .error_handler = raid5_error, | |
1da177e4 LT |
7915 | .hot_add_disk = raid5_add_disk, |
7916 | .hot_remove_disk= raid5_remove_disk, | |
7917 | .spare_active = raid5_spare_active, | |
849674e4 | 7918 | .sync_request = raid5_sync_request, |
1da177e4 | 7919 | .resize = raid5_resize, |
80c3a6ce | 7920 | .size = raid5_size, |
63c70c4f N |
7921 | .check_reshape = raid5_check_reshape, |
7922 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7923 | .finish_reshape = raid5_finish_reshape, |
72626685 | 7924 | .quiesce = raid5_quiesce, |
d562b0c4 | 7925 | .takeover = raid5_takeover, |
5c675f83 | 7926 | .congested = raid5_congested, |
1da177e4 LT |
7927 | }; |
7928 | ||
84fc4b56 | 7929 | static struct md_personality raid4_personality = |
1da177e4 | 7930 | { |
2604b703 N |
7931 | .name = "raid4", |
7932 | .level = 4, | |
7933 | .owner = THIS_MODULE, | |
849674e4 SL |
7934 | .make_request = raid5_make_request, |
7935 | .run = raid5_run, | |
afa0f557 | 7936 | .free = raid5_free, |
849674e4 SL |
7937 | .status = raid5_status, |
7938 | .error_handler = raid5_error, | |
2604b703 N |
7939 | .hot_add_disk = raid5_add_disk, |
7940 | .hot_remove_disk= raid5_remove_disk, | |
7941 | .spare_active = raid5_spare_active, | |
849674e4 | 7942 | .sync_request = raid5_sync_request, |
2604b703 | 7943 | .resize = raid5_resize, |
80c3a6ce | 7944 | .size = raid5_size, |
3d37890b N |
7945 | .check_reshape = raid5_check_reshape, |
7946 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7947 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 7948 | .quiesce = raid5_quiesce, |
a78d38a1 | 7949 | .takeover = raid4_takeover, |
5c675f83 | 7950 | .congested = raid5_congested, |
2604b703 N |
7951 | }; |
7952 | ||
7953 | static int __init raid5_init(void) | |
7954 | { | |
29c6d1bb SAS |
7955 | int ret; |
7956 | ||
851c30c9 SL |
7957 | raid5_wq = alloc_workqueue("raid5wq", |
7958 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
7959 | if (!raid5_wq) | |
7960 | return -ENOMEM; | |
29c6d1bb SAS |
7961 | |
7962 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
7963 | "md/raid5:prepare", | |
7964 | raid456_cpu_up_prepare, | |
7965 | raid456_cpu_dead); | |
7966 | if (ret) { | |
7967 | destroy_workqueue(raid5_wq); | |
7968 | return ret; | |
7969 | } | |
16a53ecc | 7970 | register_md_personality(&raid6_personality); |
2604b703 N |
7971 | register_md_personality(&raid5_personality); |
7972 | register_md_personality(&raid4_personality); | |
7973 | return 0; | |
1da177e4 LT |
7974 | } |
7975 | ||
2604b703 | 7976 | static void raid5_exit(void) |
1da177e4 | 7977 | { |
16a53ecc | 7978 | unregister_md_personality(&raid6_personality); |
2604b703 N |
7979 | unregister_md_personality(&raid5_personality); |
7980 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 7981 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 7982 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
7983 | } |
7984 | ||
7985 | module_init(raid5_init); | |
7986 | module_exit(raid5_exit); | |
7987 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7988 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 7989 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
7990 | MODULE_ALIAS("md-raid5"); |
7991 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
7992 | MODULE_ALIAS("md-level-5"); |
7993 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
7994 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
7995 | MODULE_ALIAS("md-raid6"); | |
7996 | MODULE_ALIAS("md-level-6"); | |
7997 | ||
7998 | /* This used to be two separate modules, they were: */ | |
7999 | MODULE_ALIAS("raid5"); | |
8000 | MODULE_ALIAS("raid6"); |