Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
851c30c9 | 56 | #include <linux/nodemask.h> |
46d5b785 | 57 | #include <linux/flex_array.h> |
a9add5d9 N |
58 | #include <trace/events/block.h> |
59 | ||
43b2e5d8 | 60 | #include "md.h" |
bff61975 | 61 | #include "raid5.h" |
54071b38 | 62 | #include "raid0.h" |
ef740c37 | 63 | #include "bitmap.h" |
72626685 | 64 | |
851c30c9 SL |
65 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
66 | #define ANY_GROUP NUMA_NO_NODE | |
67 | ||
8e0e99ba N |
68 | static bool devices_handle_discard_safely = false; |
69 | module_param(devices_handle_discard_safely, bool, 0644); | |
70 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
71 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 72 | static struct workqueue_struct *raid5_wq; |
1da177e4 LT |
73 | /* |
74 | * Stripe cache | |
75 | */ | |
76 | ||
77 | #define NR_STRIPES 256 | |
78 | #define STRIPE_SIZE PAGE_SIZE | |
79 | #define STRIPE_SHIFT (PAGE_SHIFT - 9) | |
80 | #define STRIPE_SECTORS (STRIPE_SIZE>>9) | |
81 | #define IO_THRESHOLD 1 | |
8b3e6cdc | 82 | #define BYPASS_THRESHOLD 1 |
fccddba0 | 83 | #define NR_HASH (PAGE_SIZE / sizeof(struct hlist_head)) |
1da177e4 | 84 | #define HASH_MASK (NR_HASH - 1) |
bfc90cb0 | 85 | #define MAX_STRIPE_BATCH 8 |
1da177e4 | 86 | |
d1688a6d | 87 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
88 | { |
89 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
90 | return &conf->stripe_hashtbl[hash]; | |
91 | } | |
1da177e4 | 92 | |
566c09c5 SL |
93 | static inline int stripe_hash_locks_hash(sector_t sect) |
94 | { | |
95 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
96 | } | |
97 | ||
98 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
99 | { | |
100 | spin_lock_irq(conf->hash_locks + hash); | |
101 | spin_lock(&conf->device_lock); | |
102 | } | |
103 | ||
104 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
105 | { | |
106 | spin_unlock(&conf->device_lock); | |
107 | spin_unlock_irq(conf->hash_locks + hash); | |
108 | } | |
109 | ||
110 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
111 | { | |
112 | int i; | |
113 | local_irq_disable(); | |
114 | spin_lock(conf->hash_locks); | |
115 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
116 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
117 | spin_lock(&conf->device_lock); | |
118 | } | |
119 | ||
120 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
121 | { | |
122 | int i; | |
123 | spin_unlock(&conf->device_lock); | |
124 | for (i = NR_STRIPE_HASH_LOCKS; i; i--) | |
125 | spin_unlock(conf->hash_locks + i - 1); | |
126 | local_irq_enable(); | |
127 | } | |
128 | ||
1da177e4 LT |
129 | /* bio's attached to a stripe+device for I/O are linked together in bi_sector |
130 | * order without overlap. There may be several bio's per stripe+device, and | |
131 | * a bio could span several devices. | |
132 | * When walking this list for a particular stripe+device, we must never proceed | |
133 | * beyond a bio that extends past this device, as the next bio might no longer | |
134 | * be valid. | |
db298e19 | 135 | * This function is used to determine the 'next' bio in the list, given the sector |
1da177e4 LT |
136 | * of the current stripe+device |
137 | */ | |
db298e19 N |
138 | static inline struct bio *r5_next_bio(struct bio *bio, sector_t sector) |
139 | { | |
aa8b57aa | 140 | int sectors = bio_sectors(bio); |
4f024f37 | 141 | if (bio->bi_iter.bi_sector + sectors < sector + STRIPE_SECTORS) |
db298e19 N |
142 | return bio->bi_next; |
143 | else | |
144 | return NULL; | |
145 | } | |
1da177e4 | 146 | |
960e739d | 147 | /* |
5b99c2ff JA |
148 | * We maintain a biased count of active stripes in the bottom 16 bits of |
149 | * bi_phys_segments, and a count of processed stripes in the upper 16 bits | |
960e739d | 150 | */ |
e7836bd6 | 151 | static inline int raid5_bi_processed_stripes(struct bio *bio) |
960e739d | 152 | { |
e7836bd6 SL |
153 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
154 | return (atomic_read(segments) >> 16) & 0xffff; | |
960e739d JA |
155 | } |
156 | ||
e7836bd6 | 157 | static inline int raid5_dec_bi_active_stripes(struct bio *bio) |
960e739d | 158 | { |
e7836bd6 SL |
159 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
160 | return atomic_sub_return(1, segments) & 0xffff; | |
960e739d JA |
161 | } |
162 | ||
e7836bd6 | 163 | static inline void raid5_inc_bi_active_stripes(struct bio *bio) |
960e739d | 164 | { |
e7836bd6 SL |
165 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
166 | atomic_inc(segments); | |
960e739d JA |
167 | } |
168 | ||
e7836bd6 SL |
169 | static inline void raid5_set_bi_processed_stripes(struct bio *bio, |
170 | unsigned int cnt) | |
960e739d | 171 | { |
e7836bd6 SL |
172 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
173 | int old, new; | |
960e739d | 174 | |
e7836bd6 SL |
175 | do { |
176 | old = atomic_read(segments); | |
177 | new = (old & 0xffff) | (cnt << 16); | |
178 | } while (atomic_cmpxchg(segments, old, new) != old); | |
960e739d JA |
179 | } |
180 | ||
e7836bd6 | 181 | static inline void raid5_set_bi_stripes(struct bio *bio, unsigned int cnt) |
960e739d | 182 | { |
e7836bd6 SL |
183 | atomic_t *segments = (atomic_t *)&bio->bi_phys_segments; |
184 | atomic_set(segments, cnt); | |
960e739d JA |
185 | } |
186 | ||
d0dabf7e N |
187 | /* Find first data disk in a raid6 stripe */ |
188 | static inline int raid6_d0(struct stripe_head *sh) | |
189 | { | |
67cc2b81 N |
190 | if (sh->ddf_layout) |
191 | /* ddf always start from first device */ | |
192 | return 0; | |
193 | /* md starts just after Q block */ | |
d0dabf7e N |
194 | if (sh->qd_idx == sh->disks - 1) |
195 | return 0; | |
196 | else | |
197 | return sh->qd_idx + 1; | |
198 | } | |
16a53ecc N |
199 | static inline int raid6_next_disk(int disk, int raid_disks) |
200 | { | |
201 | disk++; | |
202 | return (disk < raid_disks) ? disk : 0; | |
203 | } | |
a4456856 | 204 | |
d0dabf7e N |
205 | /* When walking through the disks in a raid5, starting at raid6_d0, |
206 | * We need to map each disk to a 'slot', where the data disks are slot | |
207 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
208 | * is raid_disks-1. This help does that mapping. | |
209 | */ | |
67cc2b81 N |
210 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
211 | int *count, int syndrome_disks) | |
d0dabf7e | 212 | { |
6629542e | 213 | int slot = *count; |
67cc2b81 | 214 | |
e4424fee | 215 | if (sh->ddf_layout) |
6629542e | 216 | (*count)++; |
d0dabf7e | 217 | if (idx == sh->pd_idx) |
67cc2b81 | 218 | return syndrome_disks; |
d0dabf7e | 219 | if (idx == sh->qd_idx) |
67cc2b81 | 220 | return syndrome_disks + 1; |
e4424fee | 221 | if (!sh->ddf_layout) |
6629542e | 222 | (*count)++; |
d0dabf7e N |
223 | return slot; |
224 | } | |
225 | ||
a4456856 DW |
226 | static void return_io(struct bio *return_bi) |
227 | { | |
228 | struct bio *bi = return_bi; | |
229 | while (bi) { | |
a4456856 DW |
230 | |
231 | return_bi = bi->bi_next; | |
232 | bi->bi_next = NULL; | |
4f024f37 | 233 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
234 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
235 | bi, 0); | |
0e13fe23 | 236 | bio_endio(bi, 0); |
a4456856 DW |
237 | bi = return_bi; |
238 | } | |
239 | } | |
240 | ||
d1688a6d | 241 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 242 | |
600aa109 DW |
243 | static int stripe_operations_active(struct stripe_head *sh) |
244 | { | |
245 | return sh->check_state || sh->reconstruct_state || | |
246 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
247 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
248 | } | |
249 | ||
851c30c9 SL |
250 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
251 | { | |
252 | struct r5conf *conf = sh->raid_conf; | |
253 | struct r5worker_group *group; | |
bfc90cb0 | 254 | int thread_cnt; |
851c30c9 SL |
255 | int i, cpu = sh->cpu; |
256 | ||
257 | if (!cpu_online(cpu)) { | |
258 | cpu = cpumask_any(cpu_online_mask); | |
259 | sh->cpu = cpu; | |
260 | } | |
261 | ||
262 | if (list_empty(&sh->lru)) { | |
263 | struct r5worker_group *group; | |
264 | group = conf->worker_groups + cpu_to_group(cpu); | |
265 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
266 | group->stripes_cnt++; |
267 | sh->group = group; | |
851c30c9 SL |
268 | } |
269 | ||
270 | if (conf->worker_cnt_per_group == 0) { | |
271 | md_wakeup_thread(conf->mddev->thread); | |
272 | return; | |
273 | } | |
274 | ||
275 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
276 | ||
bfc90cb0 SL |
277 | group->workers[0].working = true; |
278 | /* at least one worker should run to avoid race */ | |
279 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
280 | ||
281 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
282 | /* wakeup more workers */ | |
283 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
284 | if (group->workers[i].working == false) { | |
285 | group->workers[i].working = true; | |
286 | queue_work_on(sh->cpu, raid5_wq, | |
287 | &group->workers[i].work); | |
288 | thread_cnt--; | |
289 | } | |
290 | } | |
851c30c9 SL |
291 | } |
292 | ||
566c09c5 SL |
293 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
294 | struct list_head *temp_inactive_list) | |
1da177e4 | 295 | { |
4eb788df SL |
296 | BUG_ON(!list_empty(&sh->lru)); |
297 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
298 | if (test_bit(STRIPE_HANDLE, &sh->state)) { | |
299 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 300 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 301 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 302 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
303 | sh->bm_seq - conf->seq_write > 0) |
304 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
305 | else { | |
306 | clear_bit(STRIPE_DELAYED, &sh->state); | |
307 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 SL |
308 | if (conf->worker_cnt_per_group == 0) { |
309 | list_add_tail(&sh->lru, &conf->handle_list); | |
310 | } else { | |
311 | raid5_wakeup_stripe_thread(sh); | |
312 | return; | |
313 | } | |
4eb788df SL |
314 | } |
315 | md_wakeup_thread(conf->mddev->thread); | |
316 | } else { | |
317 | BUG_ON(stripe_operations_active(sh)); | |
318 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
319 | if (atomic_dec_return(&conf->preread_active_stripes) | |
320 | < IO_THRESHOLD) | |
321 | md_wakeup_thread(conf->mddev->thread); | |
322 | atomic_dec(&conf->active_stripes); | |
566c09c5 SL |
323 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) |
324 | list_add_tail(&sh->lru, temp_inactive_list); | |
1da177e4 LT |
325 | } |
326 | } | |
d0dabf7e | 327 | |
566c09c5 SL |
328 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
329 | struct list_head *temp_inactive_list) | |
4eb788df SL |
330 | { |
331 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
332 | do_release_stripe(conf, sh, temp_inactive_list); |
333 | } | |
334 | ||
335 | /* | |
336 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
337 | * | |
338 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
339 | * given time. Adding stripes only takes device lock, while deleting stripes | |
340 | * only takes hash lock. | |
341 | */ | |
342 | static void release_inactive_stripe_list(struct r5conf *conf, | |
343 | struct list_head *temp_inactive_list, | |
344 | int hash) | |
345 | { | |
346 | int size; | |
e9e4c377 YL |
347 | unsigned long do_wakeup = 0; |
348 | int i = 0; | |
566c09c5 SL |
349 | unsigned long flags; |
350 | ||
351 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
352 | size = NR_STRIPE_HASH_LOCKS; | |
353 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
354 | } else | |
355 | size = 1; | |
356 | while (size) { | |
357 | struct list_head *list = &temp_inactive_list[size - 1]; | |
358 | ||
359 | /* | |
360 | * We don't hold any lock here yet, get_active_stripe() might | |
361 | * remove stripes from the list | |
362 | */ | |
363 | if (!list_empty_careful(list)) { | |
364 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
365 | if (list_empty(conf->inactive_list + hash) && |
366 | !list_empty(list)) | |
367 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 368 | list_splice_tail_init(list, conf->inactive_list + hash); |
e9e4c377 | 369 | do_wakeup |= 1 << hash; |
566c09c5 SL |
370 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
371 | } | |
372 | size--; | |
373 | hash--; | |
374 | } | |
375 | ||
e9e4c377 YL |
376 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) { |
377 | if (do_wakeup & (1 << i)) | |
378 | wake_up(&conf->wait_for_stripe[i]); | |
379 | } | |
380 | ||
566c09c5 | 381 | if (do_wakeup) { |
b1b46486 YL |
382 | if (atomic_read(&conf->active_stripes) == 0) |
383 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
384 | if (conf->retry_read_aligned) |
385 | md_wakeup_thread(conf->mddev->thread); | |
386 | } | |
4eb788df SL |
387 | } |
388 | ||
773ca82f | 389 | /* should hold conf->device_lock already */ |
566c09c5 SL |
390 | static int release_stripe_list(struct r5conf *conf, |
391 | struct list_head *temp_inactive_list) | |
773ca82f SL |
392 | { |
393 | struct stripe_head *sh; | |
394 | int count = 0; | |
395 | struct llist_node *head; | |
396 | ||
397 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 398 | head = llist_reverse_order(head); |
773ca82f | 399 | while (head) { |
566c09c5 SL |
400 | int hash; |
401 | ||
773ca82f SL |
402 | sh = llist_entry(head, struct stripe_head, release_list); |
403 | head = llist_next(head); | |
404 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ | |
405 | smp_mb(); | |
406 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
407 | /* | |
408 | * Don't worry the bit is set here, because if the bit is set | |
409 | * again, the count is always > 1. This is true for | |
410 | * STRIPE_ON_UNPLUG_LIST bit too. | |
411 | */ | |
566c09c5 SL |
412 | hash = sh->hash_lock_index; |
413 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
414 | count++; |
415 | } | |
416 | ||
417 | return count; | |
418 | } | |
419 | ||
1da177e4 LT |
420 | static void release_stripe(struct stripe_head *sh) |
421 | { | |
d1688a6d | 422 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 423 | unsigned long flags; |
566c09c5 SL |
424 | struct list_head list; |
425 | int hash; | |
773ca82f | 426 | bool wakeup; |
16a53ecc | 427 | |
cf170f3f ES |
428 | /* Avoid release_list until the last reference. |
429 | */ | |
430 | if (atomic_add_unless(&sh->count, -1, 1)) | |
431 | return; | |
432 | ||
ad4068de | 433 | if (unlikely(!conf->mddev->thread) || |
434 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
435 | goto slow_path; |
436 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
437 | if (wakeup) | |
438 | md_wakeup_thread(conf->mddev->thread); | |
439 | return; | |
440 | slow_path: | |
4eb788df | 441 | local_irq_save(flags); |
773ca82f | 442 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 443 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
444 | INIT_LIST_HEAD(&list); |
445 | hash = sh->hash_lock_index; | |
446 | do_release_stripe(conf, sh, &list); | |
4eb788df | 447 | spin_unlock(&conf->device_lock); |
566c09c5 | 448 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
449 | } |
450 | local_irq_restore(flags); | |
1da177e4 LT |
451 | } |
452 | ||
fccddba0 | 453 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 454 | { |
45b4233c DW |
455 | pr_debug("remove_hash(), stripe %llu\n", |
456 | (unsigned long long)sh->sector); | |
1da177e4 | 457 | |
fccddba0 | 458 | hlist_del_init(&sh->hash); |
1da177e4 LT |
459 | } |
460 | ||
d1688a6d | 461 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 462 | { |
fccddba0 | 463 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 464 | |
45b4233c DW |
465 | pr_debug("insert_hash(), stripe %llu\n", |
466 | (unsigned long long)sh->sector); | |
1da177e4 | 467 | |
fccddba0 | 468 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
469 | } |
470 | ||
1da177e4 | 471 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 472 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
473 | { |
474 | struct stripe_head *sh = NULL; | |
475 | struct list_head *first; | |
476 | ||
566c09c5 | 477 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 478 | goto out; |
566c09c5 | 479 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
480 | sh = list_entry(first, struct stripe_head, lru); |
481 | list_del_init(first); | |
482 | remove_hash(sh); | |
483 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 484 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
485 | if (list_empty(conf->inactive_list + hash)) |
486 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
487 | out: |
488 | return sh; | |
489 | } | |
490 | ||
e4e11e38 | 491 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
492 | { |
493 | struct page *p; | |
494 | int i; | |
e4e11e38 | 495 | int num = sh->raid_conf->pool_size; |
1da177e4 | 496 | |
e4e11e38 | 497 | for (i = 0; i < num ; i++) { |
d592a996 | 498 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
499 | p = sh->dev[i].page; |
500 | if (!p) | |
501 | continue; | |
502 | sh->dev[i].page = NULL; | |
2d1f3b5d | 503 | put_page(p); |
1da177e4 LT |
504 | } |
505 | } | |
506 | ||
a9683a79 | 507 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
508 | { |
509 | int i; | |
e4e11e38 | 510 | int num = sh->raid_conf->pool_size; |
1da177e4 | 511 | |
e4e11e38 | 512 | for (i = 0; i < num; i++) { |
1da177e4 LT |
513 | struct page *page; |
514 | ||
a9683a79 | 515 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
516 | return 1; |
517 | } | |
518 | sh->dev[i].page = page; | |
d592a996 | 519 | sh->dev[i].orig_page = page; |
1da177e4 LT |
520 | } |
521 | return 0; | |
522 | } | |
523 | ||
784052ec | 524 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 525 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 526 | struct stripe_head *sh); |
1da177e4 | 527 | |
b5663ba4 | 528 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 529 | { |
d1688a6d | 530 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 531 | int i, seq; |
1da177e4 | 532 | |
78bafebd ES |
533 | BUG_ON(atomic_read(&sh->count) != 0); |
534 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 535 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 536 | BUG_ON(sh->batch_head); |
d84e0f10 | 537 | |
45b4233c | 538 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 539 | (unsigned long long)sector); |
566c09c5 SL |
540 | retry: |
541 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 542 | sh->generation = conf->generation - previous; |
b5663ba4 | 543 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 544 | sh->sector = sector; |
911d4ee8 | 545 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
546 | sh->state = 0; |
547 | ||
7ecaa1e6 | 548 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
549 | struct r5dev *dev = &sh->dev[i]; |
550 | ||
d84e0f10 | 551 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 552 | test_bit(R5_LOCKED, &dev->flags)) { |
d84e0f10 | 553 | printk(KERN_ERR "sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 554 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 555 | dev->read, dev->towrite, dev->written, |
1da177e4 | 556 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 557 | WARN_ON(1); |
1da177e4 LT |
558 | } |
559 | dev->flags = 0; | |
784052ec | 560 | raid5_build_block(sh, i, previous); |
1da177e4 | 561 | } |
566c09c5 SL |
562 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
563 | goto retry; | |
7a87f434 | 564 | sh->overwrite_disks = 0; |
1da177e4 | 565 | insert_hash(conf, sh); |
851c30c9 | 566 | sh->cpu = smp_processor_id(); |
da41ba65 | 567 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
568 | } |
569 | ||
d1688a6d | 570 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 571 | short generation) |
1da177e4 LT |
572 | { |
573 | struct stripe_head *sh; | |
574 | ||
45b4233c | 575 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 576 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 577 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 578 | return sh; |
45b4233c | 579 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
580 | return NULL; |
581 | } | |
582 | ||
674806d6 N |
583 | /* |
584 | * Need to check if array has failed when deciding whether to: | |
585 | * - start an array | |
586 | * - remove non-faulty devices | |
587 | * - add a spare | |
588 | * - allow a reshape | |
589 | * This determination is simple when no reshape is happening. | |
590 | * However if there is a reshape, we need to carefully check | |
591 | * both the before and after sections. | |
592 | * This is because some failed devices may only affect one | |
593 | * of the two sections, and some non-in_sync devices may | |
594 | * be insync in the section most affected by failed devices. | |
595 | */ | |
908f4fbd | 596 | static int calc_degraded(struct r5conf *conf) |
674806d6 | 597 | { |
908f4fbd | 598 | int degraded, degraded2; |
674806d6 | 599 | int i; |
674806d6 N |
600 | |
601 | rcu_read_lock(); | |
602 | degraded = 0; | |
603 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 604 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
605 | if (rdev && test_bit(Faulty, &rdev->flags)) |
606 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
607 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
608 | degraded++; | |
609 | else if (test_bit(In_sync, &rdev->flags)) | |
610 | ; | |
611 | else | |
612 | /* not in-sync or faulty. | |
613 | * If the reshape increases the number of devices, | |
614 | * this is being recovered by the reshape, so | |
615 | * this 'previous' section is not in_sync. | |
616 | * If the number of devices is being reduced however, | |
617 | * the device can only be part of the array if | |
618 | * we are reverting a reshape, so this section will | |
619 | * be in-sync. | |
620 | */ | |
621 | if (conf->raid_disks >= conf->previous_raid_disks) | |
622 | degraded++; | |
623 | } | |
624 | rcu_read_unlock(); | |
908f4fbd N |
625 | if (conf->raid_disks == conf->previous_raid_disks) |
626 | return degraded; | |
674806d6 | 627 | rcu_read_lock(); |
908f4fbd | 628 | degraded2 = 0; |
674806d6 | 629 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 630 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
631 | if (rdev && test_bit(Faulty, &rdev->flags)) |
632 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 633 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 634 | degraded2++; |
674806d6 N |
635 | else if (test_bit(In_sync, &rdev->flags)) |
636 | ; | |
637 | else | |
638 | /* not in-sync or faulty. | |
639 | * If reshape increases the number of devices, this | |
640 | * section has already been recovered, else it | |
641 | * almost certainly hasn't. | |
642 | */ | |
643 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 644 | degraded2++; |
674806d6 N |
645 | } |
646 | rcu_read_unlock(); | |
908f4fbd N |
647 | if (degraded2 > degraded) |
648 | return degraded2; | |
649 | return degraded; | |
650 | } | |
651 | ||
652 | static int has_failed(struct r5conf *conf) | |
653 | { | |
654 | int degraded; | |
655 | ||
656 | if (conf->mddev->reshape_position == MaxSector) | |
657 | return conf->mddev->degraded > conf->max_degraded; | |
658 | ||
659 | degraded = calc_degraded(conf); | |
674806d6 N |
660 | if (degraded > conf->max_degraded) |
661 | return 1; | |
662 | return 0; | |
663 | } | |
664 | ||
b5663ba4 | 665 | static struct stripe_head * |
d1688a6d | 666 | get_active_stripe(struct r5conf *conf, sector_t sector, |
a8c906ca | 667 | int previous, int noblock, int noquiesce) |
1da177e4 LT |
668 | { |
669 | struct stripe_head *sh; | |
566c09c5 | 670 | int hash = stripe_hash_locks_hash(sector); |
1da177e4 | 671 | |
45b4233c | 672 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 673 | |
566c09c5 | 674 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
675 | |
676 | do { | |
b1b46486 | 677 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 678 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 679 | *(conf->hash_locks + hash)); |
86b42c71 | 680 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 681 | if (!sh) { |
edbe83ab | 682 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 683 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
684 | if (!sh && !test_bit(R5_DID_ALLOC, |
685 | &conf->cache_state)) | |
edbe83ab N |
686 | set_bit(R5_ALLOC_MORE, |
687 | &conf->cache_state); | |
688 | } | |
1da177e4 LT |
689 | if (noblock && sh == NULL) |
690 | break; | |
691 | if (!sh) { | |
5423399a N |
692 | set_bit(R5_INACTIVE_BLOCKED, |
693 | &conf->cache_state); | |
e9e4c377 YL |
694 | wait_event_exclusive_cmd( |
695 | conf->wait_for_stripe[hash], | |
566c09c5 SL |
696 | !list_empty(conf->inactive_list + hash) && |
697 | (atomic_read(&conf->active_stripes) | |
698 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
699 | || !test_bit(R5_INACTIVE_BLOCKED, |
700 | &conf->cache_state)), | |
e9e4c377 YL |
701 | spin_unlock_irq(conf->hash_locks + hash), |
702 | spin_lock_irq(conf->hash_locks + hash)); | |
5423399a N |
703 | clear_bit(R5_INACTIVE_BLOCKED, |
704 | &conf->cache_state); | |
7da9d450 | 705 | } else { |
b5663ba4 | 706 | init_stripe(sh, sector, previous); |
7da9d450 N |
707 | atomic_inc(&sh->count); |
708 | } | |
e240c183 | 709 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 710 | spin_lock(&conf->device_lock); |
e240c183 | 711 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
712 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
713 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
714 | BUG_ON(list_empty(&sh->lru) && |
715 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
16a53ecc | 716 | list_del_init(&sh->lru); |
bfc90cb0 SL |
717 | if (sh->group) { |
718 | sh->group->stripes_cnt--; | |
719 | sh->group = NULL; | |
720 | } | |
1da177e4 | 721 | } |
7da9d450 | 722 | atomic_inc(&sh->count); |
6d183de4 | 723 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
724 | } |
725 | } while (sh == NULL); | |
726 | ||
e9e4c377 YL |
727 | if (!list_empty(conf->inactive_list + hash)) |
728 | wake_up(&conf->wait_for_stripe[hash]); | |
729 | ||
566c09c5 | 730 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
731 | return sh; |
732 | } | |
733 | ||
7a87f434 | 734 | static bool is_full_stripe_write(struct stripe_head *sh) |
735 | { | |
736 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
737 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
738 | } | |
739 | ||
59fc630b | 740 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
741 | { | |
742 | local_irq_disable(); | |
743 | if (sh1 > sh2) { | |
744 | spin_lock(&sh2->stripe_lock); | |
745 | spin_lock_nested(&sh1->stripe_lock, 1); | |
746 | } else { | |
747 | spin_lock(&sh1->stripe_lock); | |
748 | spin_lock_nested(&sh2->stripe_lock, 1); | |
749 | } | |
750 | } | |
751 | ||
752 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
753 | { | |
754 | spin_unlock(&sh1->stripe_lock); | |
755 | spin_unlock(&sh2->stripe_lock); | |
756 | local_irq_enable(); | |
757 | } | |
758 | ||
759 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
760 | static bool stripe_can_batch(struct stripe_head *sh) | |
761 | { | |
762 | return test_bit(STRIPE_BATCH_READY, &sh->state) && | |
d0852df5 | 763 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 764 | is_full_stripe_write(sh); |
765 | } | |
766 | ||
767 | /* we only do back search */ | |
768 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
769 | { | |
770 | struct stripe_head *head; | |
771 | sector_t head_sector, tmp_sec; | |
772 | int hash; | |
773 | int dd_idx; | |
774 | ||
775 | if (!stripe_can_batch(sh)) | |
776 | return; | |
777 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ | |
778 | tmp_sec = sh->sector; | |
779 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
780 | return; | |
781 | head_sector = sh->sector - STRIPE_SECTORS; | |
782 | ||
783 | hash = stripe_hash_locks_hash(head_sector); | |
784 | spin_lock_irq(conf->hash_locks + hash); | |
785 | head = __find_stripe(conf, head_sector, conf->generation); | |
786 | if (head && !atomic_inc_not_zero(&head->count)) { | |
787 | spin_lock(&conf->device_lock); | |
788 | if (!atomic_read(&head->count)) { | |
789 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
790 | atomic_inc(&conf->active_stripes); | |
791 | BUG_ON(list_empty(&head->lru) && | |
792 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
793 | list_del_init(&head->lru); | |
794 | if (head->group) { | |
795 | head->group->stripes_cnt--; | |
796 | head->group = NULL; | |
797 | } | |
798 | } | |
799 | atomic_inc(&head->count); | |
800 | spin_unlock(&conf->device_lock); | |
801 | } | |
802 | spin_unlock_irq(conf->hash_locks + hash); | |
803 | ||
804 | if (!head) | |
805 | return; | |
806 | if (!stripe_can_batch(head)) | |
807 | goto out; | |
808 | ||
809 | lock_two_stripes(head, sh); | |
810 | /* clear_batch_ready clear the flag */ | |
811 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
812 | goto unlock_out; | |
813 | ||
814 | if (sh->batch_head) | |
815 | goto unlock_out; | |
816 | ||
817 | dd_idx = 0; | |
818 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
819 | dd_idx++; | |
820 | if (head->dev[dd_idx].towrite->bi_rw != sh->dev[dd_idx].towrite->bi_rw) | |
821 | goto unlock_out; | |
822 | ||
823 | if (head->batch_head) { | |
824 | spin_lock(&head->batch_head->batch_lock); | |
825 | /* This batch list is already running */ | |
826 | if (!stripe_can_batch(head)) { | |
827 | spin_unlock(&head->batch_head->batch_lock); | |
828 | goto unlock_out; | |
829 | } | |
830 | ||
831 | /* | |
832 | * at this point, head's BATCH_READY could be cleared, but we | |
833 | * can still add the stripe to batch list | |
834 | */ | |
835 | list_add(&sh->batch_list, &head->batch_list); | |
836 | spin_unlock(&head->batch_head->batch_lock); | |
837 | ||
838 | sh->batch_head = head->batch_head; | |
839 | } else { | |
840 | head->batch_head = head; | |
841 | sh->batch_head = head->batch_head; | |
842 | spin_lock(&head->batch_lock); | |
843 | list_add_tail(&sh->batch_list, &head->batch_list); | |
844 | spin_unlock(&head->batch_lock); | |
845 | } | |
846 | ||
847 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
848 | if (atomic_dec_return(&conf->preread_active_stripes) | |
849 | < IO_THRESHOLD) | |
850 | md_wakeup_thread(conf->mddev->thread); | |
851 | ||
2b6b2457 N |
852 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
853 | int seq = sh->bm_seq; | |
854 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
855 | sh->batch_head->bm_seq > seq) | |
856 | seq = sh->batch_head->bm_seq; | |
857 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
858 | sh->batch_head->bm_seq = seq; | |
859 | } | |
860 | ||
59fc630b | 861 | atomic_inc(&sh->count); |
862 | unlock_out: | |
863 | unlock_two_stripes(head, sh); | |
864 | out: | |
865 | release_stripe(head); | |
866 | } | |
867 | ||
05616be5 N |
868 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
869 | * in this stripe_head. | |
870 | */ | |
871 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
872 | { | |
873 | sector_t progress = conf->reshape_progress; | |
874 | /* Need a memory barrier to make sure we see the value | |
875 | * of conf->generation, or ->data_offset that was set before | |
876 | * reshape_progress was updated. | |
877 | */ | |
878 | smp_rmb(); | |
879 | if (progress == MaxSector) | |
880 | return 0; | |
881 | if (sh->generation == conf->generation - 1) | |
882 | return 0; | |
883 | /* We are in a reshape, and this is a new-generation stripe, | |
884 | * so use new_data_offset. | |
885 | */ | |
886 | return 1; | |
887 | } | |
888 | ||
6712ecf8 N |
889 | static void |
890 | raid5_end_read_request(struct bio *bi, int error); | |
891 | static void | |
892 | raid5_end_write_request(struct bio *bi, int error); | |
91c00924 | 893 | |
c4e5ac0a | 894 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 895 | { |
d1688a6d | 896 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 897 | int i, disks = sh->disks; |
59fc630b | 898 | struct stripe_head *head_sh = sh; |
91c00924 DW |
899 | |
900 | might_sleep(); | |
901 | ||
902 | for (i = disks; i--; ) { | |
903 | int rw; | |
9a3e1101 | 904 | int replace_only = 0; |
977df362 N |
905 | struct bio *bi, *rbi; |
906 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 907 | |
908 | sh = head_sh; | |
e9c7469b TH |
909 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
910 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) | |
911 | rw = WRITE_FUA; | |
912 | else | |
913 | rw = WRITE; | |
9e444768 | 914 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
620125f2 | 915 | rw |= REQ_DISCARD; |
e9c7469b | 916 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
91c00924 | 917 | rw = READ; |
9a3e1101 N |
918 | else if (test_and_clear_bit(R5_WantReplace, |
919 | &sh->dev[i].flags)) { | |
920 | rw = WRITE; | |
921 | replace_only = 1; | |
922 | } else | |
91c00924 | 923 | continue; |
bc0934f0 SL |
924 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
925 | rw |= REQ_SYNC; | |
91c00924 | 926 | |
59fc630b | 927 | again: |
91c00924 | 928 | bi = &sh->dev[i].req; |
977df362 | 929 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 930 | |
91c00924 | 931 | rcu_read_lock(); |
9a3e1101 | 932 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
933 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
934 | rdev = rcu_dereference(conf->disks[i].rdev); | |
935 | if (!rdev) { | |
936 | rdev = rrdev; | |
937 | rrdev = NULL; | |
938 | } | |
9a3e1101 N |
939 | if (rw & WRITE) { |
940 | if (replace_only) | |
941 | rdev = NULL; | |
dd054fce N |
942 | if (rdev == rrdev) |
943 | /* We raced and saw duplicates */ | |
944 | rrdev = NULL; | |
9a3e1101 | 945 | } else { |
59fc630b | 946 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
947 | rdev = rrdev; |
948 | rrdev = NULL; | |
949 | } | |
977df362 | 950 | |
91c00924 DW |
951 | if (rdev && test_bit(Faulty, &rdev->flags)) |
952 | rdev = NULL; | |
953 | if (rdev) | |
954 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
955 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
956 | rrdev = NULL; | |
957 | if (rrdev) | |
958 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
959 | rcu_read_unlock(); |
960 | ||
73e92e51 | 961 | /* We have already checked bad blocks for reads. Now |
977df362 N |
962 | * need to check for writes. We never accept write errors |
963 | * on the replacement, so we don't to check rrdev. | |
73e92e51 N |
964 | */ |
965 | while ((rw & WRITE) && rdev && | |
966 | test_bit(WriteErrorSeen, &rdev->flags)) { | |
967 | sector_t first_bad; | |
968 | int bad_sectors; | |
969 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
970 | &first_bad, &bad_sectors); | |
971 | if (!bad) | |
972 | break; | |
973 | ||
974 | if (bad < 0) { | |
975 | set_bit(BlockedBadBlocks, &rdev->flags); | |
976 | if (!conf->mddev->external && | |
977 | conf->mddev->flags) { | |
978 | /* It is very unlikely, but we might | |
979 | * still need to write out the | |
980 | * bad block log - better give it | |
981 | * a chance*/ | |
982 | md_check_recovery(conf->mddev); | |
983 | } | |
1850753d | 984 | /* |
985 | * Because md_wait_for_blocked_rdev | |
986 | * will dec nr_pending, we must | |
987 | * increment it first. | |
988 | */ | |
989 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
990 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
991 | } else { | |
992 | /* Acknowledged bad block - skip the write */ | |
993 | rdev_dec_pending(rdev, conf->mddev); | |
994 | rdev = NULL; | |
995 | } | |
996 | } | |
997 | ||
91c00924 | 998 | if (rdev) { |
9a3e1101 N |
999 | if (s->syncing || s->expanding || s->expanded |
1000 | || s->replacing) | |
91c00924 DW |
1001 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1002 | ||
2b7497f0 DW |
1003 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1004 | ||
2f6db2a7 | 1005 | bio_reset(bi); |
91c00924 | 1006 | bi->bi_bdev = rdev->bdev; |
2f6db2a7 KO |
1007 | bi->bi_rw = rw; |
1008 | bi->bi_end_io = (rw & WRITE) | |
1009 | ? raid5_end_write_request | |
1010 | : raid5_end_read_request; | |
1011 | bi->bi_private = sh; | |
1012 | ||
91c00924 | 1013 | pr_debug("%s: for %llu schedule op %ld on disc %d\n", |
e46b272b | 1014 | __func__, (unsigned long long)sh->sector, |
91c00924 DW |
1015 | bi->bi_rw, i); |
1016 | atomic_inc(&sh->count); | |
59fc630b | 1017 | if (sh != head_sh) |
1018 | atomic_inc(&head_sh->count); | |
05616be5 | 1019 | if (use_new_offset(conf, sh)) |
4f024f37 | 1020 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1021 | + rdev->new_data_offset); |
1022 | else | |
4f024f37 | 1023 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1024 | + rdev->data_offset); |
59fc630b | 1025 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
e59aa23f | 1026 | bi->bi_rw |= REQ_NOMERGE; |
3f9e7c14 | 1027 | |
d592a996 SL |
1028 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1029 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1030 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1031 | bi->bi_vcnt = 1; |
91c00924 DW |
1032 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1033 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1034 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1035 | /* |
1036 | * If this is discard request, set bi_vcnt 0. We don't | |
1037 | * want to confuse SCSI because SCSI will replace payload | |
1038 | */ | |
1039 | if (rw & REQ_DISCARD) | |
1040 | bi->bi_vcnt = 0; | |
977df362 N |
1041 | if (rrdev) |
1042 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1043 | |
1044 | if (conf->mddev->gendisk) | |
1045 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1046 | bi, disk_devt(conf->mddev->gendisk), | |
1047 | sh->dev[i].sector); | |
91c00924 | 1048 | generic_make_request(bi); |
977df362 N |
1049 | } |
1050 | if (rrdev) { | |
9a3e1101 N |
1051 | if (s->syncing || s->expanding || s->expanded |
1052 | || s->replacing) | |
977df362 N |
1053 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1054 | ||
1055 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1056 | ||
2f6db2a7 | 1057 | bio_reset(rbi); |
977df362 | 1058 | rbi->bi_bdev = rrdev->bdev; |
2f6db2a7 KO |
1059 | rbi->bi_rw = rw; |
1060 | BUG_ON(!(rw & WRITE)); | |
1061 | rbi->bi_end_io = raid5_end_write_request; | |
1062 | rbi->bi_private = sh; | |
1063 | ||
977df362 N |
1064 | pr_debug("%s: for %llu schedule op %ld on " |
1065 | "replacement disc %d\n", | |
1066 | __func__, (unsigned long long)sh->sector, | |
1067 | rbi->bi_rw, i); | |
1068 | atomic_inc(&sh->count); | |
59fc630b | 1069 | if (sh != head_sh) |
1070 | atomic_inc(&head_sh->count); | |
05616be5 | 1071 | if (use_new_offset(conf, sh)) |
4f024f37 | 1072 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1073 | + rrdev->new_data_offset); |
1074 | else | |
4f024f37 | 1075 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1076 | + rrdev->data_offset); |
d592a996 SL |
1077 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1078 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1079 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1080 | rbi->bi_vcnt = 1; |
977df362 N |
1081 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1082 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1083 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1084 | /* |
1085 | * If this is discard request, set bi_vcnt 0. We don't | |
1086 | * want to confuse SCSI because SCSI will replace payload | |
1087 | */ | |
1088 | if (rw & REQ_DISCARD) | |
1089 | rbi->bi_vcnt = 0; | |
e3620a3a JB |
1090 | if (conf->mddev->gendisk) |
1091 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1092 | rbi, disk_devt(conf->mddev->gendisk), | |
1093 | sh->dev[i].sector); | |
977df362 N |
1094 | generic_make_request(rbi); |
1095 | } | |
1096 | if (!rdev && !rrdev) { | |
b062962e | 1097 | if (rw & WRITE) |
91c00924 DW |
1098 | set_bit(STRIPE_DEGRADED, &sh->state); |
1099 | pr_debug("skip op %ld on disc %d for sector %llu\n", | |
1100 | bi->bi_rw, i, (unsigned long long)sh->sector); | |
1101 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1102 | set_bit(STRIPE_HANDLE, &sh->state); | |
1103 | } | |
59fc630b | 1104 | |
1105 | if (!head_sh->batch_head) | |
1106 | continue; | |
1107 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1108 | batch_list); | |
1109 | if (sh != head_sh) | |
1110 | goto again; | |
91c00924 DW |
1111 | } |
1112 | } | |
1113 | ||
1114 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1115 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1116 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1117 | struct stripe_head *sh) | |
91c00924 | 1118 | { |
7988613b KO |
1119 | struct bio_vec bvl; |
1120 | struct bvec_iter iter; | |
91c00924 | 1121 | struct page *bio_page; |
91c00924 | 1122 | int page_offset; |
a08abd8c | 1123 | struct async_submit_ctl submit; |
0403e382 | 1124 | enum async_tx_flags flags = 0; |
91c00924 | 1125 | |
4f024f37 KO |
1126 | if (bio->bi_iter.bi_sector >= sector) |
1127 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1128 | else |
4f024f37 | 1129 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1130 | |
0403e382 DW |
1131 | if (frombio) |
1132 | flags |= ASYNC_TX_FENCE; | |
1133 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1134 | ||
7988613b KO |
1135 | bio_for_each_segment(bvl, bio, iter) { |
1136 | int len = bvl.bv_len; | |
91c00924 DW |
1137 | int clen; |
1138 | int b_offset = 0; | |
1139 | ||
1140 | if (page_offset < 0) { | |
1141 | b_offset = -page_offset; | |
1142 | page_offset += b_offset; | |
1143 | len -= b_offset; | |
1144 | } | |
1145 | ||
1146 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1147 | clen = STRIPE_SIZE - page_offset; | |
1148 | else | |
1149 | clen = len; | |
1150 | ||
1151 | if (clen > 0) { | |
7988613b KO |
1152 | b_offset += bvl.bv_offset; |
1153 | bio_page = bvl.bv_page; | |
d592a996 SL |
1154 | if (frombio) { |
1155 | if (sh->raid_conf->skip_copy && | |
1156 | b_offset == 0 && page_offset == 0 && | |
1157 | clen == STRIPE_SIZE) | |
1158 | *page = bio_page; | |
1159 | else | |
1160 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1161 | b_offset, clen, &submit); |
d592a996 SL |
1162 | } else |
1163 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1164 | page_offset, clen, &submit); |
91c00924 | 1165 | } |
a08abd8c DW |
1166 | /* chain the operations */ |
1167 | submit.depend_tx = tx; | |
1168 | ||
91c00924 DW |
1169 | if (clen < len) /* hit end of page */ |
1170 | break; | |
1171 | page_offset += len; | |
1172 | } | |
1173 | ||
1174 | return tx; | |
1175 | } | |
1176 | ||
1177 | static void ops_complete_biofill(void *stripe_head_ref) | |
1178 | { | |
1179 | struct stripe_head *sh = stripe_head_ref; | |
1180 | struct bio *return_bi = NULL; | |
e4d84909 | 1181 | int i; |
91c00924 | 1182 | |
e46b272b | 1183 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1184 | (unsigned long long)sh->sector); |
1185 | ||
1186 | /* clear completed biofills */ | |
1187 | for (i = sh->disks; i--; ) { | |
1188 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1189 | |
1190 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1191 | /* and check if we need to reply to a read request, |
1192 | * new R5_Wantfill requests are held off until | |
83de75cc | 1193 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1194 | */ |
1195 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1196 | struct bio *rbi, *rbi2; |
91c00924 | 1197 | |
91c00924 DW |
1198 | BUG_ON(!dev->read); |
1199 | rbi = dev->read; | |
1200 | dev->read = NULL; | |
4f024f37 | 1201 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1202 | dev->sector + STRIPE_SECTORS) { |
1203 | rbi2 = r5_next_bio(rbi, dev->sector); | |
e7836bd6 | 1204 | if (!raid5_dec_bi_active_stripes(rbi)) { |
91c00924 DW |
1205 | rbi->bi_next = return_bi; |
1206 | return_bi = rbi; | |
1207 | } | |
91c00924 DW |
1208 | rbi = rbi2; |
1209 | } | |
1210 | } | |
1211 | } | |
83de75cc | 1212 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 DW |
1213 | |
1214 | return_io(return_bi); | |
1215 | ||
e4d84909 | 1216 | set_bit(STRIPE_HANDLE, &sh->state); |
91c00924 DW |
1217 | release_stripe(sh); |
1218 | } | |
1219 | ||
1220 | static void ops_run_biofill(struct stripe_head *sh) | |
1221 | { | |
1222 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1223 | struct async_submit_ctl submit; |
91c00924 DW |
1224 | int i; |
1225 | ||
59fc630b | 1226 | BUG_ON(sh->batch_head); |
e46b272b | 1227 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1228 | (unsigned long long)sh->sector); |
1229 | ||
1230 | for (i = sh->disks; i--; ) { | |
1231 | struct r5dev *dev = &sh->dev[i]; | |
1232 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1233 | struct bio *rbi; | |
b17459c0 | 1234 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1235 | dev->read = rbi = dev->toread; |
1236 | dev->toread = NULL; | |
b17459c0 | 1237 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1238 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1239 | dev->sector + STRIPE_SECTORS) { |
d592a996 SL |
1240 | tx = async_copy_data(0, rbi, &dev->page, |
1241 | dev->sector, tx, sh); | |
91c00924 DW |
1242 | rbi = r5_next_bio(rbi, dev->sector); |
1243 | } | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | atomic_inc(&sh->count); | |
a08abd8c DW |
1248 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1249 | async_trigger_callback(&submit); | |
91c00924 DW |
1250 | } |
1251 | ||
4e7d2c0a | 1252 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1253 | { |
4e7d2c0a | 1254 | struct r5dev *tgt; |
91c00924 | 1255 | |
4e7d2c0a DW |
1256 | if (target < 0) |
1257 | return; | |
91c00924 | 1258 | |
4e7d2c0a | 1259 | tgt = &sh->dev[target]; |
91c00924 DW |
1260 | set_bit(R5_UPTODATE, &tgt->flags); |
1261 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1262 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1263 | } |
1264 | ||
ac6b53b6 | 1265 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1266 | { |
1267 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1268 | |
e46b272b | 1269 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1270 | (unsigned long long)sh->sector); |
1271 | ||
ac6b53b6 | 1272 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1273 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1274 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1275 | |
ecc65c9b DW |
1276 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1277 | if (sh->check_state == check_state_compute_run) | |
1278 | sh->check_state = check_state_compute_result; | |
91c00924 DW |
1279 | set_bit(STRIPE_HANDLE, &sh->state); |
1280 | release_stripe(sh); | |
1281 | } | |
1282 | ||
d6f38f31 DW |
1283 | /* return a pointer to the address conversion region of the scribble buffer */ |
1284 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1285 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1286 | { |
46d5b785 | 1287 | void *addr; |
1288 | ||
1289 | addr = flex_array_get(percpu->scribble, i); | |
1290 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1291 | } | |
1292 | ||
1293 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1294 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1295 | { | |
1296 | void *addr; | |
1297 | ||
1298 | addr = flex_array_get(percpu->scribble, i); | |
1299 | return addr; | |
d6f38f31 DW |
1300 | } |
1301 | ||
1302 | static struct dma_async_tx_descriptor * | |
1303 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1304 | { |
91c00924 | 1305 | int disks = sh->disks; |
46d5b785 | 1306 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1307 | int target = sh->ops.target; |
1308 | struct r5dev *tgt = &sh->dev[target]; | |
1309 | struct page *xor_dest = tgt->page; | |
1310 | int count = 0; | |
1311 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1312 | struct async_submit_ctl submit; |
91c00924 DW |
1313 | int i; |
1314 | ||
59fc630b | 1315 | BUG_ON(sh->batch_head); |
1316 | ||
91c00924 | 1317 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1318 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1319 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1320 | ||
1321 | for (i = disks; i--; ) | |
1322 | if (i != target) | |
1323 | xor_srcs[count++] = sh->dev[i].page; | |
1324 | ||
1325 | atomic_inc(&sh->count); | |
1326 | ||
0403e382 | 1327 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1328 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1329 | if (unlikely(count == 1)) |
a08abd8c | 1330 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1331 | else |
a08abd8c | 1332 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1333 | |
91c00924 DW |
1334 | return tx; |
1335 | } | |
1336 | ||
ac6b53b6 DW |
1337 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1338 | * @srcs - (struct page *) array of size sh->disks | |
1339 | * @sh - stripe_head to parse | |
1340 | * | |
1341 | * Populates srcs in proper layout order for the stripe and returns the | |
1342 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1343 | * destination buffer is recorded in srcs[count] and the Q destination | |
1344 | * is recorded in srcs[count+1]]. | |
1345 | */ | |
584acdd4 MS |
1346 | static int set_syndrome_sources(struct page **srcs, |
1347 | struct stripe_head *sh, | |
1348 | int srctype) | |
ac6b53b6 DW |
1349 | { |
1350 | int disks = sh->disks; | |
1351 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1352 | int d0_idx = raid6_d0(sh); | |
1353 | int count; | |
1354 | int i; | |
1355 | ||
1356 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1357 | srcs[i] = NULL; |
ac6b53b6 DW |
1358 | |
1359 | count = 0; | |
1360 | i = d0_idx; | |
1361 | do { | |
1362 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1363 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1364 | |
584acdd4 MS |
1365 | if (i == sh->qd_idx || i == sh->pd_idx || |
1366 | (srctype == SYNDROME_SRC_ALL) || | |
1367 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1368 | test_bit(R5_Wantdrain, &dev->flags)) || | |
1369 | (srctype == SYNDROME_SRC_WRITTEN && | |
1370 | dev->written)) | |
1371 | srcs[slot] = sh->dev[i].page; | |
ac6b53b6 DW |
1372 | i = raid6_next_disk(i, disks); |
1373 | } while (i != d0_idx); | |
ac6b53b6 | 1374 | |
e4424fee | 1375 | return syndrome_disks; |
ac6b53b6 DW |
1376 | } |
1377 | ||
1378 | static struct dma_async_tx_descriptor * | |
1379 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1380 | { | |
1381 | int disks = sh->disks; | |
46d5b785 | 1382 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1383 | int target; |
1384 | int qd_idx = sh->qd_idx; | |
1385 | struct dma_async_tx_descriptor *tx; | |
1386 | struct async_submit_ctl submit; | |
1387 | struct r5dev *tgt; | |
1388 | struct page *dest; | |
1389 | int i; | |
1390 | int count; | |
1391 | ||
59fc630b | 1392 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1393 | if (sh->ops.target < 0) |
1394 | target = sh->ops.target2; | |
1395 | else if (sh->ops.target2 < 0) | |
1396 | target = sh->ops.target; | |
91c00924 | 1397 | else |
ac6b53b6 DW |
1398 | /* we should only have one valid target */ |
1399 | BUG(); | |
1400 | BUG_ON(target < 0); | |
1401 | pr_debug("%s: stripe %llu block: %d\n", | |
1402 | __func__, (unsigned long long)sh->sector, target); | |
1403 | ||
1404 | tgt = &sh->dev[target]; | |
1405 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1406 | dest = tgt->page; | |
1407 | ||
1408 | atomic_inc(&sh->count); | |
1409 | ||
1410 | if (target == qd_idx) { | |
584acdd4 | 1411 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1412 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1413 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1414 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1415 | ops_complete_compute, sh, | |
46d5b785 | 1416 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1417 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1418 | } else { | |
1419 | /* Compute any data- or p-drive using XOR */ | |
1420 | count = 0; | |
1421 | for (i = disks; i-- ; ) { | |
1422 | if (i == target || i == qd_idx) | |
1423 | continue; | |
1424 | blocks[count++] = sh->dev[i].page; | |
1425 | } | |
1426 | ||
0403e382 DW |
1427 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1428 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1429 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1430 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1431 | } | |
91c00924 | 1432 | |
91c00924 DW |
1433 | return tx; |
1434 | } | |
1435 | ||
ac6b53b6 DW |
1436 | static struct dma_async_tx_descriptor * |
1437 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1438 | { | |
1439 | int i, count, disks = sh->disks; | |
1440 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1441 | int d0_idx = raid6_d0(sh); | |
1442 | int faila = -1, failb = -1; | |
1443 | int target = sh->ops.target; | |
1444 | int target2 = sh->ops.target2; | |
1445 | struct r5dev *tgt = &sh->dev[target]; | |
1446 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1447 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1448 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1449 | struct async_submit_ctl submit; |
1450 | ||
59fc630b | 1451 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1452 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1453 | __func__, (unsigned long long)sh->sector, target, target2); | |
1454 | BUG_ON(target < 0 || target2 < 0); | |
1455 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1456 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1457 | ||
6c910a78 | 1458 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1459 | * slot number conversion for 'faila' and 'failb' |
1460 | */ | |
1461 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1462 | blocks[i] = NULL; |
ac6b53b6 DW |
1463 | count = 0; |
1464 | i = d0_idx; | |
1465 | do { | |
1466 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1467 | ||
1468 | blocks[slot] = sh->dev[i].page; | |
1469 | ||
1470 | if (i == target) | |
1471 | faila = slot; | |
1472 | if (i == target2) | |
1473 | failb = slot; | |
1474 | i = raid6_next_disk(i, disks); | |
1475 | } while (i != d0_idx); | |
ac6b53b6 DW |
1476 | |
1477 | BUG_ON(faila == failb); | |
1478 | if (failb < faila) | |
1479 | swap(faila, failb); | |
1480 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1481 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1482 | ||
1483 | atomic_inc(&sh->count); | |
1484 | ||
1485 | if (failb == syndrome_disks+1) { | |
1486 | /* Q disk is one of the missing disks */ | |
1487 | if (faila == syndrome_disks) { | |
1488 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1489 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1490 | ops_complete_compute, sh, | |
46d5b785 | 1491 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1492 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1493 | STRIPE_SIZE, &submit); |
1494 | } else { | |
1495 | struct page *dest; | |
1496 | int data_target; | |
1497 | int qd_idx = sh->qd_idx; | |
1498 | ||
1499 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1500 | if (target == qd_idx) | |
1501 | data_target = target2; | |
1502 | else | |
1503 | data_target = target; | |
1504 | ||
1505 | count = 0; | |
1506 | for (i = disks; i-- ; ) { | |
1507 | if (i == data_target || i == qd_idx) | |
1508 | continue; | |
1509 | blocks[count++] = sh->dev[i].page; | |
1510 | } | |
1511 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1512 | init_async_submit(&submit, |
1513 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1514 | NULL, NULL, NULL, | |
46d5b785 | 1515 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1516 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1517 | &submit); | |
1518 | ||
584acdd4 | 1519 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1520 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1521 | ops_complete_compute, sh, | |
46d5b785 | 1522 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1523 | return async_gen_syndrome(blocks, 0, count+2, |
1524 | STRIPE_SIZE, &submit); | |
1525 | } | |
ac6b53b6 | 1526 | } else { |
6c910a78 DW |
1527 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1528 | ops_complete_compute, sh, | |
46d5b785 | 1529 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1530 | if (failb == syndrome_disks) { |
1531 | /* We're missing D+P. */ | |
1532 | return async_raid6_datap_recov(syndrome_disks+2, | |
1533 | STRIPE_SIZE, faila, | |
1534 | blocks, &submit); | |
1535 | } else { | |
1536 | /* We're missing D+D. */ | |
1537 | return async_raid6_2data_recov(syndrome_disks+2, | |
1538 | STRIPE_SIZE, faila, failb, | |
1539 | blocks, &submit); | |
1540 | } | |
ac6b53b6 DW |
1541 | } |
1542 | } | |
1543 | ||
91c00924 DW |
1544 | static void ops_complete_prexor(void *stripe_head_ref) |
1545 | { | |
1546 | struct stripe_head *sh = stripe_head_ref; | |
1547 | ||
e46b272b | 1548 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1549 | (unsigned long long)sh->sector); |
91c00924 DW |
1550 | } |
1551 | ||
1552 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1553 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1554 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1555 | { |
91c00924 | 1556 | int disks = sh->disks; |
46d5b785 | 1557 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1558 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1559 | struct async_submit_ctl submit; |
91c00924 DW |
1560 | |
1561 | /* existing parity data subtracted */ | |
1562 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1563 | ||
59fc630b | 1564 | BUG_ON(sh->batch_head); |
e46b272b | 1565 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1566 | (unsigned long long)sh->sector); |
1567 | ||
1568 | for (i = disks; i--; ) { | |
1569 | struct r5dev *dev = &sh->dev[i]; | |
1570 | /* Only process blocks that are known to be uptodate */ | |
d8ee0728 | 1571 | if (test_bit(R5_Wantdrain, &dev->flags)) |
91c00924 DW |
1572 | xor_srcs[count++] = dev->page; |
1573 | } | |
1574 | ||
0403e382 | 1575 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1576 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1577 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1578 | |
1579 | return tx; | |
1580 | } | |
1581 | ||
584acdd4 MS |
1582 | static struct dma_async_tx_descriptor * |
1583 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1584 | struct dma_async_tx_descriptor *tx) | |
1585 | { | |
1586 | struct page **blocks = to_addr_page(percpu, 0); | |
1587 | int count; | |
1588 | struct async_submit_ctl submit; | |
1589 | ||
1590 | pr_debug("%s: stripe %llu\n", __func__, | |
1591 | (unsigned long long)sh->sector); | |
1592 | ||
1593 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1594 | ||
1595 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1596 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1597 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1598 | ||
1599 | return tx; | |
1600 | } | |
1601 | ||
91c00924 | 1602 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1603 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 DW |
1604 | { |
1605 | int disks = sh->disks; | |
d8ee0728 | 1606 | int i; |
59fc630b | 1607 | struct stripe_head *head_sh = sh; |
91c00924 | 1608 | |
e46b272b | 1609 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1610 | (unsigned long long)sh->sector); |
1611 | ||
1612 | for (i = disks; i--; ) { | |
59fc630b | 1613 | struct r5dev *dev; |
91c00924 | 1614 | struct bio *chosen; |
91c00924 | 1615 | |
59fc630b | 1616 | sh = head_sh; |
1617 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1618 | struct bio *wbi; |
1619 | ||
59fc630b | 1620 | again: |
1621 | dev = &sh->dev[i]; | |
b17459c0 | 1622 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1623 | chosen = dev->towrite; |
1624 | dev->towrite = NULL; | |
7a87f434 | 1625 | sh->overwrite_disks = 0; |
91c00924 DW |
1626 | BUG_ON(dev->written); |
1627 | wbi = dev->written = chosen; | |
b17459c0 | 1628 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1629 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1630 | |
4f024f37 | 1631 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1632 | dev->sector + STRIPE_SECTORS) { |
e9c7469b TH |
1633 | if (wbi->bi_rw & REQ_FUA) |
1634 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1635 | if (wbi->bi_rw & REQ_SYNC) |
1636 | set_bit(R5_SyncIO, &dev->flags); | |
9e444768 | 1637 | if (wbi->bi_rw & REQ_DISCARD) |
620125f2 | 1638 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1639 | else { |
1640 | tx = async_copy_data(1, wbi, &dev->page, | |
1641 | dev->sector, tx, sh); | |
1642 | if (dev->page != dev->orig_page) { | |
1643 | set_bit(R5_SkipCopy, &dev->flags); | |
1644 | clear_bit(R5_UPTODATE, &dev->flags); | |
1645 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1646 | } | |
1647 | } | |
91c00924 DW |
1648 | wbi = r5_next_bio(wbi, dev->sector); |
1649 | } | |
59fc630b | 1650 | |
1651 | if (head_sh->batch_head) { | |
1652 | sh = list_first_entry(&sh->batch_list, | |
1653 | struct stripe_head, | |
1654 | batch_list); | |
1655 | if (sh == head_sh) | |
1656 | continue; | |
1657 | goto again; | |
1658 | } | |
91c00924 DW |
1659 | } |
1660 | } | |
1661 | ||
1662 | return tx; | |
1663 | } | |
1664 | ||
ac6b53b6 | 1665 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1666 | { |
1667 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1668 | int disks = sh->disks; |
1669 | int pd_idx = sh->pd_idx; | |
1670 | int qd_idx = sh->qd_idx; | |
1671 | int i; | |
9e444768 | 1672 | bool fua = false, sync = false, discard = false; |
91c00924 | 1673 | |
e46b272b | 1674 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1675 | (unsigned long long)sh->sector); |
1676 | ||
bc0934f0 | 1677 | for (i = disks; i--; ) { |
e9c7469b | 1678 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1679 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1680 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1681 | } |
e9c7469b | 1682 | |
91c00924 DW |
1683 | for (i = disks; i--; ) { |
1684 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1685 | |
e9c7469b | 1686 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1687 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1688 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1689 | if (fua) |
1690 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1691 | if (sync) |
1692 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1693 | } |
91c00924 DW |
1694 | } |
1695 | ||
d8ee0728 DW |
1696 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1697 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1698 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1699 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1700 | else { | |
1701 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1702 | sh->reconstruct_state = reconstruct_state_result; | |
1703 | } | |
91c00924 DW |
1704 | |
1705 | set_bit(STRIPE_HANDLE, &sh->state); | |
1706 | release_stripe(sh); | |
1707 | } | |
1708 | ||
1709 | static void | |
ac6b53b6 DW |
1710 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1711 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1712 | { |
91c00924 | 1713 | int disks = sh->disks; |
59fc630b | 1714 | struct page **xor_srcs; |
a08abd8c | 1715 | struct async_submit_ctl submit; |
59fc630b | 1716 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1717 | struct page *xor_dest; |
d8ee0728 | 1718 | int prexor = 0; |
91c00924 | 1719 | unsigned long flags; |
59fc630b | 1720 | int j = 0; |
1721 | struct stripe_head *head_sh = sh; | |
1722 | int last_stripe; | |
91c00924 | 1723 | |
e46b272b | 1724 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1725 | (unsigned long long)sh->sector); |
1726 | ||
620125f2 SL |
1727 | for (i = 0; i < sh->disks; i++) { |
1728 | if (pd_idx == i) | |
1729 | continue; | |
1730 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1731 | break; | |
1732 | } | |
1733 | if (i >= sh->disks) { | |
1734 | atomic_inc(&sh->count); | |
620125f2 SL |
1735 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1736 | ops_complete_reconstruct(sh); | |
1737 | return; | |
1738 | } | |
59fc630b | 1739 | again: |
1740 | count = 0; | |
1741 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1742 | /* check if prexor is active which means only process blocks |
1743 | * that are part of a read-modify-write (written) | |
1744 | */ | |
59fc630b | 1745 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1746 | prexor = 1; |
91c00924 DW |
1747 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1748 | for (i = disks; i--; ) { | |
1749 | struct r5dev *dev = &sh->dev[i]; | |
59fc630b | 1750 | if (head_sh->dev[i].written) |
91c00924 DW |
1751 | xor_srcs[count++] = dev->page; |
1752 | } | |
1753 | } else { | |
1754 | xor_dest = sh->dev[pd_idx].page; | |
1755 | for (i = disks; i--; ) { | |
1756 | struct r5dev *dev = &sh->dev[i]; | |
1757 | if (i != pd_idx) | |
1758 | xor_srcs[count++] = dev->page; | |
1759 | } | |
1760 | } | |
1761 | ||
91c00924 DW |
1762 | /* 1/ if we prexor'd then the dest is reused as a source |
1763 | * 2/ if we did not prexor then we are redoing the parity | |
1764 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1765 | * for the synchronous xor case | |
1766 | */ | |
59fc630b | 1767 | last_stripe = !head_sh->batch_head || |
1768 | list_first_entry(&sh->batch_list, | |
1769 | struct stripe_head, batch_list) == head_sh; | |
1770 | if (last_stripe) { | |
1771 | flags = ASYNC_TX_ACK | | |
1772 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1773 | ||
1774 | atomic_inc(&head_sh->count); | |
1775 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1776 | to_addr_conv(sh, percpu, j)); | |
1777 | } else { | |
1778 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1779 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1780 | to_addr_conv(sh, percpu, j)); | |
1781 | } | |
91c00924 | 1782 | |
a08abd8c DW |
1783 | if (unlikely(count == 1)) |
1784 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1785 | else | |
1786 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1787 | if (!last_stripe) { |
1788 | j++; | |
1789 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1790 | batch_list); | |
1791 | goto again; | |
1792 | } | |
91c00924 DW |
1793 | } |
1794 | ||
ac6b53b6 DW |
1795 | static void |
1796 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1797 | struct dma_async_tx_descriptor *tx) | |
1798 | { | |
1799 | struct async_submit_ctl submit; | |
59fc630b | 1800 | struct page **blocks; |
1801 | int count, i, j = 0; | |
1802 | struct stripe_head *head_sh = sh; | |
1803 | int last_stripe; | |
584acdd4 MS |
1804 | int synflags; |
1805 | unsigned long txflags; | |
ac6b53b6 DW |
1806 | |
1807 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1808 | ||
620125f2 SL |
1809 | for (i = 0; i < sh->disks; i++) { |
1810 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1811 | continue; | |
1812 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1813 | break; | |
1814 | } | |
1815 | if (i >= sh->disks) { | |
1816 | atomic_inc(&sh->count); | |
620125f2 SL |
1817 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1818 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1819 | ops_complete_reconstruct(sh); | |
1820 | return; | |
1821 | } | |
1822 | ||
59fc630b | 1823 | again: |
1824 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1825 | |
1826 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1827 | synflags = SYNDROME_SRC_WRITTEN; | |
1828 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1829 | } else { | |
1830 | synflags = SYNDROME_SRC_ALL; | |
1831 | txflags = ASYNC_TX_ACK; | |
1832 | } | |
1833 | ||
1834 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1835 | last_stripe = !head_sh->batch_head || |
1836 | list_first_entry(&sh->batch_list, | |
1837 | struct stripe_head, batch_list) == head_sh; | |
1838 | ||
1839 | if (last_stripe) { | |
1840 | atomic_inc(&head_sh->count); | |
584acdd4 | 1841 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1842 | head_sh, to_addr_conv(sh, percpu, j)); |
1843 | } else | |
1844 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1845 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1846 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1847 | if (!last_stripe) { |
1848 | j++; | |
1849 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1850 | batch_list); | |
1851 | goto again; | |
1852 | } | |
91c00924 DW |
1853 | } |
1854 | ||
1855 | static void ops_complete_check(void *stripe_head_ref) | |
1856 | { | |
1857 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1858 | |
e46b272b | 1859 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1860 | (unsigned long long)sh->sector); |
1861 | ||
ecc65c9b | 1862 | sh->check_state = check_state_check_result; |
91c00924 DW |
1863 | set_bit(STRIPE_HANDLE, &sh->state); |
1864 | release_stripe(sh); | |
1865 | } | |
1866 | ||
ac6b53b6 | 1867 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1868 | { |
91c00924 | 1869 | int disks = sh->disks; |
ac6b53b6 DW |
1870 | int pd_idx = sh->pd_idx; |
1871 | int qd_idx = sh->qd_idx; | |
1872 | struct page *xor_dest; | |
46d5b785 | 1873 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1874 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1875 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1876 | int count; |
1877 | int i; | |
91c00924 | 1878 | |
e46b272b | 1879 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1880 | (unsigned long long)sh->sector); |
1881 | ||
59fc630b | 1882 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1883 | count = 0; |
1884 | xor_dest = sh->dev[pd_idx].page; | |
1885 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1886 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1887 | if (i == pd_idx || i == qd_idx) |
1888 | continue; | |
1889 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1890 | } |
1891 | ||
d6f38f31 | 1892 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1893 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1894 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1895 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1896 | |
91c00924 | 1897 | atomic_inc(&sh->count); |
a08abd8c DW |
1898 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1899 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1900 | } |
1901 | ||
ac6b53b6 DW |
1902 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1903 | { | |
46d5b785 | 1904 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1905 | struct async_submit_ctl submit; |
1906 | int count; | |
1907 | ||
1908 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1909 | (unsigned long long)sh->sector, checkp); | |
1910 | ||
59fc630b | 1911 | BUG_ON(sh->batch_head); |
584acdd4 | 1912 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1913 | if (!checkp) |
1914 | srcs[count] = NULL; | |
91c00924 | 1915 | |
91c00924 | 1916 | atomic_inc(&sh->count); |
ac6b53b6 | 1917 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1918 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1919 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1920 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1921 | } |
1922 | ||
51acbcec | 1923 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1924 | { |
1925 | int overlap_clear = 0, i, disks = sh->disks; | |
1926 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1927 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1928 | int level = conf->level; |
d6f38f31 DW |
1929 | struct raid5_percpu *percpu; |
1930 | unsigned long cpu; | |
91c00924 | 1931 | |
d6f38f31 DW |
1932 | cpu = get_cpu(); |
1933 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1934 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1935 | ops_run_biofill(sh); |
1936 | overlap_clear++; | |
1937 | } | |
1938 | ||
7b3a871e | 1939 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1940 | if (level < 6) |
1941 | tx = ops_run_compute5(sh, percpu); | |
1942 | else { | |
1943 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
1944 | tx = ops_run_compute6_1(sh, percpu); | |
1945 | else | |
1946 | tx = ops_run_compute6_2(sh, percpu); | |
1947 | } | |
1948 | /* terminate the chain if reconstruct is not set to be run */ | |
1949 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
1950 | async_tx_ack(tx); |
1951 | } | |
91c00924 | 1952 | |
584acdd4 MS |
1953 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
1954 | if (level < 6) | |
1955 | tx = ops_run_prexor5(sh, percpu, tx); | |
1956 | else | |
1957 | tx = ops_run_prexor6(sh, percpu, tx); | |
1958 | } | |
91c00924 | 1959 | |
600aa109 | 1960 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 1961 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
1962 | overlap_clear++; |
1963 | } | |
1964 | ||
ac6b53b6 DW |
1965 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
1966 | if (level < 6) | |
1967 | ops_run_reconstruct5(sh, percpu, tx); | |
1968 | else | |
1969 | ops_run_reconstruct6(sh, percpu, tx); | |
1970 | } | |
91c00924 | 1971 | |
ac6b53b6 DW |
1972 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
1973 | if (sh->check_state == check_state_run) | |
1974 | ops_run_check_p(sh, percpu); | |
1975 | else if (sh->check_state == check_state_run_q) | |
1976 | ops_run_check_pq(sh, percpu, 0); | |
1977 | else if (sh->check_state == check_state_run_pq) | |
1978 | ops_run_check_pq(sh, percpu, 1); | |
1979 | else | |
1980 | BUG(); | |
1981 | } | |
91c00924 | 1982 | |
59fc630b | 1983 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
1984 | for (i = disks; i--; ) { |
1985 | struct r5dev *dev = &sh->dev[i]; | |
1986 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
1987 | wake_up(&sh->raid_conf->wait_for_overlap); | |
1988 | } | |
d6f38f31 | 1989 | put_cpu(); |
91c00924 DW |
1990 | } |
1991 | ||
f18c1a35 N |
1992 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp) |
1993 | { | |
1994 | struct stripe_head *sh; | |
1995 | ||
1996 | sh = kmem_cache_zalloc(sc, gfp); | |
1997 | if (sh) { | |
1998 | spin_lock_init(&sh->stripe_lock); | |
1999 | spin_lock_init(&sh->batch_lock); | |
2000 | INIT_LIST_HEAD(&sh->batch_list); | |
2001 | INIT_LIST_HEAD(&sh->lru); | |
2002 | atomic_set(&sh->count, 1); | |
2003 | } | |
2004 | return sh; | |
2005 | } | |
486f0644 | 2006 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2007 | { |
2008 | struct stripe_head *sh; | |
f18c1a35 N |
2009 | |
2010 | sh = alloc_stripe(conf->slab_cache, gfp); | |
3f294f4f N |
2011 | if (!sh) |
2012 | return 0; | |
6ce32846 | 2013 | |
3f294f4f | 2014 | sh->raid_conf = conf; |
3f294f4f | 2015 | |
a9683a79 | 2016 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2017 | shrink_buffers(sh); |
3f294f4f N |
2018 | kmem_cache_free(conf->slab_cache, sh); |
2019 | return 0; | |
2020 | } | |
486f0644 N |
2021 | sh->hash_lock_index = |
2022 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2023 | /* we just created an active stripe so... */ |
3f294f4f | 2024 | atomic_inc(&conf->active_stripes); |
59fc630b | 2025 | |
3f294f4f | 2026 | release_stripe(sh); |
486f0644 | 2027 | conf->max_nr_stripes++; |
3f294f4f N |
2028 | return 1; |
2029 | } | |
2030 | ||
d1688a6d | 2031 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2032 | { |
e18b890b | 2033 | struct kmem_cache *sc; |
5e5e3e78 | 2034 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2035 | |
f4be6b43 N |
2036 | if (conf->mddev->gendisk) |
2037 | sprintf(conf->cache_name[0], | |
2038 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2039 | else | |
2040 | sprintf(conf->cache_name[0], | |
2041 | "raid%d-%p", conf->level, conf->mddev); | |
2042 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2043 | ||
ad01c9e3 N |
2044 | conf->active_name = 0; |
2045 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2046 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2047 | 0, 0, NULL); |
1da177e4 LT |
2048 | if (!sc) |
2049 | return 1; | |
2050 | conf->slab_cache = sc; | |
ad01c9e3 | 2051 | conf->pool_size = devs; |
486f0644 N |
2052 | while (num--) |
2053 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2054 | return 1; |
486f0644 | 2055 | |
1da177e4 LT |
2056 | return 0; |
2057 | } | |
29269553 | 2058 | |
d6f38f31 DW |
2059 | /** |
2060 | * scribble_len - return the required size of the scribble region | |
2061 | * @num - total number of disks in the array | |
2062 | * | |
2063 | * The size must be enough to contain: | |
2064 | * 1/ a struct page pointer for each device in the array +2 | |
2065 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2066 | * (dma_map_page()) or page (page_address()) address. | |
2067 | * | |
2068 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2069 | * calculate over all devices (not just the data blocks), using zeros in place | |
2070 | * of the P and Q blocks. | |
2071 | */ | |
46d5b785 | 2072 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2073 | { |
46d5b785 | 2074 | struct flex_array *ret; |
d6f38f31 DW |
2075 | size_t len; |
2076 | ||
2077 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2078 | ret = flex_array_alloc(len, cnt, flags); |
2079 | if (!ret) | |
2080 | return NULL; | |
2081 | /* always prealloc all elements, so no locking is required */ | |
2082 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2083 | flex_array_free(ret); | |
2084 | return NULL; | |
2085 | } | |
2086 | return ret; | |
d6f38f31 DW |
2087 | } |
2088 | ||
738a2738 N |
2089 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2090 | { | |
2091 | unsigned long cpu; | |
2092 | int err = 0; | |
2093 | ||
2094 | mddev_suspend(conf->mddev); | |
2095 | get_online_cpus(); | |
2096 | for_each_present_cpu(cpu) { | |
2097 | struct raid5_percpu *percpu; | |
2098 | struct flex_array *scribble; | |
2099 | ||
2100 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2101 | scribble = scribble_alloc(new_disks, | |
2102 | new_sectors / STRIPE_SECTORS, | |
2103 | GFP_NOIO); | |
2104 | ||
2105 | if (scribble) { | |
2106 | flex_array_free(percpu->scribble); | |
2107 | percpu->scribble = scribble; | |
2108 | } else { | |
2109 | err = -ENOMEM; | |
2110 | break; | |
2111 | } | |
2112 | } | |
2113 | put_online_cpus(); | |
2114 | mddev_resume(conf->mddev); | |
2115 | return err; | |
2116 | } | |
2117 | ||
d1688a6d | 2118 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2119 | { |
2120 | /* Make all the stripes able to hold 'newsize' devices. | |
2121 | * New slots in each stripe get 'page' set to a new page. | |
2122 | * | |
2123 | * This happens in stages: | |
2124 | * 1/ create a new kmem_cache and allocate the required number of | |
2125 | * stripe_heads. | |
83f0d77a | 2126 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2127 | * to the new stripe_heads. This will have the side effect of |
2128 | * freezing the array as once all stripe_heads have been collected, | |
2129 | * no IO will be possible. Old stripe heads are freed once their | |
2130 | * pages have been transferred over, and the old kmem_cache is | |
2131 | * freed when all stripes are done. | |
2132 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2133 | * we simple return a failre status - no need to clean anything up. | |
2134 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2135 | * If this fails, we don't bother trying the shrink the | |
2136 | * stripe_heads down again, we just leave them as they are. | |
2137 | * As each stripe_head is processed the new one is released into | |
2138 | * active service. | |
2139 | * | |
2140 | * Once step2 is started, we cannot afford to wait for a write, | |
2141 | * so we use GFP_NOIO allocations. | |
2142 | */ | |
2143 | struct stripe_head *osh, *nsh; | |
2144 | LIST_HEAD(newstripes); | |
2145 | struct disk_info *ndisks; | |
b5470dc5 | 2146 | int err; |
e18b890b | 2147 | struct kmem_cache *sc; |
ad01c9e3 | 2148 | int i; |
566c09c5 | 2149 | int hash, cnt; |
ad01c9e3 N |
2150 | |
2151 | if (newsize <= conf->pool_size) | |
2152 | return 0; /* never bother to shrink */ | |
2153 | ||
b5470dc5 DW |
2154 | err = md_allow_write(conf->mddev); |
2155 | if (err) | |
2156 | return err; | |
2a2275d6 | 2157 | |
ad01c9e3 N |
2158 | /* Step 1 */ |
2159 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2160 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2161 | 0, 0, NULL); |
ad01c9e3 N |
2162 | if (!sc) |
2163 | return -ENOMEM; | |
2164 | ||
2d5b569b N |
2165 | /* Need to ensure auto-resizing doesn't interfere */ |
2166 | mutex_lock(&conf->cache_size_mutex); | |
2167 | ||
ad01c9e3 | 2168 | for (i = conf->max_nr_stripes; i; i--) { |
f18c1a35 | 2169 | nsh = alloc_stripe(sc, GFP_KERNEL); |
ad01c9e3 N |
2170 | if (!nsh) |
2171 | break; | |
2172 | ||
ad01c9e3 | 2173 | nsh->raid_conf = conf; |
ad01c9e3 N |
2174 | list_add(&nsh->lru, &newstripes); |
2175 | } | |
2176 | if (i) { | |
2177 | /* didn't get enough, give up */ | |
2178 | while (!list_empty(&newstripes)) { | |
2179 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2180 | list_del(&nsh->lru); | |
2181 | kmem_cache_free(sc, nsh); | |
2182 | } | |
2183 | kmem_cache_destroy(sc); | |
2d5b569b | 2184 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2185 | return -ENOMEM; |
2186 | } | |
2187 | /* Step 2 - Must use GFP_NOIO now. | |
2188 | * OK, we have enough stripes, start collecting inactive | |
2189 | * stripes and copying them over | |
2190 | */ | |
566c09c5 SL |
2191 | hash = 0; |
2192 | cnt = 0; | |
ad01c9e3 | 2193 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2194 | lock_device_hash_lock(conf, hash); |
e9e4c377 | 2195 | wait_event_exclusive_cmd(conf->wait_for_stripe[hash], |
566c09c5 SL |
2196 | !list_empty(conf->inactive_list + hash), |
2197 | unlock_device_hash_lock(conf, hash), | |
2198 | lock_device_hash_lock(conf, hash)); | |
2199 | osh = get_free_stripe(conf, hash); | |
2200 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2201 | |
d592a996 | 2202 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2203 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2204 | nsh->dev[i].orig_page = osh->dev[i].page; |
2205 | } | |
566c09c5 | 2206 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2207 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2208 | cnt++; |
2209 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2210 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2211 | hash++; | |
2212 | cnt = 0; | |
2213 | } | |
ad01c9e3 N |
2214 | } |
2215 | kmem_cache_destroy(conf->slab_cache); | |
2216 | ||
2217 | /* Step 3. | |
2218 | * At this point, we are holding all the stripes so the array | |
2219 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2220 | * conf->disks and the scribble region |
ad01c9e3 N |
2221 | */ |
2222 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2223 | if (ndisks) { | |
2224 | for (i=0; i<conf->raid_disks; i++) | |
2225 | ndisks[i] = conf->disks[i]; | |
2226 | kfree(conf->disks); | |
2227 | conf->disks = ndisks; | |
2228 | } else | |
2229 | err = -ENOMEM; | |
2230 | ||
2d5b569b | 2231 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2232 | /* Step 4, return new stripes to service */ |
2233 | while(!list_empty(&newstripes)) { | |
2234 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2235 | list_del_init(&nsh->lru); | |
d6f38f31 | 2236 | |
ad01c9e3 N |
2237 | for (i=conf->raid_disks; i < newsize; i++) |
2238 | if (nsh->dev[i].page == NULL) { | |
2239 | struct page *p = alloc_page(GFP_NOIO); | |
2240 | nsh->dev[i].page = p; | |
d592a996 | 2241 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2242 | if (!p) |
2243 | err = -ENOMEM; | |
2244 | } | |
2245 | release_stripe(nsh); | |
2246 | } | |
2247 | /* critical section pass, GFP_NOIO no longer needed */ | |
2248 | ||
2249 | conf->slab_cache = sc; | |
2250 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2251 | if (!err) |
2252 | conf->pool_size = newsize; | |
ad01c9e3 N |
2253 | return err; |
2254 | } | |
1da177e4 | 2255 | |
486f0644 | 2256 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2257 | { |
2258 | struct stripe_head *sh; | |
49895bcc | 2259 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2260 | |
566c09c5 SL |
2261 | spin_lock_irq(conf->hash_locks + hash); |
2262 | sh = get_free_stripe(conf, hash); | |
2263 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2264 | if (!sh) |
2265 | return 0; | |
78bafebd | 2266 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2267 | shrink_buffers(sh); |
3f294f4f N |
2268 | kmem_cache_free(conf->slab_cache, sh); |
2269 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2270 | conf->max_nr_stripes--; |
3f294f4f N |
2271 | return 1; |
2272 | } | |
2273 | ||
d1688a6d | 2274 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2275 | { |
486f0644 N |
2276 | while (conf->max_nr_stripes && |
2277 | drop_one_stripe(conf)) | |
2278 | ; | |
3f294f4f | 2279 | |
29fc7e3e N |
2280 | if (conf->slab_cache) |
2281 | kmem_cache_destroy(conf->slab_cache); | |
1da177e4 LT |
2282 | conf->slab_cache = NULL; |
2283 | } | |
2284 | ||
6712ecf8 | 2285 | static void raid5_end_read_request(struct bio * bi, int error) |
1da177e4 | 2286 | { |
99c0fb5f | 2287 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2288 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2289 | int disks = sh->disks, i; |
1da177e4 | 2290 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
d6950432 | 2291 | char b[BDEVNAME_SIZE]; |
dd054fce | 2292 | struct md_rdev *rdev = NULL; |
05616be5 | 2293 | sector_t s; |
1da177e4 LT |
2294 | |
2295 | for (i=0 ; i<disks; i++) | |
2296 | if (bi == &sh->dev[i].req) | |
2297 | break; | |
2298 | ||
45b4233c DW |
2299 | pr_debug("end_read_request %llu/%d, count: %d, uptodate %d.\n", |
2300 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), | |
1da177e4 LT |
2301 | uptodate); |
2302 | if (i == disks) { | |
2303 | BUG(); | |
6712ecf8 | 2304 | return; |
1da177e4 | 2305 | } |
14a75d3e | 2306 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2307 | /* If replacement finished while this request was outstanding, |
2308 | * 'replacement' might be NULL already. | |
2309 | * In that case it moved down to 'rdev'. | |
2310 | * rdev is not removed until all requests are finished. | |
2311 | */ | |
14a75d3e | 2312 | rdev = conf->disks[i].replacement; |
dd054fce | 2313 | if (!rdev) |
14a75d3e | 2314 | rdev = conf->disks[i].rdev; |
1da177e4 | 2315 | |
05616be5 N |
2316 | if (use_new_offset(conf, sh)) |
2317 | s = sh->sector + rdev->new_data_offset; | |
2318 | else | |
2319 | s = sh->sector + rdev->data_offset; | |
1da177e4 | 2320 | if (uptodate) { |
1da177e4 | 2321 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2322 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2323 | /* Note that this cannot happen on a |
2324 | * replacement device. We just fail those on | |
2325 | * any error | |
2326 | */ | |
8bda470e CD |
2327 | printk_ratelimited( |
2328 | KERN_INFO | |
2329 | "md/raid:%s: read error corrected" | |
2330 | " (%lu sectors at %llu on %s)\n", | |
2331 | mdname(conf->mddev), STRIPE_SECTORS, | |
05616be5 | 2332 | (unsigned long long)s, |
8bda470e | 2333 | bdevname(rdev->bdev, b)); |
ddd5115f | 2334 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2335 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2336 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2337 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2338 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2339 | ||
14a75d3e N |
2340 | if (atomic_read(&rdev->read_errors)) |
2341 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2342 | } else { |
14a75d3e | 2343 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2344 | int retry = 0; |
2e8ac303 | 2345 | int set_bad = 0; |
d6950432 | 2346 | |
1da177e4 | 2347 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2348 | atomic_inc(&rdev->read_errors); |
14a75d3e N |
2349 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
2350 | printk_ratelimited( | |
2351 | KERN_WARNING | |
2352 | "md/raid:%s: read error on replacement device " | |
2353 | "(sector %llu on %s).\n", | |
2354 | mdname(conf->mddev), | |
05616be5 | 2355 | (unsigned long long)s, |
14a75d3e | 2356 | bdn); |
2e8ac303 | 2357 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2358 | set_bad = 1; | |
8bda470e CD |
2359 | printk_ratelimited( |
2360 | KERN_WARNING | |
2361 | "md/raid:%s: read error not correctable " | |
2362 | "(sector %llu on %s).\n", | |
2363 | mdname(conf->mddev), | |
05616be5 | 2364 | (unsigned long long)s, |
8bda470e | 2365 | bdn); |
2e8ac303 | 2366 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2367 | /* Oh, no!!! */ |
2e8ac303 | 2368 | set_bad = 1; |
8bda470e CD |
2369 | printk_ratelimited( |
2370 | KERN_WARNING | |
2371 | "md/raid:%s: read error NOT corrected!! " | |
2372 | "(sector %llu on %s).\n", | |
2373 | mdname(conf->mddev), | |
05616be5 | 2374 | (unsigned long long)s, |
8bda470e | 2375 | bdn); |
2e8ac303 | 2376 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2377 | > conf->max_nr_stripes) |
14f8d26b | 2378 | printk(KERN_WARNING |
0c55e022 | 2379 | "md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2380 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2381 | else |
2382 | retry = 1; | |
edfa1f65 BY |
2383 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2384 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2385 | retry = 1; | |
ba22dcbf | 2386 | if (retry) |
3f9e7c14 | 2387 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2388 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2389 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2390 | } else | |
2391 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2392 | else { |
4e5314b5 N |
2393 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2394 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2395 | if (!(set_bad |
2396 | && test_bit(In_sync, &rdev->flags) | |
2397 | && rdev_set_badblocks( | |
2398 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2399 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2400 | } |
1da177e4 | 2401 | } |
14a75d3e | 2402 | rdev_dec_pending(rdev, conf->mddev); |
1da177e4 LT |
2403 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2404 | set_bit(STRIPE_HANDLE, &sh->state); | |
2405 | release_stripe(sh); | |
1da177e4 LT |
2406 | } |
2407 | ||
d710e138 | 2408 | static void raid5_end_write_request(struct bio *bi, int error) |
1da177e4 | 2409 | { |
99c0fb5f | 2410 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2411 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2412 | int disks = sh->disks, i; |
977df362 | 2413 | struct md_rdev *uninitialized_var(rdev); |
1da177e4 | 2414 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
b84db560 N |
2415 | sector_t first_bad; |
2416 | int bad_sectors; | |
977df362 | 2417 | int replacement = 0; |
1da177e4 | 2418 | |
977df362 N |
2419 | for (i = 0 ; i < disks; i++) { |
2420 | if (bi == &sh->dev[i].req) { | |
2421 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2422 | break; |
977df362 N |
2423 | } |
2424 | if (bi == &sh->dev[i].rreq) { | |
2425 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2426 | if (rdev) |
2427 | replacement = 1; | |
2428 | else | |
2429 | /* rdev was removed and 'replacement' | |
2430 | * replaced it. rdev is not removed | |
2431 | * until all requests are finished. | |
2432 | */ | |
2433 | rdev = conf->disks[i].rdev; | |
977df362 N |
2434 | break; |
2435 | } | |
2436 | } | |
45b4233c | 2437 | pr_debug("end_write_request %llu/%d, count %d, uptodate: %d.\n", |
1da177e4 LT |
2438 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
2439 | uptodate); | |
2440 | if (i == disks) { | |
2441 | BUG(); | |
6712ecf8 | 2442 | return; |
1da177e4 LT |
2443 | } |
2444 | ||
977df362 N |
2445 | if (replacement) { |
2446 | if (!uptodate) | |
2447 | md_error(conf->mddev, rdev); | |
2448 | else if (is_badblock(rdev, sh->sector, | |
2449 | STRIPE_SECTORS, | |
2450 | &first_bad, &bad_sectors)) | |
2451 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2452 | } else { | |
2453 | if (!uptodate) { | |
9f97e4b1 | 2454 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2455 | set_bit(WriteErrorSeen, &rdev->flags); |
2456 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2457 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2458 | set_bit(MD_RECOVERY_NEEDED, | |
2459 | &rdev->mddev->recovery); | |
977df362 N |
2460 | } else if (is_badblock(rdev, sh->sector, |
2461 | STRIPE_SECTORS, | |
c0b32972 | 2462 | &first_bad, &bad_sectors)) { |
977df362 | 2463 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2464 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2465 | /* That was a successful write so make | |
2466 | * sure it looks like we already did | |
2467 | * a re-write. | |
2468 | */ | |
2469 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2470 | } | |
977df362 N |
2471 | } |
2472 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2473 | |
bb27051f | 2474 | if (sh->batch_head && !uptodate && !replacement) |
72ac7330 | 2475 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2476 | ||
977df362 N |
2477 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2478 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2479 | set_bit(STRIPE_HANDLE, &sh->state); |
c04be0aa | 2480 | release_stripe(sh); |
59fc630b | 2481 | |
2482 | if (sh->batch_head && sh != sh->batch_head) | |
2483 | release_stripe(sh->batch_head); | |
1da177e4 LT |
2484 | } |
2485 | ||
784052ec | 2486 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous); |
d592a996 | 2487 | |
784052ec | 2488 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2489 | { |
2490 | struct r5dev *dev = &sh->dev[i]; | |
2491 | ||
2492 | bio_init(&dev->req); | |
2493 | dev->req.bi_io_vec = &dev->vec; | |
d592a996 | 2494 | dev->req.bi_max_vecs = 1; |
1da177e4 LT |
2495 | dev->req.bi_private = sh; |
2496 | ||
977df362 N |
2497 | bio_init(&dev->rreq); |
2498 | dev->rreq.bi_io_vec = &dev->rvec; | |
d592a996 | 2499 | dev->rreq.bi_max_vecs = 1; |
977df362 | 2500 | dev->rreq.bi_private = sh; |
977df362 | 2501 | |
1da177e4 | 2502 | dev->flags = 0; |
784052ec | 2503 | dev->sector = compute_blocknr(sh, i, previous); |
1da177e4 LT |
2504 | } |
2505 | ||
fd01b88c | 2506 | static void error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2507 | { |
2508 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2509 | struct r5conf *conf = mddev->private; |
908f4fbd | 2510 | unsigned long flags; |
0c55e022 | 2511 | pr_debug("raid456: error called\n"); |
1da177e4 | 2512 | |
908f4fbd N |
2513 | spin_lock_irqsave(&conf->device_lock, flags); |
2514 | clear_bit(In_sync, &rdev->flags); | |
2515 | mddev->degraded = calc_degraded(conf); | |
2516 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2517 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2518 | ||
de393cde | 2519 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 N |
2520 | set_bit(Faulty, &rdev->flags); |
2521 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2522 | printk(KERN_ALERT | |
2523 | "md/raid:%s: Disk failure on %s, disabling device.\n" | |
2524 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2525 | mdname(mddev), | |
2526 | bdevname(rdev->bdev, b), | |
2527 | mdname(mddev), | |
2528 | conf->raid_disks - mddev->degraded); | |
16a53ecc | 2529 | } |
1da177e4 LT |
2530 | |
2531 | /* | |
2532 | * Input: a 'big' sector number, | |
2533 | * Output: index of the data and parity disk, and the sector # in them. | |
2534 | */ | |
d1688a6d | 2535 | static sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
911d4ee8 N |
2536 | int previous, int *dd_idx, |
2537 | struct stripe_head *sh) | |
1da177e4 | 2538 | { |
6e3b96ed | 2539 | sector_t stripe, stripe2; |
35f2a591 | 2540 | sector_t chunk_number; |
1da177e4 | 2541 | unsigned int chunk_offset; |
911d4ee8 | 2542 | int pd_idx, qd_idx; |
67cc2b81 | 2543 | int ddf_layout = 0; |
1da177e4 | 2544 | sector_t new_sector; |
e183eaed N |
2545 | int algorithm = previous ? conf->prev_algo |
2546 | : conf->algorithm; | |
09c9e5fa AN |
2547 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2548 | : conf->chunk_sectors; | |
112bf897 N |
2549 | int raid_disks = previous ? conf->previous_raid_disks |
2550 | : conf->raid_disks; | |
2551 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2552 | |
2553 | /* First compute the information on this sector */ | |
2554 | ||
2555 | /* | |
2556 | * Compute the chunk number and the sector offset inside the chunk | |
2557 | */ | |
2558 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2559 | chunk_number = r_sector; | |
1da177e4 LT |
2560 | |
2561 | /* | |
2562 | * Compute the stripe number | |
2563 | */ | |
35f2a591 N |
2564 | stripe = chunk_number; |
2565 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2566 | stripe2 = stripe; |
1da177e4 LT |
2567 | /* |
2568 | * Select the parity disk based on the user selected algorithm. | |
2569 | */ | |
84789554 | 2570 | pd_idx = qd_idx = -1; |
16a53ecc N |
2571 | switch(conf->level) { |
2572 | case 4: | |
911d4ee8 | 2573 | pd_idx = data_disks; |
16a53ecc N |
2574 | break; |
2575 | case 5: | |
e183eaed | 2576 | switch (algorithm) { |
1da177e4 | 2577 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2578 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2579 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2580 | (*dd_idx)++; |
2581 | break; | |
2582 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2583 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2584 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2585 | (*dd_idx)++; |
2586 | break; | |
2587 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2588 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2589 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2590 | break; |
2591 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2592 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2593 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2594 | break; |
99c0fb5f N |
2595 | case ALGORITHM_PARITY_0: |
2596 | pd_idx = 0; | |
2597 | (*dd_idx)++; | |
2598 | break; | |
2599 | case ALGORITHM_PARITY_N: | |
2600 | pd_idx = data_disks; | |
2601 | break; | |
1da177e4 | 2602 | default: |
99c0fb5f | 2603 | BUG(); |
16a53ecc N |
2604 | } |
2605 | break; | |
2606 | case 6: | |
2607 | ||
e183eaed | 2608 | switch (algorithm) { |
16a53ecc | 2609 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2610 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2611 | qd_idx = pd_idx + 1; |
2612 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2613 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2614 | qd_idx = 0; |
2615 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2616 | (*dd_idx) += 2; /* D D P Q D */ |
2617 | break; | |
2618 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2619 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2620 | qd_idx = pd_idx + 1; |
2621 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2622 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2623 | qd_idx = 0; |
2624 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2625 | (*dd_idx) += 2; /* D D P Q D */ |
2626 | break; | |
2627 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2628 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2629 | qd_idx = (pd_idx + 1) % raid_disks; |
2630 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2631 | break; |
2632 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2633 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2634 | qd_idx = (pd_idx + 1) % raid_disks; |
2635 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2636 | break; |
99c0fb5f N |
2637 | |
2638 | case ALGORITHM_PARITY_0: | |
2639 | pd_idx = 0; | |
2640 | qd_idx = 1; | |
2641 | (*dd_idx) += 2; | |
2642 | break; | |
2643 | case ALGORITHM_PARITY_N: | |
2644 | pd_idx = data_disks; | |
2645 | qd_idx = data_disks + 1; | |
2646 | break; | |
2647 | ||
2648 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2649 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2650 | * of blocks for computing Q is different. | |
2651 | */ | |
6e3b96ed | 2652 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2653 | qd_idx = pd_idx + 1; |
2654 | if (pd_idx == raid_disks-1) { | |
2655 | (*dd_idx)++; /* Q D D D P */ | |
2656 | qd_idx = 0; | |
2657 | } else if (*dd_idx >= pd_idx) | |
2658 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2659 | ddf_layout = 1; |
99c0fb5f N |
2660 | break; |
2661 | ||
2662 | case ALGORITHM_ROTATING_N_RESTART: | |
2663 | /* Same a left_asymmetric, by first stripe is | |
2664 | * D D D P Q rather than | |
2665 | * Q D D D P | |
2666 | */ | |
6e3b96ed N |
2667 | stripe2 += 1; |
2668 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2669 | qd_idx = pd_idx + 1; |
2670 | if (pd_idx == raid_disks-1) { | |
2671 | (*dd_idx)++; /* Q D D D P */ | |
2672 | qd_idx = 0; | |
2673 | } else if (*dd_idx >= pd_idx) | |
2674 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2675 | ddf_layout = 1; |
99c0fb5f N |
2676 | break; |
2677 | ||
2678 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2679 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2680 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2681 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2682 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2683 | ddf_layout = 1; |
99c0fb5f N |
2684 | break; |
2685 | ||
2686 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2687 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2688 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2689 | if (*dd_idx >= pd_idx) |
2690 | (*dd_idx)++; | |
2691 | qd_idx = raid_disks - 1; | |
2692 | break; | |
2693 | ||
2694 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2695 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2696 | if (*dd_idx >= pd_idx) |
2697 | (*dd_idx)++; | |
2698 | qd_idx = raid_disks - 1; | |
2699 | break; | |
2700 | ||
2701 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2702 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2703 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2704 | qd_idx = raid_disks - 1; | |
2705 | break; | |
2706 | ||
2707 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2708 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2709 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2710 | qd_idx = raid_disks - 1; | |
2711 | break; | |
2712 | ||
2713 | case ALGORITHM_PARITY_0_6: | |
2714 | pd_idx = 0; | |
2715 | (*dd_idx)++; | |
2716 | qd_idx = raid_disks - 1; | |
2717 | break; | |
2718 | ||
16a53ecc | 2719 | default: |
99c0fb5f | 2720 | BUG(); |
16a53ecc N |
2721 | } |
2722 | break; | |
1da177e4 LT |
2723 | } |
2724 | ||
911d4ee8 N |
2725 | if (sh) { |
2726 | sh->pd_idx = pd_idx; | |
2727 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2728 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2729 | } |
1da177e4 LT |
2730 | /* |
2731 | * Finally, compute the new sector number | |
2732 | */ | |
2733 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2734 | return new_sector; | |
2735 | } | |
2736 | ||
784052ec | 2737 | static sector_t compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2738 | { |
d1688a6d | 2739 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2740 | int raid_disks = sh->disks; |
2741 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2742 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2743 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2744 | : conf->chunk_sectors; | |
e183eaed N |
2745 | int algorithm = previous ? conf->prev_algo |
2746 | : conf->algorithm; | |
1da177e4 LT |
2747 | sector_t stripe; |
2748 | int chunk_offset; | |
35f2a591 N |
2749 | sector_t chunk_number; |
2750 | int dummy1, dd_idx = i; | |
1da177e4 | 2751 | sector_t r_sector; |
911d4ee8 | 2752 | struct stripe_head sh2; |
1da177e4 LT |
2753 | |
2754 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2755 | stripe = new_sector; | |
1da177e4 | 2756 | |
16a53ecc N |
2757 | if (i == sh->pd_idx) |
2758 | return 0; | |
2759 | switch(conf->level) { | |
2760 | case 4: break; | |
2761 | case 5: | |
e183eaed | 2762 | switch (algorithm) { |
1da177e4 LT |
2763 | case ALGORITHM_LEFT_ASYMMETRIC: |
2764 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2765 | if (i > sh->pd_idx) | |
2766 | i--; | |
2767 | break; | |
2768 | case ALGORITHM_LEFT_SYMMETRIC: | |
2769 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2770 | if (i < sh->pd_idx) | |
2771 | i += raid_disks; | |
2772 | i -= (sh->pd_idx + 1); | |
2773 | break; | |
99c0fb5f N |
2774 | case ALGORITHM_PARITY_0: |
2775 | i -= 1; | |
2776 | break; | |
2777 | case ALGORITHM_PARITY_N: | |
2778 | break; | |
1da177e4 | 2779 | default: |
99c0fb5f | 2780 | BUG(); |
16a53ecc N |
2781 | } |
2782 | break; | |
2783 | case 6: | |
d0dabf7e | 2784 | if (i == sh->qd_idx) |
16a53ecc | 2785 | return 0; /* It is the Q disk */ |
e183eaed | 2786 | switch (algorithm) { |
16a53ecc N |
2787 | case ALGORITHM_LEFT_ASYMMETRIC: |
2788 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2789 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2790 | case ALGORITHM_ROTATING_N_RESTART: | |
2791 | if (sh->pd_idx == raid_disks-1) | |
2792 | i--; /* Q D D D P */ | |
16a53ecc N |
2793 | else if (i > sh->pd_idx) |
2794 | i -= 2; /* D D P Q D */ | |
2795 | break; | |
2796 | case ALGORITHM_LEFT_SYMMETRIC: | |
2797 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2798 | if (sh->pd_idx == raid_disks-1) | |
2799 | i--; /* Q D D D P */ | |
2800 | else { | |
2801 | /* D D P Q D */ | |
2802 | if (i < sh->pd_idx) | |
2803 | i += raid_disks; | |
2804 | i -= (sh->pd_idx + 2); | |
2805 | } | |
2806 | break; | |
99c0fb5f N |
2807 | case ALGORITHM_PARITY_0: |
2808 | i -= 2; | |
2809 | break; | |
2810 | case ALGORITHM_PARITY_N: | |
2811 | break; | |
2812 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2813 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2814 | if (sh->pd_idx == 0) |
2815 | i--; /* P D D D Q */ | |
e4424fee N |
2816 | else { |
2817 | /* D D Q P D */ | |
2818 | if (i < sh->pd_idx) | |
2819 | i += raid_disks; | |
2820 | i -= (sh->pd_idx + 1); | |
2821 | } | |
99c0fb5f N |
2822 | break; |
2823 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2824 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2825 | if (i > sh->pd_idx) | |
2826 | i--; | |
2827 | break; | |
2828 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2829 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2830 | if (i < sh->pd_idx) | |
2831 | i += data_disks + 1; | |
2832 | i -= (sh->pd_idx + 1); | |
2833 | break; | |
2834 | case ALGORITHM_PARITY_0_6: | |
2835 | i -= 1; | |
2836 | break; | |
16a53ecc | 2837 | default: |
99c0fb5f | 2838 | BUG(); |
16a53ecc N |
2839 | } |
2840 | break; | |
1da177e4 LT |
2841 | } |
2842 | ||
2843 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2844 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2845 | |
112bf897 | 2846 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2847 | previous, &dummy1, &sh2); |
911d4ee8 N |
2848 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2849 | || sh2.qd_idx != sh->qd_idx) { | |
0c55e022 N |
2850 | printk(KERN_ERR "md/raid:%s: compute_blocknr: map not correct\n", |
2851 | mdname(conf->mddev)); | |
1da177e4 LT |
2852 | return 0; |
2853 | } | |
2854 | return r_sector; | |
2855 | } | |
2856 | ||
600aa109 | 2857 | static void |
c0f7bddb | 2858 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2859 | int rcw, int expand) |
e33129d8 | 2860 | { |
584acdd4 | 2861 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 2862 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2863 | int level = conf->level; |
e33129d8 DW |
2864 | |
2865 | if (rcw) { | |
e33129d8 DW |
2866 | |
2867 | for (i = disks; i--; ) { | |
2868 | struct r5dev *dev = &sh->dev[i]; | |
2869 | ||
2870 | if (dev->towrite) { | |
2871 | set_bit(R5_LOCKED, &dev->flags); | |
d8ee0728 | 2872 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
2873 | if (!expand) |
2874 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2875 | s->locked++; |
e33129d8 DW |
2876 | } |
2877 | } | |
ce7d363a N |
2878 | /* if we are not expanding this is a proper write request, and |
2879 | * there will be bios with new data to be drained into the | |
2880 | * stripe cache | |
2881 | */ | |
2882 | if (!expand) { | |
2883 | if (!s->locked) | |
2884 | /* False alarm, nothing to do */ | |
2885 | return; | |
2886 | sh->reconstruct_state = reconstruct_state_drain_run; | |
2887 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2888 | } else | |
2889 | sh->reconstruct_state = reconstruct_state_run; | |
2890 | ||
2891 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
2892 | ||
c0f7bddb | 2893 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 2894 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 2895 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
2896 | } else { |
2897 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
2898 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
2899 | BUG_ON(level == 6 && |
2900 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
2901 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 2902 | |
e33129d8 DW |
2903 | for (i = disks; i--; ) { |
2904 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 2905 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
2906 | continue; |
2907 | ||
e33129d8 DW |
2908 | if (dev->towrite && |
2909 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
2910 | test_bit(R5_Wantcompute, &dev->flags))) { |
2911 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
2912 | set_bit(R5_LOCKED, &dev->flags); |
2913 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 2914 | s->locked++; |
e33129d8 DW |
2915 | } |
2916 | } | |
ce7d363a N |
2917 | if (!s->locked) |
2918 | /* False alarm - nothing to do */ | |
2919 | return; | |
2920 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
2921 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
2922 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
2923 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
2924 | } |
2925 | ||
c0f7bddb | 2926 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
2927 | * are in flight |
2928 | */ | |
2929 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
2930 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 2931 | s->locked++; |
e33129d8 | 2932 | |
c0f7bddb YT |
2933 | if (level == 6) { |
2934 | int qd_idx = sh->qd_idx; | |
2935 | struct r5dev *dev = &sh->dev[qd_idx]; | |
2936 | ||
2937 | set_bit(R5_LOCKED, &dev->flags); | |
2938 | clear_bit(R5_UPTODATE, &dev->flags); | |
2939 | s->locked++; | |
2940 | } | |
2941 | ||
600aa109 | 2942 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 2943 | __func__, (unsigned long long)sh->sector, |
600aa109 | 2944 | s->locked, s->ops_request); |
e33129d8 | 2945 | } |
16a53ecc | 2946 | |
1da177e4 LT |
2947 | /* |
2948 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 2949 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
2950 | * The bi_next chain must be in order. |
2951 | */ | |
da41ba65 | 2952 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
2953 | int forwrite, int previous) | |
1da177e4 LT |
2954 | { |
2955 | struct bio **bip; | |
d1688a6d | 2956 | struct r5conf *conf = sh->raid_conf; |
72626685 | 2957 | int firstwrite=0; |
1da177e4 | 2958 | |
cbe47ec5 | 2959 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 2960 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
2961 | (unsigned long long)sh->sector); |
2962 | ||
b17459c0 SL |
2963 | /* |
2964 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
2965 | * reference count to avoid race. The reference count should already be | |
2966 | * increased before this function is called (for example, in | |
2967 | * make_request()), so other bio sharing this stripe will not free the | |
2968 | * stripe. If a stripe is owned by one stripe, the stripe lock will | |
2969 | * protect it. | |
2970 | */ | |
2971 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 2972 | /* Don't allow new IO added to stripes in batch list */ |
2973 | if (sh->batch_head) | |
2974 | goto overlap; | |
72626685 | 2975 | if (forwrite) { |
1da177e4 | 2976 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 2977 | if (*bip == NULL) |
72626685 N |
2978 | firstwrite = 1; |
2979 | } else | |
1da177e4 | 2980 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
2981 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
2982 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
2983 | goto overlap; |
2984 | bip = & (*bip)->bi_next; | |
2985 | } | |
4f024f37 | 2986 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
2987 | goto overlap; |
2988 | ||
da41ba65 | 2989 | if (!forwrite || previous) |
2990 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
2991 | ||
78bafebd | 2992 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
2993 | if (*bip) |
2994 | bi->bi_next = *bip; | |
2995 | *bip = bi; | |
e7836bd6 | 2996 | raid5_inc_bi_active_stripes(bi); |
72626685 | 2997 | |
1da177e4 LT |
2998 | if (forwrite) { |
2999 | /* check if page is covered */ | |
3000 | sector_t sector = sh->dev[dd_idx].sector; | |
3001 | for (bi=sh->dev[dd_idx].towrite; | |
3002 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3003 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3004 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3005 | if (bio_end_sector(bi) >= sector) |
3006 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3007 | } |
3008 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3009 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3010 | sh->overwrite_disks++; | |
1da177e4 | 3011 | } |
cbe47ec5 N |
3012 | |
3013 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3014 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3015 | (unsigned long long)sh->sector, dd_idx); |
3016 | ||
3017 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3018 | /* Cannot hold spinlock over bitmap_startwrite, |
3019 | * but must ensure this isn't added to a batch until | |
3020 | * we have added to the bitmap and set bm_seq. | |
3021 | * So set STRIPE_BITMAP_PENDING to prevent | |
3022 | * batching. | |
3023 | * If multiple add_stripe_bio() calls race here they | |
3024 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3025 | * to complete "bitmap_startwrite" gets to set | |
3026 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3027 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3028 | * any more. | |
3029 | */ | |
3030 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3031 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3032 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3033 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3034 | spin_lock_irq(&sh->stripe_lock); |
3035 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3036 | if (!sh->batch_head) { | |
3037 | sh->bm_seq = conf->seq_flush+1; | |
3038 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3039 | } | |
cbe47ec5 | 3040 | } |
d0852df5 | 3041 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3042 | |
3043 | if (stripe_can_batch(sh)) | |
3044 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3045 | return 1; |
3046 | ||
3047 | overlap: | |
3048 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3049 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3050 | return 0; |
3051 | } | |
3052 | ||
d1688a6d | 3053 | static void end_reshape(struct r5conf *conf); |
29269553 | 3054 | |
d1688a6d | 3055 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3056 | struct stripe_head *sh) |
ccfcc3c1 | 3057 | { |
784052ec | 3058 | int sectors_per_chunk = |
09c9e5fa | 3059 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3060 | int dd_idx; |
2d2063ce | 3061 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3062 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3063 | |
112bf897 N |
3064 | raid5_compute_sector(conf, |
3065 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3066 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3067 | previous, |
911d4ee8 | 3068 | &dd_idx, sh); |
ccfcc3c1 N |
3069 | } |
3070 | ||
a4456856 | 3071 | static void |
d1688a6d | 3072 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3073 | struct stripe_head_state *s, int disks, |
3074 | struct bio **return_bi) | |
3075 | { | |
3076 | int i; | |
59fc630b | 3077 | BUG_ON(sh->batch_head); |
a4456856 DW |
3078 | for (i = disks; i--; ) { |
3079 | struct bio *bi; | |
3080 | int bitmap_end = 0; | |
3081 | ||
3082 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3083 | struct md_rdev *rdev; |
a4456856 DW |
3084 | rcu_read_lock(); |
3085 | rdev = rcu_dereference(conf->disks[i].rdev); | |
3086 | if (rdev && test_bit(In_sync, &rdev->flags)) | |
7f0da59b N |
3087 | atomic_inc(&rdev->nr_pending); |
3088 | else | |
3089 | rdev = NULL; | |
a4456856 | 3090 | rcu_read_unlock(); |
7f0da59b N |
3091 | if (rdev) { |
3092 | if (!rdev_set_badblocks( | |
3093 | rdev, | |
3094 | sh->sector, | |
3095 | STRIPE_SECTORS, 0)) | |
3096 | md_error(conf->mddev, rdev); | |
3097 | rdev_dec_pending(rdev, conf->mddev); | |
3098 | } | |
a4456856 | 3099 | } |
b17459c0 | 3100 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3101 | /* fail all writes first */ |
3102 | bi = sh->dev[i].towrite; | |
3103 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3104 | sh->overwrite_disks = 0; |
b17459c0 | 3105 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3106 | if (bi) |
a4456856 | 3107 | bitmap_end = 1; |
a4456856 DW |
3108 | |
3109 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
3110 | wake_up(&conf->wait_for_overlap); | |
3111 | ||
4f024f37 | 3112 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3113 | sh->dev[i].sector + STRIPE_SECTORS) { |
3114 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
3115 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3116 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3117 | md_write_end(conf->mddev); |
3118 | bi->bi_next = *return_bi; | |
3119 | *return_bi = bi; | |
3120 | } | |
3121 | bi = nextbi; | |
3122 | } | |
7eaf7e8e SL |
3123 | if (bitmap_end) |
3124 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3125 | STRIPE_SECTORS, 0, 0); | |
3126 | bitmap_end = 0; | |
a4456856 DW |
3127 | /* and fail all 'written' */ |
3128 | bi = sh->dev[i].written; | |
3129 | sh->dev[i].written = NULL; | |
d592a996 SL |
3130 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3131 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3132 | sh->dev[i].page = sh->dev[i].orig_page; | |
3133 | } | |
3134 | ||
a4456856 | 3135 | if (bi) bitmap_end = 1; |
4f024f37 | 3136 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3137 | sh->dev[i].sector + STRIPE_SECTORS) { |
3138 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
3139 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3140 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3141 | md_write_end(conf->mddev); |
3142 | bi->bi_next = *return_bi; | |
3143 | *return_bi = bi; | |
3144 | } | |
3145 | bi = bi2; | |
3146 | } | |
3147 | ||
b5e98d65 DW |
3148 | /* fail any reads if this device is non-operational and |
3149 | * the data has not reached the cache yet. | |
3150 | */ | |
3151 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
3152 | (!test_bit(R5_Insync, &sh->dev[i].flags) || | |
3153 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3154 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3155 | bi = sh->dev[i].toread; |
3156 | sh->dev[i].toread = NULL; | |
143c4d05 | 3157 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3158 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3159 | wake_up(&conf->wait_for_overlap); | |
4f024f37 | 3160 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3161 | sh->dev[i].sector + STRIPE_SECTORS) { |
3162 | struct bio *nextbi = | |
3163 | r5_next_bio(bi, sh->dev[i].sector); | |
3164 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
e7836bd6 | 3165 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 DW |
3166 | bi->bi_next = *return_bi; |
3167 | *return_bi = bi; | |
3168 | } | |
3169 | bi = nextbi; | |
3170 | } | |
3171 | } | |
a4456856 DW |
3172 | if (bitmap_end) |
3173 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3174 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3175 | /* If we were in the middle of a write the parity block might |
3176 | * still be locked - so just clear all R5_LOCKED flags | |
3177 | */ | |
3178 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 DW |
3179 | } |
3180 | ||
8b3e6cdc DW |
3181 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3182 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3183 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3184 | } |
3185 | ||
7f0da59b | 3186 | static void |
d1688a6d | 3187 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3188 | struct stripe_head_state *s) |
3189 | { | |
3190 | int abort = 0; | |
3191 | int i; | |
3192 | ||
59fc630b | 3193 | BUG_ON(sh->batch_head); |
7f0da59b | 3194 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3195 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3196 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3197 | s->syncing = 0; |
9a3e1101 | 3198 | s->replacing = 0; |
7f0da59b | 3199 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3200 | * Don't even need to abort as that is handled elsewhere |
3201 | * if needed, and not always wanted e.g. if there is a known | |
3202 | * bad block here. | |
9a3e1101 | 3203 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3204 | * non-sync devices, or abort the recovery |
3205 | */ | |
18b9837e N |
3206 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3207 | /* During recovery devices cannot be removed, so | |
3208 | * locking and refcounting of rdevs is not needed | |
3209 | */ | |
3210 | for (i = 0; i < conf->raid_disks; i++) { | |
3211 | struct md_rdev *rdev = conf->disks[i].rdev; | |
3212 | if (rdev | |
3213 | && !test_bit(Faulty, &rdev->flags) | |
3214 | && !test_bit(In_sync, &rdev->flags) | |
3215 | && !rdev_set_badblocks(rdev, sh->sector, | |
3216 | STRIPE_SECTORS, 0)) | |
3217 | abort = 1; | |
3218 | rdev = conf->disks[i].replacement; | |
3219 | if (rdev | |
3220 | && !test_bit(Faulty, &rdev->flags) | |
3221 | && !test_bit(In_sync, &rdev->flags) | |
3222 | && !rdev_set_badblocks(rdev, sh->sector, | |
3223 | STRIPE_SECTORS, 0)) | |
3224 | abort = 1; | |
3225 | } | |
3226 | if (abort) | |
3227 | conf->recovery_disabled = | |
3228 | conf->mddev->recovery_disabled; | |
7f0da59b | 3229 | } |
18b9837e | 3230 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3231 | } |
3232 | ||
9a3e1101 N |
3233 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3234 | { | |
3235 | struct md_rdev *rdev; | |
3236 | int rv = 0; | |
3237 | /* Doing recovery so rcu locking not required */ | |
3238 | rdev = sh->raid_conf->disks[disk_idx].replacement; | |
3239 | if (rdev | |
3240 | && !test_bit(Faulty, &rdev->flags) | |
3241 | && !test_bit(In_sync, &rdev->flags) | |
3242 | && (rdev->recovery_offset <= sh->sector | |
3243 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3244 | rv = 1; | |
3245 | ||
3246 | return rv; | |
3247 | } | |
3248 | ||
93b3dbce | 3249 | /* fetch_block - checks the given member device to see if its data needs |
1fe797e6 DW |
3250 | * to be read or computed to satisfy a request. |
3251 | * | |
3252 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
93b3dbce | 3253 | * 0 to tell the loop in handle_stripe_fill to continue |
f38e1219 | 3254 | */ |
2c58f06e N |
3255 | |
3256 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3257 | int disk_idx, int disks) | |
a4456856 | 3258 | { |
5599becc | 3259 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3260 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3261 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3262 | int i; |
5599becc | 3263 | |
a79cfe12 N |
3264 | |
3265 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3266 | test_bit(R5_UPTODATE, &dev->flags)) | |
3267 | /* No point reading this as we already have it or have | |
3268 | * decided to get it. | |
3269 | */ | |
3270 | return 0; | |
3271 | ||
3272 | if (dev->toread || | |
3273 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3274 | /* We need this block to directly satisfy a request */ | |
3275 | return 1; | |
3276 | ||
3277 | if (s->syncing || s->expanding || | |
3278 | (s->replacing && want_replace(sh, disk_idx))) | |
3279 | /* When syncing, or expanding we read everything. | |
3280 | * When replacing, we need the replaced block. | |
3281 | */ | |
3282 | return 1; | |
3283 | ||
3284 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3285 | (s->failed >= 2 && fdev[1]->toread)) | |
3286 | /* If we want to read from a failed device, then | |
3287 | * we need to actually read every other device. | |
3288 | */ | |
3289 | return 1; | |
3290 | ||
a9d56950 N |
3291 | /* Sometimes neither read-modify-write nor reconstruct-write |
3292 | * cycles can work. In those cases we read every block we | |
3293 | * can. Then the parity-update is certain to have enough to | |
3294 | * work with. | |
3295 | * This can only be a problem when we need to write something, | |
3296 | * and some device has failed. If either of those tests | |
3297 | * fail we need look no further. | |
3298 | */ | |
3299 | if (!s->failed || !s->to_write) | |
3300 | return 0; | |
3301 | ||
3302 | if (test_bit(R5_Insync, &dev->flags) && | |
3303 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3304 | /* Pre-reads at not permitted until after short delay | |
3305 | * to gather multiple requests. However if this | |
3306 | * device is no Insync, the block could only be be computed | |
3307 | * and there is no need to delay that. | |
3308 | */ | |
3309 | return 0; | |
ea664c82 N |
3310 | |
3311 | for (i = 0; i < s->failed; i++) { | |
3312 | if (fdev[i]->towrite && | |
3313 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3314 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3315 | /* If we have a partial write to a failed | |
3316 | * device, then we will need to reconstruct | |
3317 | * the content of that device, so all other | |
3318 | * devices must be read. | |
3319 | */ | |
3320 | return 1; | |
3321 | } | |
3322 | ||
3323 | /* If we are forced to do a reconstruct-write, either because | |
3324 | * the current RAID6 implementation only supports that, or | |
3325 | * or because parity cannot be trusted and we are currently | |
3326 | * recovering it, there is extra need to be careful. | |
3327 | * If one of the devices that we would need to read, because | |
3328 | * it is not being overwritten (and maybe not written at all) | |
3329 | * is missing/faulty, then we need to read everything we can. | |
3330 | */ | |
3331 | if (sh->raid_conf->level != 6 && | |
3332 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3333 | /* reconstruct-write isn't being forced */ | |
3334 | return 0; | |
3335 | for (i = 0; i < s->failed; i++) { | |
10d82c5f N |
3336 | if (s->failed_num[i] != sh->pd_idx && |
3337 | s->failed_num[i] != sh->qd_idx && | |
3338 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3339 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3340 | return 1; | |
3341 | } | |
3342 | ||
2c58f06e N |
3343 | return 0; |
3344 | } | |
3345 | ||
3346 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, | |
3347 | int disk_idx, int disks) | |
3348 | { | |
3349 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3350 | ||
3351 | /* is the data in this block needed, and can we get it? */ | |
3352 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3353 | /* we would like to get this block, possibly by computing it, |
3354 | * otherwise read it if the backing disk is insync | |
3355 | */ | |
3356 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3357 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3358 | BUG_ON(sh->batch_head); |
5599becc | 3359 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3360 | (s->failed && (disk_idx == s->failed_num[0] || |
3361 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3362 | /* have disk failed, and we're requested to fetch it; |
3363 | * do compute it | |
a4456856 | 3364 | */ |
5599becc YT |
3365 | pr_debug("Computing stripe %llu block %d\n", |
3366 | (unsigned long long)sh->sector, disk_idx); | |
3367 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3368 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3369 | set_bit(R5_Wantcompute, &dev->flags); | |
3370 | sh->ops.target = disk_idx; | |
3371 | sh->ops.target2 = -1; /* no 2nd target */ | |
3372 | s->req_compute = 1; | |
93b3dbce N |
3373 | /* Careful: from this point on 'uptodate' is in the eye |
3374 | * of raid_run_ops which services 'compute' operations | |
3375 | * before writes. R5_Wantcompute flags a block that will | |
3376 | * be R5_UPTODATE by the time it is needed for a | |
3377 | * subsequent operation. | |
3378 | */ | |
5599becc YT |
3379 | s->uptodate++; |
3380 | return 1; | |
3381 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3382 | /* Computing 2-failure is *very* expensive; only | |
3383 | * do it if failed >= 2 | |
3384 | */ | |
3385 | int other; | |
3386 | for (other = disks; other--; ) { | |
3387 | if (other == disk_idx) | |
3388 | continue; | |
3389 | if (!test_bit(R5_UPTODATE, | |
3390 | &sh->dev[other].flags)) | |
3391 | break; | |
a4456856 | 3392 | } |
5599becc YT |
3393 | BUG_ON(other < 0); |
3394 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3395 | (unsigned long long)sh->sector, | |
3396 | disk_idx, other); | |
3397 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3398 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3399 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3400 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3401 | sh->ops.target = disk_idx; | |
3402 | sh->ops.target2 = other; | |
3403 | s->uptodate += 2; | |
3404 | s->req_compute = 1; | |
3405 | return 1; | |
3406 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3407 | set_bit(R5_LOCKED, &dev->flags); | |
3408 | set_bit(R5_Wantread, &dev->flags); | |
3409 | s->locked++; | |
3410 | pr_debug("Reading block %d (sync=%d)\n", | |
3411 | disk_idx, s->syncing); | |
a4456856 DW |
3412 | } |
3413 | } | |
5599becc YT |
3414 | |
3415 | return 0; | |
3416 | } | |
3417 | ||
3418 | /** | |
93b3dbce | 3419 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3420 | */ |
93b3dbce N |
3421 | static void handle_stripe_fill(struct stripe_head *sh, |
3422 | struct stripe_head_state *s, | |
3423 | int disks) | |
5599becc YT |
3424 | { |
3425 | int i; | |
3426 | ||
3427 | /* look for blocks to read/compute, skip this if a compute | |
3428 | * is already in flight, or if the stripe contents are in the | |
3429 | * midst of changing due to a write | |
3430 | */ | |
3431 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
3432 | !sh->reconstruct_state) | |
3433 | for (i = disks; i--; ) | |
93b3dbce | 3434 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3435 | break; |
a4456856 DW |
3436 | set_bit(STRIPE_HANDLE, &sh->state); |
3437 | } | |
3438 | ||
787b76fa N |
3439 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3440 | unsigned long handle_flags); | |
1fe797e6 | 3441 | /* handle_stripe_clean_event |
a4456856 DW |
3442 | * any written block on an uptodate or failed drive can be returned. |
3443 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3444 | * never LOCKED, so we don't need to test 'failed' directly. | |
3445 | */ | |
d1688a6d | 3446 | static void handle_stripe_clean_event(struct r5conf *conf, |
a4456856 DW |
3447 | struct stripe_head *sh, int disks, struct bio **return_bi) |
3448 | { | |
3449 | int i; | |
3450 | struct r5dev *dev; | |
f8dfcffd | 3451 | int discard_pending = 0; |
59fc630b | 3452 | struct stripe_head *head_sh = sh; |
3453 | bool do_endio = false; | |
a4456856 DW |
3454 | |
3455 | for (i = disks; i--; ) | |
3456 | if (sh->dev[i].written) { | |
3457 | dev = &sh->dev[i]; | |
3458 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3459 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3460 | test_bit(R5_Discard, &dev->flags) || |
3461 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3462 | /* We can return any write requests */ |
3463 | struct bio *wbi, *wbi2; | |
45b4233c | 3464 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3465 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3466 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3467 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3468 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3469 | } |
59fc630b | 3470 | do_endio = true; |
3471 | ||
3472 | returnbi: | |
3473 | dev->page = dev->orig_page; | |
a4456856 DW |
3474 | wbi = dev->written; |
3475 | dev->written = NULL; | |
4f024f37 | 3476 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3477 | dev->sector + STRIPE_SECTORS) { |
3478 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3479 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 DW |
3480 | md_write_end(conf->mddev); |
3481 | wbi->bi_next = *return_bi; | |
3482 | *return_bi = wbi; | |
3483 | } | |
3484 | wbi = wbi2; | |
3485 | } | |
7eaf7e8e SL |
3486 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3487 | STRIPE_SECTORS, | |
a4456856 | 3488 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3489 | 0); |
59fc630b | 3490 | if (head_sh->batch_head) { |
3491 | sh = list_first_entry(&sh->batch_list, | |
3492 | struct stripe_head, | |
3493 | batch_list); | |
3494 | if (sh != head_sh) { | |
3495 | dev = &sh->dev[i]; | |
3496 | goto returnbi; | |
3497 | } | |
3498 | } | |
3499 | sh = head_sh; | |
3500 | dev = &sh->dev[i]; | |
f8dfcffd N |
3501 | } else if (test_bit(R5_Discard, &dev->flags)) |
3502 | discard_pending = 1; | |
d592a996 SL |
3503 | WARN_ON(test_bit(R5_SkipCopy, &dev->flags)); |
3504 | WARN_ON(dev->page != dev->orig_page); | |
f8dfcffd N |
3505 | } |
3506 | if (!discard_pending && | |
3507 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
3508 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); | |
3509 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3510 | if (sh->qd_idx >= 0) { | |
3511 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3512 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3513 | } | |
3514 | /* now that discard is done we can proceed with any sync */ | |
3515 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3516 | /* |
3517 | * SCSI discard will change some bio fields and the stripe has | |
3518 | * no updated data, so remove it from hash list and the stripe | |
3519 | * will be reinitialized | |
3520 | */ | |
3521 | spin_lock_irq(&conf->device_lock); | |
59fc630b | 3522 | unhash: |
d47648fc | 3523 | remove_hash(sh); |
59fc630b | 3524 | if (head_sh->batch_head) { |
3525 | sh = list_first_entry(&sh->batch_list, | |
3526 | struct stripe_head, batch_list); | |
3527 | if (sh != head_sh) | |
3528 | goto unhash; | |
3529 | } | |
d47648fc | 3530 | spin_unlock_irq(&conf->device_lock); |
59fc630b | 3531 | sh = head_sh; |
3532 | ||
f8dfcffd N |
3533 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3534 | set_bit(STRIPE_HANDLE, &sh->state); | |
3535 | ||
3536 | } | |
8b3e6cdc DW |
3537 | |
3538 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3539 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3540 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3541 | |
787b76fa N |
3542 | if (head_sh->batch_head && do_endio) |
3543 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3544 | } |
3545 | ||
d1688a6d | 3546 | static void handle_stripe_dirtying(struct r5conf *conf, |
c8ac1803 N |
3547 | struct stripe_head *sh, |
3548 | struct stripe_head_state *s, | |
3549 | int disks) | |
a4456856 DW |
3550 | { |
3551 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3552 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3553 | ||
584acdd4 | 3554 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3555 | * If yes, then the array is dirty (after unclean shutdown or |
3556 | * initial creation), so parity in some stripes might be inconsistent. | |
3557 | * In this case, we need to always do reconstruct-write, to ensure | |
3558 | * that in case of drive failure or read-error correction, we | |
3559 | * generate correct data from the parity. | |
3560 | */ | |
584acdd4 | 3561 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3562 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3563 | s->failed == 0)) { | |
a7854487 | 3564 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3565 | * look like rcw is cheaper |
3566 | */ | |
3567 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3568 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3569 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3570 | (unsigned long long)sh->sector); |
c8ac1803 | 3571 | } else for (i = disks; i--; ) { |
a4456856 DW |
3572 | /* would I have to read this buffer for read_modify_write */ |
3573 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3574 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3575 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3576 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3577 | test_bit(R5_Wantcompute, &dev->flags))) { | |
a4456856 DW |
3578 | if (test_bit(R5_Insync, &dev->flags)) |
3579 | rmw++; | |
3580 | else | |
3581 | rmw += 2*disks; /* cannot read it */ | |
3582 | } | |
3583 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3584 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3585 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3586 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3587 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3588 | test_bit(R5_Wantcompute, &dev->flags))) { | |
67f45548 N |
3589 | if (test_bit(R5_Insync, &dev->flags)) |
3590 | rcw++; | |
a4456856 DW |
3591 | else |
3592 | rcw += 2*disks; | |
3593 | } | |
3594 | } | |
45b4233c | 3595 | pr_debug("for sector %llu, rmw=%d rcw=%d\n", |
a4456856 DW |
3596 | (unsigned long long)sh->sector, rmw, rcw); |
3597 | set_bit(STRIPE_HANDLE, &sh->state); | |
584acdd4 | 3598 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_ENABLE_RMW)) && rmw > 0) { |
a4456856 | 3599 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3600 | if (conf->mddev->queue) |
3601 | blk_add_trace_msg(conf->mddev->queue, | |
3602 | "raid5 rmw %llu %d", | |
3603 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3604 | for (i = disks; i--; ) { |
3605 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3606 | if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx) && |
a4456856 | 3607 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 DW |
3608 | !(test_bit(R5_UPTODATE, &dev->flags) || |
3609 | test_bit(R5_Wantcompute, &dev->flags)) && | |
a4456856 | 3610 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3611 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3612 | &sh->state)) { | |
3613 | pr_debug("Read_old block %d for r-m-w\n", | |
3614 | i); | |
a4456856 DW |
3615 | set_bit(R5_LOCKED, &dev->flags); |
3616 | set_bit(R5_Wantread, &dev->flags); | |
3617 | s->locked++; | |
3618 | } else { | |
3619 | set_bit(STRIPE_DELAYED, &sh->state); | |
3620 | set_bit(STRIPE_HANDLE, &sh->state); | |
3621 | } | |
3622 | } | |
3623 | } | |
a9add5d9 | 3624 | } |
584acdd4 | 3625 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_ENABLE_RMW)) && rcw > 0) { |
a4456856 | 3626 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3627 | int qread =0; |
c8ac1803 | 3628 | rcw = 0; |
a4456856 DW |
3629 | for (i = disks; i--; ) { |
3630 | struct r5dev *dev = &sh->dev[i]; | |
3631 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3632 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3633 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3634 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3635 | test_bit(R5_Wantcompute, &dev->flags))) { |
3636 | rcw++; | |
67f45548 N |
3637 | if (test_bit(R5_Insync, &dev->flags) && |
3638 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3639 | &sh->state)) { | |
45b4233c | 3640 | pr_debug("Read_old block " |
a4456856 DW |
3641 | "%d for Reconstruct\n", i); |
3642 | set_bit(R5_LOCKED, &dev->flags); | |
3643 | set_bit(R5_Wantread, &dev->flags); | |
3644 | s->locked++; | |
a9add5d9 | 3645 | qread++; |
a4456856 DW |
3646 | } else { |
3647 | set_bit(STRIPE_DELAYED, &sh->state); | |
3648 | set_bit(STRIPE_HANDLE, &sh->state); | |
3649 | } | |
3650 | } | |
3651 | } | |
e3620a3a | 3652 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3653 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3654 | (unsigned long long)sh->sector, | |
3655 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3656 | } |
b1b02fe9 N |
3657 | |
3658 | if (rcw > disks && rmw > disks && | |
3659 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3660 | set_bit(STRIPE_DELAYED, &sh->state); | |
3661 | ||
a4456856 DW |
3662 | /* now if nothing is locked, and if we have enough data, |
3663 | * we can start a write request | |
3664 | */ | |
f38e1219 DW |
3665 | /* since handle_stripe can be called at any time we need to handle the |
3666 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3667 | * subsequent call wants to start a write request. raid_run_ops only |
3668 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3669 | * simultaneously. If this is not the case then new writes need to be |
3670 | * held off until the compute completes. | |
3671 | */ | |
976ea8d4 DW |
3672 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3673 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
3674 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) | |
c0f7bddb | 3675 | schedule_reconstruction(sh, s, rcw == 0, 0); |
a4456856 DW |
3676 | } |
3677 | ||
d1688a6d | 3678 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3679 | struct stripe_head_state *s, int disks) |
3680 | { | |
ecc65c9b | 3681 | struct r5dev *dev = NULL; |
bd2ab670 | 3682 | |
59fc630b | 3683 | BUG_ON(sh->batch_head); |
a4456856 | 3684 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3685 | |
ecc65c9b DW |
3686 | switch (sh->check_state) { |
3687 | case check_state_idle: | |
3688 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3689 | if (s->failed == 0) { |
bd2ab670 | 3690 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3691 | sh->check_state = check_state_run; |
3692 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3693 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3694 | s->uptodate--; |
ecc65c9b | 3695 | break; |
bd2ab670 | 3696 | } |
f2b3b44d | 3697 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3698 | /* fall through */ |
3699 | case check_state_compute_result: | |
3700 | sh->check_state = check_state_idle; | |
3701 | if (!dev) | |
3702 | dev = &sh->dev[sh->pd_idx]; | |
3703 | ||
3704 | /* check that a write has not made the stripe insync */ | |
3705 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3706 | break; | |
c8894419 | 3707 | |
a4456856 | 3708 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3709 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3710 | BUG_ON(s->uptodate != disks); | |
3711 | ||
3712 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3713 | s->locked++; |
a4456856 | 3714 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3715 | |
a4456856 | 3716 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3717 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3718 | break; |
3719 | case check_state_run: | |
3720 | break; /* we will be called again upon completion */ | |
3721 | case check_state_check_result: | |
3722 | sh->check_state = check_state_idle; | |
3723 | ||
3724 | /* if a failure occurred during the check operation, leave | |
3725 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3726 | */ | |
3727 | if (s->failed) | |
3728 | break; | |
3729 | ||
3730 | /* handle a successful check operation, if parity is correct | |
3731 | * we are done. Otherwise update the mismatch count and repair | |
3732 | * parity if !MD_RECOVERY_CHECK | |
3733 | */ | |
ad283ea4 | 3734 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3735 | /* parity is correct (on disc, |
3736 | * not in buffer any more) | |
3737 | */ | |
3738 | set_bit(STRIPE_INSYNC, &sh->state); | |
3739 | else { | |
7f7583d4 | 3740 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3741 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3742 | /* don't try to repair!! */ | |
3743 | set_bit(STRIPE_INSYNC, &sh->state); | |
3744 | else { | |
3745 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3746 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3747 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3748 | set_bit(R5_Wantcompute, | |
3749 | &sh->dev[sh->pd_idx].flags); | |
3750 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3751 | sh->ops.target2 = -1; |
ecc65c9b DW |
3752 | s->uptodate++; |
3753 | } | |
3754 | } | |
3755 | break; | |
3756 | case check_state_compute_run: | |
3757 | break; | |
3758 | default: | |
3759 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3760 | __func__, sh->check_state, | |
3761 | (unsigned long long) sh->sector); | |
3762 | BUG(); | |
a4456856 DW |
3763 | } |
3764 | } | |
3765 | ||
d1688a6d | 3766 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3767 | struct stripe_head_state *s, |
f2b3b44d | 3768 | int disks) |
a4456856 | 3769 | { |
a4456856 | 3770 | int pd_idx = sh->pd_idx; |
34e04e87 | 3771 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3772 | struct r5dev *dev; |
a4456856 | 3773 | |
59fc630b | 3774 | BUG_ON(sh->batch_head); |
a4456856 DW |
3775 | set_bit(STRIPE_HANDLE, &sh->state); |
3776 | ||
3777 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3778 | |
a4456856 DW |
3779 | /* Want to check and possibly repair P and Q. |
3780 | * However there could be one 'failed' device, in which | |
3781 | * case we can only check one of them, possibly using the | |
3782 | * other to generate missing data | |
3783 | */ | |
3784 | ||
d82dfee0 DW |
3785 | switch (sh->check_state) { |
3786 | case check_state_idle: | |
3787 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 3788 | if (s->failed == s->q_failed) { |
d82dfee0 | 3789 | /* The only possible failed device holds Q, so it |
a4456856 DW |
3790 | * makes sense to check P (If anything else were failed, |
3791 | * we would have used P to recreate it). | |
3792 | */ | |
d82dfee0 | 3793 | sh->check_state = check_state_run; |
a4456856 | 3794 | } |
f2b3b44d | 3795 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 3796 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
3797 | * anything, so it makes sense to check it |
3798 | */ | |
d82dfee0 DW |
3799 | if (sh->check_state == check_state_run) |
3800 | sh->check_state = check_state_run_pq; | |
3801 | else | |
3802 | sh->check_state = check_state_run_q; | |
a4456856 | 3803 | } |
a4456856 | 3804 | |
d82dfee0 DW |
3805 | /* discard potentially stale zero_sum_result */ |
3806 | sh->ops.zero_sum_result = 0; | |
a4456856 | 3807 | |
d82dfee0 DW |
3808 | if (sh->check_state == check_state_run) { |
3809 | /* async_xor_zero_sum destroys the contents of P */ | |
3810 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
3811 | s->uptodate--; | |
a4456856 | 3812 | } |
d82dfee0 DW |
3813 | if (sh->check_state >= check_state_run && |
3814 | sh->check_state <= check_state_run_pq) { | |
3815 | /* async_syndrome_zero_sum preserves P and Q, so | |
3816 | * no need to mark them !uptodate here | |
3817 | */ | |
3818 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
3819 | break; | |
a4456856 DW |
3820 | } |
3821 | ||
d82dfee0 DW |
3822 | /* we have 2-disk failure */ |
3823 | BUG_ON(s->failed != 2); | |
3824 | /* fall through */ | |
3825 | case check_state_compute_result: | |
3826 | sh->check_state = check_state_idle; | |
a4456856 | 3827 | |
d82dfee0 DW |
3828 | /* check that a write has not made the stripe insync */ |
3829 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3830 | break; | |
a4456856 DW |
3831 | |
3832 | /* now write out any block on a failed drive, | |
d82dfee0 | 3833 | * or P or Q if they were recomputed |
a4456856 | 3834 | */ |
d82dfee0 | 3835 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 3836 | if (s->failed == 2) { |
f2b3b44d | 3837 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
3838 | s->locked++; |
3839 | set_bit(R5_LOCKED, &dev->flags); | |
3840 | set_bit(R5_Wantwrite, &dev->flags); | |
3841 | } | |
3842 | if (s->failed >= 1) { | |
f2b3b44d | 3843 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
3844 | s->locked++; |
3845 | set_bit(R5_LOCKED, &dev->flags); | |
3846 | set_bit(R5_Wantwrite, &dev->flags); | |
3847 | } | |
d82dfee0 | 3848 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
3849 | dev = &sh->dev[pd_idx]; |
3850 | s->locked++; | |
3851 | set_bit(R5_LOCKED, &dev->flags); | |
3852 | set_bit(R5_Wantwrite, &dev->flags); | |
3853 | } | |
d82dfee0 | 3854 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
3855 | dev = &sh->dev[qd_idx]; |
3856 | s->locked++; | |
3857 | set_bit(R5_LOCKED, &dev->flags); | |
3858 | set_bit(R5_Wantwrite, &dev->flags); | |
3859 | } | |
3860 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
3861 | ||
3862 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
3863 | break; |
3864 | case check_state_run: | |
3865 | case check_state_run_q: | |
3866 | case check_state_run_pq: | |
3867 | break; /* we will be called again upon completion */ | |
3868 | case check_state_check_result: | |
3869 | sh->check_state = check_state_idle; | |
3870 | ||
3871 | /* handle a successful check operation, if parity is correct | |
3872 | * we are done. Otherwise update the mismatch count and repair | |
3873 | * parity if !MD_RECOVERY_CHECK | |
3874 | */ | |
3875 | if (sh->ops.zero_sum_result == 0) { | |
3876 | /* both parities are correct */ | |
3877 | if (!s->failed) | |
3878 | set_bit(STRIPE_INSYNC, &sh->state); | |
3879 | else { | |
3880 | /* in contrast to the raid5 case we can validate | |
3881 | * parity, but still have a failure to write | |
3882 | * back | |
3883 | */ | |
3884 | sh->check_state = check_state_compute_result; | |
3885 | /* Returning at this point means that we may go | |
3886 | * off and bring p and/or q uptodate again so | |
3887 | * we make sure to check zero_sum_result again | |
3888 | * to verify if p or q need writeback | |
3889 | */ | |
3890 | } | |
3891 | } else { | |
7f7583d4 | 3892 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
3893 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3894 | /* don't try to repair!! */ | |
3895 | set_bit(STRIPE_INSYNC, &sh->state); | |
3896 | else { | |
3897 | int *target = &sh->ops.target; | |
3898 | ||
3899 | sh->ops.target = -1; | |
3900 | sh->ops.target2 = -1; | |
3901 | sh->check_state = check_state_compute_run; | |
3902 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3903 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3904 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
3905 | set_bit(R5_Wantcompute, | |
3906 | &sh->dev[pd_idx].flags); | |
3907 | *target = pd_idx; | |
3908 | target = &sh->ops.target2; | |
3909 | s->uptodate++; | |
3910 | } | |
3911 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
3912 | set_bit(R5_Wantcompute, | |
3913 | &sh->dev[qd_idx].flags); | |
3914 | *target = qd_idx; | |
3915 | s->uptodate++; | |
3916 | } | |
3917 | } | |
3918 | } | |
3919 | break; | |
3920 | case check_state_compute_run: | |
3921 | break; | |
3922 | default: | |
3923 | printk(KERN_ERR "%s: unknown check_state: %d sector: %llu\n", | |
3924 | __func__, sh->check_state, | |
3925 | (unsigned long long) sh->sector); | |
3926 | BUG(); | |
a4456856 DW |
3927 | } |
3928 | } | |
3929 | ||
d1688a6d | 3930 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
3931 | { |
3932 | int i; | |
3933 | ||
3934 | /* We have read all the blocks in this stripe and now we need to | |
3935 | * copy some of them into a target stripe for expand. | |
3936 | */ | |
f0a50d37 | 3937 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 3938 | BUG_ON(sh->batch_head); |
a4456856 DW |
3939 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
3940 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 3941 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 3942 | int dd_idx, j; |
a4456856 | 3943 | struct stripe_head *sh2; |
a08abd8c | 3944 | struct async_submit_ctl submit; |
a4456856 | 3945 | |
784052ec | 3946 | sector_t bn = compute_blocknr(sh, i, 1); |
911d4ee8 N |
3947 | sector_t s = raid5_compute_sector(conf, bn, 0, |
3948 | &dd_idx, NULL); | |
a8c906ca | 3949 | sh2 = get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
3950 | if (sh2 == NULL) |
3951 | /* so far only the early blocks of this stripe | |
3952 | * have been requested. When later blocks | |
3953 | * get requested, we will try again | |
3954 | */ | |
3955 | continue; | |
3956 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
3957 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
3958 | /* must have already done this block */ | |
3959 | release_stripe(sh2); | |
3960 | continue; | |
3961 | } | |
f0a50d37 DW |
3962 | |
3963 | /* place all the copies on one channel */ | |
a08abd8c | 3964 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 3965 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 3966 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 3967 | &submit); |
f0a50d37 | 3968 | |
a4456856 DW |
3969 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
3970 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
3971 | for (j = 0; j < conf->raid_disks; j++) | |
3972 | if (j != sh2->pd_idx && | |
86c374ba | 3973 | j != sh2->qd_idx && |
a4456856 DW |
3974 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
3975 | break; | |
3976 | if (j == conf->raid_disks) { | |
3977 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
3978 | set_bit(STRIPE_HANDLE, &sh2->state); | |
3979 | } | |
3980 | release_stripe(sh2); | |
f0a50d37 | 3981 | |
a4456856 | 3982 | } |
a2e08551 | 3983 | /* done submitting copies, wait for them to complete */ |
749586b7 | 3984 | async_tx_quiesce(&tx); |
a4456856 | 3985 | } |
1da177e4 LT |
3986 | |
3987 | /* | |
3988 | * handle_stripe - do things to a stripe. | |
3989 | * | |
9a3e1101 N |
3990 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
3991 | * state of various bits to see what needs to be done. | |
1da177e4 | 3992 | * Possible results: |
9a3e1101 N |
3993 | * return some read requests which now have data |
3994 | * return some write requests which are safely on storage | |
1da177e4 LT |
3995 | * schedule a read on some buffers |
3996 | * schedule a write of some buffers | |
3997 | * return confirmation of parity correctness | |
3998 | * | |
1da177e4 | 3999 | */ |
a4456856 | 4000 | |
acfe726b | 4001 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4002 | { |
d1688a6d | 4003 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4004 | int disks = sh->disks; |
474af965 N |
4005 | struct r5dev *dev; |
4006 | int i; | |
9a3e1101 | 4007 | int do_recovery = 0; |
1da177e4 | 4008 | |
acfe726b N |
4009 | memset(s, 0, sizeof(*s)); |
4010 | ||
dabc4ec6 | 4011 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4012 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4013 | s->failed_num[0] = -1; |
4014 | s->failed_num[1] = -1; | |
1da177e4 | 4015 | |
acfe726b | 4016 | /* Now to look around and see what can be done */ |
1da177e4 | 4017 | rcu_read_lock(); |
16a53ecc | 4018 | for (i=disks; i--; ) { |
3cb03002 | 4019 | struct md_rdev *rdev; |
31c176ec N |
4020 | sector_t first_bad; |
4021 | int bad_sectors; | |
4022 | int is_bad = 0; | |
acfe726b | 4023 | |
16a53ecc | 4024 | dev = &sh->dev[i]; |
1da177e4 | 4025 | |
45b4233c | 4026 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4027 | i, dev->flags, |
4028 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4029 | /* maybe we can reply to a read |
4030 | * | |
4031 | * new wantfill requests are only permitted while | |
4032 | * ops_complete_biofill is guaranteed to be inactive | |
4033 | */ | |
4034 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4035 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4036 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4037 | |
16a53ecc | 4038 | /* now count some things */ |
cc94015a N |
4039 | if (test_bit(R5_LOCKED, &dev->flags)) |
4040 | s->locked++; | |
4041 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4042 | s->uptodate++; | |
2d6e4ecc | 4043 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4044 | s->compute++; |
4045 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4046 | } |
1da177e4 | 4047 | |
acfe726b | 4048 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4049 | s->to_fill++; |
acfe726b | 4050 | else if (dev->toread) |
cc94015a | 4051 | s->to_read++; |
16a53ecc | 4052 | if (dev->towrite) { |
cc94015a | 4053 | s->to_write++; |
16a53ecc | 4054 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4055 | s->non_overwrite++; |
16a53ecc | 4056 | } |
a4456856 | 4057 | if (dev->written) |
cc94015a | 4058 | s->written++; |
14a75d3e N |
4059 | /* Prefer to use the replacement for reads, but only |
4060 | * if it is recovered enough and has no bad blocks. | |
4061 | */ | |
4062 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4063 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4064 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4065 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4066 | &first_bad, &bad_sectors)) | |
4067 | set_bit(R5_ReadRepl, &dev->flags); | |
4068 | else { | |
e6030cb0 | 4069 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4070 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4071 | else |
4072 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4073 | rdev = rcu_dereference(conf->disks[i].rdev); |
4074 | clear_bit(R5_ReadRepl, &dev->flags); | |
4075 | } | |
9283d8c5 N |
4076 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4077 | rdev = NULL; | |
31c176ec N |
4078 | if (rdev) { |
4079 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4080 | &first_bad, &bad_sectors); | |
4081 | if (s->blocked_rdev == NULL | |
4082 | && (test_bit(Blocked, &rdev->flags) | |
4083 | || is_bad < 0)) { | |
4084 | if (is_bad < 0) | |
4085 | set_bit(BlockedBadBlocks, | |
4086 | &rdev->flags); | |
4087 | s->blocked_rdev = rdev; | |
4088 | atomic_inc(&rdev->nr_pending); | |
4089 | } | |
6bfe0b49 | 4090 | } |
415e72d0 N |
4091 | clear_bit(R5_Insync, &dev->flags); |
4092 | if (!rdev) | |
4093 | /* Not in-sync */; | |
31c176ec N |
4094 | else if (is_bad) { |
4095 | /* also not in-sync */ | |
18b9837e N |
4096 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4097 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4098 | /* treat as in-sync, but with a read error |
4099 | * which we can now try to correct | |
4100 | */ | |
4101 | set_bit(R5_Insync, &dev->flags); | |
4102 | set_bit(R5_ReadError, &dev->flags); | |
4103 | } | |
4104 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4105 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4106 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4107 | /* in sync if before recovery_offset */ |
30d7a483 N |
4108 | set_bit(R5_Insync, &dev->flags); |
4109 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4110 | test_bit(R5_Expanded, &dev->flags)) | |
4111 | /* If we've reshaped into here, we assume it is Insync. | |
4112 | * We will shortly update recovery_offset to make | |
4113 | * it official. | |
4114 | */ | |
4115 | set_bit(R5_Insync, &dev->flags); | |
4116 | ||
1cc03eb9 | 4117 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4118 | /* This flag does not apply to '.replacement' |
4119 | * only to .rdev, so make sure to check that*/ | |
4120 | struct md_rdev *rdev2 = rcu_dereference( | |
4121 | conf->disks[i].rdev); | |
4122 | if (rdev2 == rdev) | |
4123 | clear_bit(R5_Insync, &dev->flags); | |
4124 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4125 | s->handle_bad_blocks = 1; |
14a75d3e | 4126 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4127 | } else |
4128 | clear_bit(R5_WriteError, &dev->flags); | |
4129 | } | |
1cc03eb9 | 4130 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4131 | /* This flag does not apply to '.replacement' |
4132 | * only to .rdev, so make sure to check that*/ | |
4133 | struct md_rdev *rdev2 = rcu_dereference( | |
4134 | conf->disks[i].rdev); | |
4135 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4136 | s->handle_bad_blocks = 1; |
14a75d3e | 4137 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4138 | } else |
4139 | clear_bit(R5_MadeGood, &dev->flags); | |
4140 | } | |
977df362 N |
4141 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4142 | struct md_rdev *rdev2 = rcu_dereference( | |
4143 | conf->disks[i].replacement); | |
4144 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4145 | s->handle_bad_blocks = 1; | |
4146 | atomic_inc(&rdev2->nr_pending); | |
4147 | } else | |
4148 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4149 | } | |
415e72d0 | 4150 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4151 | /* The ReadError flag will just be confusing now */ |
4152 | clear_bit(R5_ReadError, &dev->flags); | |
4153 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4154 | } |
415e72d0 N |
4155 | if (test_bit(R5_ReadError, &dev->flags)) |
4156 | clear_bit(R5_Insync, &dev->flags); | |
4157 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4158 | if (s->failed < 2) |
4159 | s->failed_num[s->failed] = i; | |
4160 | s->failed++; | |
9a3e1101 N |
4161 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4162 | do_recovery = 1; | |
415e72d0 | 4163 | } |
1da177e4 | 4164 | } |
9a3e1101 N |
4165 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4166 | /* If there is a failed device being replaced, | |
4167 | * we must be recovering. | |
4168 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4169 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4170 | * else we can only be replacing |
4171 | * sync and recovery both need to read all devices, and so | |
4172 | * use the same flag. | |
4173 | */ | |
4174 | if (do_recovery || | |
c6d2e084 | 4175 | sh->sector >= conf->mddev->recovery_cp || |
4176 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4177 | s->syncing = 1; |
4178 | else | |
4179 | s->replacing = 1; | |
4180 | } | |
1da177e4 | 4181 | rcu_read_unlock(); |
cc94015a N |
4182 | } |
4183 | ||
59fc630b | 4184 | static int clear_batch_ready(struct stripe_head *sh) |
4185 | { | |
b15a9dbd N |
4186 | /* Return '1' if this is a member of batch, or |
4187 | * '0' if it is a lone stripe or a head which can now be | |
4188 | * handled. | |
4189 | */ | |
59fc630b | 4190 | struct stripe_head *tmp; |
4191 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4192 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4193 | spin_lock(&sh->stripe_lock); |
4194 | if (!sh->batch_head) { | |
4195 | spin_unlock(&sh->stripe_lock); | |
4196 | return 0; | |
4197 | } | |
4198 | ||
4199 | /* | |
4200 | * this stripe could be added to a batch list before we check | |
4201 | * BATCH_READY, skips it | |
4202 | */ | |
4203 | if (sh->batch_head != sh) { | |
4204 | spin_unlock(&sh->stripe_lock); | |
4205 | return 1; | |
4206 | } | |
4207 | spin_lock(&sh->batch_lock); | |
4208 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4209 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4210 | spin_unlock(&sh->batch_lock); | |
4211 | spin_unlock(&sh->stripe_lock); | |
4212 | ||
4213 | /* | |
4214 | * BATCH_READY is cleared, no new stripes can be added. | |
4215 | * batch_list can be accessed without lock | |
4216 | */ | |
4217 | return 0; | |
4218 | } | |
4219 | ||
3960ce79 N |
4220 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4221 | unsigned long handle_flags) | |
72ac7330 | 4222 | { |
4e3d62ff | 4223 | struct stripe_head *sh, *next; |
72ac7330 | 4224 | int i; |
fb642b92 | 4225 | int do_wakeup = 0; |
72ac7330 | 4226 | |
bb27051f N |
4227 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4228 | ||
72ac7330 | 4229 | list_del_init(&sh->batch_list); |
4230 | ||
1b956f7a N |
4231 | WARN_ON_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
4232 | (1 << STRIPE_SYNCING) | | |
4233 | (1 << STRIPE_REPLACED) | | |
4234 | (1 << STRIPE_PREREAD_ACTIVE) | | |
4235 | (1 << STRIPE_DELAYED) | | |
4236 | (1 << STRIPE_BIT_DELAY) | | |
4237 | (1 << STRIPE_FULL_WRITE) | | |
4238 | (1 << STRIPE_BIOFILL_RUN) | | |
4239 | (1 << STRIPE_COMPUTE_RUN) | | |
4240 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4241 | (1 << STRIPE_DISCARD) | | |
4242 | (1 << STRIPE_BATCH_READY) | | |
4243 | (1 << STRIPE_BATCH_ERR) | | |
4244 | (1 << STRIPE_BITMAP_PENDING))); | |
4245 | WARN_ON_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4246 | (1 << STRIPE_REPLACED))); | |
4247 | ||
4248 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
4249 | (1 << STRIPE_DEGRADED)), | |
4250 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4251 | ||
72ac7330 | 4252 | sh->check_state = head_sh->check_state; |
4253 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4254 | for (i = 0; i < sh->disks; i++) { |
4255 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4256 | do_wakeup = 1; | |
72ac7330 | 4257 | sh->dev[i].flags = head_sh->dev[i].flags & |
4258 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4259 | } |
72ac7330 | 4260 | spin_lock_irq(&sh->stripe_lock); |
4261 | sh->batch_head = NULL; | |
4262 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4263 | if (handle_flags == 0 || |
4264 | sh->state & handle_flags) | |
4265 | set_bit(STRIPE_HANDLE, &sh->state); | |
72ac7330 | 4266 | release_stripe(sh); |
72ac7330 | 4267 | } |
fb642b92 N |
4268 | spin_lock_irq(&head_sh->stripe_lock); |
4269 | head_sh->batch_head = NULL; | |
4270 | spin_unlock_irq(&head_sh->stripe_lock); | |
4271 | for (i = 0; i < head_sh->disks; i++) | |
4272 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4273 | do_wakeup = 1; | |
3960ce79 N |
4274 | if (head_sh->state & handle_flags) |
4275 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4276 | |
4277 | if (do_wakeup) | |
4278 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4279 | } |
4280 | ||
cc94015a N |
4281 | static void handle_stripe(struct stripe_head *sh) |
4282 | { | |
4283 | struct stripe_head_state s; | |
d1688a6d | 4284 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4285 | int i; |
84789554 N |
4286 | int prexor; |
4287 | int disks = sh->disks; | |
474af965 | 4288 | struct r5dev *pdev, *qdev; |
cc94015a N |
4289 | |
4290 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4291 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4292 | /* already being handled, ensure it gets handled |
4293 | * again when current action finishes */ | |
4294 | set_bit(STRIPE_HANDLE, &sh->state); | |
4295 | return; | |
4296 | } | |
4297 | ||
59fc630b | 4298 | if (clear_batch_ready(sh) ) { |
4299 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4300 | return; | |
4301 | } | |
4302 | ||
4e3d62ff | 4303 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4304 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4305 | |
dabc4ec6 | 4306 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4307 | spin_lock(&sh->stripe_lock); |
4308 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4309 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4310 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4311 | set_bit(STRIPE_SYNCING, &sh->state); | |
4312 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4313 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4314 | } |
4315 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4316 | } |
4317 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4318 | ||
4319 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4320 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4321 | (unsigned long long)sh->sector, sh->state, | |
4322 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4323 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4324 | |
acfe726b | 4325 | analyse_stripe(sh, &s); |
c5a31000 | 4326 | |
bc2607f3 N |
4327 | if (s.handle_bad_blocks) { |
4328 | set_bit(STRIPE_HANDLE, &sh->state); | |
4329 | goto finish; | |
4330 | } | |
4331 | ||
474af965 N |
4332 | if (unlikely(s.blocked_rdev)) { |
4333 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4334 | s.replacing || s.to_write || s.written) { |
474af965 N |
4335 | set_bit(STRIPE_HANDLE, &sh->state); |
4336 | goto finish; | |
4337 | } | |
4338 | /* There is nothing for the blocked_rdev to block */ | |
4339 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4340 | s.blocked_rdev = NULL; | |
4341 | } | |
4342 | ||
4343 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4344 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4345 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4346 | } | |
4347 | ||
4348 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4349 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4350 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4351 | s.failed_num[0], s.failed_num[1]); | |
4352 | /* check if the array has lost more than max_degraded devices and, | |
4353 | * if so, some requests might need to be failed. | |
4354 | */ | |
9a3f530f N |
4355 | if (s.failed > conf->max_degraded) { |
4356 | sh->check_state = 0; | |
4357 | sh->reconstruct_state = 0; | |
626f2092 | 4358 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4359 | if (s.to_read+s.to_write+s.written) |
4360 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4361 | if (s.syncing + s.replacing) |
9a3f530f N |
4362 | handle_failed_sync(conf, sh, &s); |
4363 | } | |
474af965 | 4364 | |
84789554 N |
4365 | /* Now we check to see if any write operations have recently |
4366 | * completed | |
4367 | */ | |
4368 | prexor = 0; | |
4369 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4370 | prexor = 1; | |
4371 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4372 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4373 | sh->reconstruct_state = reconstruct_state_idle; | |
4374 | ||
4375 | /* All the 'written' buffers and the parity block are ready to | |
4376 | * be written back to disk | |
4377 | */ | |
9e444768 SL |
4378 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4379 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4380 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4381 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4382 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4383 | for (i = disks; i--; ) { |
4384 | struct r5dev *dev = &sh->dev[i]; | |
4385 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4386 | (i == sh->pd_idx || i == sh->qd_idx || | |
4387 | dev->written)) { | |
4388 | pr_debug("Writing block %d\n", i); | |
4389 | set_bit(R5_Wantwrite, &dev->flags); | |
4390 | if (prexor) | |
4391 | continue; | |
9c4bdf69 N |
4392 | if (s.failed > 1) |
4393 | continue; | |
84789554 N |
4394 | if (!test_bit(R5_Insync, &dev->flags) || |
4395 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4396 | s.failed == 0)) | |
4397 | set_bit(STRIPE_INSYNC, &sh->state); | |
4398 | } | |
4399 | } | |
4400 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4401 | s.dec_preread_active = 1; | |
4402 | } | |
4403 | ||
ef5b7c69 N |
4404 | /* |
4405 | * might be able to return some write requests if the parity blocks | |
4406 | * are safe, or on a failed drive | |
4407 | */ | |
4408 | pdev = &sh->dev[sh->pd_idx]; | |
4409 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4410 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4411 | qdev = &sh->dev[sh->qd_idx]; | |
4412 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4413 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4414 | || conf->level < 6; | |
4415 | ||
4416 | if (s.written && | |
4417 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4418 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4419 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4420 | test_bit(R5_Discard, &pdev->flags))))) && | |
4421 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4422 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4423 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4424 | test_bit(R5_Discard, &qdev->flags)))))) | |
4425 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4426 | ||
4427 | /* Now we might consider reading some blocks, either to check/generate | |
4428 | * parity, or to satisfy requests | |
4429 | * or to load a block that is being partially written. | |
4430 | */ | |
4431 | if (s.to_read || s.non_overwrite | |
4432 | || (conf->level == 6 && s.to_write && s.failed) | |
4433 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4434 | || s.replacing | |
4435 | || s.expanding) | |
4436 | handle_stripe_fill(sh, &s, disks); | |
4437 | ||
84789554 N |
4438 | /* Now to consider new write requests and what else, if anything |
4439 | * should be read. We do not handle new writes when: | |
4440 | * 1/ A 'write' operation (copy+xor) is already in flight. | |
4441 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4442 | * block. | |
4443 | */ | |
4444 | if (s.to_write && !sh->reconstruct_state && !sh->check_state) | |
4445 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4446 | ||
4447 | /* maybe we need to check and possibly fix the parity for this stripe | |
4448 | * Any reads will already have been scheduled, so we just see if enough | |
4449 | * data is available. The parity check is held off while parity | |
4450 | * dependent operations are in flight. | |
4451 | */ | |
4452 | if (sh->check_state || | |
4453 | (s.syncing && s.locked == 0 && | |
4454 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4455 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4456 | if (conf->level == 6) | |
4457 | handle_parity_checks6(conf, sh, &s, disks); | |
4458 | else | |
4459 | handle_parity_checks5(conf, sh, &s, disks); | |
4460 | } | |
c5a31000 | 4461 | |
f94c0b66 N |
4462 | if ((s.replacing || s.syncing) && s.locked == 0 |
4463 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4464 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4465 | /* Write out to replacement devices where possible */ |
4466 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4467 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4468 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4469 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4470 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4471 | s.locked++; | |
4472 | } | |
f94c0b66 N |
4473 | if (s.replacing) |
4474 | set_bit(STRIPE_INSYNC, &sh->state); | |
4475 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4476 | } |
4477 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4478 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4479 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4480 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4481 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4482 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4483 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4484 | } |
4485 | ||
4486 | /* If the failed drives are just a ReadError, then we might need | |
4487 | * to progress the repair/check process | |
4488 | */ | |
4489 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4490 | for (i = 0; i < s.failed; i++) { | |
4491 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4492 | if (test_bit(R5_ReadError, &dev->flags) | |
4493 | && !test_bit(R5_LOCKED, &dev->flags) | |
4494 | && test_bit(R5_UPTODATE, &dev->flags) | |
4495 | ) { | |
4496 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4497 | set_bit(R5_Wantwrite, &dev->flags); | |
4498 | set_bit(R5_ReWrite, &dev->flags); | |
4499 | set_bit(R5_LOCKED, &dev->flags); | |
4500 | s.locked++; | |
4501 | } else { | |
4502 | /* let's read it back */ | |
4503 | set_bit(R5_Wantread, &dev->flags); | |
4504 | set_bit(R5_LOCKED, &dev->flags); | |
4505 | s.locked++; | |
4506 | } | |
4507 | } | |
4508 | } | |
4509 | ||
3687c061 N |
4510 | /* Finish reconstruct operations initiated by the expansion process */ |
4511 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4512 | struct stripe_head *sh_src | |
4513 | = get_active_stripe(conf, sh->sector, 1, 1, 1); | |
4514 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { | |
4515 | /* sh cannot be written until sh_src has been read. | |
4516 | * so arrange for sh to be delayed a little | |
4517 | */ | |
4518 | set_bit(STRIPE_DELAYED, &sh->state); | |
4519 | set_bit(STRIPE_HANDLE, &sh->state); | |
4520 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4521 | &sh_src->state)) | |
4522 | atomic_inc(&conf->preread_active_stripes); | |
4523 | release_stripe(sh_src); | |
4524 | goto finish; | |
4525 | } | |
4526 | if (sh_src) | |
4527 | release_stripe(sh_src); | |
4528 | ||
4529 | sh->reconstruct_state = reconstruct_state_idle; | |
4530 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4531 | for (i = conf->raid_disks; i--; ) { | |
4532 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4533 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4534 | s.locked++; | |
4535 | } | |
4536 | } | |
f416885e | 4537 | |
3687c061 N |
4538 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4539 | !sh->reconstruct_state) { | |
4540 | /* Need to write out all blocks after computing parity */ | |
4541 | sh->disks = conf->raid_disks; | |
4542 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4543 | schedule_reconstruction(sh, &s, 1, 1); | |
4544 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4545 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4546 | atomic_dec(&conf->reshape_stripes); | |
4547 | wake_up(&conf->wait_for_overlap); | |
4548 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4549 | } | |
4550 | ||
4551 | if (s.expanding && s.locked == 0 && | |
4552 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4553 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4554 | |
3687c061 | 4555 | finish: |
6bfe0b49 | 4556 | /* wait for this device to become unblocked */ |
5f066c63 N |
4557 | if (unlikely(s.blocked_rdev)) { |
4558 | if (conf->mddev->external) | |
4559 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4560 | conf->mddev); | |
4561 | else | |
4562 | /* Internal metadata will immediately | |
4563 | * be written by raid5d, so we don't | |
4564 | * need to wait here. | |
4565 | */ | |
4566 | rdev_dec_pending(s.blocked_rdev, | |
4567 | conf->mddev); | |
4568 | } | |
6bfe0b49 | 4569 | |
bc2607f3 N |
4570 | if (s.handle_bad_blocks) |
4571 | for (i = disks; i--; ) { | |
3cb03002 | 4572 | struct md_rdev *rdev; |
bc2607f3 N |
4573 | struct r5dev *dev = &sh->dev[i]; |
4574 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4575 | /* We own a safe reference to the rdev */ | |
4576 | rdev = conf->disks[i].rdev; | |
4577 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4578 | STRIPE_SECTORS, 0)) | |
4579 | md_error(conf->mddev, rdev); | |
4580 | rdev_dec_pending(rdev, conf->mddev); | |
4581 | } | |
b84db560 N |
4582 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4583 | rdev = conf->disks[i].rdev; | |
4584 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4585 | STRIPE_SECTORS, 0); |
b84db560 N |
4586 | rdev_dec_pending(rdev, conf->mddev); |
4587 | } | |
977df362 N |
4588 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4589 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4590 | if (!rdev) |
4591 | /* rdev have been moved down */ | |
4592 | rdev = conf->disks[i].rdev; | |
977df362 | 4593 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4594 | STRIPE_SECTORS, 0); |
977df362 N |
4595 | rdev_dec_pending(rdev, conf->mddev); |
4596 | } | |
bc2607f3 N |
4597 | } |
4598 | ||
6c0069c0 YT |
4599 | if (s.ops_request) |
4600 | raid_run_ops(sh, s.ops_request); | |
4601 | ||
f0e43bcd | 4602 | ops_run_io(sh, &s); |
16a53ecc | 4603 | |
c5709ef6 | 4604 | if (s.dec_preread_active) { |
729a1866 | 4605 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4606 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4607 | * have actually been submitted. |
4608 | */ | |
4609 | atomic_dec(&conf->preread_active_stripes); | |
4610 | if (atomic_read(&conf->preread_active_stripes) < | |
4611 | IO_THRESHOLD) | |
4612 | md_wakeup_thread(conf->mddev->thread); | |
4613 | } | |
4614 | ||
c5709ef6 | 4615 | return_io(s.return_bi); |
16a53ecc | 4616 | |
257a4b42 | 4617 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4618 | } |
4619 | ||
d1688a6d | 4620 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4621 | { |
4622 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4623 | while (!list_empty(&conf->delayed_list)) { | |
4624 | struct list_head *l = conf->delayed_list.next; | |
4625 | struct stripe_head *sh; | |
4626 | sh = list_entry(l, struct stripe_head, lru); | |
4627 | list_del_init(l); | |
4628 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4629 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4630 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4631 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4632 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4633 | } |
482c0834 | 4634 | } |
16a53ecc N |
4635 | } |
4636 | ||
566c09c5 SL |
4637 | static void activate_bit_delay(struct r5conf *conf, |
4638 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4639 | { |
4640 | /* device_lock is held */ | |
4641 | struct list_head head; | |
4642 | list_add(&head, &conf->bitmap_list); | |
4643 | list_del_init(&conf->bitmap_list); | |
4644 | while (!list_empty(&head)) { | |
4645 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4646 | int hash; |
16a53ecc N |
4647 | list_del_init(&sh->lru); |
4648 | atomic_inc(&sh->count); | |
566c09c5 SL |
4649 | hash = sh->hash_lock_index; |
4650 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4651 | } |
4652 | } | |
4653 | ||
5c675f83 | 4654 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4655 | { |
d1688a6d | 4656 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4657 | |
4658 | /* No difference between reads and writes. Just check | |
4659 | * how busy the stripe_cache is | |
4660 | */ | |
3fa841d7 | 4661 | |
5423399a | 4662 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd N |
4663 | return 1; |
4664 | if (conf->quiesce) | |
4665 | return 1; | |
4bda556a | 4666 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4667 | return 1; |
4668 | ||
4669 | return 0; | |
4670 | } | |
4671 | ||
23032a0e RBJ |
4672 | /* We want read requests to align with chunks where possible, |
4673 | * but write requests don't need to. | |
4674 | */ | |
64590f45 | 4675 | static int raid5_mergeable_bvec(struct mddev *mddev, |
cc371e66 AK |
4676 | struct bvec_merge_data *bvm, |
4677 | struct bio_vec *biovec) | |
23032a0e | 4678 | { |
3cb5edf4 | 4679 | struct r5conf *conf = mddev->private; |
cc371e66 | 4680 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
23032a0e | 4681 | int max; |
3cb5edf4 | 4682 | unsigned int chunk_sectors; |
cc371e66 | 4683 | unsigned int bio_sectors = bvm->bi_size >> 9; |
23032a0e | 4684 | |
9ffc8f7c EM |
4685 | /* |
4686 | * always allow writes to be mergeable, read as well if array | |
4687 | * is degraded as we'll go through stripe cache anyway. | |
4688 | */ | |
4689 | if ((bvm->bi_rw & 1) == WRITE || mddev->degraded) | |
4690 | return biovec->bv_len; | |
23032a0e | 4691 | |
3cb5edf4 | 4692 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
23032a0e RBJ |
4693 | max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; |
4694 | if (max < 0) max = 0; | |
4695 | if (max <= biovec->bv_len && bio_sectors == 0) | |
4696 | return biovec->bv_len; | |
4697 | else | |
4698 | return max; | |
4699 | } | |
4700 | ||
fd01b88c | 4701 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4702 | { |
3cb5edf4 | 4703 | struct r5conf *conf = mddev->private; |
4f024f37 | 4704 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 4705 | unsigned int chunk_sectors; |
aa8b57aa | 4706 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4707 | |
3cb5edf4 | 4708 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
4709 | return chunk_sectors >= |
4710 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4711 | } | |
4712 | ||
46031f9a RBJ |
4713 | /* |
4714 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4715 | * later sampled by raid5d. | |
4716 | */ | |
d1688a6d | 4717 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4718 | { |
4719 | unsigned long flags; | |
4720 | ||
4721 | spin_lock_irqsave(&conf->device_lock, flags); | |
4722 | ||
4723 | bi->bi_next = conf->retry_read_aligned_list; | |
4724 | conf->retry_read_aligned_list = bi; | |
4725 | ||
4726 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4727 | md_wakeup_thread(conf->mddev->thread); | |
4728 | } | |
4729 | ||
d1688a6d | 4730 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4731 | { |
4732 | struct bio *bi; | |
4733 | ||
4734 | bi = conf->retry_read_aligned; | |
4735 | if (bi) { | |
4736 | conf->retry_read_aligned = NULL; | |
4737 | return bi; | |
4738 | } | |
4739 | bi = conf->retry_read_aligned_list; | |
4740 | if(bi) { | |
387bb173 | 4741 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4742 | bi->bi_next = NULL; |
960e739d JA |
4743 | /* |
4744 | * this sets the active strip count to 1 and the processed | |
4745 | * strip count to zero (upper 8 bits) | |
4746 | */ | |
e7836bd6 | 4747 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
4748 | } |
4749 | ||
4750 | return bi; | |
4751 | } | |
4752 | ||
f679623f RBJ |
4753 | /* |
4754 | * The "raid5_align_endio" should check if the read succeeded and if it | |
4755 | * did, call bio_endio on the original bio (having bio_put the new bio | |
4756 | * first). | |
4757 | * If the read failed.. | |
4758 | */ | |
6712ecf8 | 4759 | static void raid5_align_endio(struct bio *bi, int error) |
f679623f RBJ |
4760 | { |
4761 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 4762 | struct mddev *mddev; |
d1688a6d | 4763 | struct r5conf *conf; |
46031f9a | 4764 | int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); |
3cb03002 | 4765 | struct md_rdev *rdev; |
46031f9a | 4766 | |
f679623f | 4767 | bio_put(bi); |
46031f9a | 4768 | |
46031f9a RBJ |
4769 | rdev = (void*)raid_bi->bi_next; |
4770 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
4771 | mddev = rdev->mddev; |
4772 | conf = mddev->private; | |
46031f9a RBJ |
4773 | |
4774 | rdev_dec_pending(rdev, conf->mddev); | |
4775 | ||
4776 | if (!error && uptodate) { | |
0a82a8d1 LT |
4777 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
4778 | raid_bi, 0); | |
6712ecf8 | 4779 | bio_endio(raid_bi, 0); |
46031f9a | 4780 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 4781 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 4782 | return; |
46031f9a RBJ |
4783 | } |
4784 | ||
45b4233c | 4785 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
4786 | |
4787 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
4788 | } |
4789 | ||
387bb173 NB |
4790 | static int bio_fits_rdev(struct bio *bi) |
4791 | { | |
165125e1 | 4792 | struct request_queue *q = bdev_get_queue(bi->bi_bdev); |
387bb173 | 4793 | |
aa8b57aa | 4794 | if (bio_sectors(bi) > queue_max_sectors(q)) |
387bb173 NB |
4795 | return 0; |
4796 | blk_recount_segments(q, bi); | |
8a78362c | 4797 | if (bi->bi_phys_segments > queue_max_segments(q)) |
387bb173 NB |
4798 | return 0; |
4799 | ||
4800 | if (q->merge_bvec_fn) | |
4801 | /* it's too hard to apply the merge_bvec_fn at this stage, | |
4802 | * just just give up | |
4803 | */ | |
4804 | return 0; | |
4805 | ||
4806 | return 1; | |
4807 | } | |
4808 | ||
fd01b88c | 4809 | static int chunk_aligned_read(struct mddev *mddev, struct bio * raid_bio) |
f679623f | 4810 | { |
d1688a6d | 4811 | struct r5conf *conf = mddev->private; |
8553fe7e | 4812 | int dd_idx; |
f679623f | 4813 | struct bio* align_bi; |
3cb03002 | 4814 | struct md_rdev *rdev; |
671488cc | 4815 | sector_t end_sector; |
f679623f RBJ |
4816 | |
4817 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
45b4233c | 4818 | pr_debug("chunk_aligned_read : non aligned\n"); |
f679623f RBJ |
4819 | return 0; |
4820 | } | |
4821 | /* | |
a167f663 | 4822 | * use bio_clone_mddev to make a copy of the bio |
f679623f | 4823 | */ |
a167f663 | 4824 | align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev); |
f679623f RBJ |
4825 | if (!align_bi) |
4826 | return 0; | |
4827 | /* | |
4828 | * set bi_end_io to a new function, and set bi_private to the | |
4829 | * original bio. | |
4830 | */ | |
4831 | align_bi->bi_end_io = raid5_align_endio; | |
4832 | align_bi->bi_private = raid_bio; | |
4833 | /* | |
4834 | * compute position | |
4835 | */ | |
4f024f37 KO |
4836 | align_bi->bi_iter.bi_sector = |
4837 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
4838 | 0, &dd_idx, NULL); | |
f679623f | 4839 | |
f73a1c7d | 4840 | end_sector = bio_end_sector(align_bi); |
f679623f | 4841 | rcu_read_lock(); |
671488cc N |
4842 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
4843 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
4844 | rdev->recovery_offset < end_sector) { | |
4845 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
4846 | if (rdev && | |
4847 | (test_bit(Faulty, &rdev->flags) || | |
4848 | !(test_bit(In_sync, &rdev->flags) || | |
4849 | rdev->recovery_offset >= end_sector))) | |
4850 | rdev = NULL; | |
4851 | } | |
4852 | if (rdev) { | |
31c176ec N |
4853 | sector_t first_bad; |
4854 | int bad_sectors; | |
4855 | ||
f679623f RBJ |
4856 | atomic_inc(&rdev->nr_pending); |
4857 | rcu_read_unlock(); | |
46031f9a RBJ |
4858 | raid_bio->bi_next = (void*)rdev; |
4859 | align_bi->bi_bdev = rdev->bdev; | |
3fd83717 | 4860 | __clear_bit(BIO_SEG_VALID, &align_bi->bi_flags); |
46031f9a | 4861 | |
31c176ec | 4862 | if (!bio_fits_rdev(align_bi) || |
4f024f37 KO |
4863 | is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4864 | bio_sectors(align_bi), | |
31c176ec N |
4865 | &first_bad, &bad_sectors)) { |
4866 | /* too big in some way, or has a known bad block */ | |
387bb173 NB |
4867 | bio_put(align_bi); |
4868 | rdev_dec_pending(rdev, mddev); | |
4869 | return 0; | |
4870 | } | |
4871 | ||
6c0544e2 | 4872 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 4873 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 4874 | |
46031f9a | 4875 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 4876 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 4877 | conf->quiesce == 0, |
eed8c02e | 4878 | conf->device_lock); |
46031f9a RBJ |
4879 | atomic_inc(&conf->active_aligned_reads); |
4880 | spin_unlock_irq(&conf->device_lock); | |
4881 | ||
e3620a3a JB |
4882 | if (mddev->gendisk) |
4883 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
4884 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 4885 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
4886 | generic_make_request(align_bi); |
4887 | return 1; | |
4888 | } else { | |
4889 | rcu_read_unlock(); | |
46031f9a | 4890 | bio_put(align_bi); |
f679623f RBJ |
4891 | return 0; |
4892 | } | |
4893 | } | |
4894 | ||
8b3e6cdc DW |
4895 | /* __get_priority_stripe - get the next stripe to process |
4896 | * | |
4897 | * Full stripe writes are allowed to pass preread active stripes up until | |
4898 | * the bypass_threshold is exceeded. In general the bypass_count | |
4899 | * increments when the handle_list is handled before the hold_list; however, it | |
4900 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
4901 | * stripe with in flight i/o. The bypass_count will be reset when the | |
4902 | * head of the hold_list has changed, i.e. the head was promoted to the | |
4903 | * handle_list. | |
4904 | */ | |
851c30c9 | 4905 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 4906 | { |
851c30c9 SL |
4907 | struct stripe_head *sh = NULL, *tmp; |
4908 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 4909 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
4910 | |
4911 | if (conf->worker_cnt_per_group == 0) { | |
4912 | handle_list = &conf->handle_list; | |
4913 | } else if (group != ANY_GROUP) { | |
4914 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 4915 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
4916 | } else { |
4917 | int i; | |
4918 | for (i = 0; i < conf->group_cnt; i++) { | |
4919 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 4920 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
4921 | if (!list_empty(handle_list)) |
4922 | break; | |
4923 | } | |
4924 | } | |
8b3e6cdc DW |
4925 | |
4926 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
4927 | __func__, | |
851c30c9 | 4928 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
4929 | list_empty(&conf->hold_list) ? "empty" : "busy", |
4930 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
4931 | ||
851c30c9 SL |
4932 | if (!list_empty(handle_list)) { |
4933 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
4934 | |
4935 | if (list_empty(&conf->hold_list)) | |
4936 | conf->bypass_count = 0; | |
4937 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
4938 | if (conf->hold_list.next == conf->last_hold) | |
4939 | conf->bypass_count++; | |
4940 | else { | |
4941 | conf->last_hold = conf->hold_list.next; | |
4942 | conf->bypass_count -= conf->bypass_threshold; | |
4943 | if (conf->bypass_count < 0) | |
4944 | conf->bypass_count = 0; | |
4945 | } | |
4946 | } | |
4947 | } else if (!list_empty(&conf->hold_list) && | |
4948 | ((conf->bypass_threshold && | |
4949 | conf->bypass_count > conf->bypass_threshold) || | |
4950 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
4951 | |
4952 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
4953 | if (conf->worker_cnt_per_group == 0 || | |
4954 | group == ANY_GROUP || | |
4955 | !cpu_online(tmp->cpu) || | |
4956 | cpu_to_group(tmp->cpu) == group) { | |
4957 | sh = tmp; | |
4958 | break; | |
4959 | } | |
4960 | } | |
4961 | ||
4962 | if (sh) { | |
4963 | conf->bypass_count -= conf->bypass_threshold; | |
4964 | if (conf->bypass_count < 0) | |
4965 | conf->bypass_count = 0; | |
4966 | } | |
bfc90cb0 | 4967 | wg = NULL; |
851c30c9 SL |
4968 | } |
4969 | ||
4970 | if (!sh) | |
8b3e6cdc DW |
4971 | return NULL; |
4972 | ||
bfc90cb0 SL |
4973 | if (wg) { |
4974 | wg->stripes_cnt--; | |
4975 | sh->group = NULL; | |
4976 | } | |
8b3e6cdc | 4977 | list_del_init(&sh->lru); |
c7a6d35e | 4978 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
4979 | return sh; |
4980 | } | |
f679623f | 4981 | |
8811b596 SL |
4982 | struct raid5_plug_cb { |
4983 | struct blk_plug_cb cb; | |
4984 | struct list_head list; | |
566c09c5 | 4985 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
4986 | }; |
4987 | ||
4988 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
4989 | { | |
4990 | struct raid5_plug_cb *cb = container_of( | |
4991 | blk_cb, struct raid5_plug_cb, cb); | |
4992 | struct stripe_head *sh; | |
4993 | struct mddev *mddev = cb->cb.data; | |
4994 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 4995 | int cnt = 0; |
566c09c5 | 4996 | int hash; |
8811b596 SL |
4997 | |
4998 | if (cb->list.next && !list_empty(&cb->list)) { | |
4999 | spin_lock_irq(&conf->device_lock); | |
5000 | while (!list_empty(&cb->list)) { | |
5001 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5002 | list_del_init(&sh->lru); | |
5003 | /* | |
5004 | * avoid race release_stripe_plug() sees | |
5005 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5006 | * is still in our list | |
5007 | */ | |
4e857c58 | 5008 | smp_mb__before_atomic(); |
8811b596 | 5009 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5010 | /* |
5011 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5012 | * case, the count is always > 1 here | |
5013 | */ | |
566c09c5 SL |
5014 | hash = sh->hash_lock_index; |
5015 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5016 | cnt++; |
8811b596 SL |
5017 | } |
5018 | spin_unlock_irq(&conf->device_lock); | |
5019 | } | |
566c09c5 SL |
5020 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5021 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5022 | if (mddev->queue) |
5023 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5024 | kfree(cb); |
5025 | } | |
5026 | ||
5027 | static void release_stripe_plug(struct mddev *mddev, | |
5028 | struct stripe_head *sh) | |
5029 | { | |
5030 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5031 | raid5_unplug, mddev, | |
5032 | sizeof(struct raid5_plug_cb)); | |
5033 | struct raid5_plug_cb *cb; | |
5034 | ||
5035 | if (!blk_cb) { | |
5036 | release_stripe(sh); | |
5037 | return; | |
5038 | } | |
5039 | ||
5040 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5041 | ||
566c09c5 SL |
5042 | if (cb->list.next == NULL) { |
5043 | int i; | |
8811b596 | 5044 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5045 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5046 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5047 | } | |
8811b596 SL |
5048 | |
5049 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5050 | list_add_tail(&sh->lru, &cb->list); | |
5051 | else | |
5052 | release_stripe(sh); | |
5053 | } | |
5054 | ||
620125f2 SL |
5055 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5056 | { | |
5057 | struct r5conf *conf = mddev->private; | |
5058 | sector_t logical_sector, last_sector; | |
5059 | struct stripe_head *sh; | |
5060 | int remaining; | |
5061 | int stripe_sectors; | |
5062 | ||
5063 | if (mddev->reshape_position != MaxSector) | |
5064 | /* Skip discard while reshape is happening */ | |
5065 | return; | |
5066 | ||
4f024f37 KO |
5067 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5068 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5069 | |
5070 | bi->bi_next = NULL; | |
5071 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5072 | ||
5073 | stripe_sectors = conf->chunk_sectors * | |
5074 | (conf->raid_disks - conf->max_degraded); | |
5075 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5076 | stripe_sectors); | |
5077 | sector_div(last_sector, stripe_sectors); | |
5078 | ||
5079 | logical_sector *= conf->chunk_sectors; | |
5080 | last_sector *= conf->chunk_sectors; | |
5081 | ||
5082 | for (; logical_sector < last_sector; | |
5083 | logical_sector += STRIPE_SECTORS) { | |
5084 | DEFINE_WAIT(w); | |
5085 | int d; | |
5086 | again: | |
5087 | sh = get_active_stripe(conf, logical_sector, 0, 0, 0); | |
5088 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5089 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5090 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5091 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
5092 | release_stripe(sh); | |
5093 | schedule(); | |
5094 | goto again; | |
5095 | } | |
5096 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5097 | spin_lock_irq(&sh->stripe_lock); |
5098 | for (d = 0; d < conf->raid_disks; d++) { | |
5099 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5100 | continue; | |
5101 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5102 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5103 | spin_unlock_irq(&sh->stripe_lock); | |
5104 | release_stripe(sh); | |
5105 | schedule(); | |
5106 | goto again; | |
5107 | } | |
5108 | } | |
f8dfcffd | 5109 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5110 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5111 | sh->overwrite_disks = 0; |
620125f2 SL |
5112 | for (d = 0; d < conf->raid_disks; d++) { |
5113 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5114 | continue; | |
5115 | sh->dev[d].towrite = bi; | |
5116 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5117 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5118 | sh->overwrite_disks++; |
620125f2 SL |
5119 | } |
5120 | spin_unlock_irq(&sh->stripe_lock); | |
5121 | if (conf->mddev->bitmap) { | |
5122 | for (d = 0; | |
5123 | d < conf->raid_disks - conf->max_degraded; | |
5124 | d++) | |
5125 | bitmap_startwrite(mddev->bitmap, | |
5126 | sh->sector, | |
5127 | STRIPE_SECTORS, | |
5128 | 0); | |
5129 | sh->bm_seq = conf->seq_flush + 1; | |
5130 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5131 | } | |
5132 | ||
5133 | set_bit(STRIPE_HANDLE, &sh->state); | |
5134 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5135 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5136 | atomic_inc(&conf->preread_active_stripes); | |
5137 | release_stripe_plug(mddev, sh); | |
5138 | } | |
5139 | ||
5140 | remaining = raid5_dec_bi_active_stripes(bi); | |
5141 | if (remaining == 0) { | |
5142 | md_write_end(mddev); | |
5143 | bio_endio(bi, 0); | |
5144 | } | |
5145 | } | |
5146 | ||
b4fdcb02 | 5147 | static void make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5148 | { |
d1688a6d | 5149 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5150 | int dd_idx; |
1da177e4 LT |
5151 | sector_t new_sector; |
5152 | sector_t logical_sector, last_sector; | |
5153 | struct stripe_head *sh; | |
a362357b | 5154 | const int rw = bio_data_dir(bi); |
49077326 | 5155 | int remaining; |
27c0f68f SL |
5156 | DEFINE_WAIT(w); |
5157 | bool do_prepare; | |
1da177e4 | 5158 | |
e9c7469b TH |
5159 | if (unlikely(bi->bi_rw & REQ_FLUSH)) { |
5160 | md_flush_request(mddev, bi); | |
5a7bbad2 | 5161 | return; |
e5dcdd80 N |
5162 | } |
5163 | ||
3d310eb7 | 5164 | md_write_start(mddev, bi); |
06d91a5f | 5165 | |
9ffc8f7c EM |
5166 | /* |
5167 | * If array is degraded, better not do chunk aligned read because | |
5168 | * later we might have to read it again in order to reconstruct | |
5169 | * data on failed drives. | |
5170 | */ | |
5171 | if (rw == READ && mddev->degraded == 0 && | |
52488615 | 5172 | mddev->reshape_position == MaxSector && |
21a52c6d | 5173 | chunk_aligned_read(mddev,bi)) |
5a7bbad2 | 5174 | return; |
52488615 | 5175 | |
620125f2 SL |
5176 | if (unlikely(bi->bi_rw & REQ_DISCARD)) { |
5177 | make_discard_request(mddev, bi); | |
5178 | return; | |
5179 | } | |
5180 | ||
4f024f37 | 5181 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5182 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5183 | bi->bi_next = NULL; |
5184 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5185 | |
27c0f68f | 5186 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5187 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5188 | int previous; |
c46501b2 | 5189 | int seq; |
b578d55f | 5190 | |
27c0f68f | 5191 | do_prepare = false; |
7ecaa1e6 | 5192 | retry: |
c46501b2 | 5193 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5194 | previous = 0; |
27c0f68f SL |
5195 | if (do_prepare) |
5196 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5197 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5198 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5199 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5200 | * 64bit on a 32bit platform, and so it might be |
5201 | * possible to see a half-updated value | |
aeb878b0 | 5202 | * Of course reshape_progress could change after |
df8e7f76 N |
5203 | * the lock is dropped, so once we get a reference |
5204 | * to the stripe that we think it is, we will have | |
5205 | * to check again. | |
5206 | */ | |
7ecaa1e6 | 5207 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5208 | if (mddev->reshape_backwards |
fef9c61f N |
5209 | ? logical_sector < conf->reshape_progress |
5210 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5211 | previous = 1; |
5212 | } else { | |
2c810cdd | 5213 | if (mddev->reshape_backwards |
fef9c61f N |
5214 | ? logical_sector < conf->reshape_safe |
5215 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5216 | spin_unlock_irq(&conf->device_lock); |
5217 | schedule(); | |
27c0f68f | 5218 | do_prepare = true; |
b578d55f N |
5219 | goto retry; |
5220 | } | |
5221 | } | |
7ecaa1e6 N |
5222 | spin_unlock_irq(&conf->device_lock); |
5223 | } | |
16a53ecc | 5224 | |
112bf897 N |
5225 | new_sector = raid5_compute_sector(conf, logical_sector, |
5226 | previous, | |
911d4ee8 | 5227 | &dd_idx, NULL); |
0c55e022 | 5228 | pr_debug("raid456: make_request, sector %llu logical %llu\n", |
c46501b2 | 5229 | (unsigned long long)new_sector, |
1da177e4 LT |
5230 | (unsigned long long)logical_sector); |
5231 | ||
b5663ba4 | 5232 | sh = get_active_stripe(conf, new_sector, previous, |
a8c906ca | 5233 | (bi->bi_rw&RWA_MASK), 0); |
1da177e4 | 5234 | if (sh) { |
b0f9ec04 | 5235 | if (unlikely(previous)) { |
7ecaa1e6 | 5236 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5237 | * stripe, so we must do the range check again. |
5238 | * Expansion could still move past after this | |
5239 | * test, but as we are holding a reference to | |
5240 | * 'sh', we know that if that happens, | |
5241 | * STRIPE_EXPANDING will get set and the expansion | |
5242 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5243 | */ |
5244 | int must_retry = 0; | |
5245 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5246 | if (mddev->reshape_backwards |
b0f9ec04 N |
5247 | ? logical_sector >= conf->reshape_progress |
5248 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5249 | /* mismatch, need to try again */ |
5250 | must_retry = 1; | |
5251 | spin_unlock_irq(&conf->device_lock); | |
5252 | if (must_retry) { | |
5253 | release_stripe(sh); | |
7a3ab908 | 5254 | schedule(); |
27c0f68f | 5255 | do_prepare = true; |
7ecaa1e6 N |
5256 | goto retry; |
5257 | } | |
5258 | } | |
c46501b2 N |
5259 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5260 | /* Might have got the wrong stripe_head | |
5261 | * by accident | |
5262 | */ | |
5263 | release_stripe(sh); | |
5264 | goto retry; | |
5265 | } | |
e62e58a5 | 5266 | |
ffd96e35 | 5267 | if (rw == WRITE && |
a5c308d4 | 5268 | logical_sector >= mddev->suspend_lo && |
e464eafd N |
5269 | logical_sector < mddev->suspend_hi) { |
5270 | release_stripe(sh); | |
e62e58a5 N |
5271 | /* As the suspend_* range is controlled by |
5272 | * userspace, we want an interruptible | |
5273 | * wait. | |
5274 | */ | |
5275 | flush_signals(current); | |
5276 | prepare_to_wait(&conf->wait_for_overlap, | |
5277 | &w, TASK_INTERRUPTIBLE); | |
5278 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5279 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5280 | schedule(); |
27c0f68f SL |
5281 | do_prepare = true; |
5282 | } | |
e464eafd N |
5283 | goto retry; |
5284 | } | |
7ecaa1e6 N |
5285 | |
5286 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5287 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5288 | /* Stripe is busy expanding or |
5289 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5290 | * and wait a while |
5291 | */ | |
482c0834 | 5292 | md_wakeup_thread(mddev->thread); |
1da177e4 LT |
5293 | release_stripe(sh); |
5294 | schedule(); | |
27c0f68f | 5295 | do_prepare = true; |
1da177e4 LT |
5296 | goto retry; |
5297 | } | |
6ed3003c N |
5298 | set_bit(STRIPE_HANDLE, &sh->state); |
5299 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5300 | if ((!sh->batch_head || sh == sh->batch_head) && |
5301 | (bi->bi_rw & REQ_SYNC) && | |
729a1866 N |
5302 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5303 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5304 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5305 | } else { |
5306 | /* cannot get stripe for read-ahead, just give-up */ | |
5307 | clear_bit(BIO_UPTODATE, &bi->bi_flags); | |
1da177e4 LT |
5308 | break; |
5309 | } | |
1da177e4 | 5310 | } |
27c0f68f | 5311 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5312 | |
e7836bd6 | 5313 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5314 | if (remaining == 0) { |
1da177e4 | 5315 | |
16a53ecc | 5316 | if ( rw == WRITE ) |
1da177e4 | 5317 | md_write_end(mddev); |
6712ecf8 | 5318 | |
0a82a8d1 LT |
5319 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5320 | bi, 0); | |
0e13fe23 | 5321 | bio_endio(bi, 0); |
1da177e4 | 5322 | } |
1da177e4 LT |
5323 | } |
5324 | ||
fd01b88c | 5325 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5326 | |
fd01b88c | 5327 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5328 | { |
52c03291 N |
5329 | /* reshaping is quite different to recovery/resync so it is |
5330 | * handled quite separately ... here. | |
5331 | * | |
5332 | * On each call to sync_request, we gather one chunk worth of | |
5333 | * destination stripes and flag them as expanding. | |
5334 | * Then we find all the source stripes and request reads. | |
5335 | * As the reads complete, handle_stripe will copy the data | |
5336 | * into the destination stripe and release that stripe. | |
5337 | */ | |
d1688a6d | 5338 | struct r5conf *conf = mddev->private; |
1da177e4 | 5339 | struct stripe_head *sh; |
ccfcc3c1 | 5340 | sector_t first_sector, last_sector; |
f416885e N |
5341 | int raid_disks = conf->previous_raid_disks; |
5342 | int data_disks = raid_disks - conf->max_degraded; | |
5343 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5344 | int i; |
5345 | int dd_idx; | |
c8f517c4 | 5346 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5347 | sector_t stripe_addr; |
7a661381 | 5348 | int reshape_sectors; |
ab69ae12 | 5349 | struct list_head stripes; |
92140480 | 5350 | sector_t retn; |
52c03291 | 5351 | |
fef9c61f N |
5352 | if (sector_nr == 0) { |
5353 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5354 | if (mddev->reshape_backwards && |
fef9c61f N |
5355 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5356 | sector_nr = raid5_size(mddev, 0, 0) | |
5357 | - conf->reshape_progress; | |
2c810cdd | 5358 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5359 | conf->reshape_progress > 0) |
5360 | sector_nr = conf->reshape_progress; | |
f416885e | 5361 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5362 | if (sector_nr) { |
8dee7211 N |
5363 | mddev->curr_resync_completed = sector_nr; |
5364 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5365 | *skipped = 1; |
92140480 N |
5366 | retn = sector_nr; |
5367 | goto finish; | |
fef9c61f | 5368 | } |
52c03291 N |
5369 | } |
5370 | ||
7a661381 N |
5371 | /* We need to process a full chunk at a time. |
5372 | * If old and new chunk sizes differ, we need to process the | |
5373 | * largest of these | |
5374 | */ | |
3cb5edf4 N |
5375 | |
5376 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5377 | |
b5254dd5 N |
5378 | /* We update the metadata at least every 10 seconds, or when |
5379 | * the data about to be copied would over-write the source of | |
5380 | * the data at the front of the range. i.e. one new_stripe | |
5381 | * along from reshape_progress new_maps to after where | |
5382 | * reshape_safe old_maps to | |
52c03291 | 5383 | */ |
fef9c61f | 5384 | writepos = conf->reshape_progress; |
f416885e | 5385 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5386 | readpos = conf->reshape_progress; |
5387 | sector_div(readpos, data_disks); | |
fef9c61f | 5388 | safepos = conf->reshape_safe; |
f416885e | 5389 | sector_div(safepos, data_disks); |
2c810cdd | 5390 | if (mddev->reshape_backwards) { |
ed37d83e | 5391 | writepos -= min_t(sector_t, reshape_sectors, writepos); |
c8f517c4 | 5392 | readpos += reshape_sectors; |
7a661381 | 5393 | safepos += reshape_sectors; |
fef9c61f | 5394 | } else { |
7a661381 | 5395 | writepos += reshape_sectors; |
ed37d83e N |
5396 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5397 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5398 | } |
52c03291 | 5399 | |
b5254dd5 N |
5400 | /* Having calculated the 'writepos' possibly use it |
5401 | * to set 'stripe_addr' which is where we will write to. | |
5402 | */ | |
5403 | if (mddev->reshape_backwards) { | |
5404 | BUG_ON(conf->reshape_progress == 0); | |
5405 | stripe_addr = writepos; | |
5406 | BUG_ON((mddev->dev_sectors & | |
5407 | ~((sector_t)reshape_sectors - 1)) | |
5408 | - reshape_sectors - stripe_addr | |
5409 | != sector_nr); | |
5410 | } else { | |
5411 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5412 | stripe_addr = sector_nr; | |
5413 | } | |
5414 | ||
c8f517c4 N |
5415 | /* 'writepos' is the most advanced device address we might write. |
5416 | * 'readpos' is the least advanced device address we might read. | |
5417 | * 'safepos' is the least address recorded in the metadata as having | |
5418 | * been reshaped. | |
b5254dd5 N |
5419 | * If there is a min_offset_diff, these are adjusted either by |
5420 | * increasing the safepos/readpos if diff is negative, or | |
5421 | * increasing writepos if diff is positive. | |
5422 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5423 | * ensure safety in the face of a crash - that must be done by userspace |
5424 | * making a backup of the data. So in that case there is no particular | |
5425 | * rush to update metadata. | |
5426 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5427 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5428 | * we can be safe in the event of a crash. | |
5429 | * So we insist on updating metadata if safepos is behind writepos and | |
5430 | * readpos is beyond writepos. | |
5431 | * In any case, update the metadata every 10 seconds. | |
5432 | * Maybe that number should be configurable, but I'm not sure it is | |
5433 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5434 | */ | |
b5254dd5 N |
5435 | if (conf->min_offset_diff < 0) { |
5436 | safepos += -conf->min_offset_diff; | |
5437 | readpos += -conf->min_offset_diff; | |
5438 | } else | |
5439 | writepos += conf->min_offset_diff; | |
5440 | ||
2c810cdd | 5441 | if ((mddev->reshape_backwards |
c8f517c4 N |
5442 | ? (safepos > writepos && readpos < writepos) |
5443 | : (safepos < writepos && readpos > writepos)) || | |
5444 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5445 | /* Cannot proceed until we've updated the superblock... */ |
5446 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5447 | atomic_read(&conf->reshape_stripes)==0 |
5448 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5449 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5450 | return 0; | |
fef9c61f | 5451 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5452 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5453 | conf->reshape_checkpoint = jiffies; |
850b2b42 | 5454 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
52c03291 | 5455 | md_wakeup_thread(mddev->thread); |
850b2b42 | 5456 | wait_event(mddev->sb_wait, mddev->flags == 0 || |
c91abf5a N |
5457 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5458 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5459 | return 0; | |
52c03291 | 5460 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5461 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5462 | spin_unlock_irq(&conf->device_lock); |
5463 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5464 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5465 | } |
5466 | ||
ab69ae12 | 5467 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5468 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5469 | int j; |
a9f326eb | 5470 | int skipped_disk = 0; |
a8c906ca | 5471 | sh = get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5472 | set_bit(STRIPE_EXPANDING, &sh->state); |
5473 | atomic_inc(&conf->reshape_stripes); | |
5474 | /* If any of this stripe is beyond the end of the old | |
5475 | * array, then we need to zero those blocks | |
5476 | */ | |
5477 | for (j=sh->disks; j--;) { | |
5478 | sector_t s; | |
5479 | if (j == sh->pd_idx) | |
5480 | continue; | |
f416885e | 5481 | if (conf->level == 6 && |
d0dabf7e | 5482 | j == sh->qd_idx) |
f416885e | 5483 | continue; |
784052ec | 5484 | s = compute_blocknr(sh, j, 0); |
b522adcd | 5485 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5486 | skipped_disk = 1; |
52c03291 N |
5487 | continue; |
5488 | } | |
5489 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5490 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5491 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5492 | } | |
a9f326eb | 5493 | if (!skipped_disk) { |
52c03291 N |
5494 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5495 | set_bit(STRIPE_HANDLE, &sh->state); | |
5496 | } | |
ab69ae12 | 5497 | list_add(&sh->lru, &stripes); |
52c03291 N |
5498 | } |
5499 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5500 | if (mddev->reshape_backwards) |
7a661381 | 5501 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5502 | else |
7a661381 | 5503 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5504 | spin_unlock_irq(&conf->device_lock); |
5505 | /* Ok, those stripe are ready. We can start scheduling | |
5506 | * reads on the source stripes. | |
5507 | * The source stripes are determined by mapping the first and last | |
5508 | * block on the destination stripes. | |
5509 | */ | |
52c03291 | 5510 | first_sector = |
ec32a2bd | 5511 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5512 | 1, &dd_idx, NULL); |
52c03291 | 5513 | last_sector = |
0e6e0271 | 5514 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5515 | * new_data_disks - 1), |
911d4ee8 | 5516 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5517 | if (last_sector >= mddev->dev_sectors) |
5518 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5519 | while (first_sector <= last_sector) { |
a8c906ca | 5520 | sh = get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5521 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5522 | set_bit(STRIPE_HANDLE, &sh->state); | |
5523 | release_stripe(sh); | |
5524 | first_sector += STRIPE_SECTORS; | |
5525 | } | |
ab69ae12 N |
5526 | /* Now that the sources are clearly marked, we can release |
5527 | * the destination stripes | |
5528 | */ | |
5529 | while (!list_empty(&stripes)) { | |
5530 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5531 | list_del_init(&sh->lru); | |
5532 | release_stripe(sh); | |
5533 | } | |
c6207277 N |
5534 | /* If this takes us to the resync_max point where we have to pause, |
5535 | * then we need to write out the superblock. | |
5536 | */ | |
7a661381 | 5537 | sector_nr += reshape_sectors; |
92140480 N |
5538 | retn = reshape_sectors; |
5539 | finish: | |
c03f6a19 N |
5540 | if ((sector_nr - mddev->curr_resync_completed) * 2 |
5541 | >= mddev->resync_max - mddev->curr_resync_completed) { | |
c6207277 N |
5542 | /* Cannot proceed until we've updated the superblock... */ |
5543 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5544 | atomic_read(&conf->reshape_stripes) == 0 |
5545 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5546 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5547 | goto ret; | |
fef9c61f | 5548 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5549 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5550 | conf->reshape_checkpoint = jiffies; |
c6207277 N |
5551 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
5552 | md_wakeup_thread(mddev->thread); | |
5553 | wait_event(mddev->sb_wait, | |
5554 | !test_bit(MD_CHANGE_DEVS, &mddev->flags) | |
c91abf5a N |
5555 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5556 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5557 | goto ret; | |
c6207277 | 5558 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5559 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5560 | spin_unlock_irq(&conf->device_lock); |
5561 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5562 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5563 | } |
c91abf5a | 5564 | ret: |
92140480 | 5565 | return retn; |
52c03291 N |
5566 | } |
5567 | ||
09314799 | 5568 | static inline sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
52c03291 | 5569 | { |
d1688a6d | 5570 | struct r5conf *conf = mddev->private; |
52c03291 | 5571 | struct stripe_head *sh; |
58c0fed4 | 5572 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5573 | sector_t sync_blocks; |
16a53ecc N |
5574 | int still_degraded = 0; |
5575 | int i; | |
1da177e4 | 5576 | |
72626685 | 5577 | if (sector_nr >= max_sector) { |
1da177e4 | 5578 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5579 | |
29269553 N |
5580 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5581 | end_reshape(conf); | |
5582 | return 0; | |
5583 | } | |
72626685 N |
5584 | |
5585 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5586 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5587 | &sync_blocks, 1); | |
16a53ecc | 5588 | else /* completed sync */ |
72626685 N |
5589 | conf->fullsync = 0; |
5590 | bitmap_close_sync(mddev->bitmap); | |
5591 | ||
1da177e4 LT |
5592 | return 0; |
5593 | } | |
ccfcc3c1 | 5594 | |
64bd660b N |
5595 | /* Allow raid5_quiesce to complete */ |
5596 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5597 | ||
52c03291 N |
5598 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5599 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5600 | |
c6207277 N |
5601 | /* No need to check resync_max as we never do more than one |
5602 | * stripe, and as resync_max will always be on a chunk boundary, | |
5603 | * if the check in md_do_sync didn't fire, there is no chance | |
5604 | * of overstepping resync_max here | |
5605 | */ | |
5606 | ||
16a53ecc | 5607 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5608 | * to resync, then assert that we are finished, because there is |
5609 | * nothing we can do. | |
5610 | */ | |
3285edf1 | 5611 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5612 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5613 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5614 | *skipped = 1; |
1da177e4 LT |
5615 | return rv; |
5616 | } | |
6f608040 | 5617 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5618 | !conf->fullsync && | |
5619 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5620 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5621 | /* we can skip this block, and probably more */ |
5622 | sync_blocks /= STRIPE_SECTORS; | |
5623 | *skipped = 1; | |
5624 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5625 | } | |
1da177e4 | 5626 | |
b47490c9 N |
5627 | bitmap_cond_end_sync(mddev->bitmap, sector_nr); |
5628 | ||
a8c906ca | 5629 | sh = get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5630 | if (sh == NULL) { |
a8c906ca | 5631 | sh = get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5632 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5633 | * is trying to get access |
1da177e4 | 5634 | */ |
66c006a5 | 5635 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5636 | } |
16a53ecc | 5637 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5638 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5639 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5640 | */ |
16d9cfab EM |
5641 | rcu_read_lock(); |
5642 | for (i = 0; i < conf->raid_disks; i++) { | |
5643 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5644 | ||
5645 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5646 | still_degraded = 1; |
16d9cfab EM |
5647 | } |
5648 | rcu_read_unlock(); | |
16a53ecc N |
5649 | |
5650 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5651 | ||
83206d66 | 5652 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5653 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5654 | |
1da177e4 LT |
5655 | release_stripe(sh); |
5656 | ||
5657 | return STRIPE_SECTORS; | |
5658 | } | |
5659 | ||
d1688a6d | 5660 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5661 | { |
5662 | /* We may not be able to submit a whole bio at once as there | |
5663 | * may not be enough stripe_heads available. | |
5664 | * We cannot pre-allocate enough stripe_heads as we may need | |
5665 | * more than exist in the cache (if we allow ever large chunks). | |
5666 | * So we do one stripe head at a time and record in | |
5667 | * ->bi_hw_segments how many have been done. | |
5668 | * | |
5669 | * We *know* that this entire raid_bio is in one chunk, so | |
5670 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5671 | */ | |
5672 | struct stripe_head *sh; | |
911d4ee8 | 5673 | int dd_idx; |
46031f9a RBJ |
5674 | sector_t sector, logical_sector, last_sector; |
5675 | int scnt = 0; | |
5676 | int remaining; | |
5677 | int handled = 0; | |
5678 | ||
4f024f37 KO |
5679 | logical_sector = raid_bio->bi_iter.bi_sector & |
5680 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5681 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5682 | 0, &dd_idx, NULL); |
f73a1c7d | 5683 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5684 | |
5685 | for (; logical_sector < last_sector; | |
387bb173 NB |
5686 | logical_sector += STRIPE_SECTORS, |
5687 | sector += STRIPE_SECTORS, | |
5688 | scnt++) { | |
46031f9a | 5689 | |
e7836bd6 | 5690 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5691 | /* already done this stripe */ |
5692 | continue; | |
5693 | ||
2844dc32 | 5694 | sh = get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5695 | |
5696 | if (!sh) { | |
5697 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5698 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5699 | conf->retry_read_aligned = raid_bio; |
5700 | return handled; | |
5701 | } | |
5702 | ||
da41ba65 | 5703 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
387bb173 | 5704 | release_stripe(sh); |
e7836bd6 | 5705 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
5706 | conf->retry_read_aligned = raid_bio; |
5707 | return handled; | |
5708 | } | |
5709 | ||
3f9e7c14 | 5710 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 5711 | handle_stripe(sh); |
46031f9a RBJ |
5712 | release_stripe(sh); |
5713 | handled++; | |
5714 | } | |
e7836bd6 | 5715 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
5716 | if (remaining == 0) { |
5717 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
5718 | raid_bio, 0); | |
0e13fe23 | 5719 | bio_endio(raid_bio, 0); |
0a82a8d1 | 5720 | } |
46031f9a | 5721 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5722 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
5723 | return handled; |
5724 | } | |
5725 | ||
bfc90cb0 | 5726 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
5727 | struct r5worker *worker, |
5728 | struct list_head *temp_inactive_list) | |
46a06401 SL |
5729 | { |
5730 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
5731 | int i, batch_size = 0, hash; |
5732 | bool release_inactive = false; | |
46a06401 SL |
5733 | |
5734 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 5735 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
5736 | batch[batch_size++] = sh; |
5737 | ||
566c09c5 SL |
5738 | if (batch_size == 0) { |
5739 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
5740 | if (!list_empty(temp_inactive_list + i)) | |
5741 | break; | |
5742 | if (i == NR_STRIPE_HASH_LOCKS) | |
5743 | return batch_size; | |
5744 | release_inactive = true; | |
5745 | } | |
46a06401 SL |
5746 | spin_unlock_irq(&conf->device_lock); |
5747 | ||
566c09c5 SL |
5748 | release_inactive_stripe_list(conf, temp_inactive_list, |
5749 | NR_STRIPE_HASH_LOCKS); | |
5750 | ||
5751 | if (release_inactive) { | |
5752 | spin_lock_irq(&conf->device_lock); | |
5753 | return 0; | |
5754 | } | |
5755 | ||
46a06401 SL |
5756 | for (i = 0; i < batch_size; i++) |
5757 | handle_stripe(batch[i]); | |
5758 | ||
5759 | cond_resched(); | |
5760 | ||
5761 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
5762 | for (i = 0; i < batch_size; i++) { |
5763 | hash = batch[i]->hash_lock_index; | |
5764 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
5765 | } | |
46a06401 SL |
5766 | return batch_size; |
5767 | } | |
46031f9a | 5768 | |
851c30c9 SL |
5769 | static void raid5_do_work(struct work_struct *work) |
5770 | { | |
5771 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
5772 | struct r5worker_group *group = worker->group; | |
5773 | struct r5conf *conf = group->conf; | |
5774 | int group_id = group - conf->worker_groups; | |
5775 | int handled; | |
5776 | struct blk_plug plug; | |
5777 | ||
5778 | pr_debug("+++ raid5worker active\n"); | |
5779 | ||
5780 | blk_start_plug(&plug); | |
5781 | handled = 0; | |
5782 | spin_lock_irq(&conf->device_lock); | |
5783 | while (1) { | |
5784 | int batch_size, released; | |
5785 | ||
566c09c5 | 5786 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 5787 | |
566c09c5 SL |
5788 | batch_size = handle_active_stripes(conf, group_id, worker, |
5789 | worker->temp_inactive_list); | |
bfc90cb0 | 5790 | worker->working = false; |
851c30c9 SL |
5791 | if (!batch_size && !released) |
5792 | break; | |
5793 | handled += batch_size; | |
5794 | } | |
5795 | pr_debug("%d stripes handled\n", handled); | |
5796 | ||
5797 | spin_unlock_irq(&conf->device_lock); | |
5798 | blk_finish_plug(&plug); | |
5799 | ||
5800 | pr_debug("--- raid5worker inactive\n"); | |
5801 | } | |
5802 | ||
1da177e4 LT |
5803 | /* |
5804 | * This is our raid5 kernel thread. | |
5805 | * | |
5806 | * We scan the hash table for stripes which can be handled now. | |
5807 | * During the scan, completed stripes are saved for us by the interrupt | |
5808 | * handler, so that they will not have to wait for our next wakeup. | |
5809 | */ | |
4ed8731d | 5810 | static void raid5d(struct md_thread *thread) |
1da177e4 | 5811 | { |
4ed8731d | 5812 | struct mddev *mddev = thread->mddev; |
d1688a6d | 5813 | struct r5conf *conf = mddev->private; |
1da177e4 | 5814 | int handled; |
e1dfa0a2 | 5815 | struct blk_plug plug; |
1da177e4 | 5816 | |
45b4233c | 5817 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
5818 | |
5819 | md_check_recovery(mddev); | |
1da177e4 | 5820 | |
e1dfa0a2 | 5821 | blk_start_plug(&plug); |
1da177e4 LT |
5822 | handled = 0; |
5823 | spin_lock_irq(&conf->device_lock); | |
5824 | while (1) { | |
46031f9a | 5825 | struct bio *bio; |
773ca82f SL |
5826 | int batch_size, released; |
5827 | ||
566c09c5 | 5828 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
5829 | if (released) |
5830 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 5831 | |
0021b7bc | 5832 | if ( |
7c13edc8 N |
5833 | !list_empty(&conf->bitmap_list)) { |
5834 | /* Now is a good time to flush some bitmap updates */ | |
5835 | conf->seq_flush++; | |
700e432d | 5836 | spin_unlock_irq(&conf->device_lock); |
72626685 | 5837 | bitmap_unplug(mddev->bitmap); |
700e432d | 5838 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 5839 | conf->seq_write = conf->seq_flush; |
566c09c5 | 5840 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 5841 | } |
0021b7bc | 5842 | raid5_activate_delayed(conf); |
72626685 | 5843 | |
46031f9a RBJ |
5844 | while ((bio = remove_bio_from_retry(conf))) { |
5845 | int ok; | |
5846 | spin_unlock_irq(&conf->device_lock); | |
5847 | ok = retry_aligned_read(conf, bio); | |
5848 | spin_lock_irq(&conf->device_lock); | |
5849 | if (!ok) | |
5850 | break; | |
5851 | handled++; | |
5852 | } | |
5853 | ||
566c09c5 SL |
5854 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
5855 | conf->temp_inactive_list); | |
773ca82f | 5856 | if (!batch_size && !released) |
1da177e4 | 5857 | break; |
46a06401 | 5858 | handled += batch_size; |
1da177e4 | 5859 | |
46a06401 SL |
5860 | if (mddev->flags & ~(1<<MD_CHANGE_PENDING)) { |
5861 | spin_unlock_irq(&conf->device_lock); | |
de393cde | 5862 | md_check_recovery(mddev); |
46a06401 SL |
5863 | spin_lock_irq(&conf->device_lock); |
5864 | } | |
1da177e4 | 5865 | } |
45b4233c | 5866 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
5867 | |
5868 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
5869 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
5870 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
5871 | grow_one_stripe(conf, __GFP_NOWARN); |
5872 | /* Set flag even if allocation failed. This helps | |
5873 | * slow down allocation requests when mem is short | |
5874 | */ | |
5875 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 5876 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 5877 | } |
1da177e4 | 5878 | |
c9f21aaf | 5879 | async_tx_issue_pending_all(); |
e1dfa0a2 | 5880 | blk_finish_plug(&plug); |
1da177e4 | 5881 | |
45b4233c | 5882 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
5883 | } |
5884 | ||
3f294f4f | 5885 | static ssize_t |
fd01b88c | 5886 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 5887 | { |
7b1485ba N |
5888 | struct r5conf *conf; |
5889 | int ret = 0; | |
5890 | spin_lock(&mddev->lock); | |
5891 | conf = mddev->private; | |
96de1e66 | 5892 | if (conf) |
edbe83ab | 5893 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
5894 | spin_unlock(&mddev->lock); |
5895 | return ret; | |
3f294f4f N |
5896 | } |
5897 | ||
c41d4ac4 | 5898 | int |
fd01b88c | 5899 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 5900 | { |
d1688a6d | 5901 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
5902 | int err; |
5903 | ||
c41d4ac4 | 5904 | if (size <= 16 || size > 32768) |
3f294f4f | 5905 | return -EINVAL; |
486f0644 | 5906 | |
edbe83ab | 5907 | conf->min_nr_stripes = size; |
2d5b569b | 5908 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5909 | while (size < conf->max_nr_stripes && |
5910 | drop_one_stripe(conf)) | |
5911 | ; | |
2d5b569b | 5912 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5913 | |
edbe83ab | 5914 | |
b5470dc5 DW |
5915 | err = md_allow_write(mddev); |
5916 | if (err) | |
5917 | return err; | |
486f0644 | 5918 | |
2d5b569b | 5919 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
5920 | while (size > conf->max_nr_stripes) |
5921 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
5922 | break; | |
2d5b569b | 5923 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 5924 | |
c41d4ac4 N |
5925 | return 0; |
5926 | } | |
5927 | EXPORT_SYMBOL(raid5_set_cache_size); | |
5928 | ||
5929 | static ssize_t | |
fd01b88c | 5930 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 5931 | { |
6791875e | 5932 | struct r5conf *conf; |
c41d4ac4 N |
5933 | unsigned long new; |
5934 | int err; | |
5935 | ||
5936 | if (len >= PAGE_SIZE) | |
5937 | return -EINVAL; | |
b29bebd6 | 5938 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 5939 | return -EINVAL; |
6791875e | 5940 | err = mddev_lock(mddev); |
c41d4ac4 N |
5941 | if (err) |
5942 | return err; | |
6791875e N |
5943 | conf = mddev->private; |
5944 | if (!conf) | |
5945 | err = -ENODEV; | |
5946 | else | |
5947 | err = raid5_set_cache_size(mddev, new); | |
5948 | mddev_unlock(mddev); | |
5949 | ||
5950 | return err ?: len; | |
3f294f4f | 5951 | } |
007583c9 | 5952 | |
96de1e66 N |
5953 | static struct md_sysfs_entry |
5954 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
5955 | raid5_show_stripe_cache_size, | |
5956 | raid5_store_stripe_cache_size); | |
3f294f4f | 5957 | |
d06f191f MS |
5958 | static ssize_t |
5959 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
5960 | { | |
5961 | struct r5conf *conf = mddev->private; | |
5962 | if (conf) | |
5963 | return sprintf(page, "%d\n", conf->rmw_level); | |
5964 | else | |
5965 | return 0; | |
5966 | } | |
5967 | ||
5968 | static ssize_t | |
5969 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
5970 | { | |
5971 | struct r5conf *conf = mddev->private; | |
5972 | unsigned long new; | |
5973 | ||
5974 | if (!conf) | |
5975 | return -ENODEV; | |
5976 | ||
5977 | if (len >= PAGE_SIZE) | |
5978 | return -EINVAL; | |
5979 | ||
5980 | if (kstrtoul(page, 10, &new)) | |
5981 | return -EINVAL; | |
5982 | ||
5983 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
5984 | return -EINVAL; | |
5985 | ||
5986 | if (new != PARITY_DISABLE_RMW && | |
5987 | new != PARITY_ENABLE_RMW && | |
5988 | new != PARITY_PREFER_RMW) | |
5989 | return -EINVAL; | |
5990 | ||
5991 | conf->rmw_level = new; | |
5992 | return len; | |
5993 | } | |
5994 | ||
5995 | static struct md_sysfs_entry | |
5996 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
5997 | raid5_show_rmw_level, | |
5998 | raid5_store_rmw_level); | |
5999 | ||
6000 | ||
8b3e6cdc | 6001 | static ssize_t |
fd01b88c | 6002 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6003 | { |
7b1485ba N |
6004 | struct r5conf *conf; |
6005 | int ret = 0; | |
6006 | spin_lock(&mddev->lock); | |
6007 | conf = mddev->private; | |
8b3e6cdc | 6008 | if (conf) |
7b1485ba N |
6009 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6010 | spin_unlock(&mddev->lock); | |
6011 | return ret; | |
8b3e6cdc DW |
6012 | } |
6013 | ||
6014 | static ssize_t | |
fd01b88c | 6015 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6016 | { |
6791875e | 6017 | struct r5conf *conf; |
4ef197d8 | 6018 | unsigned long new; |
6791875e N |
6019 | int err; |
6020 | ||
8b3e6cdc DW |
6021 | if (len >= PAGE_SIZE) |
6022 | return -EINVAL; | |
b29bebd6 | 6023 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6024 | return -EINVAL; |
6791875e N |
6025 | |
6026 | err = mddev_lock(mddev); | |
6027 | if (err) | |
6028 | return err; | |
6029 | conf = mddev->private; | |
6030 | if (!conf) | |
6031 | err = -ENODEV; | |
edbe83ab | 6032 | else if (new > conf->min_nr_stripes) |
6791875e N |
6033 | err = -EINVAL; |
6034 | else | |
6035 | conf->bypass_threshold = new; | |
6036 | mddev_unlock(mddev); | |
6037 | return err ?: len; | |
8b3e6cdc DW |
6038 | } |
6039 | ||
6040 | static struct md_sysfs_entry | |
6041 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6042 | S_IRUGO | S_IWUSR, | |
6043 | raid5_show_preread_threshold, | |
6044 | raid5_store_preread_threshold); | |
6045 | ||
d592a996 SL |
6046 | static ssize_t |
6047 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6048 | { | |
7b1485ba N |
6049 | struct r5conf *conf; |
6050 | int ret = 0; | |
6051 | spin_lock(&mddev->lock); | |
6052 | conf = mddev->private; | |
d592a996 | 6053 | if (conf) |
7b1485ba N |
6054 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6055 | spin_unlock(&mddev->lock); | |
6056 | return ret; | |
d592a996 SL |
6057 | } |
6058 | ||
6059 | static ssize_t | |
6060 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6061 | { | |
6791875e | 6062 | struct r5conf *conf; |
d592a996 | 6063 | unsigned long new; |
6791875e N |
6064 | int err; |
6065 | ||
d592a996 SL |
6066 | if (len >= PAGE_SIZE) |
6067 | return -EINVAL; | |
d592a996 SL |
6068 | if (kstrtoul(page, 10, &new)) |
6069 | return -EINVAL; | |
6070 | new = !!new; | |
6791875e N |
6071 | |
6072 | err = mddev_lock(mddev); | |
6073 | if (err) | |
6074 | return err; | |
6075 | conf = mddev->private; | |
6076 | if (!conf) | |
6077 | err = -ENODEV; | |
6078 | else if (new != conf->skip_copy) { | |
6079 | mddev_suspend(mddev); | |
6080 | conf->skip_copy = new; | |
6081 | if (new) | |
6082 | mddev->queue->backing_dev_info.capabilities |= | |
6083 | BDI_CAP_STABLE_WRITES; | |
6084 | else | |
6085 | mddev->queue->backing_dev_info.capabilities &= | |
6086 | ~BDI_CAP_STABLE_WRITES; | |
6087 | mddev_resume(mddev); | |
6088 | } | |
6089 | mddev_unlock(mddev); | |
6090 | return err ?: len; | |
d592a996 SL |
6091 | } |
6092 | ||
6093 | static struct md_sysfs_entry | |
6094 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6095 | raid5_show_skip_copy, | |
6096 | raid5_store_skip_copy); | |
6097 | ||
3f294f4f | 6098 | static ssize_t |
fd01b88c | 6099 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6100 | { |
d1688a6d | 6101 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6102 | if (conf) |
6103 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6104 | else | |
6105 | return 0; | |
3f294f4f N |
6106 | } |
6107 | ||
96de1e66 N |
6108 | static struct md_sysfs_entry |
6109 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6110 | |
b721420e SL |
6111 | static ssize_t |
6112 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6113 | { | |
7b1485ba N |
6114 | struct r5conf *conf; |
6115 | int ret = 0; | |
6116 | spin_lock(&mddev->lock); | |
6117 | conf = mddev->private; | |
b721420e | 6118 | if (conf) |
7b1485ba N |
6119 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6120 | spin_unlock(&mddev->lock); | |
6121 | return ret; | |
b721420e SL |
6122 | } |
6123 | ||
60aaf933 | 6124 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6125 | int *group_cnt, | |
6126 | int *worker_cnt_per_group, | |
6127 | struct r5worker_group **worker_groups); | |
b721420e SL |
6128 | static ssize_t |
6129 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6130 | { | |
6791875e | 6131 | struct r5conf *conf; |
b721420e SL |
6132 | unsigned long new; |
6133 | int err; | |
60aaf933 | 6134 | struct r5worker_group *new_groups, *old_groups; |
6135 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6136 | |
6137 | if (len >= PAGE_SIZE) | |
6138 | return -EINVAL; | |
b721420e SL |
6139 | if (kstrtoul(page, 10, &new)) |
6140 | return -EINVAL; | |
6141 | ||
6791875e N |
6142 | err = mddev_lock(mddev); |
6143 | if (err) | |
6144 | return err; | |
6145 | conf = mddev->private; | |
6146 | if (!conf) | |
6147 | err = -ENODEV; | |
6148 | else if (new != conf->worker_cnt_per_group) { | |
6149 | mddev_suspend(mddev); | |
b721420e | 6150 | |
6791875e N |
6151 | old_groups = conf->worker_groups; |
6152 | if (old_groups) | |
6153 | flush_workqueue(raid5_wq); | |
d206dcfa | 6154 | |
6791875e N |
6155 | err = alloc_thread_groups(conf, new, |
6156 | &group_cnt, &worker_cnt_per_group, | |
6157 | &new_groups); | |
6158 | if (!err) { | |
6159 | spin_lock_irq(&conf->device_lock); | |
6160 | conf->group_cnt = group_cnt; | |
6161 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6162 | conf->worker_groups = new_groups; | |
6163 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6164 | |
6791875e N |
6165 | if (old_groups) |
6166 | kfree(old_groups[0].workers); | |
6167 | kfree(old_groups); | |
6168 | } | |
6169 | mddev_resume(mddev); | |
b721420e | 6170 | } |
6791875e | 6171 | mddev_unlock(mddev); |
b721420e | 6172 | |
6791875e | 6173 | return err ?: len; |
b721420e SL |
6174 | } |
6175 | ||
6176 | static struct md_sysfs_entry | |
6177 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6178 | raid5_show_group_thread_cnt, | |
6179 | raid5_store_group_thread_cnt); | |
6180 | ||
007583c9 | 6181 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6182 | &raid5_stripecache_size.attr, |
6183 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6184 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6185 | &raid5_group_thread_cnt.attr, |
d592a996 | 6186 | &raid5_skip_copy.attr, |
d06f191f | 6187 | &raid5_rmw_level.attr, |
3f294f4f N |
6188 | NULL, |
6189 | }; | |
007583c9 N |
6190 | static struct attribute_group raid5_attrs_group = { |
6191 | .name = NULL, | |
6192 | .attrs = raid5_attrs, | |
3f294f4f N |
6193 | }; |
6194 | ||
60aaf933 | 6195 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6196 | int *group_cnt, | |
6197 | int *worker_cnt_per_group, | |
6198 | struct r5worker_group **worker_groups) | |
851c30c9 | 6199 | { |
566c09c5 | 6200 | int i, j, k; |
851c30c9 SL |
6201 | ssize_t size; |
6202 | struct r5worker *workers; | |
6203 | ||
60aaf933 | 6204 | *worker_cnt_per_group = cnt; |
851c30c9 | 6205 | if (cnt == 0) { |
60aaf933 | 6206 | *group_cnt = 0; |
6207 | *worker_groups = NULL; | |
851c30c9 SL |
6208 | return 0; |
6209 | } | |
60aaf933 | 6210 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6211 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6212 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6213 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6214 | *group_cnt, GFP_NOIO); | |
6215 | if (!*worker_groups || !workers) { | |
851c30c9 | 6216 | kfree(workers); |
60aaf933 | 6217 | kfree(*worker_groups); |
851c30c9 SL |
6218 | return -ENOMEM; |
6219 | } | |
6220 | ||
60aaf933 | 6221 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6222 | struct r5worker_group *group; |
6223 | ||
0c775d52 | 6224 | group = &(*worker_groups)[i]; |
851c30c9 SL |
6225 | INIT_LIST_HEAD(&group->handle_list); |
6226 | group->conf = conf; | |
6227 | group->workers = workers + i * cnt; | |
6228 | ||
6229 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6230 | struct r5worker *worker = group->workers + j; |
6231 | worker->group = group; | |
6232 | INIT_WORK(&worker->work, raid5_do_work); | |
6233 | ||
6234 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6235 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6236 | } |
6237 | } | |
6238 | ||
6239 | return 0; | |
6240 | } | |
6241 | ||
6242 | static void free_thread_groups(struct r5conf *conf) | |
6243 | { | |
6244 | if (conf->worker_groups) | |
6245 | kfree(conf->worker_groups[0].workers); | |
6246 | kfree(conf->worker_groups); | |
6247 | conf->worker_groups = NULL; | |
6248 | } | |
6249 | ||
80c3a6ce | 6250 | static sector_t |
fd01b88c | 6251 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6252 | { |
d1688a6d | 6253 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6254 | |
6255 | if (!sectors) | |
6256 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6257 | if (!raid_disks) |
7ec05478 | 6258 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6259 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6260 | |
3cb5edf4 N |
6261 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6262 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6263 | return sectors * (raid_disks - conf->max_degraded); |
6264 | } | |
6265 | ||
789b5e03 ON |
6266 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6267 | { | |
6268 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6269 | if (percpu->scribble) |
6270 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6271 | percpu->spare_page = NULL; |
6272 | percpu->scribble = NULL; | |
6273 | } | |
6274 | ||
6275 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6276 | { | |
6277 | if (conf->level == 6 && !percpu->spare_page) | |
6278 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6279 | if (!percpu->scribble) | |
46d5b785 | 6280 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6281 | conf->previous_raid_disks), |
6282 | max(conf->chunk_sectors, | |
6283 | conf->prev_chunk_sectors) | |
6284 | / STRIPE_SECTORS, | |
6285 | GFP_KERNEL); | |
789b5e03 ON |
6286 | |
6287 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6288 | free_scratch_buffer(conf, percpu); | |
6289 | return -ENOMEM; | |
6290 | } | |
6291 | ||
6292 | return 0; | |
6293 | } | |
6294 | ||
d1688a6d | 6295 | static void raid5_free_percpu(struct r5conf *conf) |
36d1c647 | 6296 | { |
36d1c647 DW |
6297 | unsigned long cpu; |
6298 | ||
6299 | if (!conf->percpu) | |
6300 | return; | |
6301 | ||
36d1c647 DW |
6302 | #ifdef CONFIG_HOTPLUG_CPU |
6303 | unregister_cpu_notifier(&conf->cpu_notify); | |
6304 | #endif | |
789b5e03 ON |
6305 | |
6306 | get_online_cpus(); | |
6307 | for_each_possible_cpu(cpu) | |
6308 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
36d1c647 DW |
6309 | put_online_cpus(); |
6310 | ||
6311 | free_percpu(conf->percpu); | |
6312 | } | |
6313 | ||
d1688a6d | 6314 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6315 | { |
edbe83ab N |
6316 | if (conf->shrinker.seeks) |
6317 | unregister_shrinker(&conf->shrinker); | |
851c30c9 | 6318 | free_thread_groups(conf); |
95fc17aa | 6319 | shrink_stripes(conf); |
36d1c647 | 6320 | raid5_free_percpu(conf); |
95fc17aa DW |
6321 | kfree(conf->disks); |
6322 | kfree(conf->stripe_hashtbl); | |
6323 | kfree(conf); | |
6324 | } | |
6325 | ||
36d1c647 DW |
6326 | #ifdef CONFIG_HOTPLUG_CPU |
6327 | static int raid456_cpu_notify(struct notifier_block *nfb, unsigned long action, | |
6328 | void *hcpu) | |
6329 | { | |
d1688a6d | 6330 | struct r5conf *conf = container_of(nfb, struct r5conf, cpu_notify); |
36d1c647 DW |
6331 | long cpu = (long)hcpu; |
6332 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); | |
6333 | ||
6334 | switch (action) { | |
6335 | case CPU_UP_PREPARE: | |
6336 | case CPU_UP_PREPARE_FROZEN: | |
789b5e03 | 6337 | if (alloc_scratch_buffer(conf, percpu)) { |
36d1c647 DW |
6338 | pr_err("%s: failed memory allocation for cpu%ld\n", |
6339 | __func__, cpu); | |
55af6bb5 | 6340 | return notifier_from_errno(-ENOMEM); |
36d1c647 DW |
6341 | } |
6342 | break; | |
6343 | case CPU_DEAD: | |
6344 | case CPU_DEAD_FROZEN: | |
789b5e03 | 6345 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
36d1c647 DW |
6346 | break; |
6347 | default: | |
6348 | break; | |
6349 | } | |
6350 | return NOTIFY_OK; | |
6351 | } | |
6352 | #endif | |
6353 | ||
d1688a6d | 6354 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 DW |
6355 | { |
6356 | unsigned long cpu; | |
789b5e03 | 6357 | int err = 0; |
36d1c647 | 6358 | |
789b5e03 ON |
6359 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6360 | if (!conf->percpu) | |
36d1c647 | 6361 | return -ENOMEM; |
789b5e03 ON |
6362 | |
6363 | #ifdef CONFIG_HOTPLUG_CPU | |
6364 | conf->cpu_notify.notifier_call = raid456_cpu_notify; | |
6365 | conf->cpu_notify.priority = 0; | |
6366 | err = register_cpu_notifier(&conf->cpu_notify); | |
6367 | if (err) | |
6368 | return err; | |
6369 | #endif | |
36d1c647 DW |
6370 | |
6371 | get_online_cpus(); | |
36d1c647 | 6372 | for_each_present_cpu(cpu) { |
789b5e03 ON |
6373 | err = alloc_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); |
6374 | if (err) { | |
6375 | pr_err("%s: failed memory allocation for cpu%ld\n", | |
6376 | __func__, cpu); | |
36d1c647 DW |
6377 | break; |
6378 | } | |
36d1c647 | 6379 | } |
36d1c647 DW |
6380 | put_online_cpus(); |
6381 | ||
6382 | return err; | |
6383 | } | |
6384 | ||
edbe83ab N |
6385 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6386 | struct shrink_control *sc) | |
6387 | { | |
6388 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6389 | unsigned long ret = SHRINK_STOP; |
6390 | ||
6391 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6392 | ret= 0; | |
49895bcc N |
6393 | while (ret < sc->nr_to_scan && |
6394 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6395 | if (drop_one_stripe(conf) == 0) { |
6396 | ret = SHRINK_STOP; | |
6397 | break; | |
6398 | } | |
6399 | ret++; | |
6400 | } | |
6401 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6402 | } |
6403 | return ret; | |
6404 | } | |
6405 | ||
6406 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6407 | struct shrink_control *sc) | |
6408 | { | |
6409 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6410 | ||
6411 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6412 | /* unlikely, but not impossible */ | |
6413 | return 0; | |
6414 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6415 | } | |
6416 | ||
d1688a6d | 6417 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6418 | { |
d1688a6d | 6419 | struct r5conf *conf; |
5e5e3e78 | 6420 | int raid_disk, memory, max_disks; |
3cb03002 | 6421 | struct md_rdev *rdev; |
1da177e4 | 6422 | struct disk_info *disk; |
0232605d | 6423 | char pers_name[6]; |
566c09c5 | 6424 | int i; |
60aaf933 | 6425 | int group_cnt, worker_cnt_per_group; |
6426 | struct r5worker_group *new_group; | |
1da177e4 | 6427 | |
91adb564 N |
6428 | if (mddev->new_level != 5 |
6429 | && mddev->new_level != 4 | |
6430 | && mddev->new_level != 6) { | |
0c55e022 | 6431 | printk(KERN_ERR "md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
91adb564 N |
6432 | mdname(mddev), mddev->new_level); |
6433 | return ERR_PTR(-EIO); | |
1da177e4 | 6434 | } |
91adb564 N |
6435 | if ((mddev->new_level == 5 |
6436 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6437 | (mddev->new_level == 6 | |
6438 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
0c55e022 | 6439 | printk(KERN_ERR "md/raid:%s: layout %d not supported\n", |
91adb564 N |
6440 | mdname(mddev), mddev->new_layout); |
6441 | return ERR_PTR(-EIO); | |
99c0fb5f | 6442 | } |
91adb564 | 6443 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
0c55e022 | 6444 | printk(KERN_ERR "md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
91adb564 N |
6445 | mdname(mddev), mddev->raid_disks); |
6446 | return ERR_PTR(-EINVAL); | |
4bbf3771 N |
6447 | } |
6448 | ||
664e7c41 AN |
6449 | if (!mddev->new_chunk_sectors || |
6450 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6451 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
0c55e022 N |
6452 | printk(KERN_ERR "md/raid:%s: invalid chunk size %d\n", |
6453 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6454 | return ERR_PTR(-EINVAL); |
f6705578 N |
6455 | } |
6456 | ||
d1688a6d | 6457 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6458 | if (conf == NULL) |
1da177e4 | 6459 | goto abort; |
851c30c9 | 6460 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6461 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6462 | &new_group)) { | |
6463 | conf->group_cnt = group_cnt; | |
6464 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6465 | conf->worker_groups = new_group; | |
6466 | } else | |
851c30c9 | 6467 | goto abort; |
f5efd45a | 6468 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6469 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6470 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6471 | init_waitqueue_head(&conf->wait_for_quiescent); |
e9e4c377 YL |
6472 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) { |
6473 | init_waitqueue_head(&conf->wait_for_stripe[i]); | |
6474 | } | |
f5efd45a DW |
6475 | init_waitqueue_head(&conf->wait_for_overlap); |
6476 | INIT_LIST_HEAD(&conf->handle_list); | |
6477 | INIT_LIST_HEAD(&conf->hold_list); | |
6478 | INIT_LIST_HEAD(&conf->delayed_list); | |
6479 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 6480 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6481 | atomic_set(&conf->active_stripes, 0); |
6482 | atomic_set(&conf->preread_active_stripes, 0); | |
6483 | atomic_set(&conf->active_aligned_reads, 0); | |
6484 | conf->bypass_threshold = BYPASS_THRESHOLD; | |
d890fa2b | 6485 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6486 | |
6487 | conf->raid_disks = mddev->raid_disks; | |
6488 | if (mddev->reshape_position == MaxSector) | |
6489 | conf->previous_raid_disks = mddev->raid_disks; | |
6490 | else | |
f6705578 | 6491 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6492 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6493 | |
5e5e3e78 | 6494 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc N |
6495 | GFP_KERNEL); |
6496 | if (!conf->disks) | |
6497 | goto abort; | |
9ffae0cf | 6498 | |
1da177e4 LT |
6499 | conf->mddev = mddev; |
6500 | ||
fccddba0 | 6501 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6502 | goto abort; |
1da177e4 | 6503 | |
566c09c5 SL |
6504 | /* We init hash_locks[0] separately to that it can be used |
6505 | * as the reference lock in the spin_lock_nest_lock() call | |
6506 | * in lock_all_device_hash_locks_irq in order to convince | |
6507 | * lockdep that we know what we are doing. | |
6508 | */ | |
6509 | spin_lock_init(conf->hash_locks); | |
6510 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6511 | spin_lock_init(conf->hash_locks + i); | |
6512 | ||
6513 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6514 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6515 | ||
6516 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6517 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6518 | ||
36d1c647 | 6519 | conf->level = mddev->new_level; |
46d5b785 | 6520 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6521 | if (raid5_alloc_percpu(conf) != 0) |
6522 | goto abort; | |
6523 | ||
0c55e022 | 6524 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6525 | |
dafb20fa | 6526 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6527 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6528 | if (raid_disk >= max_disks |
1da177e4 LT |
6529 | || raid_disk < 0) |
6530 | continue; | |
6531 | disk = conf->disks + raid_disk; | |
6532 | ||
17045f52 N |
6533 | if (test_bit(Replacement, &rdev->flags)) { |
6534 | if (disk->replacement) | |
6535 | goto abort; | |
6536 | disk->replacement = rdev; | |
6537 | } else { | |
6538 | if (disk->rdev) | |
6539 | goto abort; | |
6540 | disk->rdev = rdev; | |
6541 | } | |
1da177e4 | 6542 | |
b2d444d7 | 6543 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6544 | char b[BDEVNAME_SIZE]; |
0c55e022 N |
6545 | printk(KERN_INFO "md/raid:%s: device %s operational as raid" |
6546 | " disk %d\n", | |
6547 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6548 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6549 | /* Cannot rely on bitmap to complete recovery */ |
6550 | conf->fullsync = 1; | |
1da177e4 LT |
6551 | } |
6552 | ||
91adb564 | 6553 | conf->level = mddev->new_level; |
584acdd4 | 6554 | if (conf->level == 6) { |
16a53ecc | 6555 | conf->max_degraded = 2; |
584acdd4 MS |
6556 | if (raid6_call.xor_syndrome) |
6557 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6558 | else | |
6559 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6560 | } else { | |
16a53ecc | 6561 | conf->max_degraded = 1; |
584acdd4 MS |
6562 | conf->rmw_level = PARITY_ENABLE_RMW; |
6563 | } | |
91adb564 | 6564 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6565 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6566 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6567 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 6568 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
6569 | } else { |
6570 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
6571 | conf->prev_algo = conf->algorithm; | |
e183eaed | 6572 | } |
1da177e4 | 6573 | |
edbe83ab N |
6574 | conf->min_nr_stripes = NR_STRIPES; |
6575 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + | |
5e5e3e78 | 6576 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6577 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6578 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
91adb564 | 6579 | printk(KERN_ERR |
0c55e022 N |
6580 | "md/raid:%s: couldn't allocate %dkB for buffers\n", |
6581 | mdname(mddev), memory); | |
91adb564 N |
6582 | goto abort; |
6583 | } else | |
0c55e022 N |
6584 | printk(KERN_INFO "md/raid:%s: allocated %dkB\n", |
6585 | mdname(mddev), memory); | |
edbe83ab N |
6586 | /* |
6587 | * Losing a stripe head costs more than the time to refill it, | |
6588 | * it reduces the queue depth and so can hurt throughput. | |
6589 | * So set it rather large, scaled by number of devices. | |
6590 | */ | |
6591 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6592 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6593 | conf->shrinker.count_objects = raid5_cache_count; | |
6594 | conf->shrinker.batch = 128; | |
6595 | conf->shrinker.flags = 0; | |
6596 | register_shrinker(&conf->shrinker); | |
1da177e4 | 6597 | |
0232605d N |
6598 | sprintf(pers_name, "raid%d", mddev->new_level); |
6599 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 N |
6600 | if (!conf->thread) { |
6601 | printk(KERN_ERR | |
0c55e022 | 6602 | "md/raid:%s: couldn't allocate thread.\n", |
91adb564 | 6603 | mdname(mddev)); |
16a53ecc N |
6604 | goto abort; |
6605 | } | |
91adb564 N |
6606 | |
6607 | return conf; | |
6608 | ||
6609 | abort: | |
6610 | if (conf) { | |
95fc17aa | 6611 | free_conf(conf); |
91adb564 N |
6612 | return ERR_PTR(-EIO); |
6613 | } else | |
6614 | return ERR_PTR(-ENOMEM); | |
6615 | } | |
6616 | ||
c148ffdc N |
6617 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6618 | { | |
6619 | switch (algo) { | |
6620 | case ALGORITHM_PARITY_0: | |
6621 | if (raid_disk < max_degraded) | |
6622 | return 1; | |
6623 | break; | |
6624 | case ALGORITHM_PARITY_N: | |
6625 | if (raid_disk >= raid_disks - max_degraded) | |
6626 | return 1; | |
6627 | break; | |
6628 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6629 | if (raid_disk == 0 || |
c148ffdc N |
6630 | raid_disk == raid_disks - 1) |
6631 | return 1; | |
6632 | break; | |
6633 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6634 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6635 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6636 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6637 | if (raid_disk == raid_disks - 1) | |
6638 | return 1; | |
6639 | } | |
6640 | return 0; | |
6641 | } | |
6642 | ||
fd01b88c | 6643 | static int run(struct mddev *mddev) |
91adb564 | 6644 | { |
d1688a6d | 6645 | struct r5conf *conf; |
9f7c2220 | 6646 | int working_disks = 0; |
c148ffdc | 6647 | int dirty_parity_disks = 0; |
3cb03002 | 6648 | struct md_rdev *rdev; |
c148ffdc | 6649 | sector_t reshape_offset = 0; |
17045f52 | 6650 | int i; |
b5254dd5 N |
6651 | long long min_offset_diff = 0; |
6652 | int first = 1; | |
91adb564 | 6653 | |
8c6ac868 | 6654 | if (mddev->recovery_cp != MaxSector) |
0c55e022 | 6655 | printk(KERN_NOTICE "md/raid:%s: not clean" |
8c6ac868 AN |
6656 | " -- starting background reconstruction\n", |
6657 | mdname(mddev)); | |
b5254dd5 N |
6658 | |
6659 | rdev_for_each(rdev, mddev) { | |
6660 | long long diff; | |
6661 | if (rdev->raid_disk < 0) | |
6662 | continue; | |
6663 | diff = (rdev->new_data_offset - rdev->data_offset); | |
6664 | if (first) { | |
6665 | min_offset_diff = diff; | |
6666 | first = 0; | |
6667 | } else if (mddev->reshape_backwards && | |
6668 | diff < min_offset_diff) | |
6669 | min_offset_diff = diff; | |
6670 | else if (!mddev->reshape_backwards && | |
6671 | diff > min_offset_diff) | |
6672 | min_offset_diff = diff; | |
6673 | } | |
6674 | ||
91adb564 N |
6675 | if (mddev->reshape_position != MaxSector) { |
6676 | /* Check that we can continue the reshape. | |
b5254dd5 N |
6677 | * Difficulties arise if the stripe we would write to |
6678 | * next is at or after the stripe we would read from next. | |
6679 | * For a reshape that changes the number of devices, this | |
6680 | * is only possible for a very short time, and mdadm makes | |
6681 | * sure that time appears to have past before assembling | |
6682 | * the array. So we fail if that time hasn't passed. | |
6683 | * For a reshape that keeps the number of devices the same | |
6684 | * mdadm must be monitoring the reshape can keeping the | |
6685 | * critical areas read-only and backed up. It will start | |
6686 | * the array in read-only mode, so we check for that. | |
91adb564 N |
6687 | */ |
6688 | sector_t here_new, here_old; | |
6689 | int old_disks; | |
18b00334 | 6690 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
91adb564 | 6691 | |
88ce4930 | 6692 | if (mddev->new_level != mddev->level) { |
0c55e022 | 6693 | printk(KERN_ERR "md/raid:%s: unsupported reshape " |
91adb564 N |
6694 | "required - aborting.\n", |
6695 | mdname(mddev)); | |
6696 | return -EINVAL; | |
6697 | } | |
91adb564 N |
6698 | old_disks = mddev->raid_disks - mddev->delta_disks; |
6699 | /* reshape_position must be on a new-stripe boundary, and one | |
6700 | * further up in new geometry must map after here in old | |
6701 | * geometry. | |
6702 | */ | |
6703 | here_new = mddev->reshape_position; | |
664e7c41 | 6704 | if (sector_div(here_new, mddev->new_chunk_sectors * |
91adb564 | 6705 | (mddev->raid_disks - max_degraded))) { |
0c55e022 N |
6706 | printk(KERN_ERR "md/raid:%s: reshape_position not " |
6707 | "on a stripe boundary\n", mdname(mddev)); | |
91adb564 N |
6708 | return -EINVAL; |
6709 | } | |
c148ffdc | 6710 | reshape_offset = here_new * mddev->new_chunk_sectors; |
91adb564 N |
6711 | /* here_new is the stripe we will write to */ |
6712 | here_old = mddev->reshape_position; | |
9d8f0363 | 6713 | sector_div(here_old, mddev->chunk_sectors * |
91adb564 N |
6714 | (old_disks-max_degraded)); |
6715 | /* here_old is the first stripe that we might need to read | |
6716 | * from */ | |
67ac6011 | 6717 | if (mddev->delta_disks == 0) { |
b5254dd5 N |
6718 | if ((here_new * mddev->new_chunk_sectors != |
6719 | here_old * mddev->chunk_sectors)) { | |
6720 | printk(KERN_ERR "md/raid:%s: reshape position is" | |
6721 | " confused - aborting\n", mdname(mddev)); | |
6722 | return -EINVAL; | |
6723 | } | |
67ac6011 | 6724 | /* We cannot be sure it is safe to start an in-place |
b5254dd5 | 6725 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
6726 | * and taking constant backups. |
6727 | * mdadm always starts a situation like this in | |
6728 | * readonly mode so it can take control before | |
6729 | * allowing any writes. So just check for that. | |
6730 | */ | |
b5254dd5 N |
6731 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
6732 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
6733 | /* not really in-place - so OK */; | |
6734 | else if (mddev->ro == 0) { | |
6735 | printk(KERN_ERR "md/raid:%s: in-place reshape " | |
6736 | "must be started in read-only mode " | |
6737 | "- aborting\n", | |
0c55e022 | 6738 | mdname(mddev)); |
67ac6011 N |
6739 | return -EINVAL; |
6740 | } | |
2c810cdd | 6741 | } else if (mddev->reshape_backwards |
b5254dd5 | 6742 | ? (here_new * mddev->new_chunk_sectors + min_offset_diff <= |
67ac6011 N |
6743 | here_old * mddev->chunk_sectors) |
6744 | : (here_new * mddev->new_chunk_sectors >= | |
b5254dd5 | 6745 | here_old * mddev->chunk_sectors + (-min_offset_diff))) { |
91adb564 | 6746 | /* Reading from the same stripe as writing to - bad */ |
0c55e022 N |
6747 | printk(KERN_ERR "md/raid:%s: reshape_position too early for " |
6748 | "auto-recovery - aborting.\n", | |
6749 | mdname(mddev)); | |
91adb564 N |
6750 | return -EINVAL; |
6751 | } | |
0c55e022 N |
6752 | printk(KERN_INFO "md/raid:%s: reshape will continue\n", |
6753 | mdname(mddev)); | |
91adb564 N |
6754 | /* OK, we should be able to continue; */ |
6755 | } else { | |
6756 | BUG_ON(mddev->level != mddev->new_level); | |
6757 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 6758 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 6759 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 6760 | } |
91adb564 | 6761 | |
245f46c2 N |
6762 | if (mddev->private == NULL) |
6763 | conf = setup_conf(mddev); | |
6764 | else | |
6765 | conf = mddev->private; | |
6766 | ||
91adb564 N |
6767 | if (IS_ERR(conf)) |
6768 | return PTR_ERR(conf); | |
6769 | ||
b5254dd5 | 6770 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
6771 | mddev->thread = conf->thread; |
6772 | conf->thread = NULL; | |
6773 | mddev->private = conf; | |
6774 | ||
17045f52 N |
6775 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
6776 | i++) { | |
6777 | rdev = conf->disks[i].rdev; | |
6778 | if (!rdev && conf->disks[i].replacement) { | |
6779 | /* The replacement is all we have yet */ | |
6780 | rdev = conf->disks[i].replacement; | |
6781 | conf->disks[i].replacement = NULL; | |
6782 | clear_bit(Replacement, &rdev->flags); | |
6783 | conf->disks[i].rdev = rdev; | |
6784 | } | |
6785 | if (!rdev) | |
c148ffdc | 6786 | continue; |
17045f52 N |
6787 | if (conf->disks[i].replacement && |
6788 | conf->reshape_progress != MaxSector) { | |
6789 | /* replacements and reshape simply do not mix. */ | |
6790 | printk(KERN_ERR "md: cannot handle concurrent " | |
6791 | "replacement and reshape.\n"); | |
6792 | goto abort; | |
6793 | } | |
2f115882 | 6794 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 6795 | working_disks++; |
2f115882 N |
6796 | continue; |
6797 | } | |
c148ffdc N |
6798 | /* This disc is not fully in-sync. However if it |
6799 | * just stored parity (beyond the recovery_offset), | |
6800 | * when we don't need to be concerned about the | |
6801 | * array being dirty. | |
6802 | * When reshape goes 'backwards', we never have | |
6803 | * partially completed devices, so we only need | |
6804 | * to worry about reshape going forwards. | |
6805 | */ | |
6806 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
6807 | if (mddev->major_version == 0 && | |
6808 | mddev->minor_version > 90) | |
6809 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 6810 | |
c148ffdc N |
6811 | if (rdev->recovery_offset < reshape_offset) { |
6812 | /* We need to check old and new layout */ | |
6813 | if (!only_parity(rdev->raid_disk, | |
6814 | conf->algorithm, | |
6815 | conf->raid_disks, | |
6816 | conf->max_degraded)) | |
6817 | continue; | |
6818 | } | |
6819 | if (!only_parity(rdev->raid_disk, | |
6820 | conf->prev_algo, | |
6821 | conf->previous_raid_disks, | |
6822 | conf->max_degraded)) | |
6823 | continue; | |
6824 | dirty_parity_disks++; | |
6825 | } | |
91adb564 | 6826 | |
17045f52 N |
6827 | /* |
6828 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
6829 | */ | |
908f4fbd | 6830 | mddev->degraded = calc_degraded(conf); |
91adb564 | 6831 | |
674806d6 | 6832 | if (has_failed(conf)) { |
0c55e022 | 6833 | printk(KERN_ERR "md/raid:%s: not enough operational devices" |
1da177e4 | 6834 | " (%d/%d failed)\n", |
02c2de8c | 6835 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
6836 | goto abort; |
6837 | } | |
6838 | ||
91adb564 | 6839 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 6840 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
6841 | mddev->resync_max_sectors = mddev->dev_sectors; |
6842 | ||
c148ffdc | 6843 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 6844 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec N |
6845 | if (mddev->ok_start_degraded) |
6846 | printk(KERN_WARNING | |
0c55e022 N |
6847 | "md/raid:%s: starting dirty degraded array" |
6848 | " - data corruption possible.\n", | |
6ff8d8ec N |
6849 | mdname(mddev)); |
6850 | else { | |
6851 | printk(KERN_ERR | |
0c55e022 | 6852 | "md/raid:%s: cannot start dirty degraded array.\n", |
6ff8d8ec N |
6853 | mdname(mddev)); |
6854 | goto abort; | |
6855 | } | |
1da177e4 LT |
6856 | } |
6857 | ||
1da177e4 | 6858 | if (mddev->degraded == 0) |
0c55e022 N |
6859 | printk(KERN_INFO "md/raid:%s: raid level %d active with %d out of %d" |
6860 | " devices, algorithm %d\n", mdname(mddev), conf->level, | |
e183eaed N |
6861 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, |
6862 | mddev->new_layout); | |
1da177e4 | 6863 | else |
0c55e022 N |
6864 | printk(KERN_ALERT "md/raid:%s: raid level %d active with %d" |
6865 | " out of %d devices, algorithm %d\n", | |
6866 | mdname(mddev), conf->level, | |
6867 | mddev->raid_disks - mddev->degraded, | |
6868 | mddev->raid_disks, mddev->new_layout); | |
1da177e4 LT |
6869 | |
6870 | print_raid5_conf(conf); | |
6871 | ||
fef9c61f | 6872 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 6873 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
6874 | atomic_set(&conf->reshape_stripes, 0); |
6875 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
6876 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
6877 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
6878 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
6879 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 6880 | "reshape"); |
f6705578 N |
6881 | } |
6882 | ||
1da177e4 | 6883 | /* Ok, everything is just fine now */ |
a64c876f N |
6884 | if (mddev->to_remove == &raid5_attrs_group) |
6885 | mddev->to_remove = NULL; | |
00bcb4ac N |
6886 | else if (mddev->kobj.sd && |
6887 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
5e55e2f5 | 6888 | printk(KERN_WARNING |
4a5add49 | 6889 | "raid5: failed to create sysfs attributes for %s\n", |
5e55e2f5 | 6890 | mdname(mddev)); |
4a5add49 | 6891 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 6892 | |
4a5add49 | 6893 | if (mddev->queue) { |
9f7c2220 | 6894 | int chunk_size; |
620125f2 | 6895 | bool discard_supported = true; |
4a5add49 N |
6896 | /* read-ahead size must cover two whole stripes, which |
6897 | * is 2 * (datadisks) * chunksize where 'n' is the | |
6898 | * number of raid devices | |
6899 | */ | |
6900 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
6901 | int stripe = data_disks * | |
6902 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
6903 | if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) | |
6904 | mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
91adb564 | 6905 | |
9f7c2220 N |
6906 | chunk_size = mddev->chunk_sectors << 9; |
6907 | blk_queue_io_min(mddev->queue, chunk_size); | |
6908 | blk_queue_io_opt(mddev->queue, chunk_size * | |
6909 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 6910 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
6911 | /* |
6912 | * We can only discard a whole stripe. It doesn't make sense to | |
6913 | * discard data disk but write parity disk | |
6914 | */ | |
6915 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
6916 | /* Round up to power of 2, as discard handling |
6917 | * currently assumes that */ | |
6918 | while ((stripe-1) & stripe) | |
6919 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
6920 | mddev->queue->limits.discard_alignment = stripe; |
6921 | mddev->queue->limits.discard_granularity = stripe; | |
6922 | /* | |
6923 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 6924 | * guarantee discard_zeroes_data |
620125f2 SL |
6925 | */ |
6926 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 6927 | |
5026d7a9 PA |
6928 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
6929 | ||
05616be5 | 6930 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
6931 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6932 | rdev->data_offset << 9); | |
05616be5 N |
6933 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
6934 | rdev->new_data_offset << 9); | |
620125f2 SL |
6935 | /* |
6936 | * discard_zeroes_data is required, otherwise data | |
6937 | * could be lost. Consider a scenario: discard a stripe | |
6938 | * (the stripe could be inconsistent if | |
6939 | * discard_zeroes_data is 0); write one disk of the | |
6940 | * stripe (the stripe could be inconsistent again | |
6941 | * depending on which disks are used to calculate | |
6942 | * parity); the disk is broken; The stripe data of this | |
6943 | * disk is lost. | |
6944 | */ | |
6945 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
6946 | !bdev_get_queue(rdev->bdev)-> | |
6947 | limits.discard_zeroes_data) | |
6948 | discard_supported = false; | |
8e0e99ba N |
6949 | /* Unfortunately, discard_zeroes_data is not currently |
6950 | * a guarantee - just a hint. So we only allow DISCARD | |
6951 | * if the sysadmin has confirmed that only safe devices | |
6952 | * are in use by setting a module parameter. | |
6953 | */ | |
6954 | if (!devices_handle_discard_safely) { | |
6955 | if (discard_supported) { | |
6956 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
6957 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
6958 | } | |
6959 | discard_supported = false; | |
6960 | } | |
05616be5 | 6961 | } |
620125f2 SL |
6962 | |
6963 | if (discard_supported && | |
6964 | mddev->queue->limits.max_discard_sectors >= stripe && | |
6965 | mddev->queue->limits.discard_granularity >= stripe) | |
6966 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
6967 | mddev->queue); | |
6968 | else | |
6969 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
6970 | mddev->queue); | |
9f7c2220 | 6971 | } |
23032a0e | 6972 | |
1da177e4 LT |
6973 | return 0; |
6974 | abort: | |
01f96c0a | 6975 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
6976 | print_raid5_conf(conf); |
6977 | free_conf(conf); | |
1da177e4 | 6978 | mddev->private = NULL; |
0c55e022 | 6979 | printk(KERN_ALERT "md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
6980 | return -EIO; |
6981 | } | |
6982 | ||
afa0f557 | 6983 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 6984 | { |
afa0f557 | 6985 | struct r5conf *conf = priv; |
1da177e4 | 6986 | |
95fc17aa | 6987 | free_conf(conf); |
a64c876f | 6988 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
6989 | } |
6990 | ||
fd01b88c | 6991 | static void status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 6992 | { |
d1688a6d | 6993 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
6994 | int i; |
6995 | ||
9d8f0363 | 6996 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 6997 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 6998 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
1da177e4 LT |
6999 | for (i = 0; i < conf->raid_disks; i++) |
7000 | seq_printf (seq, "%s", | |
7001 | conf->disks[i].rdev && | |
b2d444d7 | 7002 | test_bit(In_sync, &conf->disks[i].rdev->flags) ? "U" : "_"); |
1da177e4 | 7003 | seq_printf (seq, "]"); |
1da177e4 LT |
7004 | } |
7005 | ||
d1688a6d | 7006 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7007 | { |
7008 | int i; | |
7009 | struct disk_info *tmp; | |
7010 | ||
0c55e022 | 7011 | printk(KERN_DEBUG "RAID conf printout:\n"); |
1da177e4 LT |
7012 | if (!conf) { |
7013 | printk("(conf==NULL)\n"); | |
7014 | return; | |
7015 | } | |
0c55e022 N |
7016 | printk(KERN_DEBUG " --- level:%d rd:%d wd:%d\n", conf->level, |
7017 | conf->raid_disks, | |
7018 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7019 | |
7020 | for (i = 0; i < conf->raid_disks; i++) { | |
7021 | char b[BDEVNAME_SIZE]; | |
7022 | tmp = conf->disks + i; | |
7023 | if (tmp->rdev) | |
0c55e022 N |
7024 | printk(KERN_DEBUG " disk %d, o:%d, dev:%s\n", |
7025 | i, !test_bit(Faulty, &tmp->rdev->flags), | |
7026 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7027 | } |
7028 | } | |
7029 | ||
fd01b88c | 7030 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7031 | { |
7032 | int i; | |
d1688a6d | 7033 | struct r5conf *conf = mddev->private; |
1da177e4 | 7034 | struct disk_info *tmp; |
6b965620 N |
7035 | int count = 0; |
7036 | unsigned long flags; | |
1da177e4 LT |
7037 | |
7038 | for (i = 0; i < conf->raid_disks; i++) { | |
7039 | tmp = conf->disks + i; | |
dd054fce N |
7040 | if (tmp->replacement |
7041 | && tmp->replacement->recovery_offset == MaxSector | |
7042 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7043 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7044 | /* Replacement has just become active. */ | |
7045 | if (!tmp->rdev | |
7046 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7047 | count++; | |
7048 | if (tmp->rdev) { | |
7049 | /* Replaced device not technically faulty, | |
7050 | * but we need to be sure it gets removed | |
7051 | * and never re-added. | |
7052 | */ | |
7053 | set_bit(Faulty, &tmp->rdev->flags); | |
7054 | sysfs_notify_dirent_safe( | |
7055 | tmp->rdev->sysfs_state); | |
7056 | } | |
7057 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7058 | } else if (tmp->rdev | |
70fffd0b | 7059 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7060 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7061 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7062 | count++; |
43c73ca4 | 7063 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7064 | } |
7065 | } | |
6b965620 | 7066 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7067 | mddev->degraded = calc_degraded(conf); |
6b965620 | 7068 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7069 | print_raid5_conf(conf); |
6b965620 | 7070 | return count; |
1da177e4 LT |
7071 | } |
7072 | ||
b8321b68 | 7073 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7074 | { |
d1688a6d | 7075 | struct r5conf *conf = mddev->private; |
1da177e4 | 7076 | int err = 0; |
b8321b68 | 7077 | int number = rdev->raid_disk; |
657e3e4d | 7078 | struct md_rdev **rdevp; |
1da177e4 LT |
7079 | struct disk_info *p = conf->disks + number; |
7080 | ||
7081 | print_raid5_conf(conf); | |
657e3e4d N |
7082 | if (rdev == p->rdev) |
7083 | rdevp = &p->rdev; | |
7084 | else if (rdev == p->replacement) | |
7085 | rdevp = &p->replacement; | |
7086 | else | |
7087 | return 0; | |
7088 | ||
7089 | if (number >= conf->raid_disks && | |
7090 | conf->reshape_progress == MaxSector) | |
7091 | clear_bit(In_sync, &rdev->flags); | |
7092 | ||
7093 | if (test_bit(In_sync, &rdev->flags) || | |
7094 | atomic_read(&rdev->nr_pending)) { | |
7095 | err = -EBUSY; | |
7096 | goto abort; | |
7097 | } | |
7098 | /* Only remove non-faulty devices if recovery | |
7099 | * isn't possible. | |
7100 | */ | |
7101 | if (!test_bit(Faulty, &rdev->flags) && | |
7102 | mddev->recovery_disabled != conf->recovery_disabled && | |
7103 | !has_failed(conf) && | |
dd054fce | 7104 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7105 | number < conf->raid_disks) { |
7106 | err = -EBUSY; | |
7107 | goto abort; | |
7108 | } | |
7109 | *rdevp = NULL; | |
7110 | synchronize_rcu(); | |
7111 | if (atomic_read(&rdev->nr_pending)) { | |
7112 | /* lost the race, try later */ | |
7113 | err = -EBUSY; | |
7114 | *rdevp = rdev; | |
dd054fce N |
7115 | } else if (p->replacement) { |
7116 | /* We must have just cleared 'rdev' */ | |
7117 | p->rdev = p->replacement; | |
7118 | clear_bit(Replacement, &p->replacement->flags); | |
7119 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7120 | * but will never see neither - if they are careful | |
7121 | */ | |
7122 | p->replacement = NULL; | |
7123 | clear_bit(WantReplacement, &rdev->flags); | |
7124 | } else | |
7125 | /* We might have just removed the Replacement as faulty- | |
7126 | * clear the bit just in case | |
7127 | */ | |
7128 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7129 | abort: |
7130 | ||
7131 | print_raid5_conf(conf); | |
7132 | return err; | |
7133 | } | |
7134 | ||
fd01b88c | 7135 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7136 | { |
d1688a6d | 7137 | struct r5conf *conf = mddev->private; |
199050ea | 7138 | int err = -EEXIST; |
1da177e4 LT |
7139 | int disk; |
7140 | struct disk_info *p; | |
6c2fce2e NB |
7141 | int first = 0; |
7142 | int last = conf->raid_disks - 1; | |
1da177e4 | 7143 | |
7f0da59b N |
7144 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7145 | return -EBUSY; | |
7146 | ||
dc10c643 | 7147 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7148 | /* no point adding a device */ |
199050ea | 7149 | return -EINVAL; |
1da177e4 | 7150 | |
6c2fce2e NB |
7151 | if (rdev->raid_disk >= 0) |
7152 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7153 | |
7154 | /* | |
16a53ecc N |
7155 | * find the disk ... but prefer rdev->saved_raid_disk |
7156 | * if possible. | |
1da177e4 | 7157 | */ |
16a53ecc | 7158 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7159 | rdev->saved_raid_disk >= first && |
16a53ecc | 7160 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7161 | first = rdev->saved_raid_disk; |
7162 | ||
7163 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7164 | p = conf->disks + disk; |
7165 | if (p->rdev == NULL) { | |
b2d444d7 | 7166 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7167 | rdev->raid_disk = disk; |
199050ea | 7168 | err = 0; |
72626685 N |
7169 | if (rdev->saved_raid_disk != disk) |
7170 | conf->fullsync = 1; | |
d6065f7b | 7171 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7172 | goto out; |
1da177e4 | 7173 | } |
5cfb22a1 N |
7174 | } |
7175 | for (disk = first; disk <= last; disk++) { | |
7176 | p = conf->disks + disk; | |
7bfec5f3 N |
7177 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7178 | p->replacement == NULL) { | |
7179 | clear_bit(In_sync, &rdev->flags); | |
7180 | set_bit(Replacement, &rdev->flags); | |
7181 | rdev->raid_disk = disk; | |
7182 | err = 0; | |
7183 | conf->fullsync = 1; | |
7184 | rcu_assign_pointer(p->replacement, rdev); | |
7185 | break; | |
7186 | } | |
7187 | } | |
5cfb22a1 | 7188 | out: |
1da177e4 | 7189 | print_raid5_conf(conf); |
199050ea | 7190 | return err; |
1da177e4 LT |
7191 | } |
7192 | ||
fd01b88c | 7193 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7194 | { |
7195 | /* no resync is happening, and there is enough space | |
7196 | * on all devices, so we can resize. | |
7197 | * We need to make sure resync covers any new space. | |
7198 | * If the array is shrinking we should possibly wait until | |
7199 | * any io in the removed space completes, but it hardly seems | |
7200 | * worth it. | |
7201 | */ | |
a4a6125a | 7202 | sector_t newsize; |
3cb5edf4 N |
7203 | struct r5conf *conf = mddev->private; |
7204 | ||
7205 | sectors &= ~((sector_t)conf->chunk_sectors - 1); | |
a4a6125a N |
7206 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7207 | if (mddev->external_size && | |
7208 | mddev->array_sectors > newsize) | |
b522adcd | 7209 | return -EINVAL; |
a4a6125a N |
7210 | if (mddev->bitmap) { |
7211 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7212 | if (ret) | |
7213 | return ret; | |
7214 | } | |
7215 | md_set_array_sectors(mddev, newsize); | |
f233ea5c | 7216 | set_capacity(mddev->gendisk, mddev->array_sectors); |
449aad3e | 7217 | revalidate_disk(mddev->gendisk); |
b098636c N |
7218 | if (sectors > mddev->dev_sectors && |
7219 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7220 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7221 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7222 | } | |
58c0fed4 | 7223 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7224 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7225 | return 0; |
7226 | } | |
7227 | ||
fd01b88c | 7228 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7229 | { |
7230 | /* Can only proceed if there are plenty of stripe_heads. | |
7231 | * We need a minimum of one full stripe,, and for sensible progress | |
7232 | * it is best to have about 4 times that. | |
7233 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7234 | * allow for chunk sizes up to 256K, which is probably OK. | |
7235 | * If the chunk size is greater, user-space should request more | |
7236 | * stripe_heads first. | |
7237 | */ | |
d1688a6d | 7238 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7239 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7240 | > conf->min_nr_stripes || |
01ee22b4 | 7241 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7242 | > conf->min_nr_stripes) { |
0c55e022 N |
7243 | printk(KERN_WARNING "md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7244 | mdname(mddev), | |
01ee22b4 N |
7245 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) |
7246 | / STRIPE_SIZE)*4); | |
7247 | return 0; | |
7248 | } | |
7249 | return 1; | |
7250 | } | |
7251 | ||
fd01b88c | 7252 | static int check_reshape(struct mddev *mddev) |
29269553 | 7253 | { |
d1688a6d | 7254 | struct r5conf *conf = mddev->private; |
29269553 | 7255 | |
88ce4930 N |
7256 | if (mddev->delta_disks == 0 && |
7257 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7258 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7259 | return 0; /* nothing to do */ |
674806d6 | 7260 | if (has_failed(conf)) |
ec32a2bd | 7261 | return -EINVAL; |
fdcfbbb6 | 7262 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7263 | /* We might be able to shrink, but the devices must |
7264 | * be made bigger first. | |
7265 | * For raid6, 4 is the minimum size. | |
7266 | * Otherwise 2 is the minimum | |
7267 | */ | |
7268 | int min = 2; | |
7269 | if (mddev->level == 6) | |
7270 | min = 4; | |
7271 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7272 | return -EINVAL; | |
7273 | } | |
29269553 | 7274 | |
01ee22b4 | 7275 | if (!check_stripe_cache(mddev)) |
29269553 | 7276 | return -ENOSPC; |
29269553 | 7277 | |
738a2738 N |
7278 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7279 | mddev->delta_disks > 0) | |
7280 | if (resize_chunks(conf, | |
7281 | conf->previous_raid_disks | |
7282 | + max(0, mddev->delta_disks), | |
7283 | max(mddev->new_chunk_sectors, | |
7284 | mddev->chunk_sectors) | |
7285 | ) < 0) | |
7286 | return -ENOMEM; | |
e56108d6 N |
7287 | return resize_stripes(conf, (conf->previous_raid_disks |
7288 | + mddev->delta_disks)); | |
63c70c4f N |
7289 | } |
7290 | ||
fd01b88c | 7291 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7292 | { |
d1688a6d | 7293 | struct r5conf *conf = mddev->private; |
3cb03002 | 7294 | struct md_rdev *rdev; |
63c70c4f | 7295 | int spares = 0; |
c04be0aa | 7296 | unsigned long flags; |
63c70c4f | 7297 | |
f416885e | 7298 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7299 | return -EBUSY; |
7300 | ||
01ee22b4 N |
7301 | if (!check_stripe_cache(mddev)) |
7302 | return -ENOSPC; | |
7303 | ||
30b67645 N |
7304 | if (has_failed(conf)) |
7305 | return -EINVAL; | |
7306 | ||
c6563a8c | 7307 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7308 | if (!test_bit(In_sync, &rdev->flags) |
7309 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7310 | spares++; |
c6563a8c | 7311 | } |
63c70c4f | 7312 | |
f416885e | 7313 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7314 | /* Not enough devices even to make a degraded array |
7315 | * of that size | |
7316 | */ | |
7317 | return -EINVAL; | |
7318 | ||
ec32a2bd N |
7319 | /* Refuse to reduce size of the array. Any reductions in |
7320 | * array size must be through explicit setting of array_size | |
7321 | * attribute. | |
7322 | */ | |
7323 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7324 | < mddev->array_sectors) { | |
0c55e022 | 7325 | printk(KERN_ERR "md/raid:%s: array size must be reduced " |
ec32a2bd N |
7326 | "before number of disks\n", mdname(mddev)); |
7327 | return -EINVAL; | |
7328 | } | |
7329 | ||
f6705578 | 7330 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7331 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7332 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7333 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7334 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7335 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7336 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7337 | conf->prev_algo = conf->algorithm; |
7338 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7339 | conf->generation++; |
7340 | /* Code that selects data_offset needs to see the generation update | |
7341 | * if reshape_progress has been set - so a memory barrier needed. | |
7342 | */ | |
7343 | smp_mb(); | |
2c810cdd | 7344 | if (mddev->reshape_backwards) |
fef9c61f N |
7345 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7346 | else | |
7347 | conf->reshape_progress = 0; | |
7348 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7349 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7350 | spin_unlock_irq(&conf->device_lock); |
7351 | ||
4d77e3ba N |
7352 | /* Now make sure any requests that proceeded on the assumption |
7353 | * the reshape wasn't running - like Discard or Read - have | |
7354 | * completed. | |
7355 | */ | |
7356 | mddev_suspend(mddev); | |
7357 | mddev_resume(mddev); | |
7358 | ||
29269553 N |
7359 | /* Add some new drives, as many as will fit. |
7360 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7361 | * Don't add devices if we are reducing the number of |
7362 | * devices in the array. This is because it is not possible | |
7363 | * to correctly record the "partially reconstructed" state of | |
7364 | * such devices during the reshape and confusion could result. | |
29269553 | 7365 | */ |
87a8dec9 | 7366 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7367 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7368 | if (rdev->raid_disk < 0 && |
7369 | !test_bit(Faulty, &rdev->flags)) { | |
7370 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7371 | if (rdev->raid_disk |
9d4c7d87 | 7372 | >= conf->previous_raid_disks) |
87a8dec9 | 7373 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7374 | else |
87a8dec9 | 7375 | rdev->recovery_offset = 0; |
36fad858 NK |
7376 | |
7377 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7378 | /* Failure here is OK */; |
50da0840 | 7379 | } |
87a8dec9 N |
7380 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7381 | && !test_bit(Faulty, &rdev->flags)) { | |
7382 | /* This is a spare that was manually added */ | |
7383 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7384 | } |
29269553 | 7385 | |
87a8dec9 N |
7386 | /* When a reshape changes the number of devices, |
7387 | * ->degraded is measured against the larger of the | |
7388 | * pre and post number of devices. | |
7389 | */ | |
ec32a2bd | 7390 | spin_lock_irqsave(&conf->device_lock, flags); |
908f4fbd | 7391 | mddev->degraded = calc_degraded(conf); |
ec32a2bd N |
7392 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7393 | } | |
63c70c4f | 7394 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7395 | mddev->reshape_position = conf->reshape_progress; |
850b2b42 | 7396 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
f6705578 | 7397 | |
29269553 N |
7398 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7399 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7400 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7401 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7402 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7403 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7404 | "reshape"); |
29269553 N |
7405 | if (!mddev->sync_thread) { |
7406 | mddev->recovery = 0; | |
7407 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7408 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7409 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7410 | mddev->new_chunk_sectors = |
7411 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7412 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7413 | rdev_for_each(rdev, mddev) |
7414 | rdev->new_data_offset = rdev->data_offset; | |
7415 | smp_wmb(); | |
ba8805b9 | 7416 | conf->generation --; |
fef9c61f | 7417 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7418 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7419 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7420 | spin_unlock_irq(&conf->device_lock); |
7421 | return -EAGAIN; | |
7422 | } | |
c8f517c4 | 7423 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7424 | md_wakeup_thread(mddev->sync_thread); |
7425 | md_new_event(mddev); | |
7426 | return 0; | |
7427 | } | |
29269553 | 7428 | |
ec32a2bd N |
7429 | /* This is called from the reshape thread and should make any |
7430 | * changes needed in 'conf' | |
7431 | */ | |
d1688a6d | 7432 | static void end_reshape(struct r5conf *conf) |
29269553 | 7433 | { |
29269553 | 7434 | |
f6705578 | 7435 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7436 | struct md_rdev *rdev; |
f6705578 | 7437 | |
f6705578 | 7438 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7439 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7440 | rdev_for_each(rdev, conf->mddev) |
7441 | rdev->data_offset = rdev->new_data_offset; | |
7442 | smp_wmb(); | |
fef9c61f | 7443 | conf->reshape_progress = MaxSector; |
f6705578 | 7444 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7445 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7446 | |
7447 | /* read-ahead size must cover two whole stripes, which is | |
7448 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7449 | */ | |
4a5add49 | 7450 | if (conf->mddev->queue) { |
cea9c228 | 7451 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7452 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7453 | / PAGE_SIZE); |
16a53ecc N |
7454 | if (conf->mddev->queue->backing_dev_info.ra_pages < 2 * stripe) |
7455 | conf->mddev->queue->backing_dev_info.ra_pages = 2 * stripe; | |
7456 | } | |
29269553 | 7457 | } |
29269553 N |
7458 | } |
7459 | ||
ec32a2bd N |
7460 | /* This is called from the raid5d thread with mddev_lock held. |
7461 | * It makes config changes to the device. | |
7462 | */ | |
fd01b88c | 7463 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7464 | { |
d1688a6d | 7465 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7466 | |
7467 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7468 | ||
ec32a2bd N |
7469 | if (mddev->delta_disks > 0) { |
7470 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
7471 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
449aad3e | 7472 | revalidate_disk(mddev->gendisk); |
ec32a2bd N |
7473 | } else { |
7474 | int d; | |
908f4fbd N |
7475 | spin_lock_irq(&conf->device_lock); |
7476 | mddev->degraded = calc_degraded(conf); | |
7477 | spin_unlock_irq(&conf->device_lock); | |
ec32a2bd N |
7478 | for (d = conf->raid_disks ; |
7479 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7480 | d++) { |
3cb03002 | 7481 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7482 | if (rdev) |
7483 | clear_bit(In_sync, &rdev->flags); | |
7484 | rdev = conf->disks[d].replacement; | |
7485 | if (rdev) | |
7486 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7487 | } |
cea9c228 | 7488 | } |
88ce4930 | 7489 | mddev->layout = conf->algorithm; |
09c9e5fa | 7490 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7491 | mddev->reshape_position = MaxSector; |
7492 | mddev->delta_disks = 0; | |
2c810cdd | 7493 | mddev->reshape_backwards = 0; |
cea9c228 N |
7494 | } |
7495 | } | |
7496 | ||
fd01b88c | 7497 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7498 | { |
d1688a6d | 7499 | struct r5conf *conf = mddev->private; |
72626685 N |
7500 | |
7501 | switch(state) { | |
e464eafd N |
7502 | case 2: /* resume for a suspend */ |
7503 | wake_up(&conf->wait_for_overlap); | |
7504 | break; | |
7505 | ||
72626685 | 7506 | case 1: /* stop all writes */ |
566c09c5 | 7507 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7508 | /* '2' tells resync/reshape to pause so that all |
7509 | * active stripes can drain | |
7510 | */ | |
7511 | conf->quiesce = 2; | |
b1b46486 | 7512 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7513 | atomic_read(&conf->active_stripes) == 0 && |
7514 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7515 | unlock_all_device_hash_locks_irq(conf), |
7516 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7517 | conf->quiesce = 1; |
566c09c5 | 7518 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7519 | /* allow reshape to continue */ |
7520 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7521 | break; |
7522 | ||
7523 | case 0: /* re-enable writes */ | |
566c09c5 | 7524 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7525 | conf->quiesce = 0; |
b1b46486 | 7526 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7527 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7528 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7529 | break; |
7530 | } | |
72626685 | 7531 | } |
b15c2e57 | 7532 | |
fd01b88c | 7533 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7534 | { |
e373ab10 | 7535 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7536 | sector_t sectors; |
54071b38 | 7537 | |
f1b29bca | 7538 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7539 | if (raid0_conf->nr_strip_zones > 1) { |
0c55e022 N |
7540 | printk(KERN_ERR "md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7541 | mdname(mddev)); | |
f1b29bca DW |
7542 | return ERR_PTR(-EINVAL); |
7543 | } | |
7544 | ||
e373ab10 N |
7545 | sectors = raid0_conf->strip_zone[0].zone_end; |
7546 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7547 | mddev->dev_sectors = sectors; |
f1b29bca | 7548 | mddev->new_level = level; |
54071b38 TM |
7549 | mddev->new_layout = ALGORITHM_PARITY_N; |
7550 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7551 | mddev->raid_disks += 1; | |
7552 | mddev->delta_disks = 1; | |
7553 | /* make sure it will be not marked as dirty */ | |
7554 | mddev->recovery_cp = MaxSector; | |
7555 | ||
7556 | return setup_conf(mddev); | |
7557 | } | |
7558 | ||
fd01b88c | 7559 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7560 | { |
7561 | int chunksect; | |
7562 | ||
7563 | if (mddev->raid_disks != 2 || | |
7564 | mddev->degraded > 1) | |
7565 | return ERR_PTR(-EINVAL); | |
7566 | ||
7567 | /* Should check if there are write-behind devices? */ | |
7568 | ||
7569 | chunksect = 64*2; /* 64K by default */ | |
7570 | ||
7571 | /* The array must be an exact multiple of chunksize */ | |
7572 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
7573 | chunksect >>= 1; | |
7574 | ||
7575 | if ((chunksect<<9) < STRIPE_SIZE) | |
7576 | /* array size does not allow a suitable chunk size */ | |
7577 | return ERR_PTR(-EINVAL); | |
7578 | ||
7579 | mddev->new_level = 5; | |
7580 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 7581 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 N |
7582 | |
7583 | return setup_conf(mddev); | |
7584 | } | |
7585 | ||
fd01b88c | 7586 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
7587 | { |
7588 | int new_layout; | |
7589 | ||
7590 | switch (mddev->layout) { | |
7591 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7592 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7593 | break; | |
7594 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7595 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
7596 | break; | |
7597 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7598 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
7599 | break; | |
7600 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7601 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
7602 | break; | |
7603 | case ALGORITHM_PARITY_0_6: | |
7604 | new_layout = ALGORITHM_PARITY_0; | |
7605 | break; | |
7606 | case ALGORITHM_PARITY_N: | |
7607 | new_layout = ALGORITHM_PARITY_N; | |
7608 | break; | |
7609 | default: | |
7610 | return ERR_PTR(-EINVAL); | |
7611 | } | |
7612 | mddev->new_level = 5; | |
7613 | mddev->new_layout = new_layout; | |
7614 | mddev->delta_disks = -1; | |
7615 | mddev->raid_disks -= 1; | |
7616 | return setup_conf(mddev); | |
7617 | } | |
7618 | ||
fd01b88c | 7619 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 7620 | { |
88ce4930 N |
7621 | /* For a 2-drive array, the layout and chunk size can be changed |
7622 | * immediately as not restriping is needed. | |
7623 | * For larger arrays we record the new value - after validation | |
7624 | * to be used by a reshape pass. | |
b3546035 | 7625 | */ |
d1688a6d | 7626 | struct r5conf *conf = mddev->private; |
597a711b | 7627 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 7628 | |
597a711b | 7629 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
7630 | return -EINVAL; |
7631 | if (new_chunk > 0) { | |
0ba459d2 | 7632 | if (!is_power_of_2(new_chunk)) |
b3546035 | 7633 | return -EINVAL; |
597a711b | 7634 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 7635 | return -EINVAL; |
597a711b | 7636 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
7637 | /* not factor of array size */ |
7638 | return -EINVAL; | |
7639 | } | |
7640 | ||
7641 | /* They look valid */ | |
7642 | ||
88ce4930 | 7643 | if (mddev->raid_disks == 2) { |
597a711b N |
7644 | /* can make the change immediately */ |
7645 | if (mddev->new_layout >= 0) { | |
7646 | conf->algorithm = mddev->new_layout; | |
7647 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
7648 | } |
7649 | if (new_chunk > 0) { | |
597a711b N |
7650 | conf->chunk_sectors = new_chunk ; |
7651 | mddev->chunk_sectors = new_chunk; | |
88ce4930 N |
7652 | } |
7653 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
7654 | md_wakeup_thread(mddev->thread); | |
b3546035 | 7655 | } |
50ac168a | 7656 | return check_reshape(mddev); |
88ce4930 N |
7657 | } |
7658 | ||
fd01b88c | 7659 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 7660 | { |
597a711b | 7661 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 7662 | |
597a711b | 7663 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 7664 | return -EINVAL; |
b3546035 | 7665 | if (new_chunk > 0) { |
0ba459d2 | 7666 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 7667 | return -EINVAL; |
597a711b | 7668 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 7669 | return -EINVAL; |
597a711b | 7670 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
7671 | /* not factor of array size */ |
7672 | return -EINVAL; | |
b3546035 | 7673 | } |
88ce4930 N |
7674 | |
7675 | /* They look valid */ | |
50ac168a | 7676 | return check_reshape(mddev); |
b3546035 N |
7677 | } |
7678 | ||
fd01b88c | 7679 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
7680 | { |
7681 | /* raid5 can take over: | |
f1b29bca | 7682 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
7683 | * raid1 - if there are two drives. We need to know the chunk size |
7684 | * raid4 - trivial - just use a raid4 layout. | |
7685 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 7686 | */ |
f1b29bca DW |
7687 | if (mddev->level == 0) |
7688 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
7689 | if (mddev->level == 1) |
7690 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
7691 | if (mddev->level == 4) { |
7692 | mddev->new_layout = ALGORITHM_PARITY_N; | |
7693 | mddev->new_level = 5; | |
7694 | return setup_conf(mddev); | |
7695 | } | |
fc9739c6 N |
7696 | if (mddev->level == 6) |
7697 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
7698 | |
7699 | return ERR_PTR(-EINVAL); | |
7700 | } | |
7701 | ||
fd01b88c | 7702 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 7703 | { |
f1b29bca DW |
7704 | /* raid4 can take over: |
7705 | * raid0 - if there is only one strip zone | |
7706 | * raid5 - if layout is right | |
a78d38a1 | 7707 | */ |
f1b29bca DW |
7708 | if (mddev->level == 0) |
7709 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
7710 | if (mddev->level == 5 && |
7711 | mddev->layout == ALGORITHM_PARITY_N) { | |
7712 | mddev->new_layout = 0; | |
7713 | mddev->new_level = 4; | |
7714 | return setup_conf(mddev); | |
7715 | } | |
7716 | return ERR_PTR(-EINVAL); | |
7717 | } | |
d562b0c4 | 7718 | |
84fc4b56 | 7719 | static struct md_personality raid5_personality; |
245f46c2 | 7720 | |
fd01b88c | 7721 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
7722 | { |
7723 | /* Currently can only take over a raid5. We map the | |
7724 | * personality to an equivalent raid6 personality | |
7725 | * with the Q block at the end. | |
7726 | */ | |
7727 | int new_layout; | |
7728 | ||
7729 | if (mddev->pers != &raid5_personality) | |
7730 | return ERR_PTR(-EINVAL); | |
7731 | if (mddev->degraded > 1) | |
7732 | return ERR_PTR(-EINVAL); | |
7733 | if (mddev->raid_disks > 253) | |
7734 | return ERR_PTR(-EINVAL); | |
7735 | if (mddev->raid_disks < 3) | |
7736 | return ERR_PTR(-EINVAL); | |
7737 | ||
7738 | switch (mddev->layout) { | |
7739 | case ALGORITHM_LEFT_ASYMMETRIC: | |
7740 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
7741 | break; | |
7742 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
7743 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
7744 | break; | |
7745 | case ALGORITHM_LEFT_SYMMETRIC: | |
7746 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
7747 | break; | |
7748 | case ALGORITHM_RIGHT_SYMMETRIC: | |
7749 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
7750 | break; | |
7751 | case ALGORITHM_PARITY_0: | |
7752 | new_layout = ALGORITHM_PARITY_0_6; | |
7753 | break; | |
7754 | case ALGORITHM_PARITY_N: | |
7755 | new_layout = ALGORITHM_PARITY_N; | |
7756 | break; | |
7757 | default: | |
7758 | return ERR_PTR(-EINVAL); | |
7759 | } | |
7760 | mddev->new_level = 6; | |
7761 | mddev->new_layout = new_layout; | |
7762 | mddev->delta_disks = 1; | |
7763 | mddev->raid_disks += 1; | |
7764 | return setup_conf(mddev); | |
7765 | } | |
7766 | ||
84fc4b56 | 7767 | static struct md_personality raid6_personality = |
16a53ecc N |
7768 | { |
7769 | .name = "raid6", | |
7770 | .level = 6, | |
7771 | .owner = THIS_MODULE, | |
7772 | .make_request = make_request, | |
7773 | .run = run, | |
afa0f557 | 7774 | .free = raid5_free, |
16a53ecc N |
7775 | .status = status, |
7776 | .error_handler = error, | |
7777 | .hot_add_disk = raid5_add_disk, | |
7778 | .hot_remove_disk= raid5_remove_disk, | |
7779 | .spare_active = raid5_spare_active, | |
7780 | .sync_request = sync_request, | |
7781 | .resize = raid5_resize, | |
80c3a6ce | 7782 | .size = raid5_size, |
50ac168a | 7783 | .check_reshape = raid6_check_reshape, |
f416885e | 7784 | .start_reshape = raid5_start_reshape, |
cea9c228 | 7785 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 7786 | .quiesce = raid5_quiesce, |
245f46c2 | 7787 | .takeover = raid6_takeover, |
5c675f83 | 7788 | .congested = raid5_congested, |
64590f45 | 7789 | .mergeable_bvec = raid5_mergeable_bvec, |
16a53ecc | 7790 | }; |
84fc4b56 | 7791 | static struct md_personality raid5_personality = |
1da177e4 LT |
7792 | { |
7793 | .name = "raid5", | |
2604b703 | 7794 | .level = 5, |
1da177e4 LT |
7795 | .owner = THIS_MODULE, |
7796 | .make_request = make_request, | |
7797 | .run = run, | |
afa0f557 | 7798 | .free = raid5_free, |
1da177e4 LT |
7799 | .status = status, |
7800 | .error_handler = error, | |
7801 | .hot_add_disk = raid5_add_disk, | |
7802 | .hot_remove_disk= raid5_remove_disk, | |
7803 | .spare_active = raid5_spare_active, | |
7804 | .sync_request = sync_request, | |
7805 | .resize = raid5_resize, | |
80c3a6ce | 7806 | .size = raid5_size, |
63c70c4f N |
7807 | .check_reshape = raid5_check_reshape, |
7808 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7809 | .finish_reshape = raid5_finish_reshape, |
72626685 | 7810 | .quiesce = raid5_quiesce, |
d562b0c4 | 7811 | .takeover = raid5_takeover, |
5c675f83 | 7812 | .congested = raid5_congested, |
64590f45 | 7813 | .mergeable_bvec = raid5_mergeable_bvec, |
1da177e4 LT |
7814 | }; |
7815 | ||
84fc4b56 | 7816 | static struct md_personality raid4_personality = |
1da177e4 | 7817 | { |
2604b703 N |
7818 | .name = "raid4", |
7819 | .level = 4, | |
7820 | .owner = THIS_MODULE, | |
7821 | .make_request = make_request, | |
7822 | .run = run, | |
afa0f557 | 7823 | .free = raid5_free, |
2604b703 N |
7824 | .status = status, |
7825 | .error_handler = error, | |
7826 | .hot_add_disk = raid5_add_disk, | |
7827 | .hot_remove_disk= raid5_remove_disk, | |
7828 | .spare_active = raid5_spare_active, | |
7829 | .sync_request = sync_request, | |
7830 | .resize = raid5_resize, | |
80c3a6ce | 7831 | .size = raid5_size, |
3d37890b N |
7832 | .check_reshape = raid5_check_reshape, |
7833 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 7834 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 7835 | .quiesce = raid5_quiesce, |
a78d38a1 | 7836 | .takeover = raid4_takeover, |
5c675f83 | 7837 | .congested = raid5_congested, |
64590f45 | 7838 | .mergeable_bvec = raid5_mergeable_bvec, |
2604b703 N |
7839 | }; |
7840 | ||
7841 | static int __init raid5_init(void) | |
7842 | { | |
851c30c9 SL |
7843 | raid5_wq = alloc_workqueue("raid5wq", |
7844 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
7845 | if (!raid5_wq) | |
7846 | return -ENOMEM; | |
16a53ecc | 7847 | register_md_personality(&raid6_personality); |
2604b703 N |
7848 | register_md_personality(&raid5_personality); |
7849 | register_md_personality(&raid4_personality); | |
7850 | return 0; | |
1da177e4 LT |
7851 | } |
7852 | ||
2604b703 | 7853 | static void raid5_exit(void) |
1da177e4 | 7854 | { |
16a53ecc | 7855 | unregister_md_personality(&raid6_personality); |
2604b703 N |
7856 | unregister_md_personality(&raid5_personality); |
7857 | unregister_md_personality(&raid4_personality); | |
851c30c9 | 7858 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
7859 | } |
7860 | ||
7861 | module_init(raid5_init); | |
7862 | module_exit(raid5_exit); | |
7863 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 7864 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 7865 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
7866 | MODULE_ALIAS("md-raid5"); |
7867 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
7868 | MODULE_ALIAS("md-level-5"); |
7869 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
7870 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
7871 | MODULE_ALIAS("md-raid6"); | |
7872 | MODULE_ALIAS("md-level-6"); | |
7873 | ||
7874 | /* This used to be two separate modules, they were: */ | |
7875 | MODULE_ALIAS("raid5"); | |
7876 | MODULE_ALIAS("raid6"); |