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> |
3f07c014 IM |
58 | #include <linux/sched/signal.h> |
59 | ||
a9add5d9 | 60 | #include <trace/events/block.h> |
aaf9f12e | 61 | #include <linux/list_sort.h> |
a9add5d9 | 62 | |
43b2e5d8 | 63 | #include "md.h" |
bff61975 | 64 | #include "raid5.h" |
54071b38 | 65 | #include "raid0.h" |
ef740c37 | 66 | #include "bitmap.h" |
ff875738 | 67 | #include "raid5-log.h" |
72626685 | 68 | |
394ed8e4 SL |
69 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
70 | ||
851c30c9 SL |
71 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
72 | #define ANY_GROUP NUMA_NO_NODE | |
73 | ||
8e0e99ba N |
74 | static bool devices_handle_discard_safely = false; |
75 | module_param(devices_handle_discard_safely, bool, 0644); | |
76 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
77 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 78 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 79 | |
d1688a6d | 80 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
81 | { |
82 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
83 | return &conf->stripe_hashtbl[hash]; | |
84 | } | |
1da177e4 | 85 | |
566c09c5 SL |
86 | static inline int stripe_hash_locks_hash(sector_t sect) |
87 | { | |
88 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
89 | } | |
90 | ||
91 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
92 | { | |
93 | spin_lock_irq(conf->hash_locks + hash); | |
94 | spin_lock(&conf->device_lock); | |
95 | } | |
96 | ||
97 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
98 | { | |
99 | spin_unlock(&conf->device_lock); | |
100 | spin_unlock_irq(conf->hash_locks + hash); | |
101 | } | |
102 | ||
103 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
104 | { | |
105 | int i; | |
106 | local_irq_disable(); | |
107 | spin_lock(conf->hash_locks); | |
108 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
109 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
110 | spin_lock(&conf->device_lock); | |
111 | } | |
112 | ||
113 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
114 | { | |
115 | int i; | |
116 | spin_unlock(&conf->device_lock); | |
117 | for (i = NR_STRIPE_HASH_LOCKS; i; i--) | |
118 | spin_unlock(conf->hash_locks + i - 1); | |
119 | local_irq_enable(); | |
120 | } | |
121 | ||
d0dabf7e N |
122 | /* Find first data disk in a raid6 stripe */ |
123 | static inline int raid6_d0(struct stripe_head *sh) | |
124 | { | |
67cc2b81 N |
125 | if (sh->ddf_layout) |
126 | /* ddf always start from first device */ | |
127 | return 0; | |
128 | /* md starts just after Q block */ | |
d0dabf7e N |
129 | if (sh->qd_idx == sh->disks - 1) |
130 | return 0; | |
131 | else | |
132 | return sh->qd_idx + 1; | |
133 | } | |
16a53ecc N |
134 | static inline int raid6_next_disk(int disk, int raid_disks) |
135 | { | |
136 | disk++; | |
137 | return (disk < raid_disks) ? disk : 0; | |
138 | } | |
a4456856 | 139 | |
d0dabf7e N |
140 | /* When walking through the disks in a raid5, starting at raid6_d0, |
141 | * We need to map each disk to a 'slot', where the data disks are slot | |
142 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
143 | * is raid_disks-1. This help does that mapping. | |
144 | */ | |
67cc2b81 N |
145 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
146 | int *count, int syndrome_disks) | |
d0dabf7e | 147 | { |
6629542e | 148 | int slot = *count; |
67cc2b81 | 149 | |
e4424fee | 150 | if (sh->ddf_layout) |
6629542e | 151 | (*count)++; |
d0dabf7e | 152 | if (idx == sh->pd_idx) |
67cc2b81 | 153 | return syndrome_disks; |
d0dabf7e | 154 | if (idx == sh->qd_idx) |
67cc2b81 | 155 | return syndrome_disks + 1; |
e4424fee | 156 | if (!sh->ddf_layout) |
6629542e | 157 | (*count)++; |
d0dabf7e N |
158 | return slot; |
159 | } | |
160 | ||
34a6f80e | 161 | static void return_io(struct bio_list *return_bi) |
a4456856 | 162 | { |
34a6f80e N |
163 | struct bio *bi; |
164 | while ((bi = bio_list_pop(return_bi)) != NULL) { | |
4f024f37 | 165 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
166 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
167 | bi, 0); | |
4246a0b6 | 168 | bio_endio(bi); |
a4456856 DW |
169 | } |
170 | } | |
171 | ||
d1688a6d | 172 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 173 | |
600aa109 DW |
174 | static int stripe_operations_active(struct stripe_head *sh) |
175 | { | |
176 | return sh->check_state || sh->reconstruct_state || | |
177 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
178 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
179 | } | |
180 | ||
535ae4eb SL |
181 | static bool stripe_is_lowprio(struct stripe_head *sh) |
182 | { | |
183 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
184 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
185 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
186 | } | |
187 | ||
851c30c9 SL |
188 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
189 | { | |
190 | struct r5conf *conf = sh->raid_conf; | |
191 | struct r5worker_group *group; | |
bfc90cb0 | 192 | int thread_cnt; |
851c30c9 SL |
193 | int i, cpu = sh->cpu; |
194 | ||
195 | if (!cpu_online(cpu)) { | |
196 | cpu = cpumask_any(cpu_online_mask); | |
197 | sh->cpu = cpu; | |
198 | } | |
199 | ||
200 | if (list_empty(&sh->lru)) { | |
201 | struct r5worker_group *group; | |
202 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
203 | if (stripe_is_lowprio(sh)) |
204 | list_add_tail(&sh->lru, &group->loprio_list); | |
205 | else | |
206 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
207 | group->stripes_cnt++; |
208 | sh->group = group; | |
851c30c9 SL |
209 | } |
210 | ||
211 | if (conf->worker_cnt_per_group == 0) { | |
212 | md_wakeup_thread(conf->mddev->thread); | |
213 | return; | |
214 | } | |
215 | ||
216 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
217 | ||
bfc90cb0 SL |
218 | group->workers[0].working = true; |
219 | /* at least one worker should run to avoid race */ | |
220 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
221 | ||
222 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
223 | /* wakeup more workers */ | |
224 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
225 | if (group->workers[i].working == false) { | |
226 | group->workers[i].working = true; | |
227 | queue_work_on(sh->cpu, raid5_wq, | |
228 | &group->workers[i].work); | |
229 | thread_cnt--; | |
230 | } | |
231 | } | |
851c30c9 SL |
232 | } |
233 | ||
566c09c5 SL |
234 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
235 | struct list_head *temp_inactive_list) | |
1da177e4 | 236 | { |
1e6d690b SL |
237 | int i; |
238 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
239 | ||
4eb788df SL |
240 | BUG_ON(!list_empty(&sh->lru)); |
241 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
242 | |
243 | if (r5c_is_writeback(conf->log)) | |
244 | for (i = sh->disks; i--; ) | |
245 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
246 | injournal++; | |
a39f7afd SL |
247 | /* |
248 | * When quiesce in r5c write back, set STRIPE_HANDLE for stripes with | |
249 | * data in journal, so they are not released to cached lists | |
250 | */ | |
251 | if (conf->quiesce && r5c_is_writeback(conf->log) && | |
252 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0) { | |
253 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
254 | r5c_make_stripe_write_out(sh); | |
255 | set_bit(STRIPE_HANDLE, &sh->state); | |
256 | } | |
1e6d690b | 257 | |
4eb788df SL |
258 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
259 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 260 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 261 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 262 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
263 | sh->bm_seq - conf->seq_write > 0) |
264 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
265 | else { | |
266 | clear_bit(STRIPE_DELAYED, &sh->state); | |
267 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 268 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
269 | if (stripe_is_lowprio(sh)) |
270 | list_add_tail(&sh->lru, | |
271 | &conf->loprio_list); | |
272 | else | |
273 | list_add_tail(&sh->lru, | |
274 | &conf->handle_list); | |
851c30c9 SL |
275 | } else { |
276 | raid5_wakeup_stripe_thread(sh); | |
277 | return; | |
278 | } | |
4eb788df SL |
279 | } |
280 | md_wakeup_thread(conf->mddev->thread); | |
281 | } else { | |
282 | BUG_ON(stripe_operations_active(sh)); | |
283 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
284 | if (atomic_dec_return(&conf->preread_active_stripes) | |
285 | < IO_THRESHOLD) | |
286 | md_wakeup_thread(conf->mddev->thread); | |
287 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
288 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
289 | if (!r5c_is_writeback(conf->log)) | |
290 | list_add_tail(&sh->lru, temp_inactive_list); | |
291 | else { | |
292 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
293 | if (injournal == 0) | |
294 | list_add_tail(&sh->lru, temp_inactive_list); | |
295 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
296 | /* full stripe */ | |
297 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
298 | atomic_inc(&conf->r5c_cached_full_stripes); | |
299 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
300 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
301 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 302 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
303 | } else |
304 | /* | |
305 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
306 | * r5c_try_caching_write(). No need to | |
307 | * set it again. | |
308 | */ | |
1e6d690b | 309 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
310 | } |
311 | } | |
1da177e4 LT |
312 | } |
313 | } | |
d0dabf7e | 314 | |
566c09c5 SL |
315 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
316 | struct list_head *temp_inactive_list) | |
4eb788df SL |
317 | { |
318 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
319 | do_release_stripe(conf, sh, temp_inactive_list); |
320 | } | |
321 | ||
322 | /* | |
323 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
324 | * | |
325 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
326 | * given time. Adding stripes only takes device lock, while deleting stripes | |
327 | * only takes hash lock. | |
328 | */ | |
329 | static void release_inactive_stripe_list(struct r5conf *conf, | |
330 | struct list_head *temp_inactive_list, | |
331 | int hash) | |
332 | { | |
333 | int size; | |
6ab2a4b8 | 334 | bool do_wakeup = false; |
566c09c5 SL |
335 | unsigned long flags; |
336 | ||
337 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
338 | size = NR_STRIPE_HASH_LOCKS; | |
339 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
340 | } else | |
341 | size = 1; | |
342 | while (size) { | |
343 | struct list_head *list = &temp_inactive_list[size - 1]; | |
344 | ||
345 | /* | |
6d036f7d | 346 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
347 | * remove stripes from the list |
348 | */ | |
349 | if (!list_empty_careful(list)) { | |
350 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
351 | if (list_empty(conf->inactive_list + hash) && |
352 | !list_empty(list)) | |
353 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 354 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 355 | do_wakeup = true; |
566c09c5 SL |
356 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
357 | } | |
358 | size--; | |
359 | hash--; | |
360 | } | |
361 | ||
362 | if (do_wakeup) { | |
6ab2a4b8 | 363 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
364 | if (atomic_read(&conf->active_stripes) == 0) |
365 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
366 | if (conf->retry_read_aligned) |
367 | md_wakeup_thread(conf->mddev->thread); | |
368 | } | |
4eb788df SL |
369 | } |
370 | ||
773ca82f | 371 | /* should hold conf->device_lock already */ |
566c09c5 SL |
372 | static int release_stripe_list(struct r5conf *conf, |
373 | struct list_head *temp_inactive_list) | |
773ca82f | 374 | { |
eae8263f | 375 | struct stripe_head *sh, *t; |
773ca82f SL |
376 | int count = 0; |
377 | struct llist_node *head; | |
378 | ||
379 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 380 | head = llist_reverse_order(head); |
eae8263f | 381 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
382 | int hash; |
383 | ||
773ca82f SL |
384 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
385 | smp_mb(); | |
386 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
387 | /* | |
388 | * Don't worry the bit is set here, because if the bit is set | |
389 | * again, the count is always > 1. This is true for | |
390 | * STRIPE_ON_UNPLUG_LIST bit too. | |
391 | */ | |
566c09c5 SL |
392 | hash = sh->hash_lock_index; |
393 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
394 | count++; |
395 | } | |
396 | ||
397 | return count; | |
398 | } | |
399 | ||
6d036f7d | 400 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 401 | { |
d1688a6d | 402 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 403 | unsigned long flags; |
566c09c5 SL |
404 | struct list_head list; |
405 | int hash; | |
773ca82f | 406 | bool wakeup; |
16a53ecc | 407 | |
cf170f3f ES |
408 | /* Avoid release_list until the last reference. |
409 | */ | |
410 | if (atomic_add_unless(&sh->count, -1, 1)) | |
411 | return; | |
412 | ||
ad4068de | 413 | if (unlikely(!conf->mddev->thread) || |
414 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
415 | goto slow_path; |
416 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
417 | if (wakeup) | |
418 | md_wakeup_thread(conf->mddev->thread); | |
419 | return; | |
420 | slow_path: | |
4eb788df | 421 | local_irq_save(flags); |
773ca82f | 422 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 423 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
424 | INIT_LIST_HEAD(&list); |
425 | hash = sh->hash_lock_index; | |
426 | do_release_stripe(conf, sh, &list); | |
4eb788df | 427 | spin_unlock(&conf->device_lock); |
566c09c5 | 428 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
429 | } |
430 | local_irq_restore(flags); | |
1da177e4 LT |
431 | } |
432 | ||
fccddba0 | 433 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 434 | { |
45b4233c DW |
435 | pr_debug("remove_hash(), stripe %llu\n", |
436 | (unsigned long long)sh->sector); | |
1da177e4 | 437 | |
fccddba0 | 438 | hlist_del_init(&sh->hash); |
1da177e4 LT |
439 | } |
440 | ||
d1688a6d | 441 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 442 | { |
fccddba0 | 443 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 444 | |
45b4233c DW |
445 | pr_debug("insert_hash(), stripe %llu\n", |
446 | (unsigned long long)sh->sector); | |
1da177e4 | 447 | |
fccddba0 | 448 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
449 | } |
450 | ||
1da177e4 | 451 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 452 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
453 | { |
454 | struct stripe_head *sh = NULL; | |
455 | struct list_head *first; | |
456 | ||
566c09c5 | 457 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 458 | goto out; |
566c09c5 | 459 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
460 | sh = list_entry(first, struct stripe_head, lru); |
461 | list_del_init(first); | |
462 | remove_hash(sh); | |
463 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 464 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
465 | if (list_empty(conf->inactive_list + hash)) |
466 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
467 | out: |
468 | return sh; | |
469 | } | |
470 | ||
e4e11e38 | 471 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
472 | { |
473 | struct page *p; | |
474 | int i; | |
e4e11e38 | 475 | int num = sh->raid_conf->pool_size; |
1da177e4 | 476 | |
e4e11e38 | 477 | for (i = 0; i < num ; i++) { |
d592a996 | 478 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
479 | p = sh->dev[i].page; |
480 | if (!p) | |
481 | continue; | |
482 | sh->dev[i].page = NULL; | |
2d1f3b5d | 483 | put_page(p); |
1da177e4 | 484 | } |
3418d036 AP |
485 | |
486 | if (sh->ppl_page) { | |
487 | put_page(sh->ppl_page); | |
488 | sh->ppl_page = NULL; | |
489 | } | |
1da177e4 LT |
490 | } |
491 | ||
a9683a79 | 492 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
493 | { |
494 | int i; | |
e4e11e38 | 495 | int num = sh->raid_conf->pool_size; |
1da177e4 | 496 | |
e4e11e38 | 497 | for (i = 0; i < num; i++) { |
1da177e4 LT |
498 | struct page *page; |
499 | ||
a9683a79 | 500 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
501 | return 1; |
502 | } | |
503 | sh->dev[i].page = page; | |
d592a996 | 504 | sh->dev[i].orig_page = page; |
1da177e4 | 505 | } |
3418d036 AP |
506 | |
507 | if (raid5_has_ppl(sh->raid_conf)) { | |
508 | sh->ppl_page = alloc_page(gfp); | |
509 | if (!sh->ppl_page) | |
510 | return 1; | |
511 | } | |
512 | ||
1da177e4 LT |
513 | return 0; |
514 | } | |
515 | ||
784052ec | 516 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 517 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 518 | struct stripe_head *sh); |
1da177e4 | 519 | |
b5663ba4 | 520 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 521 | { |
d1688a6d | 522 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 523 | int i, seq; |
1da177e4 | 524 | |
78bafebd ES |
525 | BUG_ON(atomic_read(&sh->count) != 0); |
526 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 527 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 528 | BUG_ON(sh->batch_head); |
d84e0f10 | 529 | |
45b4233c | 530 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 531 | (unsigned long long)sector); |
566c09c5 SL |
532 | retry: |
533 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 534 | sh->generation = conf->generation - previous; |
b5663ba4 | 535 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 536 | sh->sector = sector; |
911d4ee8 | 537 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
538 | sh->state = 0; |
539 | ||
7ecaa1e6 | 540 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
541 | struct r5dev *dev = &sh->dev[i]; |
542 | ||
d84e0f10 | 543 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 544 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 545 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 546 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 547 | dev->read, dev->towrite, dev->written, |
1da177e4 | 548 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 549 | WARN_ON(1); |
1da177e4 LT |
550 | } |
551 | dev->flags = 0; | |
784052ec | 552 | raid5_build_block(sh, i, previous); |
1da177e4 | 553 | } |
566c09c5 SL |
554 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
555 | goto retry; | |
7a87f434 | 556 | sh->overwrite_disks = 0; |
1da177e4 | 557 | insert_hash(conf, sh); |
851c30c9 | 558 | sh->cpu = smp_processor_id(); |
da41ba65 | 559 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
560 | } |
561 | ||
d1688a6d | 562 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 563 | short generation) |
1da177e4 LT |
564 | { |
565 | struct stripe_head *sh; | |
566 | ||
45b4233c | 567 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 568 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 569 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 570 | return sh; |
45b4233c | 571 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
572 | return NULL; |
573 | } | |
574 | ||
674806d6 N |
575 | /* |
576 | * Need to check if array has failed when deciding whether to: | |
577 | * - start an array | |
578 | * - remove non-faulty devices | |
579 | * - add a spare | |
580 | * - allow a reshape | |
581 | * This determination is simple when no reshape is happening. | |
582 | * However if there is a reshape, we need to carefully check | |
583 | * both the before and after sections. | |
584 | * This is because some failed devices may only affect one | |
585 | * of the two sections, and some non-in_sync devices may | |
586 | * be insync in the section most affected by failed devices. | |
587 | */ | |
2e38a37f | 588 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 589 | { |
908f4fbd | 590 | int degraded, degraded2; |
674806d6 | 591 | int i; |
674806d6 N |
592 | |
593 | rcu_read_lock(); | |
594 | degraded = 0; | |
595 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 596 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
597 | if (rdev && test_bit(Faulty, &rdev->flags)) |
598 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
599 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
600 | degraded++; | |
601 | else if (test_bit(In_sync, &rdev->flags)) | |
602 | ; | |
603 | else | |
604 | /* not in-sync or faulty. | |
605 | * If the reshape increases the number of devices, | |
606 | * this is being recovered by the reshape, so | |
607 | * this 'previous' section is not in_sync. | |
608 | * If the number of devices is being reduced however, | |
609 | * the device can only be part of the array if | |
610 | * we are reverting a reshape, so this section will | |
611 | * be in-sync. | |
612 | */ | |
613 | if (conf->raid_disks >= conf->previous_raid_disks) | |
614 | degraded++; | |
615 | } | |
616 | rcu_read_unlock(); | |
908f4fbd N |
617 | if (conf->raid_disks == conf->previous_raid_disks) |
618 | return degraded; | |
674806d6 | 619 | rcu_read_lock(); |
908f4fbd | 620 | degraded2 = 0; |
674806d6 | 621 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 622 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
623 | if (rdev && test_bit(Faulty, &rdev->flags)) |
624 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 625 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 626 | degraded2++; |
674806d6 N |
627 | else if (test_bit(In_sync, &rdev->flags)) |
628 | ; | |
629 | else | |
630 | /* not in-sync or faulty. | |
631 | * If reshape increases the number of devices, this | |
632 | * section has already been recovered, else it | |
633 | * almost certainly hasn't. | |
634 | */ | |
635 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 636 | degraded2++; |
674806d6 N |
637 | } |
638 | rcu_read_unlock(); | |
908f4fbd N |
639 | if (degraded2 > degraded) |
640 | return degraded2; | |
641 | return degraded; | |
642 | } | |
643 | ||
644 | static int has_failed(struct r5conf *conf) | |
645 | { | |
646 | int degraded; | |
647 | ||
648 | if (conf->mddev->reshape_position == MaxSector) | |
649 | return conf->mddev->degraded > conf->max_degraded; | |
650 | ||
2e38a37f | 651 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
652 | if (degraded > conf->max_degraded) |
653 | return 1; | |
654 | return 0; | |
655 | } | |
656 | ||
6d036f7d SL |
657 | struct stripe_head * |
658 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
659 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
660 | { |
661 | struct stripe_head *sh; | |
566c09c5 | 662 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 663 | int inc_empty_inactive_list_flag; |
1da177e4 | 664 | |
45b4233c | 665 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 666 | |
566c09c5 | 667 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
668 | |
669 | do { | |
b1b46486 | 670 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 671 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 672 | *(conf->hash_locks + hash)); |
86b42c71 | 673 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 674 | if (!sh) { |
edbe83ab | 675 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 676 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
677 | if (!sh && !test_bit(R5_DID_ALLOC, |
678 | &conf->cache_state)) | |
edbe83ab N |
679 | set_bit(R5_ALLOC_MORE, |
680 | &conf->cache_state); | |
681 | } | |
1da177e4 LT |
682 | if (noblock && sh == NULL) |
683 | break; | |
a39f7afd SL |
684 | |
685 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 686 | if (!sh) { |
5423399a N |
687 | set_bit(R5_INACTIVE_BLOCKED, |
688 | &conf->cache_state); | |
a39f7afd | 689 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
690 | wait_event_lock_irq( |
691 | conf->wait_for_stripe, | |
566c09c5 SL |
692 | !list_empty(conf->inactive_list + hash) && |
693 | (atomic_read(&conf->active_stripes) | |
694 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
695 | || !test_bit(R5_INACTIVE_BLOCKED, |
696 | &conf->cache_state)), | |
6ab2a4b8 | 697 | *(conf->hash_locks + hash)); |
5423399a N |
698 | clear_bit(R5_INACTIVE_BLOCKED, |
699 | &conf->cache_state); | |
7da9d450 | 700 | } else { |
b5663ba4 | 701 | init_stripe(sh, sector, previous); |
7da9d450 N |
702 | atomic_inc(&sh->count); |
703 | } | |
e240c183 | 704 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 705 | spin_lock(&conf->device_lock); |
e240c183 | 706 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
707 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
708 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
709 | BUG_ON(list_empty(&sh->lru) && |
710 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
711 | inc_empty_inactive_list_flag = 0; |
712 | if (!list_empty(conf->inactive_list + hash)) | |
713 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 714 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
715 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
716 | atomic_inc(&conf->empty_inactive_list_nr); | |
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 | ||
566c09c5 | 727 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
728 | return sh; |
729 | } | |
730 | ||
7a87f434 | 731 | static bool is_full_stripe_write(struct stripe_head *sh) |
732 | { | |
733 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
734 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
735 | } | |
736 | ||
59fc630b | 737 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
738 | { | |
739 | local_irq_disable(); | |
740 | if (sh1 > sh2) { | |
741 | spin_lock(&sh2->stripe_lock); | |
742 | spin_lock_nested(&sh1->stripe_lock, 1); | |
743 | } else { | |
744 | spin_lock(&sh1->stripe_lock); | |
745 | spin_lock_nested(&sh2->stripe_lock, 1); | |
746 | } | |
747 | } | |
748 | ||
749 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
750 | { | |
751 | spin_unlock(&sh1->stripe_lock); | |
752 | spin_unlock(&sh2->stripe_lock); | |
753 | local_irq_enable(); | |
754 | } | |
755 | ||
756 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
757 | static bool stripe_can_batch(struct stripe_head *sh) | |
758 | { | |
9c3e333d SL |
759 | struct r5conf *conf = sh->raid_conf; |
760 | ||
3418d036 | 761 | if (conf->log || raid5_has_ppl(conf)) |
9c3e333d | 762 | return false; |
59fc630b | 763 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 764 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 765 | is_full_stripe_write(sh); |
766 | } | |
767 | ||
768 | /* we only do back search */ | |
769 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
770 | { | |
771 | struct stripe_head *head; | |
772 | sector_t head_sector, tmp_sec; | |
773 | int hash; | |
774 | int dd_idx; | |
ff00d3b4 | 775 | int inc_empty_inactive_list_flag; |
59fc630b | 776 | |
59fc630b | 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)); | |
ff00d3b4 ZL |
793 | inc_empty_inactive_list_flag = 0; |
794 | if (!list_empty(conf->inactive_list + hash)) | |
795 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 796 | list_del_init(&head->lru); |
ff00d3b4 ZL |
797 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
798 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 799 | if (head->group) { |
800 | head->group->stripes_cnt--; | |
801 | head->group = NULL; | |
802 | } | |
803 | } | |
804 | atomic_inc(&head->count); | |
805 | spin_unlock(&conf->device_lock); | |
806 | } | |
807 | spin_unlock_irq(conf->hash_locks + hash); | |
808 | ||
809 | if (!head) | |
810 | return; | |
811 | if (!stripe_can_batch(head)) | |
812 | goto out; | |
813 | ||
814 | lock_two_stripes(head, sh); | |
815 | /* clear_batch_ready clear the flag */ | |
816 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
817 | goto unlock_out; | |
818 | ||
819 | if (sh->batch_head) | |
820 | goto unlock_out; | |
821 | ||
822 | dd_idx = 0; | |
823 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
824 | dd_idx++; | |
1eff9d32 | 825 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 826 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 827 | goto unlock_out; |
828 | ||
829 | if (head->batch_head) { | |
830 | spin_lock(&head->batch_head->batch_lock); | |
831 | /* This batch list is already running */ | |
832 | if (!stripe_can_batch(head)) { | |
833 | spin_unlock(&head->batch_head->batch_lock); | |
834 | goto unlock_out; | |
835 | } | |
836 | ||
837 | /* | |
838 | * at this point, head's BATCH_READY could be cleared, but we | |
839 | * can still add the stripe to batch list | |
840 | */ | |
841 | list_add(&sh->batch_list, &head->batch_list); | |
842 | spin_unlock(&head->batch_head->batch_lock); | |
843 | ||
844 | sh->batch_head = head->batch_head; | |
845 | } else { | |
846 | head->batch_head = head; | |
847 | sh->batch_head = head->batch_head; | |
848 | spin_lock(&head->batch_lock); | |
849 | list_add_tail(&sh->batch_list, &head->batch_list); | |
850 | spin_unlock(&head->batch_lock); | |
851 | } | |
852 | ||
853 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
854 | if (atomic_dec_return(&conf->preread_active_stripes) | |
855 | < IO_THRESHOLD) | |
856 | md_wakeup_thread(conf->mddev->thread); | |
857 | ||
2b6b2457 N |
858 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
859 | int seq = sh->bm_seq; | |
860 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
861 | sh->batch_head->bm_seq > seq) | |
862 | seq = sh->batch_head->bm_seq; | |
863 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
864 | sh->batch_head->bm_seq = seq; | |
865 | } | |
866 | ||
59fc630b | 867 | atomic_inc(&sh->count); |
868 | unlock_out: | |
869 | unlock_two_stripes(head, sh); | |
870 | out: | |
6d036f7d | 871 | raid5_release_stripe(head); |
59fc630b | 872 | } |
873 | ||
05616be5 N |
874 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
875 | * in this stripe_head. | |
876 | */ | |
877 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
878 | { | |
879 | sector_t progress = conf->reshape_progress; | |
880 | /* Need a memory barrier to make sure we see the value | |
881 | * of conf->generation, or ->data_offset that was set before | |
882 | * reshape_progress was updated. | |
883 | */ | |
884 | smp_rmb(); | |
885 | if (progress == MaxSector) | |
886 | return 0; | |
887 | if (sh->generation == conf->generation - 1) | |
888 | return 0; | |
889 | /* We are in a reshape, and this is a new-generation stripe, | |
890 | * so use new_data_offset. | |
891 | */ | |
892 | return 1; | |
893 | } | |
894 | ||
aaf9f12e | 895 | static void dispatch_bio_list(struct bio_list *tmp) |
765d704d | 896 | { |
765d704d SL |
897 | struct bio *bio; |
898 | ||
aaf9f12e SL |
899 | while ((bio = bio_list_pop(tmp))) |
900 | generic_make_request(bio); | |
901 | } | |
902 | ||
903 | static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b) | |
904 | { | |
905 | const struct r5pending_data *da = list_entry(a, | |
906 | struct r5pending_data, sibling); | |
907 | const struct r5pending_data *db = list_entry(b, | |
908 | struct r5pending_data, sibling); | |
909 | if (da->sector > db->sector) | |
910 | return 1; | |
911 | if (da->sector < db->sector) | |
912 | return -1; | |
913 | return 0; | |
914 | } | |
915 | ||
916 | static void dispatch_defer_bios(struct r5conf *conf, int target, | |
917 | struct bio_list *list) | |
918 | { | |
919 | struct r5pending_data *data; | |
920 | struct list_head *first, *next = NULL; | |
921 | int cnt = 0; | |
922 | ||
923 | if (conf->pending_data_cnt == 0) | |
924 | return; | |
925 | ||
926 | list_sort(NULL, &conf->pending_list, cmp_stripe); | |
927 | ||
928 | first = conf->pending_list.next; | |
929 | ||
930 | /* temporarily move the head */ | |
931 | if (conf->next_pending_data) | |
932 | list_move_tail(&conf->pending_list, | |
933 | &conf->next_pending_data->sibling); | |
934 | ||
935 | while (!list_empty(&conf->pending_list)) { | |
936 | data = list_first_entry(&conf->pending_list, | |
937 | struct r5pending_data, sibling); | |
938 | if (&data->sibling == first) | |
939 | first = data->sibling.next; | |
940 | next = data->sibling.next; | |
941 | ||
942 | bio_list_merge(list, &data->bios); | |
943 | list_move(&data->sibling, &conf->free_list); | |
944 | cnt++; | |
945 | if (cnt >= target) | |
946 | break; | |
947 | } | |
948 | conf->pending_data_cnt -= cnt; | |
949 | BUG_ON(conf->pending_data_cnt < 0 || cnt < target); | |
950 | ||
951 | if (next != &conf->pending_list) | |
952 | conf->next_pending_data = list_entry(next, | |
953 | struct r5pending_data, sibling); | |
954 | else | |
955 | conf->next_pending_data = NULL; | |
956 | /* list isn't empty */ | |
957 | if (first != &conf->pending_list) | |
958 | list_move_tail(&conf->pending_list, first); | |
959 | } | |
960 | ||
961 | static void flush_deferred_bios(struct r5conf *conf) | |
962 | { | |
963 | struct bio_list tmp = BIO_EMPTY_LIST; | |
964 | ||
965 | if (conf->pending_data_cnt == 0) | |
765d704d SL |
966 | return; |
967 | ||
765d704d | 968 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
969 | dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp); |
970 | BUG_ON(conf->pending_data_cnt != 0); | |
765d704d SL |
971 | spin_unlock(&conf->pending_bios_lock); |
972 | ||
aaf9f12e | 973 | dispatch_bio_list(&tmp); |
765d704d SL |
974 | } |
975 | ||
aaf9f12e SL |
976 | static void defer_issue_bios(struct r5conf *conf, sector_t sector, |
977 | struct bio_list *bios) | |
765d704d | 978 | { |
aaf9f12e SL |
979 | struct bio_list tmp = BIO_EMPTY_LIST; |
980 | struct r5pending_data *ent; | |
981 | ||
765d704d | 982 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
983 | ent = list_first_entry(&conf->free_list, struct r5pending_data, |
984 | sibling); | |
985 | list_move_tail(&ent->sibling, &conf->pending_list); | |
986 | ent->sector = sector; | |
987 | bio_list_init(&ent->bios); | |
988 | bio_list_merge(&ent->bios, bios); | |
989 | conf->pending_data_cnt++; | |
990 | if (conf->pending_data_cnt >= PENDING_IO_MAX) | |
991 | dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp); | |
992 | ||
765d704d | 993 | spin_unlock(&conf->pending_bios_lock); |
aaf9f12e SL |
994 | |
995 | dispatch_bio_list(&tmp); | |
765d704d SL |
996 | } |
997 | ||
6712ecf8 | 998 | static void |
4246a0b6 | 999 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 1000 | static void |
4246a0b6 | 1001 | raid5_end_write_request(struct bio *bi); |
91c00924 | 1002 | |
c4e5ac0a | 1003 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 1004 | { |
d1688a6d | 1005 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1006 | int i, disks = sh->disks; |
59fc630b | 1007 | struct stripe_head *head_sh = sh; |
aaf9f12e SL |
1008 | struct bio_list pending_bios = BIO_EMPTY_LIST; |
1009 | bool should_defer; | |
91c00924 DW |
1010 | |
1011 | might_sleep(); | |
1012 | ||
ff875738 AP |
1013 | if (log_stripe(sh, s) == 0) |
1014 | return; | |
1e6d690b | 1015 | |
aaf9f12e SL |
1016 | should_defer = conf->batch_bio_dispatch && conf->group_cnt; |
1017 | ||
91c00924 | 1018 | for (i = disks; i--; ) { |
796a5cf0 | 1019 | int op, op_flags = 0; |
9a3e1101 | 1020 | int replace_only = 0; |
977df362 N |
1021 | struct bio *bi, *rbi; |
1022 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 1023 | |
1024 | sh = head_sh; | |
e9c7469b | 1025 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 1026 | op = REQ_OP_WRITE; |
e9c7469b | 1027 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 1028 | op_flags = REQ_FUA; |
9e444768 | 1029 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 1030 | op = REQ_OP_DISCARD; |
e9c7469b | 1031 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 1032 | op = REQ_OP_READ; |
9a3e1101 N |
1033 | else if (test_and_clear_bit(R5_WantReplace, |
1034 | &sh->dev[i].flags)) { | |
796a5cf0 | 1035 | op = REQ_OP_WRITE; |
9a3e1101 N |
1036 | replace_only = 1; |
1037 | } else | |
91c00924 | 1038 | continue; |
bc0934f0 | 1039 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 1040 | op_flags |= REQ_SYNC; |
91c00924 | 1041 | |
59fc630b | 1042 | again: |
91c00924 | 1043 | bi = &sh->dev[i].req; |
977df362 | 1044 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 1045 | |
91c00924 | 1046 | rcu_read_lock(); |
9a3e1101 | 1047 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
1048 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
1049 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1050 | if (!rdev) { | |
1051 | rdev = rrdev; | |
1052 | rrdev = NULL; | |
1053 | } | |
796a5cf0 | 1054 | if (op_is_write(op)) { |
9a3e1101 N |
1055 | if (replace_only) |
1056 | rdev = NULL; | |
dd054fce N |
1057 | if (rdev == rrdev) |
1058 | /* We raced and saw duplicates */ | |
1059 | rrdev = NULL; | |
9a3e1101 | 1060 | } else { |
59fc630b | 1061 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
1062 | rdev = rrdev; |
1063 | rrdev = NULL; | |
1064 | } | |
977df362 | 1065 | |
91c00924 DW |
1066 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1067 | rdev = NULL; | |
1068 | if (rdev) | |
1069 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
1070 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
1071 | rrdev = NULL; | |
1072 | if (rrdev) | |
1073 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1074 | rcu_read_unlock(); |
1075 | ||
73e92e51 | 1076 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1077 | * need to check for writes. We never accept write errors |
1078 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1079 | */ |
796a5cf0 | 1080 | while (op_is_write(op) && rdev && |
73e92e51 N |
1081 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1082 | sector_t first_bad; | |
1083 | int bad_sectors; | |
1084 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
1085 | &first_bad, &bad_sectors); | |
1086 | if (!bad) | |
1087 | break; | |
1088 | ||
1089 | if (bad < 0) { | |
1090 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1091 | if (!conf->mddev->external && | |
2953079c | 1092 | conf->mddev->sb_flags) { |
73e92e51 N |
1093 | /* It is very unlikely, but we might |
1094 | * still need to write out the | |
1095 | * bad block log - better give it | |
1096 | * a chance*/ | |
1097 | md_check_recovery(conf->mddev); | |
1098 | } | |
1850753d | 1099 | /* |
1100 | * Because md_wait_for_blocked_rdev | |
1101 | * will dec nr_pending, we must | |
1102 | * increment it first. | |
1103 | */ | |
1104 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1105 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1106 | } else { | |
1107 | /* Acknowledged bad block - skip the write */ | |
1108 | rdev_dec_pending(rdev, conf->mddev); | |
1109 | rdev = NULL; | |
1110 | } | |
1111 | } | |
1112 | ||
91c00924 | 1113 | if (rdev) { |
9a3e1101 N |
1114 | if (s->syncing || s->expanding || s->expanded |
1115 | || s->replacing) | |
91c00924 DW |
1116 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1117 | ||
2b7497f0 DW |
1118 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1119 | ||
91c00924 | 1120 | bi->bi_bdev = rdev->bdev; |
796a5cf0 MC |
1121 | bio_set_op_attrs(bi, op, op_flags); |
1122 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1123 | ? raid5_end_write_request |
1124 | : raid5_end_read_request; | |
1125 | bi->bi_private = sh; | |
1126 | ||
6296b960 | 1127 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1128 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1129 | bi->bi_opf, i); |
91c00924 | 1130 | atomic_inc(&sh->count); |
59fc630b | 1131 | if (sh != head_sh) |
1132 | atomic_inc(&head_sh->count); | |
05616be5 | 1133 | if (use_new_offset(conf, sh)) |
4f024f37 | 1134 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1135 | + rdev->new_data_offset); |
1136 | else | |
4f024f37 | 1137 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1138 | + rdev->data_offset); |
59fc630b | 1139 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1140 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1141 | |
d592a996 SL |
1142 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1143 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1144 | |
1145 | if (!op_is_write(op) && | |
1146 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1147 | /* | |
1148 | * issuing read for a page in journal, this | |
1149 | * must be preparing for prexor in rmw; read | |
1150 | * the data into orig_page | |
1151 | */ | |
1152 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1153 | else | |
1154 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1155 | bi->bi_vcnt = 1; |
91c00924 DW |
1156 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1157 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1158 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1159 | /* |
1160 | * If this is discard request, set bi_vcnt 0. We don't | |
1161 | * want to confuse SCSI because SCSI will replace payload | |
1162 | */ | |
796a5cf0 | 1163 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1164 | bi->bi_vcnt = 0; |
977df362 N |
1165 | if (rrdev) |
1166 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1167 | |
1168 | if (conf->mddev->gendisk) | |
1169 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1170 | bi, disk_devt(conf->mddev->gendisk), | |
1171 | sh->dev[i].sector); | |
aaf9f12e SL |
1172 | if (should_defer && op_is_write(op)) |
1173 | bio_list_add(&pending_bios, bi); | |
1174 | else | |
1175 | generic_make_request(bi); | |
977df362 N |
1176 | } |
1177 | if (rrdev) { | |
9a3e1101 N |
1178 | if (s->syncing || s->expanding || s->expanded |
1179 | || s->replacing) | |
977df362 N |
1180 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1181 | ||
1182 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1183 | ||
1184 | rbi->bi_bdev = rrdev->bdev; | |
796a5cf0 MC |
1185 | bio_set_op_attrs(rbi, op, op_flags); |
1186 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1187 | rbi->bi_end_io = raid5_end_write_request; |
1188 | rbi->bi_private = sh; | |
1189 | ||
6296b960 | 1190 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1191 | "replacement disc %d\n", |
1192 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1193 | rbi->bi_opf, i); |
977df362 | 1194 | atomic_inc(&sh->count); |
59fc630b | 1195 | if (sh != head_sh) |
1196 | atomic_inc(&head_sh->count); | |
05616be5 | 1197 | if (use_new_offset(conf, sh)) |
4f024f37 | 1198 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1199 | + rrdev->new_data_offset); |
1200 | else | |
4f024f37 | 1201 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1202 | + rrdev->data_offset); |
d592a996 SL |
1203 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1204 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1205 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1206 | rbi->bi_vcnt = 1; |
977df362 N |
1207 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1208 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1209 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1210 | /* |
1211 | * If this is discard request, set bi_vcnt 0. We don't | |
1212 | * want to confuse SCSI because SCSI will replace payload | |
1213 | */ | |
796a5cf0 | 1214 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1215 | rbi->bi_vcnt = 0; |
e3620a3a JB |
1216 | if (conf->mddev->gendisk) |
1217 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1218 | rbi, disk_devt(conf->mddev->gendisk), | |
1219 | sh->dev[i].sector); | |
aaf9f12e SL |
1220 | if (should_defer && op_is_write(op)) |
1221 | bio_list_add(&pending_bios, rbi); | |
1222 | else | |
1223 | generic_make_request(rbi); | |
977df362 N |
1224 | } |
1225 | if (!rdev && !rrdev) { | |
796a5cf0 | 1226 | if (op_is_write(op)) |
91c00924 | 1227 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1228 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1229 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1230 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1231 | set_bit(STRIPE_HANDLE, &sh->state); | |
1232 | } | |
59fc630b | 1233 | |
1234 | if (!head_sh->batch_head) | |
1235 | continue; | |
1236 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1237 | batch_list); | |
1238 | if (sh != head_sh) | |
1239 | goto again; | |
91c00924 | 1240 | } |
aaf9f12e SL |
1241 | |
1242 | if (should_defer && !bio_list_empty(&pending_bios)) | |
1243 | defer_issue_bios(conf, head_sh->sector, &pending_bios); | |
91c00924 DW |
1244 | } |
1245 | ||
1246 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1247 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1248 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1249 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1250 | { |
7988613b KO |
1251 | struct bio_vec bvl; |
1252 | struct bvec_iter iter; | |
91c00924 | 1253 | struct page *bio_page; |
91c00924 | 1254 | int page_offset; |
a08abd8c | 1255 | struct async_submit_ctl submit; |
0403e382 | 1256 | enum async_tx_flags flags = 0; |
91c00924 | 1257 | |
4f024f37 KO |
1258 | if (bio->bi_iter.bi_sector >= sector) |
1259 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1260 | else |
4f024f37 | 1261 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1262 | |
0403e382 DW |
1263 | if (frombio) |
1264 | flags |= ASYNC_TX_FENCE; | |
1265 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1266 | ||
7988613b KO |
1267 | bio_for_each_segment(bvl, bio, iter) { |
1268 | int len = bvl.bv_len; | |
91c00924 DW |
1269 | int clen; |
1270 | int b_offset = 0; | |
1271 | ||
1272 | if (page_offset < 0) { | |
1273 | b_offset = -page_offset; | |
1274 | page_offset += b_offset; | |
1275 | len -= b_offset; | |
1276 | } | |
1277 | ||
1278 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1279 | clen = STRIPE_SIZE - page_offset; | |
1280 | else | |
1281 | clen = len; | |
1282 | ||
1283 | if (clen > 0) { | |
7988613b KO |
1284 | b_offset += bvl.bv_offset; |
1285 | bio_page = bvl.bv_page; | |
d592a996 SL |
1286 | if (frombio) { |
1287 | if (sh->raid_conf->skip_copy && | |
1288 | b_offset == 0 && page_offset == 0 && | |
1e6d690b SL |
1289 | clen == STRIPE_SIZE && |
1290 | !no_skipcopy) | |
d592a996 SL |
1291 | *page = bio_page; |
1292 | else | |
1293 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1294 | b_offset, clen, &submit); |
d592a996 SL |
1295 | } else |
1296 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1297 | page_offset, clen, &submit); |
91c00924 | 1298 | } |
a08abd8c DW |
1299 | /* chain the operations */ |
1300 | submit.depend_tx = tx; | |
1301 | ||
91c00924 DW |
1302 | if (clen < len) /* hit end of page */ |
1303 | break; | |
1304 | page_offset += len; | |
1305 | } | |
1306 | ||
1307 | return tx; | |
1308 | } | |
1309 | ||
1310 | static void ops_complete_biofill(void *stripe_head_ref) | |
1311 | { | |
1312 | struct stripe_head *sh = stripe_head_ref; | |
34a6f80e | 1313 | struct bio_list return_bi = BIO_EMPTY_LIST; |
e4d84909 | 1314 | int i; |
91c00924 | 1315 | |
e46b272b | 1316 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1317 | (unsigned long long)sh->sector); |
1318 | ||
1319 | /* clear completed biofills */ | |
1320 | for (i = sh->disks; i--; ) { | |
1321 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1322 | |
1323 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1324 | /* and check if we need to reply to a read request, |
1325 | * new R5_Wantfill requests are held off until | |
83de75cc | 1326 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1327 | */ |
1328 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1329 | struct bio *rbi, *rbi2; |
91c00924 | 1330 | |
91c00924 DW |
1331 | BUG_ON(!dev->read); |
1332 | rbi = dev->read; | |
1333 | dev->read = NULL; | |
4f024f37 | 1334 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1335 | dev->sector + STRIPE_SECTORS) { |
1336 | rbi2 = r5_next_bio(rbi, dev->sector); | |
34a6f80e N |
1337 | if (!raid5_dec_bi_active_stripes(rbi)) |
1338 | bio_list_add(&return_bi, rbi); | |
91c00924 DW |
1339 | rbi = rbi2; |
1340 | } | |
1341 | } | |
1342 | } | |
83de75cc | 1343 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1344 | |
34a6f80e | 1345 | return_io(&return_bi); |
91c00924 | 1346 | |
e4d84909 | 1347 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1348 | raid5_release_stripe(sh); |
91c00924 DW |
1349 | } |
1350 | ||
1351 | static void ops_run_biofill(struct stripe_head *sh) | |
1352 | { | |
1353 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1354 | struct async_submit_ctl submit; |
91c00924 DW |
1355 | int i; |
1356 | ||
59fc630b | 1357 | BUG_ON(sh->batch_head); |
e46b272b | 1358 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1359 | (unsigned long long)sh->sector); |
1360 | ||
1361 | for (i = sh->disks; i--; ) { | |
1362 | struct r5dev *dev = &sh->dev[i]; | |
1363 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1364 | struct bio *rbi; | |
b17459c0 | 1365 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1366 | dev->read = rbi = dev->toread; |
1367 | dev->toread = NULL; | |
b17459c0 | 1368 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1369 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1370 | dev->sector + STRIPE_SECTORS) { |
d592a996 | 1371 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1372 | dev->sector, tx, sh, 0); |
91c00924 DW |
1373 | rbi = r5_next_bio(rbi, dev->sector); |
1374 | } | |
1375 | } | |
1376 | } | |
1377 | ||
1378 | atomic_inc(&sh->count); | |
a08abd8c DW |
1379 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1380 | async_trigger_callback(&submit); | |
91c00924 DW |
1381 | } |
1382 | ||
4e7d2c0a | 1383 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1384 | { |
4e7d2c0a | 1385 | struct r5dev *tgt; |
91c00924 | 1386 | |
4e7d2c0a DW |
1387 | if (target < 0) |
1388 | return; | |
91c00924 | 1389 | |
4e7d2c0a | 1390 | tgt = &sh->dev[target]; |
91c00924 DW |
1391 | set_bit(R5_UPTODATE, &tgt->flags); |
1392 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1393 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1394 | } |
1395 | ||
ac6b53b6 | 1396 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1397 | { |
1398 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1399 | |
e46b272b | 1400 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1401 | (unsigned long long)sh->sector); |
1402 | ||
ac6b53b6 | 1403 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1404 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1405 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1406 | |
ecc65c9b DW |
1407 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1408 | if (sh->check_state == check_state_compute_run) | |
1409 | sh->check_state = check_state_compute_result; | |
91c00924 | 1410 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1411 | raid5_release_stripe(sh); |
91c00924 DW |
1412 | } |
1413 | ||
d6f38f31 DW |
1414 | /* return a pointer to the address conversion region of the scribble buffer */ |
1415 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1416 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1417 | { |
46d5b785 | 1418 | void *addr; |
1419 | ||
1420 | addr = flex_array_get(percpu->scribble, i); | |
1421 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1422 | } | |
1423 | ||
1424 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1425 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1426 | { | |
1427 | void *addr; | |
1428 | ||
1429 | addr = flex_array_get(percpu->scribble, i); | |
1430 | return addr; | |
d6f38f31 DW |
1431 | } |
1432 | ||
1433 | static struct dma_async_tx_descriptor * | |
1434 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1435 | { |
91c00924 | 1436 | int disks = sh->disks; |
46d5b785 | 1437 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1438 | int target = sh->ops.target; |
1439 | struct r5dev *tgt = &sh->dev[target]; | |
1440 | struct page *xor_dest = tgt->page; | |
1441 | int count = 0; | |
1442 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1443 | struct async_submit_ctl submit; |
91c00924 DW |
1444 | int i; |
1445 | ||
59fc630b | 1446 | BUG_ON(sh->batch_head); |
1447 | ||
91c00924 | 1448 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1449 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1450 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1451 | ||
1452 | for (i = disks; i--; ) | |
1453 | if (i != target) | |
1454 | xor_srcs[count++] = sh->dev[i].page; | |
1455 | ||
1456 | atomic_inc(&sh->count); | |
1457 | ||
0403e382 | 1458 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1459 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1460 | if (unlikely(count == 1)) |
a08abd8c | 1461 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1462 | else |
a08abd8c | 1463 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1464 | |
91c00924 DW |
1465 | return tx; |
1466 | } | |
1467 | ||
ac6b53b6 DW |
1468 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1469 | * @srcs - (struct page *) array of size sh->disks | |
1470 | * @sh - stripe_head to parse | |
1471 | * | |
1472 | * Populates srcs in proper layout order for the stripe and returns the | |
1473 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1474 | * destination buffer is recorded in srcs[count] and the Q destination | |
1475 | * is recorded in srcs[count+1]]. | |
1476 | */ | |
584acdd4 MS |
1477 | static int set_syndrome_sources(struct page **srcs, |
1478 | struct stripe_head *sh, | |
1479 | int srctype) | |
ac6b53b6 DW |
1480 | { |
1481 | int disks = sh->disks; | |
1482 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1483 | int d0_idx = raid6_d0(sh); | |
1484 | int count; | |
1485 | int i; | |
1486 | ||
1487 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1488 | srcs[i] = NULL; |
ac6b53b6 DW |
1489 | |
1490 | count = 0; | |
1491 | i = d0_idx; | |
1492 | do { | |
1493 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1494 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1495 | |
584acdd4 MS |
1496 | if (i == sh->qd_idx || i == sh->pd_idx || |
1497 | (srctype == SYNDROME_SRC_ALL) || | |
1498 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1499 | (test_bit(R5_Wantdrain, &dev->flags) || |
1500 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1501 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1502 | (dev->written || |
1503 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1504 | if (test_bit(R5_InJournal, &dev->flags)) |
1505 | srcs[slot] = sh->dev[i].orig_page; | |
1506 | else | |
1507 | srcs[slot] = sh->dev[i].page; | |
1508 | } | |
ac6b53b6 DW |
1509 | i = raid6_next_disk(i, disks); |
1510 | } while (i != d0_idx); | |
ac6b53b6 | 1511 | |
e4424fee | 1512 | return syndrome_disks; |
ac6b53b6 DW |
1513 | } |
1514 | ||
1515 | static struct dma_async_tx_descriptor * | |
1516 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1517 | { | |
1518 | int disks = sh->disks; | |
46d5b785 | 1519 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1520 | int target; |
1521 | int qd_idx = sh->qd_idx; | |
1522 | struct dma_async_tx_descriptor *tx; | |
1523 | struct async_submit_ctl submit; | |
1524 | struct r5dev *tgt; | |
1525 | struct page *dest; | |
1526 | int i; | |
1527 | int count; | |
1528 | ||
59fc630b | 1529 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1530 | if (sh->ops.target < 0) |
1531 | target = sh->ops.target2; | |
1532 | else if (sh->ops.target2 < 0) | |
1533 | target = sh->ops.target; | |
91c00924 | 1534 | else |
ac6b53b6 DW |
1535 | /* we should only have one valid target */ |
1536 | BUG(); | |
1537 | BUG_ON(target < 0); | |
1538 | pr_debug("%s: stripe %llu block: %d\n", | |
1539 | __func__, (unsigned long long)sh->sector, target); | |
1540 | ||
1541 | tgt = &sh->dev[target]; | |
1542 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1543 | dest = tgt->page; | |
1544 | ||
1545 | atomic_inc(&sh->count); | |
1546 | ||
1547 | if (target == qd_idx) { | |
584acdd4 | 1548 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1549 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1550 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1551 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1552 | ops_complete_compute, sh, | |
46d5b785 | 1553 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1554 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1555 | } else { | |
1556 | /* Compute any data- or p-drive using XOR */ | |
1557 | count = 0; | |
1558 | for (i = disks; i-- ; ) { | |
1559 | if (i == target || i == qd_idx) | |
1560 | continue; | |
1561 | blocks[count++] = sh->dev[i].page; | |
1562 | } | |
1563 | ||
0403e382 DW |
1564 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1565 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1566 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1567 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1568 | } | |
91c00924 | 1569 | |
91c00924 DW |
1570 | return tx; |
1571 | } | |
1572 | ||
ac6b53b6 DW |
1573 | static struct dma_async_tx_descriptor * |
1574 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1575 | { | |
1576 | int i, count, disks = sh->disks; | |
1577 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1578 | int d0_idx = raid6_d0(sh); | |
1579 | int faila = -1, failb = -1; | |
1580 | int target = sh->ops.target; | |
1581 | int target2 = sh->ops.target2; | |
1582 | struct r5dev *tgt = &sh->dev[target]; | |
1583 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1584 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1585 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1586 | struct async_submit_ctl submit; |
1587 | ||
59fc630b | 1588 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1589 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1590 | __func__, (unsigned long long)sh->sector, target, target2); | |
1591 | BUG_ON(target < 0 || target2 < 0); | |
1592 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1593 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1594 | ||
6c910a78 | 1595 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1596 | * slot number conversion for 'faila' and 'failb' |
1597 | */ | |
1598 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1599 | blocks[i] = NULL; |
ac6b53b6 DW |
1600 | count = 0; |
1601 | i = d0_idx; | |
1602 | do { | |
1603 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1604 | ||
1605 | blocks[slot] = sh->dev[i].page; | |
1606 | ||
1607 | if (i == target) | |
1608 | faila = slot; | |
1609 | if (i == target2) | |
1610 | failb = slot; | |
1611 | i = raid6_next_disk(i, disks); | |
1612 | } while (i != d0_idx); | |
ac6b53b6 DW |
1613 | |
1614 | BUG_ON(faila == failb); | |
1615 | if (failb < faila) | |
1616 | swap(faila, failb); | |
1617 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1618 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1619 | ||
1620 | atomic_inc(&sh->count); | |
1621 | ||
1622 | if (failb == syndrome_disks+1) { | |
1623 | /* Q disk is one of the missing disks */ | |
1624 | if (faila == syndrome_disks) { | |
1625 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1626 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1627 | ops_complete_compute, sh, | |
46d5b785 | 1628 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1629 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1630 | STRIPE_SIZE, &submit); |
1631 | } else { | |
1632 | struct page *dest; | |
1633 | int data_target; | |
1634 | int qd_idx = sh->qd_idx; | |
1635 | ||
1636 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1637 | if (target == qd_idx) | |
1638 | data_target = target2; | |
1639 | else | |
1640 | data_target = target; | |
1641 | ||
1642 | count = 0; | |
1643 | for (i = disks; i-- ; ) { | |
1644 | if (i == data_target || i == qd_idx) | |
1645 | continue; | |
1646 | blocks[count++] = sh->dev[i].page; | |
1647 | } | |
1648 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1649 | init_async_submit(&submit, |
1650 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1651 | NULL, NULL, NULL, | |
46d5b785 | 1652 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1653 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1654 | &submit); | |
1655 | ||
584acdd4 | 1656 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1657 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1658 | ops_complete_compute, sh, | |
46d5b785 | 1659 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1660 | return async_gen_syndrome(blocks, 0, count+2, |
1661 | STRIPE_SIZE, &submit); | |
1662 | } | |
ac6b53b6 | 1663 | } else { |
6c910a78 DW |
1664 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1665 | ops_complete_compute, sh, | |
46d5b785 | 1666 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1667 | if (failb == syndrome_disks) { |
1668 | /* We're missing D+P. */ | |
1669 | return async_raid6_datap_recov(syndrome_disks+2, | |
1670 | STRIPE_SIZE, faila, | |
1671 | blocks, &submit); | |
1672 | } else { | |
1673 | /* We're missing D+D. */ | |
1674 | return async_raid6_2data_recov(syndrome_disks+2, | |
1675 | STRIPE_SIZE, faila, failb, | |
1676 | blocks, &submit); | |
1677 | } | |
ac6b53b6 DW |
1678 | } |
1679 | } | |
1680 | ||
91c00924 DW |
1681 | static void ops_complete_prexor(void *stripe_head_ref) |
1682 | { | |
1683 | struct stripe_head *sh = stripe_head_ref; | |
1684 | ||
e46b272b | 1685 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1686 | (unsigned long long)sh->sector); |
1e6d690b SL |
1687 | |
1688 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1689 | /* | |
1690 | * raid5-cache write back uses orig_page during prexor. | |
1691 | * After prexor, it is time to free orig_page | |
1692 | */ | |
1693 | r5c_release_extra_page(sh); | |
91c00924 DW |
1694 | } |
1695 | ||
1696 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1697 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1698 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1699 | { |
91c00924 | 1700 | int disks = sh->disks; |
46d5b785 | 1701 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1702 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1703 | struct async_submit_ctl submit; |
91c00924 DW |
1704 | |
1705 | /* existing parity data subtracted */ | |
1706 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1707 | ||
59fc630b | 1708 | BUG_ON(sh->batch_head); |
e46b272b | 1709 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1710 | (unsigned long long)sh->sector); |
1711 | ||
1712 | for (i = disks; i--; ) { | |
1713 | struct r5dev *dev = &sh->dev[i]; | |
1714 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1715 | if (test_bit(R5_InJournal, &dev->flags)) |
1716 | xor_srcs[count++] = dev->orig_page; | |
1717 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1718 | xor_srcs[count++] = dev->page; |
1719 | } | |
1720 | ||
0403e382 | 1721 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1722 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1723 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1724 | |
1725 | return tx; | |
1726 | } | |
1727 | ||
584acdd4 MS |
1728 | static struct dma_async_tx_descriptor * |
1729 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1730 | struct dma_async_tx_descriptor *tx) | |
1731 | { | |
1732 | struct page **blocks = to_addr_page(percpu, 0); | |
1733 | int count; | |
1734 | struct async_submit_ctl submit; | |
1735 | ||
1736 | pr_debug("%s: stripe %llu\n", __func__, | |
1737 | (unsigned long long)sh->sector); | |
1738 | ||
1739 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1740 | ||
1741 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1742 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1743 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1744 | ||
1745 | return tx; | |
1746 | } | |
1747 | ||
91c00924 | 1748 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1749 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1750 | { |
1e6d690b | 1751 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1752 | int disks = sh->disks; |
d8ee0728 | 1753 | int i; |
59fc630b | 1754 | struct stripe_head *head_sh = sh; |
91c00924 | 1755 | |
e46b272b | 1756 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1757 | (unsigned long long)sh->sector); |
1758 | ||
1759 | for (i = disks; i--; ) { | |
59fc630b | 1760 | struct r5dev *dev; |
91c00924 | 1761 | struct bio *chosen; |
91c00924 | 1762 | |
59fc630b | 1763 | sh = head_sh; |
1764 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1765 | struct bio *wbi; |
1766 | ||
59fc630b | 1767 | again: |
1768 | dev = &sh->dev[i]; | |
1e6d690b SL |
1769 | /* |
1770 | * clear R5_InJournal, so when rewriting a page in | |
1771 | * journal, it is not skipped by r5l_log_stripe() | |
1772 | */ | |
1773 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1774 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1775 | chosen = dev->towrite; |
1776 | dev->towrite = NULL; | |
7a87f434 | 1777 | sh->overwrite_disks = 0; |
91c00924 DW |
1778 | BUG_ON(dev->written); |
1779 | wbi = dev->written = chosen; | |
b17459c0 | 1780 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1781 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1782 | |
4f024f37 | 1783 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1784 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1785 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1786 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1787 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1788 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1789 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1790 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1791 | else { |
1792 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1793 | dev->sector, tx, sh, |
1794 | r5c_is_writeback(conf->log)); | |
1795 | if (dev->page != dev->orig_page && | |
1796 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1797 | set_bit(R5_SkipCopy, &dev->flags); |
1798 | clear_bit(R5_UPTODATE, &dev->flags); | |
1799 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1800 | } | |
1801 | } | |
91c00924 DW |
1802 | wbi = r5_next_bio(wbi, dev->sector); |
1803 | } | |
59fc630b | 1804 | |
1805 | if (head_sh->batch_head) { | |
1806 | sh = list_first_entry(&sh->batch_list, | |
1807 | struct stripe_head, | |
1808 | batch_list); | |
1809 | if (sh == head_sh) | |
1810 | continue; | |
1811 | goto again; | |
1812 | } | |
91c00924 DW |
1813 | } |
1814 | } | |
1815 | ||
1816 | return tx; | |
1817 | } | |
1818 | ||
ac6b53b6 | 1819 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1820 | { |
1821 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1822 | int disks = sh->disks; |
1823 | int pd_idx = sh->pd_idx; | |
1824 | int qd_idx = sh->qd_idx; | |
1825 | int i; | |
9e444768 | 1826 | bool fua = false, sync = false, discard = false; |
91c00924 | 1827 | |
e46b272b | 1828 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1829 | (unsigned long long)sh->sector); |
1830 | ||
bc0934f0 | 1831 | for (i = disks; i--; ) { |
e9c7469b | 1832 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1833 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1834 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1835 | } |
e9c7469b | 1836 | |
91c00924 DW |
1837 | for (i = disks; i--; ) { |
1838 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1839 | |
e9c7469b | 1840 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1841 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1842 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1843 | if (fua) |
1844 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1845 | if (sync) |
1846 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1847 | } |
91c00924 DW |
1848 | } |
1849 | ||
d8ee0728 DW |
1850 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1851 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1852 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1853 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1854 | else { | |
1855 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1856 | sh->reconstruct_state = reconstruct_state_result; | |
1857 | } | |
91c00924 DW |
1858 | |
1859 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1860 | raid5_release_stripe(sh); |
91c00924 DW |
1861 | } |
1862 | ||
1863 | static void | |
ac6b53b6 DW |
1864 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1865 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1866 | { |
91c00924 | 1867 | int disks = sh->disks; |
59fc630b | 1868 | struct page **xor_srcs; |
a08abd8c | 1869 | struct async_submit_ctl submit; |
59fc630b | 1870 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1871 | struct page *xor_dest; |
d8ee0728 | 1872 | int prexor = 0; |
91c00924 | 1873 | unsigned long flags; |
59fc630b | 1874 | int j = 0; |
1875 | struct stripe_head *head_sh = sh; | |
1876 | int last_stripe; | |
91c00924 | 1877 | |
e46b272b | 1878 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1879 | (unsigned long long)sh->sector); |
1880 | ||
620125f2 SL |
1881 | for (i = 0; i < sh->disks; i++) { |
1882 | if (pd_idx == i) | |
1883 | continue; | |
1884 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1885 | break; | |
1886 | } | |
1887 | if (i >= sh->disks) { | |
1888 | atomic_inc(&sh->count); | |
620125f2 SL |
1889 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1890 | ops_complete_reconstruct(sh); | |
1891 | return; | |
1892 | } | |
59fc630b | 1893 | again: |
1894 | count = 0; | |
1895 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1896 | /* check if prexor is active which means only process blocks |
1897 | * that are part of a read-modify-write (written) | |
1898 | */ | |
59fc630b | 1899 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1900 | prexor = 1; |
91c00924 DW |
1901 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1902 | for (i = disks; i--; ) { | |
1903 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1904 | if (head_sh->dev[i].written || |
1905 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1906 | xor_srcs[count++] = dev->page; |
1907 | } | |
1908 | } else { | |
1909 | xor_dest = sh->dev[pd_idx].page; | |
1910 | for (i = disks; i--; ) { | |
1911 | struct r5dev *dev = &sh->dev[i]; | |
1912 | if (i != pd_idx) | |
1913 | xor_srcs[count++] = dev->page; | |
1914 | } | |
1915 | } | |
1916 | ||
91c00924 DW |
1917 | /* 1/ if we prexor'd then the dest is reused as a source |
1918 | * 2/ if we did not prexor then we are redoing the parity | |
1919 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1920 | * for the synchronous xor case | |
1921 | */ | |
59fc630b | 1922 | last_stripe = !head_sh->batch_head || |
1923 | list_first_entry(&sh->batch_list, | |
1924 | struct stripe_head, batch_list) == head_sh; | |
1925 | if (last_stripe) { | |
1926 | flags = ASYNC_TX_ACK | | |
1927 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1928 | ||
1929 | atomic_inc(&head_sh->count); | |
1930 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1931 | to_addr_conv(sh, percpu, j)); | |
1932 | } else { | |
1933 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1934 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1935 | to_addr_conv(sh, percpu, j)); | |
1936 | } | |
91c00924 | 1937 | |
a08abd8c DW |
1938 | if (unlikely(count == 1)) |
1939 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1940 | else | |
1941 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1942 | if (!last_stripe) { |
1943 | j++; | |
1944 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1945 | batch_list); | |
1946 | goto again; | |
1947 | } | |
91c00924 DW |
1948 | } |
1949 | ||
ac6b53b6 DW |
1950 | static void |
1951 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1952 | struct dma_async_tx_descriptor *tx) | |
1953 | { | |
1954 | struct async_submit_ctl submit; | |
59fc630b | 1955 | struct page **blocks; |
1956 | int count, i, j = 0; | |
1957 | struct stripe_head *head_sh = sh; | |
1958 | int last_stripe; | |
584acdd4 MS |
1959 | int synflags; |
1960 | unsigned long txflags; | |
ac6b53b6 DW |
1961 | |
1962 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1963 | ||
620125f2 SL |
1964 | for (i = 0; i < sh->disks; i++) { |
1965 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1966 | continue; | |
1967 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1968 | break; | |
1969 | } | |
1970 | if (i >= sh->disks) { | |
1971 | atomic_inc(&sh->count); | |
620125f2 SL |
1972 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1973 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1974 | ops_complete_reconstruct(sh); | |
1975 | return; | |
1976 | } | |
1977 | ||
59fc630b | 1978 | again: |
1979 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1980 | |
1981 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1982 | synflags = SYNDROME_SRC_WRITTEN; | |
1983 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1984 | } else { | |
1985 | synflags = SYNDROME_SRC_ALL; | |
1986 | txflags = ASYNC_TX_ACK; | |
1987 | } | |
1988 | ||
1989 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1990 | last_stripe = !head_sh->batch_head || |
1991 | list_first_entry(&sh->batch_list, | |
1992 | struct stripe_head, batch_list) == head_sh; | |
1993 | ||
1994 | if (last_stripe) { | |
1995 | atomic_inc(&head_sh->count); | |
584acdd4 | 1996 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1997 | head_sh, to_addr_conv(sh, percpu, j)); |
1998 | } else | |
1999 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
2000 | to_addr_conv(sh, percpu, j)); | |
48769695 | 2001 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 2002 | if (!last_stripe) { |
2003 | j++; | |
2004 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
2005 | batch_list); | |
2006 | goto again; | |
2007 | } | |
91c00924 DW |
2008 | } |
2009 | ||
2010 | static void ops_complete_check(void *stripe_head_ref) | |
2011 | { | |
2012 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 2013 | |
e46b272b | 2014 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2015 | (unsigned long long)sh->sector); |
2016 | ||
ecc65c9b | 2017 | sh->check_state = check_state_check_result; |
91c00924 | 2018 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2019 | raid5_release_stripe(sh); |
91c00924 DW |
2020 | } |
2021 | ||
ac6b53b6 | 2022 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 2023 | { |
91c00924 | 2024 | int disks = sh->disks; |
ac6b53b6 DW |
2025 | int pd_idx = sh->pd_idx; |
2026 | int qd_idx = sh->qd_idx; | |
2027 | struct page *xor_dest; | |
46d5b785 | 2028 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 2029 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 2030 | struct async_submit_ctl submit; |
ac6b53b6 DW |
2031 | int count; |
2032 | int i; | |
91c00924 | 2033 | |
e46b272b | 2034 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2035 | (unsigned long long)sh->sector); |
2036 | ||
59fc630b | 2037 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
2038 | count = 0; |
2039 | xor_dest = sh->dev[pd_idx].page; | |
2040 | xor_srcs[count++] = xor_dest; | |
91c00924 | 2041 | for (i = disks; i--; ) { |
ac6b53b6 DW |
2042 | if (i == pd_idx || i == qd_idx) |
2043 | continue; | |
2044 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
2045 | } |
2046 | ||
d6f38f31 | 2047 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 2048 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 2049 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 2050 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 2051 | |
91c00924 | 2052 | atomic_inc(&sh->count); |
a08abd8c DW |
2053 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
2054 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
2055 | } |
2056 | ||
ac6b53b6 DW |
2057 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
2058 | { | |
46d5b785 | 2059 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
2060 | struct async_submit_ctl submit; |
2061 | int count; | |
2062 | ||
2063 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
2064 | (unsigned long long)sh->sector, checkp); | |
2065 | ||
59fc630b | 2066 | BUG_ON(sh->batch_head); |
584acdd4 | 2067 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
2068 | if (!checkp) |
2069 | srcs[count] = NULL; | |
91c00924 | 2070 | |
91c00924 | 2071 | atomic_inc(&sh->count); |
ac6b53b6 | 2072 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 2073 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
2074 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
2075 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
2076 | } |
2077 | ||
51acbcec | 2078 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
2079 | { |
2080 | int overlap_clear = 0, i, disks = sh->disks; | |
2081 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 2082 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2083 | int level = conf->level; |
d6f38f31 DW |
2084 | struct raid5_percpu *percpu; |
2085 | unsigned long cpu; | |
91c00924 | 2086 | |
d6f38f31 DW |
2087 | cpu = get_cpu(); |
2088 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2089 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2090 | ops_run_biofill(sh); |
2091 | overlap_clear++; | |
2092 | } | |
2093 | ||
7b3a871e | 2094 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2095 | if (level < 6) |
2096 | tx = ops_run_compute5(sh, percpu); | |
2097 | else { | |
2098 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2099 | tx = ops_run_compute6_1(sh, percpu); | |
2100 | else | |
2101 | tx = ops_run_compute6_2(sh, percpu); | |
2102 | } | |
2103 | /* terminate the chain if reconstruct is not set to be run */ | |
2104 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2105 | async_tx_ack(tx); |
2106 | } | |
91c00924 | 2107 | |
3418d036 AP |
2108 | if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request)) |
2109 | tx = ops_run_partial_parity(sh, percpu, tx); | |
2110 | ||
584acdd4 MS |
2111 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2112 | if (level < 6) | |
2113 | tx = ops_run_prexor5(sh, percpu, tx); | |
2114 | else | |
2115 | tx = ops_run_prexor6(sh, percpu, tx); | |
2116 | } | |
91c00924 | 2117 | |
600aa109 | 2118 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2119 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2120 | overlap_clear++; |
2121 | } | |
2122 | ||
ac6b53b6 DW |
2123 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2124 | if (level < 6) | |
2125 | ops_run_reconstruct5(sh, percpu, tx); | |
2126 | else | |
2127 | ops_run_reconstruct6(sh, percpu, tx); | |
2128 | } | |
91c00924 | 2129 | |
ac6b53b6 DW |
2130 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2131 | if (sh->check_state == check_state_run) | |
2132 | ops_run_check_p(sh, percpu); | |
2133 | else if (sh->check_state == check_state_run_q) | |
2134 | ops_run_check_pq(sh, percpu, 0); | |
2135 | else if (sh->check_state == check_state_run_pq) | |
2136 | ops_run_check_pq(sh, percpu, 1); | |
2137 | else | |
2138 | BUG(); | |
2139 | } | |
91c00924 | 2140 | |
59fc630b | 2141 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2142 | for (i = disks; i--; ) { |
2143 | struct r5dev *dev = &sh->dev[i]; | |
2144 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2145 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2146 | } | |
d6f38f31 | 2147 | put_cpu(); |
91c00924 DW |
2148 | } |
2149 | ||
5f9d1fde SL |
2150 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
2151 | int disks) | |
f18c1a35 N |
2152 | { |
2153 | struct stripe_head *sh; | |
5f9d1fde | 2154 | int i; |
f18c1a35 N |
2155 | |
2156 | sh = kmem_cache_zalloc(sc, gfp); | |
2157 | if (sh) { | |
2158 | spin_lock_init(&sh->stripe_lock); | |
2159 | spin_lock_init(&sh->batch_lock); | |
2160 | INIT_LIST_HEAD(&sh->batch_list); | |
2161 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2162 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2163 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2164 | atomic_set(&sh->count, 1); |
a39f7afd | 2165 | sh->log_start = MaxSector; |
5f9d1fde SL |
2166 | for (i = 0; i < disks; i++) { |
2167 | struct r5dev *dev = &sh->dev[i]; | |
2168 | ||
3a83f467 ML |
2169 | bio_init(&dev->req, &dev->vec, 1); |
2170 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2171 | } |
f18c1a35 N |
2172 | } |
2173 | return sh; | |
2174 | } | |
486f0644 | 2175 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2176 | { |
2177 | struct stripe_head *sh; | |
f18c1a35 | 2178 | |
5f9d1fde | 2179 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size); |
3f294f4f N |
2180 | if (!sh) |
2181 | return 0; | |
6ce32846 | 2182 | |
3f294f4f | 2183 | sh->raid_conf = conf; |
3f294f4f | 2184 | |
a9683a79 | 2185 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2186 | shrink_buffers(sh); |
3f294f4f N |
2187 | kmem_cache_free(conf->slab_cache, sh); |
2188 | return 0; | |
2189 | } | |
486f0644 N |
2190 | sh->hash_lock_index = |
2191 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2192 | /* we just created an active stripe so... */ |
3f294f4f | 2193 | atomic_inc(&conf->active_stripes); |
59fc630b | 2194 | |
6d036f7d | 2195 | raid5_release_stripe(sh); |
486f0644 | 2196 | conf->max_nr_stripes++; |
3f294f4f N |
2197 | return 1; |
2198 | } | |
2199 | ||
d1688a6d | 2200 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2201 | { |
e18b890b | 2202 | struct kmem_cache *sc; |
5e5e3e78 | 2203 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2204 | |
f4be6b43 N |
2205 | if (conf->mddev->gendisk) |
2206 | sprintf(conf->cache_name[0], | |
2207 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2208 | else | |
2209 | sprintf(conf->cache_name[0], | |
2210 | "raid%d-%p", conf->level, conf->mddev); | |
2211 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2212 | ||
ad01c9e3 N |
2213 | conf->active_name = 0; |
2214 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2215 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2216 | 0, 0, NULL); |
1da177e4 LT |
2217 | if (!sc) |
2218 | return 1; | |
2219 | conf->slab_cache = sc; | |
ad01c9e3 | 2220 | conf->pool_size = devs; |
486f0644 N |
2221 | while (num--) |
2222 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2223 | return 1; |
486f0644 | 2224 | |
1da177e4 LT |
2225 | return 0; |
2226 | } | |
29269553 | 2227 | |
d6f38f31 DW |
2228 | /** |
2229 | * scribble_len - return the required size of the scribble region | |
2230 | * @num - total number of disks in the array | |
2231 | * | |
2232 | * The size must be enough to contain: | |
2233 | * 1/ a struct page pointer for each device in the array +2 | |
2234 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2235 | * (dma_map_page()) or page (page_address()) address. | |
2236 | * | |
2237 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2238 | * calculate over all devices (not just the data blocks), using zeros in place | |
2239 | * of the P and Q blocks. | |
2240 | */ | |
46d5b785 | 2241 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2242 | { |
46d5b785 | 2243 | struct flex_array *ret; |
d6f38f31 DW |
2244 | size_t len; |
2245 | ||
2246 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2247 | ret = flex_array_alloc(len, cnt, flags); |
2248 | if (!ret) | |
2249 | return NULL; | |
2250 | /* always prealloc all elements, so no locking is required */ | |
2251 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2252 | flex_array_free(ret); | |
2253 | return NULL; | |
2254 | } | |
2255 | return ret; | |
d6f38f31 DW |
2256 | } |
2257 | ||
738a2738 N |
2258 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2259 | { | |
2260 | unsigned long cpu; | |
2261 | int err = 0; | |
2262 | ||
27a353c0 SL |
2263 | /* |
2264 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2265 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2266 | * should equal to new_disks and new_sectors | |
2267 | */ | |
2268 | if (conf->scribble_disks >= new_disks && | |
2269 | conf->scribble_sectors >= new_sectors) | |
2270 | return 0; | |
738a2738 N |
2271 | mddev_suspend(conf->mddev); |
2272 | get_online_cpus(); | |
2273 | for_each_present_cpu(cpu) { | |
2274 | struct raid5_percpu *percpu; | |
2275 | struct flex_array *scribble; | |
2276 | ||
2277 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2278 | scribble = scribble_alloc(new_disks, | |
2279 | new_sectors / STRIPE_SECTORS, | |
2280 | GFP_NOIO); | |
2281 | ||
2282 | if (scribble) { | |
2283 | flex_array_free(percpu->scribble); | |
2284 | percpu->scribble = scribble; | |
2285 | } else { | |
2286 | err = -ENOMEM; | |
2287 | break; | |
2288 | } | |
2289 | } | |
2290 | put_online_cpus(); | |
2291 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2292 | if (!err) { |
2293 | conf->scribble_disks = new_disks; | |
2294 | conf->scribble_sectors = new_sectors; | |
2295 | } | |
738a2738 N |
2296 | return err; |
2297 | } | |
2298 | ||
d1688a6d | 2299 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2300 | { |
2301 | /* Make all the stripes able to hold 'newsize' devices. | |
2302 | * New slots in each stripe get 'page' set to a new page. | |
2303 | * | |
2304 | * This happens in stages: | |
2305 | * 1/ create a new kmem_cache and allocate the required number of | |
2306 | * stripe_heads. | |
83f0d77a | 2307 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2308 | * to the new stripe_heads. This will have the side effect of |
2309 | * freezing the array as once all stripe_heads have been collected, | |
2310 | * no IO will be possible. Old stripe heads are freed once their | |
2311 | * pages have been transferred over, and the old kmem_cache is | |
2312 | * freed when all stripes are done. | |
2313 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2314 | * we simple return a failre status - no need to clean anything up. | |
2315 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2316 | * If this fails, we don't bother trying the shrink the | |
2317 | * stripe_heads down again, we just leave them as they are. | |
2318 | * As each stripe_head is processed the new one is released into | |
2319 | * active service. | |
2320 | * | |
2321 | * Once step2 is started, we cannot afford to wait for a write, | |
2322 | * so we use GFP_NOIO allocations. | |
2323 | */ | |
2324 | struct stripe_head *osh, *nsh; | |
2325 | LIST_HEAD(newstripes); | |
2326 | struct disk_info *ndisks; | |
b5470dc5 | 2327 | int err; |
e18b890b | 2328 | struct kmem_cache *sc; |
ad01c9e3 | 2329 | int i; |
566c09c5 | 2330 | int hash, cnt; |
ad01c9e3 N |
2331 | |
2332 | if (newsize <= conf->pool_size) | |
2333 | return 0; /* never bother to shrink */ | |
2334 | ||
b5470dc5 DW |
2335 | err = md_allow_write(conf->mddev); |
2336 | if (err) | |
2337 | return err; | |
2a2275d6 | 2338 | |
ad01c9e3 N |
2339 | /* Step 1 */ |
2340 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2341 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2342 | 0, 0, NULL); |
ad01c9e3 N |
2343 | if (!sc) |
2344 | return -ENOMEM; | |
2345 | ||
2d5b569b N |
2346 | /* Need to ensure auto-resizing doesn't interfere */ |
2347 | mutex_lock(&conf->cache_size_mutex); | |
2348 | ||
ad01c9e3 | 2349 | for (i = conf->max_nr_stripes; i; i--) { |
5f9d1fde | 2350 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize); |
ad01c9e3 N |
2351 | if (!nsh) |
2352 | break; | |
2353 | ||
ad01c9e3 | 2354 | nsh->raid_conf = conf; |
ad01c9e3 N |
2355 | list_add(&nsh->lru, &newstripes); |
2356 | } | |
2357 | if (i) { | |
2358 | /* didn't get enough, give up */ | |
2359 | while (!list_empty(&newstripes)) { | |
2360 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2361 | list_del(&nsh->lru); | |
2362 | kmem_cache_free(sc, nsh); | |
2363 | } | |
2364 | kmem_cache_destroy(sc); | |
2d5b569b | 2365 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2366 | return -ENOMEM; |
2367 | } | |
2368 | /* Step 2 - Must use GFP_NOIO now. | |
2369 | * OK, we have enough stripes, start collecting inactive | |
2370 | * stripes and copying them over | |
2371 | */ | |
566c09c5 SL |
2372 | hash = 0; |
2373 | cnt = 0; | |
ad01c9e3 | 2374 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2375 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2376 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2377 | !list_empty(conf->inactive_list + hash), |
2378 | unlock_device_hash_lock(conf, hash), | |
2379 | lock_device_hash_lock(conf, hash)); | |
2380 | osh = get_free_stripe(conf, hash); | |
2381 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2382 | |
d592a996 | 2383 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2384 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2385 | nsh->dev[i].orig_page = osh->dev[i].page; |
2386 | } | |
566c09c5 | 2387 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2388 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2389 | cnt++; |
2390 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2391 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2392 | hash++; | |
2393 | cnt = 0; | |
2394 | } | |
ad01c9e3 N |
2395 | } |
2396 | kmem_cache_destroy(conf->slab_cache); | |
2397 | ||
2398 | /* Step 3. | |
2399 | * At this point, we are holding all the stripes so the array | |
2400 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2401 | * conf->disks and the scribble region |
ad01c9e3 N |
2402 | */ |
2403 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2404 | if (ndisks) { | |
d7bd398e | 2405 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2406 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2407 | |
2408 | for (i = conf->pool_size; i < newsize; i++) { | |
2409 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2410 | if (!ndisks[i].extra_page) | |
2411 | err = -ENOMEM; | |
2412 | } | |
2413 | ||
2414 | if (err) { | |
2415 | for (i = conf->pool_size; i < newsize; i++) | |
2416 | if (ndisks[i].extra_page) | |
2417 | put_page(ndisks[i].extra_page); | |
2418 | kfree(ndisks); | |
2419 | } else { | |
2420 | kfree(conf->disks); | |
2421 | conf->disks = ndisks; | |
2422 | } | |
ad01c9e3 N |
2423 | } else |
2424 | err = -ENOMEM; | |
2425 | ||
2d5b569b | 2426 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2427 | /* Step 4, return new stripes to service */ |
2428 | while(!list_empty(&newstripes)) { | |
2429 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2430 | list_del_init(&nsh->lru); | |
d6f38f31 | 2431 | |
ad01c9e3 N |
2432 | for (i=conf->raid_disks; i < newsize; i++) |
2433 | if (nsh->dev[i].page == NULL) { | |
2434 | struct page *p = alloc_page(GFP_NOIO); | |
2435 | nsh->dev[i].page = p; | |
d592a996 | 2436 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2437 | if (!p) |
2438 | err = -ENOMEM; | |
2439 | } | |
6d036f7d | 2440 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2441 | } |
2442 | /* critical section pass, GFP_NOIO no longer needed */ | |
2443 | ||
2444 | conf->slab_cache = sc; | |
2445 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2446 | if (!err) |
2447 | conf->pool_size = newsize; | |
ad01c9e3 N |
2448 | return err; |
2449 | } | |
1da177e4 | 2450 | |
486f0644 | 2451 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2452 | { |
2453 | struct stripe_head *sh; | |
49895bcc | 2454 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2455 | |
566c09c5 SL |
2456 | spin_lock_irq(conf->hash_locks + hash); |
2457 | sh = get_free_stripe(conf, hash); | |
2458 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2459 | if (!sh) |
2460 | return 0; | |
78bafebd | 2461 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2462 | shrink_buffers(sh); |
3f294f4f N |
2463 | kmem_cache_free(conf->slab_cache, sh); |
2464 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2465 | conf->max_nr_stripes--; |
3f294f4f N |
2466 | return 1; |
2467 | } | |
2468 | ||
d1688a6d | 2469 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2470 | { |
486f0644 N |
2471 | while (conf->max_nr_stripes && |
2472 | drop_one_stripe(conf)) | |
2473 | ; | |
3f294f4f | 2474 | |
644df1a8 | 2475 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2476 | conf->slab_cache = NULL; |
2477 | } | |
2478 | ||
4246a0b6 | 2479 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2480 | { |
99c0fb5f | 2481 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2482 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2483 | int disks = sh->disks, i; |
d6950432 | 2484 | char b[BDEVNAME_SIZE]; |
dd054fce | 2485 | struct md_rdev *rdev = NULL; |
05616be5 | 2486 | sector_t s; |
1da177e4 LT |
2487 | |
2488 | for (i=0 ; i<disks; i++) | |
2489 | if (bi == &sh->dev[i].req) | |
2490 | break; | |
2491 | ||
4246a0b6 | 2492 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2493 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2494 | bi->bi_error); |
1da177e4 | 2495 | if (i == disks) { |
5f9d1fde | 2496 | bio_reset(bi); |
1da177e4 | 2497 | BUG(); |
6712ecf8 | 2498 | return; |
1da177e4 | 2499 | } |
14a75d3e | 2500 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2501 | /* If replacement finished while this request was outstanding, |
2502 | * 'replacement' might be NULL already. | |
2503 | * In that case it moved down to 'rdev'. | |
2504 | * rdev is not removed until all requests are finished. | |
2505 | */ | |
14a75d3e | 2506 | rdev = conf->disks[i].replacement; |
dd054fce | 2507 | if (!rdev) |
14a75d3e | 2508 | rdev = conf->disks[i].rdev; |
1da177e4 | 2509 | |
05616be5 N |
2510 | if (use_new_offset(conf, sh)) |
2511 | s = sh->sector + rdev->new_data_offset; | |
2512 | else | |
2513 | s = sh->sector + rdev->data_offset; | |
4246a0b6 | 2514 | if (!bi->bi_error) { |
1da177e4 | 2515 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2516 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2517 | /* Note that this cannot happen on a |
2518 | * replacement device. We just fail those on | |
2519 | * any error | |
2520 | */ | |
cc6167b4 N |
2521 | pr_info_ratelimited( |
2522 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
8bda470e | 2523 | mdname(conf->mddev), STRIPE_SECTORS, |
05616be5 | 2524 | (unsigned long long)s, |
8bda470e | 2525 | bdevname(rdev->bdev, b)); |
ddd5115f | 2526 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2527 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2528 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2529 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2530 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2531 | ||
86aa1397 SL |
2532 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2533 | /* | |
2534 | * end read for a page in journal, this | |
2535 | * must be preparing for prexor in rmw | |
2536 | */ | |
2537 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2538 | ||
14a75d3e N |
2539 | if (atomic_read(&rdev->read_errors)) |
2540 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2541 | } else { |
14a75d3e | 2542 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2543 | int retry = 0; |
2e8ac303 | 2544 | int set_bad = 0; |
d6950432 | 2545 | |
1da177e4 | 2546 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2547 | atomic_inc(&rdev->read_errors); |
14a75d3e | 2548 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2549 | pr_warn_ratelimited( |
2550 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2551 | mdname(conf->mddev), |
05616be5 | 2552 | (unsigned long long)s, |
14a75d3e | 2553 | bdn); |
2e8ac303 | 2554 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2555 | set_bad = 1; | |
cc6167b4 N |
2556 | pr_warn_ratelimited( |
2557 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2558 | mdname(conf->mddev), |
05616be5 | 2559 | (unsigned long long)s, |
8bda470e | 2560 | bdn); |
2e8ac303 | 2561 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2562 | /* Oh, no!!! */ |
2e8ac303 | 2563 | set_bad = 1; |
cc6167b4 N |
2564 | pr_warn_ratelimited( |
2565 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2566 | mdname(conf->mddev), |
05616be5 | 2567 | (unsigned long long)s, |
8bda470e | 2568 | bdn); |
2e8ac303 | 2569 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2570 | > conf->max_nr_stripes) |
cc6167b4 | 2571 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2572 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2573 | else |
2574 | retry = 1; | |
edfa1f65 BY |
2575 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2576 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2577 | retry = 1; | |
ba22dcbf | 2578 | if (retry) |
3f9e7c14 | 2579 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2580 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2581 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2582 | } else | |
2583 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2584 | else { |
4e5314b5 N |
2585 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2586 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2587 | if (!(set_bad |
2588 | && test_bit(In_sync, &rdev->flags) | |
2589 | && rdev_set_badblocks( | |
2590 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2591 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2592 | } |
1da177e4 | 2593 | } |
14a75d3e | 2594 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2595 | bio_reset(bi); |
1da177e4 LT |
2596 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2597 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2598 | raid5_release_stripe(sh); |
1da177e4 LT |
2599 | } |
2600 | ||
4246a0b6 | 2601 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2602 | { |
99c0fb5f | 2603 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2604 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2605 | int disks = sh->disks, i; |
977df362 | 2606 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2607 | sector_t first_bad; |
2608 | int bad_sectors; | |
977df362 | 2609 | int replacement = 0; |
1da177e4 | 2610 | |
977df362 N |
2611 | for (i = 0 ; i < disks; i++) { |
2612 | if (bi == &sh->dev[i].req) { | |
2613 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2614 | break; |
977df362 N |
2615 | } |
2616 | if (bi == &sh->dev[i].rreq) { | |
2617 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2618 | if (rdev) |
2619 | replacement = 1; | |
2620 | else | |
2621 | /* rdev was removed and 'replacement' | |
2622 | * replaced it. rdev is not removed | |
2623 | * until all requests are finished. | |
2624 | */ | |
2625 | rdev = conf->disks[i].rdev; | |
977df362 N |
2626 | break; |
2627 | } | |
2628 | } | |
4246a0b6 | 2629 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2630 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2631 | bi->bi_error); |
1da177e4 | 2632 | if (i == disks) { |
5f9d1fde | 2633 | bio_reset(bi); |
1da177e4 | 2634 | BUG(); |
6712ecf8 | 2635 | return; |
1da177e4 LT |
2636 | } |
2637 | ||
977df362 | 2638 | if (replacement) { |
4246a0b6 | 2639 | if (bi->bi_error) |
977df362 N |
2640 | md_error(conf->mddev, rdev); |
2641 | else if (is_badblock(rdev, sh->sector, | |
2642 | STRIPE_SECTORS, | |
2643 | &first_bad, &bad_sectors)) | |
2644 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2645 | } else { | |
4246a0b6 | 2646 | if (bi->bi_error) { |
9f97e4b1 | 2647 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2648 | set_bit(WriteErrorSeen, &rdev->flags); |
2649 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2650 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2651 | set_bit(MD_RECOVERY_NEEDED, | |
2652 | &rdev->mddev->recovery); | |
977df362 N |
2653 | } else if (is_badblock(rdev, sh->sector, |
2654 | STRIPE_SECTORS, | |
c0b32972 | 2655 | &first_bad, &bad_sectors)) { |
977df362 | 2656 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2657 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2658 | /* That was a successful write so make | |
2659 | * sure it looks like we already did | |
2660 | * a re-write. | |
2661 | */ | |
2662 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2663 | } | |
977df362 N |
2664 | } |
2665 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2666 | |
4246a0b6 | 2667 | if (sh->batch_head && bi->bi_error && !replacement) |
72ac7330 | 2668 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2669 | ||
c9445555 | 2670 | bio_reset(bi); |
977df362 N |
2671 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2672 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2673 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2674 | raid5_release_stripe(sh); |
59fc630b | 2675 | |
2676 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2677 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2678 | } |
2679 | ||
784052ec | 2680 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2681 | { |
2682 | struct r5dev *dev = &sh->dev[i]; | |
2683 | ||
1da177e4 | 2684 | dev->flags = 0; |
6d036f7d | 2685 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 LT |
2686 | } |
2687 | ||
849674e4 | 2688 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2689 | { |
2690 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2691 | struct r5conf *conf = mddev->private; |
908f4fbd | 2692 | unsigned long flags; |
0c55e022 | 2693 | pr_debug("raid456: error called\n"); |
1da177e4 | 2694 | |
908f4fbd N |
2695 | spin_lock_irqsave(&conf->device_lock, flags); |
2696 | clear_bit(In_sync, &rdev->flags); | |
2e38a37f | 2697 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2698 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2699 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2700 | ||
de393cde | 2701 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 | 2702 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
2703 | set_mask_bits(&mddev->sb_flags, 0, |
2704 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2705 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2706 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2707 | mdname(mddev), | |
2708 | bdevname(rdev->bdev, b), | |
2709 | mdname(mddev), | |
2710 | conf->raid_disks - mddev->degraded); | |
2e38a37f | 2711 | r5c_update_on_rdev_error(mddev); |
16a53ecc | 2712 | } |
1da177e4 LT |
2713 | |
2714 | /* | |
2715 | * Input: a 'big' sector number, | |
2716 | * Output: index of the data and parity disk, and the sector # in them. | |
2717 | */ | |
6d036f7d SL |
2718 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2719 | int previous, int *dd_idx, | |
2720 | struct stripe_head *sh) | |
1da177e4 | 2721 | { |
6e3b96ed | 2722 | sector_t stripe, stripe2; |
35f2a591 | 2723 | sector_t chunk_number; |
1da177e4 | 2724 | unsigned int chunk_offset; |
911d4ee8 | 2725 | int pd_idx, qd_idx; |
67cc2b81 | 2726 | int ddf_layout = 0; |
1da177e4 | 2727 | sector_t new_sector; |
e183eaed N |
2728 | int algorithm = previous ? conf->prev_algo |
2729 | : conf->algorithm; | |
09c9e5fa AN |
2730 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2731 | : conf->chunk_sectors; | |
112bf897 N |
2732 | int raid_disks = previous ? conf->previous_raid_disks |
2733 | : conf->raid_disks; | |
2734 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2735 | |
2736 | /* First compute the information on this sector */ | |
2737 | ||
2738 | /* | |
2739 | * Compute the chunk number and the sector offset inside the chunk | |
2740 | */ | |
2741 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2742 | chunk_number = r_sector; | |
1da177e4 LT |
2743 | |
2744 | /* | |
2745 | * Compute the stripe number | |
2746 | */ | |
35f2a591 N |
2747 | stripe = chunk_number; |
2748 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2749 | stripe2 = stripe; |
1da177e4 LT |
2750 | /* |
2751 | * Select the parity disk based on the user selected algorithm. | |
2752 | */ | |
84789554 | 2753 | pd_idx = qd_idx = -1; |
16a53ecc N |
2754 | switch(conf->level) { |
2755 | case 4: | |
911d4ee8 | 2756 | pd_idx = data_disks; |
16a53ecc N |
2757 | break; |
2758 | case 5: | |
e183eaed | 2759 | switch (algorithm) { |
1da177e4 | 2760 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2761 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2762 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2763 | (*dd_idx)++; |
2764 | break; | |
2765 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2766 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2767 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2768 | (*dd_idx)++; |
2769 | break; | |
2770 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2771 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2772 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2773 | break; |
2774 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2775 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2776 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2777 | break; |
99c0fb5f N |
2778 | case ALGORITHM_PARITY_0: |
2779 | pd_idx = 0; | |
2780 | (*dd_idx)++; | |
2781 | break; | |
2782 | case ALGORITHM_PARITY_N: | |
2783 | pd_idx = data_disks; | |
2784 | break; | |
1da177e4 | 2785 | default: |
99c0fb5f | 2786 | BUG(); |
16a53ecc N |
2787 | } |
2788 | break; | |
2789 | case 6: | |
2790 | ||
e183eaed | 2791 | switch (algorithm) { |
16a53ecc | 2792 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2793 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2794 | qd_idx = pd_idx + 1; |
2795 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2796 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2797 | qd_idx = 0; |
2798 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2799 | (*dd_idx) += 2; /* D D P Q D */ |
2800 | break; | |
2801 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2802 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2803 | qd_idx = pd_idx + 1; |
2804 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2805 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2806 | qd_idx = 0; |
2807 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2808 | (*dd_idx) += 2; /* D D P Q D */ |
2809 | break; | |
2810 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2811 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2812 | qd_idx = (pd_idx + 1) % raid_disks; |
2813 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2814 | break; |
2815 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2816 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2817 | qd_idx = (pd_idx + 1) % raid_disks; |
2818 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2819 | break; |
99c0fb5f N |
2820 | |
2821 | case ALGORITHM_PARITY_0: | |
2822 | pd_idx = 0; | |
2823 | qd_idx = 1; | |
2824 | (*dd_idx) += 2; | |
2825 | break; | |
2826 | case ALGORITHM_PARITY_N: | |
2827 | pd_idx = data_disks; | |
2828 | qd_idx = data_disks + 1; | |
2829 | break; | |
2830 | ||
2831 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2832 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2833 | * of blocks for computing Q is different. | |
2834 | */ | |
6e3b96ed | 2835 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2836 | qd_idx = pd_idx + 1; |
2837 | if (pd_idx == raid_disks-1) { | |
2838 | (*dd_idx)++; /* Q D D D P */ | |
2839 | qd_idx = 0; | |
2840 | } else if (*dd_idx >= pd_idx) | |
2841 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2842 | ddf_layout = 1; |
99c0fb5f N |
2843 | break; |
2844 | ||
2845 | case ALGORITHM_ROTATING_N_RESTART: | |
2846 | /* Same a left_asymmetric, by first stripe is | |
2847 | * D D D P Q rather than | |
2848 | * Q D D D P | |
2849 | */ | |
6e3b96ed N |
2850 | stripe2 += 1; |
2851 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2852 | qd_idx = pd_idx + 1; |
2853 | if (pd_idx == raid_disks-1) { | |
2854 | (*dd_idx)++; /* Q D D D P */ | |
2855 | qd_idx = 0; | |
2856 | } else if (*dd_idx >= pd_idx) | |
2857 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2858 | ddf_layout = 1; |
99c0fb5f N |
2859 | break; |
2860 | ||
2861 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2862 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2863 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2864 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2865 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2866 | ddf_layout = 1; |
99c0fb5f N |
2867 | break; |
2868 | ||
2869 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2870 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2871 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2872 | if (*dd_idx >= pd_idx) |
2873 | (*dd_idx)++; | |
2874 | qd_idx = raid_disks - 1; | |
2875 | break; | |
2876 | ||
2877 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2878 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2879 | if (*dd_idx >= pd_idx) |
2880 | (*dd_idx)++; | |
2881 | qd_idx = raid_disks - 1; | |
2882 | break; | |
2883 | ||
2884 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2885 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2886 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2887 | qd_idx = raid_disks - 1; | |
2888 | break; | |
2889 | ||
2890 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2891 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2892 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2893 | qd_idx = raid_disks - 1; | |
2894 | break; | |
2895 | ||
2896 | case ALGORITHM_PARITY_0_6: | |
2897 | pd_idx = 0; | |
2898 | (*dd_idx)++; | |
2899 | qd_idx = raid_disks - 1; | |
2900 | break; | |
2901 | ||
16a53ecc | 2902 | default: |
99c0fb5f | 2903 | BUG(); |
16a53ecc N |
2904 | } |
2905 | break; | |
1da177e4 LT |
2906 | } |
2907 | ||
911d4ee8 N |
2908 | if (sh) { |
2909 | sh->pd_idx = pd_idx; | |
2910 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2911 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2912 | } |
1da177e4 LT |
2913 | /* |
2914 | * Finally, compute the new sector number | |
2915 | */ | |
2916 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2917 | return new_sector; | |
2918 | } | |
2919 | ||
6d036f7d | 2920 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2921 | { |
d1688a6d | 2922 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2923 | int raid_disks = sh->disks; |
2924 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2925 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2926 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2927 | : conf->chunk_sectors; | |
e183eaed N |
2928 | int algorithm = previous ? conf->prev_algo |
2929 | : conf->algorithm; | |
1da177e4 LT |
2930 | sector_t stripe; |
2931 | int chunk_offset; | |
35f2a591 N |
2932 | sector_t chunk_number; |
2933 | int dummy1, dd_idx = i; | |
1da177e4 | 2934 | sector_t r_sector; |
911d4ee8 | 2935 | struct stripe_head sh2; |
1da177e4 LT |
2936 | |
2937 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2938 | stripe = new_sector; | |
1da177e4 | 2939 | |
16a53ecc N |
2940 | if (i == sh->pd_idx) |
2941 | return 0; | |
2942 | switch(conf->level) { | |
2943 | case 4: break; | |
2944 | case 5: | |
e183eaed | 2945 | switch (algorithm) { |
1da177e4 LT |
2946 | case ALGORITHM_LEFT_ASYMMETRIC: |
2947 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2948 | if (i > sh->pd_idx) | |
2949 | i--; | |
2950 | break; | |
2951 | case ALGORITHM_LEFT_SYMMETRIC: | |
2952 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2953 | if (i < sh->pd_idx) | |
2954 | i += raid_disks; | |
2955 | i -= (sh->pd_idx + 1); | |
2956 | break; | |
99c0fb5f N |
2957 | case ALGORITHM_PARITY_0: |
2958 | i -= 1; | |
2959 | break; | |
2960 | case ALGORITHM_PARITY_N: | |
2961 | break; | |
1da177e4 | 2962 | default: |
99c0fb5f | 2963 | BUG(); |
16a53ecc N |
2964 | } |
2965 | break; | |
2966 | case 6: | |
d0dabf7e | 2967 | if (i == sh->qd_idx) |
16a53ecc | 2968 | return 0; /* It is the Q disk */ |
e183eaed | 2969 | switch (algorithm) { |
16a53ecc N |
2970 | case ALGORITHM_LEFT_ASYMMETRIC: |
2971 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2972 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2973 | case ALGORITHM_ROTATING_N_RESTART: | |
2974 | if (sh->pd_idx == raid_disks-1) | |
2975 | i--; /* Q D D D P */ | |
16a53ecc N |
2976 | else if (i > sh->pd_idx) |
2977 | i -= 2; /* D D P Q D */ | |
2978 | break; | |
2979 | case ALGORITHM_LEFT_SYMMETRIC: | |
2980 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2981 | if (sh->pd_idx == raid_disks-1) | |
2982 | i--; /* Q D D D P */ | |
2983 | else { | |
2984 | /* D D P Q D */ | |
2985 | if (i < sh->pd_idx) | |
2986 | i += raid_disks; | |
2987 | i -= (sh->pd_idx + 2); | |
2988 | } | |
2989 | break; | |
99c0fb5f N |
2990 | case ALGORITHM_PARITY_0: |
2991 | i -= 2; | |
2992 | break; | |
2993 | case ALGORITHM_PARITY_N: | |
2994 | break; | |
2995 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2996 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2997 | if (sh->pd_idx == 0) |
2998 | i--; /* P D D D Q */ | |
e4424fee N |
2999 | else { |
3000 | /* D D Q P D */ | |
3001 | if (i < sh->pd_idx) | |
3002 | i += raid_disks; | |
3003 | i -= (sh->pd_idx + 1); | |
3004 | } | |
99c0fb5f N |
3005 | break; |
3006 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
3007 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
3008 | if (i > sh->pd_idx) | |
3009 | i--; | |
3010 | break; | |
3011 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
3012 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
3013 | if (i < sh->pd_idx) | |
3014 | i += data_disks + 1; | |
3015 | i -= (sh->pd_idx + 1); | |
3016 | break; | |
3017 | case ALGORITHM_PARITY_0_6: | |
3018 | i -= 1; | |
3019 | break; | |
16a53ecc | 3020 | default: |
99c0fb5f | 3021 | BUG(); |
16a53ecc N |
3022 | } |
3023 | break; | |
1da177e4 LT |
3024 | } |
3025 | ||
3026 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 3027 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 3028 | |
112bf897 | 3029 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 3030 | previous, &dummy1, &sh2); |
911d4ee8 N |
3031 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
3032 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
3033 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
3034 | mdname(conf->mddev)); | |
1da177e4 LT |
3035 | return 0; |
3036 | } | |
3037 | return r_sector; | |
3038 | } | |
3039 | ||
07e83364 SL |
3040 | /* |
3041 | * There are cases where we want handle_stripe_dirtying() and | |
3042 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
3043 | * | |
3044 | * This function checks whether we want to delay the towrite. Specifically, | |
3045 | * we delay the towrite when: | |
3046 | * | |
3047 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
3048 | * stripe has data in journal (for other devices). | |
3049 | * | |
3050 | * In this case, when reading data for the non-overwrite dev, it is | |
3051 | * necessary to handle complex rmw of write back cache (prexor with | |
3052 | * orig_page, and xor with page). To keep read path simple, we would | |
3053 | * like to flush data in journal to RAID disks first, so complex rmw | |
3054 | * is handled in the write patch (handle_stripe_dirtying). | |
3055 | * | |
39b99586 SL |
3056 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
3057 | * | |
3058 | * It is important to be able to flush all stripes in raid5-cache. | |
3059 | * Therefore, we need reserve some space on the journal device for | |
3060 | * these flushes. If flush operation includes pending writes to the | |
3061 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
3062 | * for the flush out. If we exclude these pending writes from flush | |
3063 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
3064 | * Therefore, excluding pending writes in these cases enables more | |
3065 | * efficient use of the journal device. | |
3066 | * | |
3067 | * Note: To make sure the stripe makes progress, we only delay | |
3068 | * towrite for stripes with data already in journal (injournal > 0). | |
3069 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
3070 | * no_space_stripes list. | |
3071 | * | |
07e83364 | 3072 | */ |
39b99586 SL |
3073 | static inline bool delay_towrite(struct r5conf *conf, |
3074 | struct r5dev *dev, | |
3075 | struct stripe_head_state *s) | |
07e83364 | 3076 | { |
39b99586 SL |
3077 | /* case 1 above */ |
3078 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
3079 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
3080 | return true; | |
3081 | /* case 2 above */ | |
3082 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
3083 | s->injournal > 0) | |
3084 | return true; | |
3085 | return false; | |
07e83364 SL |
3086 | } |
3087 | ||
600aa109 | 3088 | static void |
c0f7bddb | 3089 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3090 | int rcw, int expand) |
e33129d8 | 3091 | { |
584acdd4 | 3092 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3093 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3094 | int level = conf->level; |
e33129d8 DW |
3095 | |
3096 | if (rcw) { | |
1e6d690b SL |
3097 | /* |
3098 | * In some cases, handle_stripe_dirtying initially decided to | |
3099 | * run rmw and allocates extra page for prexor. However, rcw is | |
3100 | * cheaper later on. We need to free the extra page now, | |
3101 | * because we won't be able to do that in ops_complete_prexor(). | |
3102 | */ | |
3103 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3104 | |
3105 | for (i = disks; i--; ) { | |
3106 | struct r5dev *dev = &sh->dev[i]; | |
3107 | ||
39b99586 | 3108 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3109 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3110 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3111 | if (!expand) |
3112 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3113 | s->locked++; |
1e6d690b SL |
3114 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3115 | set_bit(R5_LOCKED, &dev->flags); | |
3116 | s->locked++; | |
e33129d8 DW |
3117 | } |
3118 | } | |
ce7d363a N |
3119 | /* if we are not expanding this is a proper write request, and |
3120 | * there will be bios with new data to be drained into the | |
3121 | * stripe cache | |
3122 | */ | |
3123 | if (!expand) { | |
3124 | if (!s->locked) | |
3125 | /* False alarm, nothing to do */ | |
3126 | return; | |
3127 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3128 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3129 | } else | |
3130 | sh->reconstruct_state = reconstruct_state_run; | |
3131 | ||
3132 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3133 | ||
c0f7bddb | 3134 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3135 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3136 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3137 | } else { |
3138 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3139 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3140 | BUG_ON(level == 6 && |
3141 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3142 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3143 | |
e33129d8 DW |
3144 | for (i = disks; i--; ) { |
3145 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3146 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3147 | continue; |
3148 | ||
e33129d8 DW |
3149 | if (dev->towrite && |
3150 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3151 | test_bit(R5_Wantcompute, &dev->flags))) { |
3152 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3153 | set_bit(R5_LOCKED, &dev->flags); |
3154 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3155 | s->locked++; |
1e6d690b SL |
3156 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3157 | set_bit(R5_LOCKED, &dev->flags); | |
3158 | s->locked++; | |
e33129d8 DW |
3159 | } |
3160 | } | |
ce7d363a N |
3161 | if (!s->locked) |
3162 | /* False alarm - nothing to do */ | |
3163 | return; | |
3164 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3165 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3166 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3167 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3168 | } |
3169 | ||
c0f7bddb | 3170 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3171 | * are in flight |
3172 | */ | |
3173 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3174 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3175 | s->locked++; |
e33129d8 | 3176 | |
c0f7bddb YT |
3177 | if (level == 6) { |
3178 | int qd_idx = sh->qd_idx; | |
3179 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3180 | ||
3181 | set_bit(R5_LOCKED, &dev->flags); | |
3182 | clear_bit(R5_UPTODATE, &dev->flags); | |
3183 | s->locked++; | |
3184 | } | |
3185 | ||
3418d036 AP |
3186 | if (raid5_has_ppl(sh->raid_conf) && |
3187 | test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) && | |
3188 | !test_bit(STRIPE_FULL_WRITE, &sh->state) && | |
3189 | test_bit(R5_Insync, &sh->dev[pd_idx].flags)) | |
3190 | set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request); | |
3191 | ||
600aa109 | 3192 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3193 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3194 | s->locked, s->ops_request); |
e33129d8 | 3195 | } |
16a53ecc | 3196 | |
1da177e4 LT |
3197 | /* |
3198 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3199 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3200 | * The bi_next chain must be in order. |
3201 | */ | |
da41ba65 | 3202 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3203 | int forwrite, int previous) | |
1da177e4 LT |
3204 | { |
3205 | struct bio **bip; | |
d1688a6d | 3206 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3207 | int firstwrite=0; |
1da177e4 | 3208 | |
cbe47ec5 | 3209 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3210 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3211 | (unsigned long long)sh->sector); |
3212 | ||
b17459c0 SL |
3213 | /* |
3214 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
3215 | * reference count to avoid race. The reference count should already be | |
3216 | * increased before this function is called (for example, in | |
849674e4 | 3217 | * raid5_make_request()), so other bio sharing this stripe will not free the |
b17459c0 SL |
3218 | * stripe. If a stripe is owned by one stripe, the stripe lock will |
3219 | * protect it. | |
3220 | */ | |
3221 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 3222 | /* Don't allow new IO added to stripes in batch list */ |
3223 | if (sh->batch_head) | |
3224 | goto overlap; | |
72626685 | 3225 | if (forwrite) { |
1da177e4 | 3226 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3227 | if (*bip == NULL) |
72626685 N |
3228 | firstwrite = 1; |
3229 | } else | |
1da177e4 | 3230 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3231 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3232 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3233 | goto overlap; |
3234 | bip = & (*bip)->bi_next; | |
3235 | } | |
4f024f37 | 3236 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3237 | goto overlap; |
3238 | ||
3418d036 AP |
3239 | if (forwrite && raid5_has_ppl(conf)) { |
3240 | /* | |
3241 | * With PPL only writes to consecutive data chunks within a | |
3242 | * stripe are allowed because for a single stripe_head we can | |
3243 | * only have one PPL entry at a time, which describes one data | |
3244 | * range. Not really an overlap, but wait_for_overlap can be | |
3245 | * used to handle this. | |
3246 | */ | |
3247 | sector_t sector; | |
3248 | sector_t first = 0; | |
3249 | sector_t last = 0; | |
3250 | int count = 0; | |
3251 | int i; | |
3252 | ||
3253 | for (i = 0; i < sh->disks; i++) { | |
3254 | if (i != sh->pd_idx && | |
3255 | (i == dd_idx || sh->dev[i].towrite)) { | |
3256 | sector = sh->dev[i].sector; | |
3257 | if (count == 0 || sector < first) | |
3258 | first = sector; | |
3259 | if (sector > last) | |
3260 | last = sector; | |
3261 | count++; | |
3262 | } | |
3263 | } | |
3264 | ||
3265 | if (first + conf->chunk_sectors * (count - 1) != last) | |
3266 | goto overlap; | |
3267 | } | |
3268 | ||
da41ba65 | 3269 | if (!forwrite || previous) |
3270 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3271 | ||
78bafebd | 3272 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3273 | if (*bip) |
3274 | bi->bi_next = *bip; | |
3275 | *bip = bi; | |
e7836bd6 | 3276 | raid5_inc_bi_active_stripes(bi); |
49728050 | 3277 | md_write_inc(conf->mddev, bi); |
72626685 | 3278 | |
1da177e4 LT |
3279 | if (forwrite) { |
3280 | /* check if page is covered */ | |
3281 | sector_t sector = sh->dev[dd_idx].sector; | |
3282 | for (bi=sh->dev[dd_idx].towrite; | |
3283 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3284 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3285 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3286 | if (bio_end_sector(bi) >= sector) |
3287 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3288 | } |
3289 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3290 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3291 | sh->overwrite_disks++; | |
1da177e4 | 3292 | } |
cbe47ec5 N |
3293 | |
3294 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3295 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3296 | (unsigned long long)sh->sector, dd_idx); |
3297 | ||
3298 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3299 | /* Cannot hold spinlock over bitmap_startwrite, |
3300 | * but must ensure this isn't added to a batch until | |
3301 | * we have added to the bitmap and set bm_seq. | |
3302 | * So set STRIPE_BITMAP_PENDING to prevent | |
3303 | * batching. | |
3304 | * If multiple add_stripe_bio() calls race here they | |
3305 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3306 | * to complete "bitmap_startwrite" gets to set | |
3307 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3308 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3309 | * any more. | |
3310 | */ | |
3311 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3312 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3313 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3314 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3315 | spin_lock_irq(&sh->stripe_lock); |
3316 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3317 | if (!sh->batch_head) { | |
3318 | sh->bm_seq = conf->seq_flush+1; | |
3319 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3320 | } | |
cbe47ec5 | 3321 | } |
d0852df5 | 3322 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3323 | |
3324 | if (stripe_can_batch(sh)) | |
3325 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3326 | return 1; |
3327 | ||
3328 | overlap: | |
3329 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3330 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3331 | return 0; |
3332 | } | |
3333 | ||
d1688a6d | 3334 | static void end_reshape(struct r5conf *conf); |
29269553 | 3335 | |
d1688a6d | 3336 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3337 | struct stripe_head *sh) |
ccfcc3c1 | 3338 | { |
784052ec | 3339 | int sectors_per_chunk = |
09c9e5fa | 3340 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3341 | int dd_idx; |
2d2063ce | 3342 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3343 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3344 | |
112bf897 N |
3345 | raid5_compute_sector(conf, |
3346 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3347 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3348 | previous, |
911d4ee8 | 3349 | &dd_idx, sh); |
ccfcc3c1 N |
3350 | } |
3351 | ||
a4456856 | 3352 | static void |
d1688a6d | 3353 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 | 3354 | struct stripe_head_state *s, int disks, |
34a6f80e | 3355 | struct bio_list *return_bi) |
a4456856 DW |
3356 | { |
3357 | int i; | |
59fc630b | 3358 | BUG_ON(sh->batch_head); |
a4456856 DW |
3359 | for (i = disks; i--; ) { |
3360 | struct bio *bi; | |
3361 | int bitmap_end = 0; | |
3362 | ||
3363 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3364 | struct md_rdev *rdev; |
a4456856 DW |
3365 | rcu_read_lock(); |
3366 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3367 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3368 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3369 | atomic_inc(&rdev->nr_pending); |
3370 | else | |
3371 | rdev = NULL; | |
a4456856 | 3372 | rcu_read_unlock(); |
7f0da59b N |
3373 | if (rdev) { |
3374 | if (!rdev_set_badblocks( | |
3375 | rdev, | |
3376 | sh->sector, | |
3377 | STRIPE_SECTORS, 0)) | |
3378 | md_error(conf->mddev, rdev); | |
3379 | rdev_dec_pending(rdev, conf->mddev); | |
3380 | } | |
a4456856 | 3381 | } |
b17459c0 | 3382 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3383 | /* fail all writes first */ |
3384 | bi = sh->dev[i].towrite; | |
3385 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3386 | sh->overwrite_disks = 0; |
b17459c0 | 3387 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3388 | if (bi) |
a4456856 | 3389 | bitmap_end = 1; |
a4456856 | 3390 | |
ff875738 | 3391 | log_stripe_write_finished(sh); |
0576b1c6 | 3392 | |
a4456856 DW |
3393 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3394 | wake_up(&conf->wait_for_overlap); | |
3395 | ||
4f024f37 | 3396 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3397 | sh->dev[i].sector + STRIPE_SECTORS) { |
3398 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3399 | |
3400 | bi->bi_error = -EIO; | |
49728050 N |
3401 | md_write_end(conf->mddev); |
3402 | if (!raid5_dec_bi_active_stripes(bi)) | |
34a6f80e | 3403 | bio_list_add(return_bi, bi); |
a4456856 DW |
3404 | bi = nextbi; |
3405 | } | |
7eaf7e8e SL |
3406 | if (bitmap_end) |
3407 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3408 | STRIPE_SECTORS, 0, 0); | |
3409 | bitmap_end = 0; | |
a4456856 DW |
3410 | /* and fail all 'written' */ |
3411 | bi = sh->dev[i].written; | |
3412 | sh->dev[i].written = NULL; | |
d592a996 SL |
3413 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3414 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3415 | sh->dev[i].page = sh->dev[i].orig_page; | |
3416 | } | |
3417 | ||
a4456856 | 3418 | if (bi) bitmap_end = 1; |
4f024f37 | 3419 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3420 | sh->dev[i].sector + STRIPE_SECTORS) { |
3421 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3422 | |
3423 | bi->bi_error = -EIO; | |
49728050 N |
3424 | md_write_end(conf->mddev); |
3425 | if (!raid5_dec_bi_active_stripes(bi)) | |
34a6f80e | 3426 | bio_list_add(return_bi, bi); |
a4456856 DW |
3427 | bi = bi2; |
3428 | } | |
3429 | ||
b5e98d65 DW |
3430 | /* fail any reads if this device is non-operational and |
3431 | * the data has not reached the cache yet. | |
3432 | */ | |
3433 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3434 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3435 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3436 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3437 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3438 | bi = sh->dev[i].toread; |
3439 | sh->dev[i].toread = NULL; | |
143c4d05 | 3440 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3441 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3442 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3443 | if (bi) |
3444 | s->to_read--; | |
4f024f37 | 3445 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3446 | sh->dev[i].sector + STRIPE_SECTORS) { |
3447 | struct bio *nextbi = | |
3448 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3449 | |
3450 | bi->bi_error = -EIO; | |
34a6f80e N |
3451 | if (!raid5_dec_bi_active_stripes(bi)) |
3452 | bio_list_add(return_bi, bi); | |
a4456856 DW |
3453 | bi = nextbi; |
3454 | } | |
3455 | } | |
a4456856 DW |
3456 | if (bitmap_end) |
3457 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3458 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3459 | /* If we were in the middle of a write the parity block might |
3460 | * still be locked - so just clear all R5_LOCKED flags | |
3461 | */ | |
3462 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3463 | } |
ebda780b SL |
3464 | s->to_write = 0; |
3465 | s->written = 0; | |
a4456856 | 3466 | |
8b3e6cdc DW |
3467 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3468 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3469 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3470 | } |
3471 | ||
7f0da59b | 3472 | static void |
d1688a6d | 3473 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3474 | struct stripe_head_state *s) |
3475 | { | |
3476 | int abort = 0; | |
3477 | int i; | |
3478 | ||
59fc630b | 3479 | BUG_ON(sh->batch_head); |
7f0da59b | 3480 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3481 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3482 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3483 | s->syncing = 0; |
9a3e1101 | 3484 | s->replacing = 0; |
7f0da59b | 3485 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3486 | * Don't even need to abort as that is handled elsewhere |
3487 | * if needed, and not always wanted e.g. if there is a known | |
3488 | * bad block here. | |
9a3e1101 | 3489 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3490 | * non-sync devices, or abort the recovery |
3491 | */ | |
18b9837e N |
3492 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3493 | /* During recovery devices cannot be removed, so | |
3494 | * locking and refcounting of rdevs is not needed | |
3495 | */ | |
e50d3992 | 3496 | rcu_read_lock(); |
18b9837e | 3497 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3498 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3499 | if (rdev |
3500 | && !test_bit(Faulty, &rdev->flags) | |
3501 | && !test_bit(In_sync, &rdev->flags) | |
3502 | && !rdev_set_badblocks(rdev, sh->sector, | |
3503 | STRIPE_SECTORS, 0)) | |
3504 | abort = 1; | |
e50d3992 | 3505 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3506 | if (rdev |
3507 | && !test_bit(Faulty, &rdev->flags) | |
3508 | && !test_bit(In_sync, &rdev->flags) | |
3509 | && !rdev_set_badblocks(rdev, sh->sector, | |
3510 | STRIPE_SECTORS, 0)) | |
3511 | abort = 1; | |
3512 | } | |
e50d3992 | 3513 | rcu_read_unlock(); |
18b9837e N |
3514 | if (abort) |
3515 | conf->recovery_disabled = | |
3516 | conf->mddev->recovery_disabled; | |
7f0da59b | 3517 | } |
18b9837e | 3518 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3519 | } |
3520 | ||
9a3e1101 N |
3521 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3522 | { | |
3523 | struct md_rdev *rdev; | |
3524 | int rv = 0; | |
3f232d6a N |
3525 | |
3526 | rcu_read_lock(); | |
3527 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3528 | if (rdev |
3529 | && !test_bit(Faulty, &rdev->flags) | |
3530 | && !test_bit(In_sync, &rdev->flags) | |
3531 | && (rdev->recovery_offset <= sh->sector | |
3532 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3533 | rv = 1; | |
3f232d6a | 3534 | rcu_read_unlock(); |
9a3e1101 N |
3535 | return rv; |
3536 | } | |
3537 | ||
2c58f06e N |
3538 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3539 | int disk_idx, int disks) | |
a4456856 | 3540 | { |
5599becc | 3541 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3542 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3543 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3544 | int i; |
5599becc | 3545 | |
a79cfe12 N |
3546 | |
3547 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3548 | test_bit(R5_UPTODATE, &dev->flags)) | |
3549 | /* No point reading this as we already have it or have | |
3550 | * decided to get it. | |
3551 | */ | |
3552 | return 0; | |
3553 | ||
3554 | if (dev->toread || | |
3555 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3556 | /* We need this block to directly satisfy a request */ | |
3557 | return 1; | |
3558 | ||
3559 | if (s->syncing || s->expanding || | |
3560 | (s->replacing && want_replace(sh, disk_idx))) | |
3561 | /* When syncing, or expanding we read everything. | |
3562 | * When replacing, we need the replaced block. | |
3563 | */ | |
3564 | return 1; | |
3565 | ||
3566 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3567 | (s->failed >= 2 && fdev[1]->toread)) | |
3568 | /* If we want to read from a failed device, then | |
3569 | * we need to actually read every other device. | |
3570 | */ | |
3571 | return 1; | |
3572 | ||
a9d56950 N |
3573 | /* Sometimes neither read-modify-write nor reconstruct-write |
3574 | * cycles can work. In those cases we read every block we | |
3575 | * can. Then the parity-update is certain to have enough to | |
3576 | * work with. | |
3577 | * This can only be a problem when we need to write something, | |
3578 | * and some device has failed. If either of those tests | |
3579 | * fail we need look no further. | |
3580 | */ | |
3581 | if (!s->failed || !s->to_write) | |
3582 | return 0; | |
3583 | ||
3584 | if (test_bit(R5_Insync, &dev->flags) && | |
3585 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3586 | /* Pre-reads at not permitted until after short delay | |
3587 | * to gather multiple requests. However if this | |
3588 | * device is no Insync, the block could only be be computed | |
3589 | * and there is no need to delay that. | |
3590 | */ | |
3591 | return 0; | |
ea664c82 | 3592 | |
36707bb2 | 3593 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3594 | if (fdev[i]->towrite && |
3595 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3596 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3597 | /* If we have a partial write to a failed | |
3598 | * device, then we will need to reconstruct | |
3599 | * the content of that device, so all other | |
3600 | * devices must be read. | |
3601 | */ | |
3602 | return 1; | |
3603 | } | |
3604 | ||
3605 | /* If we are forced to do a reconstruct-write, either because | |
3606 | * the current RAID6 implementation only supports that, or | |
3607 | * or because parity cannot be trusted and we are currently | |
3608 | * recovering it, there is extra need to be careful. | |
3609 | * If one of the devices that we would need to read, because | |
3610 | * it is not being overwritten (and maybe not written at all) | |
3611 | * is missing/faulty, then we need to read everything we can. | |
3612 | */ | |
3613 | if (sh->raid_conf->level != 6 && | |
3614 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3615 | /* reconstruct-write isn't being forced */ | |
3616 | return 0; | |
36707bb2 | 3617 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3618 | if (s->failed_num[i] != sh->pd_idx && |
3619 | s->failed_num[i] != sh->qd_idx && | |
3620 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3621 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3622 | return 1; | |
3623 | } | |
3624 | ||
2c58f06e N |
3625 | return 0; |
3626 | } | |
3627 | ||
ba02684d SL |
3628 | /* fetch_block - checks the given member device to see if its data needs |
3629 | * to be read or computed to satisfy a request. | |
3630 | * | |
3631 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3632 | * 0 to tell the loop in handle_stripe_fill to continue | |
3633 | */ | |
2c58f06e N |
3634 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3635 | int disk_idx, int disks) | |
3636 | { | |
3637 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3638 | ||
3639 | /* is the data in this block needed, and can we get it? */ | |
3640 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3641 | /* we would like to get this block, possibly by computing it, |
3642 | * otherwise read it if the backing disk is insync | |
3643 | */ | |
3644 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3645 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3646 | BUG_ON(sh->batch_head); |
5599becc | 3647 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3648 | (s->failed && (disk_idx == s->failed_num[0] || |
3649 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3650 | /* have disk failed, and we're requested to fetch it; |
3651 | * do compute it | |
a4456856 | 3652 | */ |
5599becc YT |
3653 | pr_debug("Computing stripe %llu block %d\n", |
3654 | (unsigned long long)sh->sector, disk_idx); | |
3655 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3656 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3657 | set_bit(R5_Wantcompute, &dev->flags); | |
3658 | sh->ops.target = disk_idx; | |
3659 | sh->ops.target2 = -1; /* no 2nd target */ | |
3660 | s->req_compute = 1; | |
93b3dbce N |
3661 | /* Careful: from this point on 'uptodate' is in the eye |
3662 | * of raid_run_ops which services 'compute' operations | |
3663 | * before writes. R5_Wantcompute flags a block that will | |
3664 | * be R5_UPTODATE by the time it is needed for a | |
3665 | * subsequent operation. | |
3666 | */ | |
5599becc YT |
3667 | s->uptodate++; |
3668 | return 1; | |
3669 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3670 | /* Computing 2-failure is *very* expensive; only | |
3671 | * do it if failed >= 2 | |
3672 | */ | |
3673 | int other; | |
3674 | for (other = disks; other--; ) { | |
3675 | if (other == disk_idx) | |
3676 | continue; | |
3677 | if (!test_bit(R5_UPTODATE, | |
3678 | &sh->dev[other].flags)) | |
3679 | break; | |
a4456856 | 3680 | } |
5599becc YT |
3681 | BUG_ON(other < 0); |
3682 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3683 | (unsigned long long)sh->sector, | |
3684 | disk_idx, other); | |
3685 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3686 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3687 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3688 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3689 | sh->ops.target = disk_idx; | |
3690 | sh->ops.target2 = other; | |
3691 | s->uptodate += 2; | |
3692 | s->req_compute = 1; | |
3693 | return 1; | |
3694 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3695 | set_bit(R5_LOCKED, &dev->flags); | |
3696 | set_bit(R5_Wantread, &dev->flags); | |
3697 | s->locked++; | |
3698 | pr_debug("Reading block %d (sync=%d)\n", | |
3699 | disk_idx, s->syncing); | |
a4456856 DW |
3700 | } |
3701 | } | |
5599becc YT |
3702 | |
3703 | return 0; | |
3704 | } | |
3705 | ||
3706 | /** | |
93b3dbce | 3707 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3708 | */ |
93b3dbce N |
3709 | static void handle_stripe_fill(struct stripe_head *sh, |
3710 | struct stripe_head_state *s, | |
3711 | int disks) | |
5599becc YT |
3712 | { |
3713 | int i; | |
3714 | ||
3715 | /* look for blocks to read/compute, skip this if a compute | |
3716 | * is already in flight, or if the stripe contents are in the | |
3717 | * midst of changing due to a write | |
3718 | */ | |
3719 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3720 | !sh->reconstruct_state) { |
3721 | ||
3722 | /* | |
3723 | * For degraded stripe with data in journal, do not handle | |
3724 | * read requests yet, instead, flush the stripe to raid | |
3725 | * disks first, this avoids handling complex rmw of write | |
3726 | * back cache (prexor with orig_page, and then xor with | |
3727 | * page) in the read path | |
3728 | */ | |
3729 | if (s->injournal && s->failed) { | |
3730 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3731 | r5c_make_stripe_write_out(sh); | |
3732 | goto out; | |
3733 | } | |
3734 | ||
5599becc | 3735 | for (i = disks; i--; ) |
93b3dbce | 3736 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3737 | break; |
07e83364 SL |
3738 | } |
3739 | out: | |
a4456856 DW |
3740 | set_bit(STRIPE_HANDLE, &sh->state); |
3741 | } | |
3742 | ||
787b76fa N |
3743 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3744 | unsigned long handle_flags); | |
1fe797e6 | 3745 | /* handle_stripe_clean_event |
a4456856 DW |
3746 | * any written block on an uptodate or failed drive can be returned. |
3747 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3748 | * never LOCKED, so we don't need to test 'failed' directly. | |
3749 | */ | |
d1688a6d | 3750 | static void handle_stripe_clean_event(struct r5conf *conf, |
34a6f80e | 3751 | struct stripe_head *sh, int disks, struct bio_list *return_bi) |
a4456856 DW |
3752 | { |
3753 | int i; | |
3754 | struct r5dev *dev; | |
f8dfcffd | 3755 | int discard_pending = 0; |
59fc630b | 3756 | struct stripe_head *head_sh = sh; |
3757 | bool do_endio = false; | |
a4456856 DW |
3758 | |
3759 | for (i = disks; i--; ) | |
3760 | if (sh->dev[i].written) { | |
3761 | dev = &sh->dev[i]; | |
3762 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3763 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3764 | test_bit(R5_Discard, &dev->flags) || |
3765 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3766 | /* We can return any write requests */ |
3767 | struct bio *wbi, *wbi2; | |
45b4233c | 3768 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3769 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3770 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3771 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3772 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3773 | } |
59fc630b | 3774 | do_endio = true; |
3775 | ||
3776 | returnbi: | |
3777 | dev->page = dev->orig_page; | |
a4456856 DW |
3778 | wbi = dev->written; |
3779 | dev->written = NULL; | |
4f024f37 | 3780 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3781 | dev->sector + STRIPE_SECTORS) { |
3782 | wbi2 = r5_next_bio(wbi, dev->sector); | |
49728050 N |
3783 | md_write_end(conf->mddev); |
3784 | if (!raid5_dec_bi_active_stripes(wbi)) | |
34a6f80e | 3785 | bio_list_add(return_bi, wbi); |
a4456856 DW |
3786 | wbi = wbi2; |
3787 | } | |
7eaf7e8e SL |
3788 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3789 | STRIPE_SECTORS, | |
a4456856 | 3790 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3791 | 0); |
59fc630b | 3792 | if (head_sh->batch_head) { |
3793 | sh = list_first_entry(&sh->batch_list, | |
3794 | struct stripe_head, | |
3795 | batch_list); | |
3796 | if (sh != head_sh) { | |
3797 | dev = &sh->dev[i]; | |
3798 | goto returnbi; | |
3799 | } | |
3800 | } | |
3801 | sh = head_sh; | |
3802 | dev = &sh->dev[i]; | |
f8dfcffd N |
3803 | } else if (test_bit(R5_Discard, &dev->flags)) |
3804 | discard_pending = 1; | |
3805 | } | |
f6bed0ef | 3806 | |
ff875738 | 3807 | log_stripe_write_finished(sh); |
0576b1c6 | 3808 | |
f8dfcffd N |
3809 | if (!discard_pending && |
3810 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3811 | int hash; |
f8dfcffd N |
3812 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3813 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3814 | if (sh->qd_idx >= 0) { | |
3815 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3816 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3817 | } | |
3818 | /* now that discard is done we can proceed with any sync */ | |
3819 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3820 | /* |
3821 | * SCSI discard will change some bio fields and the stripe has | |
3822 | * no updated data, so remove it from hash list and the stripe | |
3823 | * will be reinitialized | |
3824 | */ | |
59fc630b | 3825 | unhash: |
b8a9d66d RG |
3826 | hash = sh->hash_lock_index; |
3827 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3828 | remove_hash(sh); |
b8a9d66d | 3829 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3830 | if (head_sh->batch_head) { |
3831 | sh = list_first_entry(&sh->batch_list, | |
3832 | struct stripe_head, batch_list); | |
3833 | if (sh != head_sh) | |
3834 | goto unhash; | |
3835 | } | |
59fc630b | 3836 | sh = head_sh; |
3837 | ||
f8dfcffd N |
3838 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3839 | set_bit(STRIPE_HANDLE, &sh->state); | |
3840 | ||
3841 | } | |
8b3e6cdc DW |
3842 | |
3843 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3844 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3845 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3846 | |
787b76fa N |
3847 | if (head_sh->batch_head && do_endio) |
3848 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3849 | } |
3850 | ||
86aa1397 SL |
3851 | /* |
3852 | * For RMW in write back cache, we need extra page in prexor to store the | |
3853 | * old data. This page is stored in dev->orig_page. | |
3854 | * | |
3855 | * This function checks whether we have data for prexor. The exact logic | |
3856 | * is: | |
3857 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3858 | */ | |
3859 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3860 | { | |
3861 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3862 | (!test_bit(R5_InJournal, &dev->flags) || | |
3863 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3864 | } | |
3865 | ||
d7bd398e SL |
3866 | static int handle_stripe_dirtying(struct r5conf *conf, |
3867 | struct stripe_head *sh, | |
3868 | struct stripe_head_state *s, | |
3869 | int disks) | |
a4456856 DW |
3870 | { |
3871 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3872 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3873 | ||
584acdd4 | 3874 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3875 | * If yes, then the array is dirty (after unclean shutdown or |
3876 | * initial creation), so parity in some stripes might be inconsistent. | |
3877 | * In this case, we need to always do reconstruct-write, to ensure | |
3878 | * that in case of drive failure or read-error correction, we | |
3879 | * generate correct data from the parity. | |
3880 | */ | |
584acdd4 | 3881 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3882 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3883 | s->failed == 0)) { | |
a7854487 | 3884 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3885 | * look like rcw is cheaper |
3886 | */ | |
3887 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3888 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3889 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3890 | (unsigned long long)sh->sector); |
c8ac1803 | 3891 | } else for (i = disks; i--; ) { |
a4456856 DW |
3892 | /* would I have to read this buffer for read_modify_write */ |
3893 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3894 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3895 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3896 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3897 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3898 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3899 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3900 | if (test_bit(R5_Insync, &dev->flags)) |
3901 | rmw++; | |
3902 | else | |
3903 | rmw += 2*disks; /* cannot read it */ | |
3904 | } | |
3905 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3906 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3907 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3908 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3909 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3910 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3911 | if (test_bit(R5_Insync, &dev->flags)) |
3912 | rcw++; | |
a4456856 DW |
3913 | else |
3914 | rcw += 2*disks; | |
3915 | } | |
3916 | } | |
1e6d690b | 3917 | |
39b99586 SL |
3918 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3919 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3920 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3921 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3922 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3923 | if (conf->mddev->queue) |
3924 | blk_add_trace_msg(conf->mddev->queue, | |
3925 | "raid5 rmw %llu %d", | |
3926 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3927 | for (i = disks; i--; ) { |
3928 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3929 | if (test_bit(R5_InJournal, &dev->flags) && |
3930 | dev->page == dev->orig_page && | |
3931 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3932 | /* alloc page for prexor */ | |
d7bd398e SL |
3933 | struct page *p = alloc_page(GFP_NOIO); |
3934 | ||
3935 | if (p) { | |
3936 | dev->orig_page = p; | |
3937 | continue; | |
3938 | } | |
3939 | ||
3940 | /* | |
3941 | * alloc_page() failed, try use | |
3942 | * disk_info->extra_page | |
3943 | */ | |
3944 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3945 | &conf->cache_state)) { | |
3946 | r5c_use_extra_page(sh); | |
3947 | break; | |
3948 | } | |
1e6d690b | 3949 | |
d7bd398e SL |
3950 | /* extra_page in use, add to delayed_list */ |
3951 | set_bit(STRIPE_DELAYED, &sh->state); | |
3952 | s->waiting_extra_page = 1; | |
3953 | return -EAGAIN; | |
1e6d690b | 3954 | } |
d7bd398e | 3955 | } |
1e6d690b | 3956 | |
d7bd398e SL |
3957 | for (i = disks; i--; ) { |
3958 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3959 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3960 | i == sh->pd_idx || i == sh->qd_idx || |
3961 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3962 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3963 | !(uptodate_for_rmw(dev) || |
1e6d690b | 3964 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 3965 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3966 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3967 | &sh->state)) { | |
3968 | pr_debug("Read_old block %d for r-m-w\n", | |
3969 | i); | |
a4456856 DW |
3970 | set_bit(R5_LOCKED, &dev->flags); |
3971 | set_bit(R5_Wantread, &dev->flags); | |
3972 | s->locked++; | |
3973 | } else { | |
3974 | set_bit(STRIPE_DELAYED, &sh->state); | |
3975 | set_bit(STRIPE_HANDLE, &sh->state); | |
3976 | } | |
3977 | } | |
3978 | } | |
a9add5d9 | 3979 | } |
41257580 | 3980 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3981 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3982 | int qread =0; |
c8ac1803 | 3983 | rcw = 0; |
a4456856 DW |
3984 | for (i = disks; i--; ) { |
3985 | struct r5dev *dev = &sh->dev[i]; | |
3986 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3987 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3988 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3989 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3990 | test_bit(R5_Wantcompute, &dev->flags))) { |
3991 | rcw++; | |
67f45548 N |
3992 | if (test_bit(R5_Insync, &dev->flags) && |
3993 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3994 | &sh->state)) { | |
45b4233c | 3995 | pr_debug("Read_old block " |
a4456856 DW |
3996 | "%d for Reconstruct\n", i); |
3997 | set_bit(R5_LOCKED, &dev->flags); | |
3998 | set_bit(R5_Wantread, &dev->flags); | |
3999 | s->locked++; | |
a9add5d9 | 4000 | qread++; |
a4456856 DW |
4001 | } else { |
4002 | set_bit(STRIPE_DELAYED, &sh->state); | |
4003 | set_bit(STRIPE_HANDLE, &sh->state); | |
4004 | } | |
4005 | } | |
4006 | } | |
e3620a3a | 4007 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
4008 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
4009 | (unsigned long long)sh->sector, | |
4010 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 4011 | } |
b1b02fe9 N |
4012 | |
4013 | if (rcw > disks && rmw > disks && | |
4014 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4015 | set_bit(STRIPE_DELAYED, &sh->state); | |
4016 | ||
a4456856 DW |
4017 | /* now if nothing is locked, and if we have enough data, |
4018 | * we can start a write request | |
4019 | */ | |
f38e1219 DW |
4020 | /* since handle_stripe can be called at any time we need to handle the |
4021 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
4022 | * subsequent call wants to start a write request. raid_run_ops only |
4023 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
4024 | * simultaneously. If this is not the case then new writes need to be |
4025 | * held off until the compute completes. | |
4026 | */ | |
976ea8d4 DW |
4027 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
4028 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 4029 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 4030 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 4031 | return 0; |
a4456856 DW |
4032 | } |
4033 | ||
d1688a6d | 4034 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
4035 | struct stripe_head_state *s, int disks) |
4036 | { | |
ecc65c9b | 4037 | struct r5dev *dev = NULL; |
bd2ab670 | 4038 | |
59fc630b | 4039 | BUG_ON(sh->batch_head); |
a4456856 | 4040 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 4041 | |
ecc65c9b DW |
4042 | switch (sh->check_state) { |
4043 | case check_state_idle: | |
4044 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 4045 | if (s->failed == 0) { |
bd2ab670 | 4046 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
4047 | sh->check_state = check_state_run; |
4048 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 4049 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 4050 | s->uptodate--; |
ecc65c9b | 4051 | break; |
bd2ab670 | 4052 | } |
f2b3b44d | 4053 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
4054 | /* fall through */ |
4055 | case check_state_compute_result: | |
4056 | sh->check_state = check_state_idle; | |
4057 | if (!dev) | |
4058 | dev = &sh->dev[sh->pd_idx]; | |
4059 | ||
4060 | /* check that a write has not made the stripe insync */ | |
4061 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4062 | break; | |
c8894419 | 4063 | |
a4456856 | 4064 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
4065 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
4066 | BUG_ON(s->uptodate != disks); | |
4067 | ||
4068 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 4069 | s->locked++; |
a4456856 | 4070 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 4071 | |
a4456856 | 4072 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 4073 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
4074 | break; |
4075 | case check_state_run: | |
4076 | break; /* we will be called again upon completion */ | |
4077 | case check_state_check_result: | |
4078 | sh->check_state = check_state_idle; | |
4079 | ||
4080 | /* if a failure occurred during the check operation, leave | |
4081 | * STRIPE_INSYNC not set and let the stripe be handled again | |
4082 | */ | |
4083 | if (s->failed) | |
4084 | break; | |
4085 | ||
4086 | /* handle a successful check operation, if parity is correct | |
4087 | * we are done. Otherwise update the mismatch count and repair | |
4088 | * parity if !MD_RECOVERY_CHECK | |
4089 | */ | |
ad283ea4 | 4090 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
4091 | /* parity is correct (on disc, |
4092 | * not in buffer any more) | |
4093 | */ | |
4094 | set_bit(STRIPE_INSYNC, &sh->state); | |
4095 | else { | |
7f7583d4 | 4096 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
4097 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
4098 | /* don't try to repair!! */ | |
4099 | set_bit(STRIPE_INSYNC, &sh->state); | |
4100 | else { | |
4101 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 4102 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
4103 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
4104 | set_bit(R5_Wantcompute, | |
4105 | &sh->dev[sh->pd_idx].flags); | |
4106 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 4107 | sh->ops.target2 = -1; |
ecc65c9b DW |
4108 | s->uptodate++; |
4109 | } | |
4110 | } | |
4111 | break; | |
4112 | case check_state_compute_run: | |
4113 | break; | |
4114 | default: | |
cc6167b4 | 4115 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
4116 | __func__, sh->check_state, |
4117 | (unsigned long long) sh->sector); | |
4118 | BUG(); | |
a4456856 DW |
4119 | } |
4120 | } | |
4121 | ||
d1688a6d | 4122 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4123 | struct stripe_head_state *s, |
f2b3b44d | 4124 | int disks) |
a4456856 | 4125 | { |
a4456856 | 4126 | int pd_idx = sh->pd_idx; |
34e04e87 | 4127 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4128 | struct r5dev *dev; |
a4456856 | 4129 | |
59fc630b | 4130 | BUG_ON(sh->batch_head); |
a4456856 DW |
4131 | set_bit(STRIPE_HANDLE, &sh->state); |
4132 | ||
4133 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4134 | |
a4456856 DW |
4135 | /* Want to check and possibly repair P and Q. |
4136 | * However there could be one 'failed' device, in which | |
4137 | * case we can only check one of them, possibly using the | |
4138 | * other to generate missing data | |
4139 | */ | |
4140 | ||
d82dfee0 DW |
4141 | switch (sh->check_state) { |
4142 | case check_state_idle: | |
4143 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4144 | if (s->failed == s->q_failed) { |
d82dfee0 | 4145 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4146 | * makes sense to check P (If anything else were failed, |
4147 | * we would have used P to recreate it). | |
4148 | */ | |
d82dfee0 | 4149 | sh->check_state = check_state_run; |
a4456856 | 4150 | } |
f2b3b44d | 4151 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4152 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4153 | * anything, so it makes sense to check it |
4154 | */ | |
d82dfee0 DW |
4155 | if (sh->check_state == check_state_run) |
4156 | sh->check_state = check_state_run_pq; | |
4157 | else | |
4158 | sh->check_state = check_state_run_q; | |
a4456856 | 4159 | } |
a4456856 | 4160 | |
d82dfee0 DW |
4161 | /* discard potentially stale zero_sum_result */ |
4162 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4163 | |
d82dfee0 DW |
4164 | if (sh->check_state == check_state_run) { |
4165 | /* async_xor_zero_sum destroys the contents of P */ | |
4166 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4167 | s->uptodate--; | |
a4456856 | 4168 | } |
d82dfee0 DW |
4169 | if (sh->check_state >= check_state_run && |
4170 | sh->check_state <= check_state_run_pq) { | |
4171 | /* async_syndrome_zero_sum preserves P and Q, so | |
4172 | * no need to mark them !uptodate here | |
4173 | */ | |
4174 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4175 | break; | |
a4456856 DW |
4176 | } |
4177 | ||
d82dfee0 DW |
4178 | /* we have 2-disk failure */ |
4179 | BUG_ON(s->failed != 2); | |
4180 | /* fall through */ | |
4181 | case check_state_compute_result: | |
4182 | sh->check_state = check_state_idle; | |
a4456856 | 4183 | |
d82dfee0 DW |
4184 | /* check that a write has not made the stripe insync */ |
4185 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4186 | break; | |
a4456856 DW |
4187 | |
4188 | /* now write out any block on a failed drive, | |
d82dfee0 | 4189 | * or P or Q if they were recomputed |
a4456856 | 4190 | */ |
d82dfee0 | 4191 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 4192 | if (s->failed == 2) { |
f2b3b44d | 4193 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4194 | s->locked++; |
4195 | set_bit(R5_LOCKED, &dev->flags); | |
4196 | set_bit(R5_Wantwrite, &dev->flags); | |
4197 | } | |
4198 | if (s->failed >= 1) { | |
f2b3b44d | 4199 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4200 | s->locked++; |
4201 | set_bit(R5_LOCKED, &dev->flags); | |
4202 | set_bit(R5_Wantwrite, &dev->flags); | |
4203 | } | |
d82dfee0 | 4204 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4205 | dev = &sh->dev[pd_idx]; |
4206 | s->locked++; | |
4207 | set_bit(R5_LOCKED, &dev->flags); | |
4208 | set_bit(R5_Wantwrite, &dev->flags); | |
4209 | } | |
d82dfee0 | 4210 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4211 | dev = &sh->dev[qd_idx]; |
4212 | s->locked++; | |
4213 | set_bit(R5_LOCKED, &dev->flags); | |
4214 | set_bit(R5_Wantwrite, &dev->flags); | |
4215 | } | |
4216 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
4217 | ||
4218 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4219 | break; |
4220 | case check_state_run: | |
4221 | case check_state_run_q: | |
4222 | case check_state_run_pq: | |
4223 | break; /* we will be called again upon completion */ | |
4224 | case check_state_check_result: | |
4225 | sh->check_state = check_state_idle; | |
4226 | ||
4227 | /* handle a successful check operation, if parity is correct | |
4228 | * we are done. Otherwise update the mismatch count and repair | |
4229 | * parity if !MD_RECOVERY_CHECK | |
4230 | */ | |
4231 | if (sh->ops.zero_sum_result == 0) { | |
4232 | /* both parities are correct */ | |
4233 | if (!s->failed) | |
4234 | set_bit(STRIPE_INSYNC, &sh->state); | |
4235 | else { | |
4236 | /* in contrast to the raid5 case we can validate | |
4237 | * parity, but still have a failure to write | |
4238 | * back | |
4239 | */ | |
4240 | sh->check_state = check_state_compute_result; | |
4241 | /* Returning at this point means that we may go | |
4242 | * off and bring p and/or q uptodate again so | |
4243 | * we make sure to check zero_sum_result again | |
4244 | * to verify if p or q need writeback | |
4245 | */ | |
4246 | } | |
4247 | } else { | |
7f7583d4 | 4248 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
4249 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
4250 | /* don't try to repair!! */ | |
4251 | set_bit(STRIPE_INSYNC, &sh->state); | |
4252 | else { | |
4253 | int *target = &sh->ops.target; | |
4254 | ||
4255 | sh->ops.target = -1; | |
4256 | sh->ops.target2 = -1; | |
4257 | sh->check_state = check_state_compute_run; | |
4258 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4259 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4260 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4261 | set_bit(R5_Wantcompute, | |
4262 | &sh->dev[pd_idx].flags); | |
4263 | *target = pd_idx; | |
4264 | target = &sh->ops.target2; | |
4265 | s->uptodate++; | |
4266 | } | |
4267 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4268 | set_bit(R5_Wantcompute, | |
4269 | &sh->dev[qd_idx].flags); | |
4270 | *target = qd_idx; | |
4271 | s->uptodate++; | |
4272 | } | |
4273 | } | |
4274 | } | |
4275 | break; | |
4276 | case check_state_compute_run: | |
4277 | break; | |
4278 | default: | |
cc6167b4 N |
4279 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4280 | __func__, sh->check_state, | |
4281 | (unsigned long long) sh->sector); | |
d82dfee0 | 4282 | BUG(); |
a4456856 DW |
4283 | } |
4284 | } | |
4285 | ||
d1688a6d | 4286 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4287 | { |
4288 | int i; | |
4289 | ||
4290 | /* We have read all the blocks in this stripe and now we need to | |
4291 | * copy some of them into a target stripe for expand. | |
4292 | */ | |
f0a50d37 | 4293 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4294 | BUG_ON(sh->batch_head); |
a4456856 DW |
4295 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4296 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4297 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4298 | int dd_idx, j; |
a4456856 | 4299 | struct stripe_head *sh2; |
a08abd8c | 4300 | struct async_submit_ctl submit; |
a4456856 | 4301 | |
6d036f7d | 4302 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4303 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4304 | &dd_idx, NULL); | |
6d036f7d | 4305 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4306 | if (sh2 == NULL) |
4307 | /* so far only the early blocks of this stripe | |
4308 | * have been requested. When later blocks | |
4309 | * get requested, we will try again | |
4310 | */ | |
4311 | continue; | |
4312 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4313 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4314 | /* must have already done this block */ | |
6d036f7d | 4315 | raid5_release_stripe(sh2); |
a4456856 DW |
4316 | continue; |
4317 | } | |
f0a50d37 DW |
4318 | |
4319 | /* place all the copies on one channel */ | |
a08abd8c | 4320 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4321 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 4322 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 4323 | &submit); |
f0a50d37 | 4324 | |
a4456856 DW |
4325 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4326 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4327 | for (j = 0; j < conf->raid_disks; j++) | |
4328 | if (j != sh2->pd_idx && | |
86c374ba | 4329 | j != sh2->qd_idx && |
a4456856 DW |
4330 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4331 | break; | |
4332 | if (j == conf->raid_disks) { | |
4333 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4334 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4335 | } | |
6d036f7d | 4336 | raid5_release_stripe(sh2); |
f0a50d37 | 4337 | |
a4456856 | 4338 | } |
a2e08551 | 4339 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4340 | async_tx_quiesce(&tx); |
a4456856 | 4341 | } |
1da177e4 LT |
4342 | |
4343 | /* | |
4344 | * handle_stripe - do things to a stripe. | |
4345 | * | |
9a3e1101 N |
4346 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4347 | * state of various bits to see what needs to be done. | |
1da177e4 | 4348 | * Possible results: |
9a3e1101 N |
4349 | * return some read requests which now have data |
4350 | * return some write requests which are safely on storage | |
1da177e4 LT |
4351 | * schedule a read on some buffers |
4352 | * schedule a write of some buffers | |
4353 | * return confirmation of parity correctness | |
4354 | * | |
1da177e4 | 4355 | */ |
a4456856 | 4356 | |
acfe726b | 4357 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4358 | { |
d1688a6d | 4359 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4360 | int disks = sh->disks; |
474af965 N |
4361 | struct r5dev *dev; |
4362 | int i; | |
9a3e1101 | 4363 | int do_recovery = 0; |
1da177e4 | 4364 | |
acfe726b N |
4365 | memset(s, 0, sizeof(*s)); |
4366 | ||
dabc4ec6 | 4367 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4368 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4369 | s->failed_num[0] = -1; |
4370 | s->failed_num[1] = -1; | |
6e74a9cf | 4371 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4372 | |
acfe726b | 4373 | /* Now to look around and see what can be done */ |
1da177e4 | 4374 | rcu_read_lock(); |
16a53ecc | 4375 | for (i=disks; i--; ) { |
3cb03002 | 4376 | struct md_rdev *rdev; |
31c176ec N |
4377 | sector_t first_bad; |
4378 | int bad_sectors; | |
4379 | int is_bad = 0; | |
acfe726b | 4380 | |
16a53ecc | 4381 | dev = &sh->dev[i]; |
1da177e4 | 4382 | |
45b4233c | 4383 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4384 | i, dev->flags, |
4385 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4386 | /* maybe we can reply to a read |
4387 | * | |
4388 | * new wantfill requests are only permitted while | |
4389 | * ops_complete_biofill is guaranteed to be inactive | |
4390 | */ | |
4391 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4392 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4393 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4394 | |
16a53ecc | 4395 | /* now count some things */ |
cc94015a N |
4396 | if (test_bit(R5_LOCKED, &dev->flags)) |
4397 | s->locked++; | |
4398 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4399 | s->uptodate++; | |
2d6e4ecc | 4400 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4401 | s->compute++; |
4402 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4403 | } |
1da177e4 | 4404 | |
acfe726b | 4405 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4406 | s->to_fill++; |
acfe726b | 4407 | else if (dev->toread) |
cc94015a | 4408 | s->to_read++; |
16a53ecc | 4409 | if (dev->towrite) { |
cc94015a | 4410 | s->to_write++; |
16a53ecc | 4411 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4412 | s->non_overwrite++; |
16a53ecc | 4413 | } |
a4456856 | 4414 | if (dev->written) |
cc94015a | 4415 | s->written++; |
14a75d3e N |
4416 | /* Prefer to use the replacement for reads, but only |
4417 | * if it is recovered enough and has no bad blocks. | |
4418 | */ | |
4419 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4420 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4421 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4422 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4423 | &first_bad, &bad_sectors)) | |
4424 | set_bit(R5_ReadRepl, &dev->flags); | |
4425 | else { | |
e6030cb0 | 4426 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4427 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4428 | else |
4429 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4430 | rdev = rcu_dereference(conf->disks[i].rdev); |
4431 | clear_bit(R5_ReadRepl, &dev->flags); | |
4432 | } | |
9283d8c5 N |
4433 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4434 | rdev = NULL; | |
31c176ec N |
4435 | if (rdev) { |
4436 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4437 | &first_bad, &bad_sectors); | |
4438 | if (s->blocked_rdev == NULL | |
4439 | && (test_bit(Blocked, &rdev->flags) | |
4440 | || is_bad < 0)) { | |
4441 | if (is_bad < 0) | |
4442 | set_bit(BlockedBadBlocks, | |
4443 | &rdev->flags); | |
4444 | s->blocked_rdev = rdev; | |
4445 | atomic_inc(&rdev->nr_pending); | |
4446 | } | |
6bfe0b49 | 4447 | } |
415e72d0 N |
4448 | clear_bit(R5_Insync, &dev->flags); |
4449 | if (!rdev) | |
4450 | /* Not in-sync */; | |
31c176ec N |
4451 | else if (is_bad) { |
4452 | /* also not in-sync */ | |
18b9837e N |
4453 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4454 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4455 | /* treat as in-sync, but with a read error |
4456 | * which we can now try to correct | |
4457 | */ | |
4458 | set_bit(R5_Insync, &dev->flags); | |
4459 | set_bit(R5_ReadError, &dev->flags); | |
4460 | } | |
4461 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4462 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4463 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4464 | /* in sync if before recovery_offset */ |
30d7a483 N |
4465 | set_bit(R5_Insync, &dev->flags); |
4466 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4467 | test_bit(R5_Expanded, &dev->flags)) | |
4468 | /* If we've reshaped into here, we assume it is Insync. | |
4469 | * We will shortly update recovery_offset to make | |
4470 | * it official. | |
4471 | */ | |
4472 | set_bit(R5_Insync, &dev->flags); | |
4473 | ||
1cc03eb9 | 4474 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4475 | /* This flag does not apply to '.replacement' |
4476 | * only to .rdev, so make sure to check that*/ | |
4477 | struct md_rdev *rdev2 = rcu_dereference( | |
4478 | conf->disks[i].rdev); | |
4479 | if (rdev2 == rdev) | |
4480 | clear_bit(R5_Insync, &dev->flags); | |
4481 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4482 | s->handle_bad_blocks = 1; |
14a75d3e | 4483 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4484 | } else |
4485 | clear_bit(R5_WriteError, &dev->flags); | |
4486 | } | |
1cc03eb9 | 4487 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4488 | /* This flag does not apply to '.replacement' |
4489 | * only to .rdev, so make sure to check that*/ | |
4490 | struct md_rdev *rdev2 = rcu_dereference( | |
4491 | conf->disks[i].rdev); | |
4492 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4493 | s->handle_bad_blocks = 1; |
14a75d3e | 4494 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4495 | } else |
4496 | clear_bit(R5_MadeGood, &dev->flags); | |
4497 | } | |
977df362 N |
4498 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4499 | struct md_rdev *rdev2 = rcu_dereference( | |
4500 | conf->disks[i].replacement); | |
4501 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4502 | s->handle_bad_blocks = 1; | |
4503 | atomic_inc(&rdev2->nr_pending); | |
4504 | } else | |
4505 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4506 | } | |
415e72d0 | 4507 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4508 | /* The ReadError flag will just be confusing now */ |
4509 | clear_bit(R5_ReadError, &dev->flags); | |
4510 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4511 | } |
415e72d0 N |
4512 | if (test_bit(R5_ReadError, &dev->flags)) |
4513 | clear_bit(R5_Insync, &dev->flags); | |
4514 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4515 | if (s->failed < 2) |
4516 | s->failed_num[s->failed] = i; | |
4517 | s->failed++; | |
9a3e1101 N |
4518 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4519 | do_recovery = 1; | |
415e72d0 | 4520 | } |
2ded3703 SL |
4521 | |
4522 | if (test_bit(R5_InJournal, &dev->flags)) | |
4523 | s->injournal++; | |
1e6d690b SL |
4524 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4525 | s->just_cached++; | |
1da177e4 | 4526 | } |
9a3e1101 N |
4527 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4528 | /* If there is a failed device being replaced, | |
4529 | * we must be recovering. | |
4530 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4531 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4532 | * else we can only be replacing |
4533 | * sync and recovery both need to read all devices, and so | |
4534 | * use the same flag. | |
4535 | */ | |
4536 | if (do_recovery || | |
c6d2e084 | 4537 | sh->sector >= conf->mddev->recovery_cp || |
4538 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4539 | s->syncing = 1; |
4540 | else | |
4541 | s->replacing = 1; | |
4542 | } | |
1da177e4 | 4543 | rcu_read_unlock(); |
cc94015a N |
4544 | } |
4545 | ||
59fc630b | 4546 | static int clear_batch_ready(struct stripe_head *sh) |
4547 | { | |
b15a9dbd N |
4548 | /* Return '1' if this is a member of batch, or |
4549 | * '0' if it is a lone stripe or a head which can now be | |
4550 | * handled. | |
4551 | */ | |
59fc630b | 4552 | struct stripe_head *tmp; |
4553 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4554 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4555 | spin_lock(&sh->stripe_lock); |
4556 | if (!sh->batch_head) { | |
4557 | spin_unlock(&sh->stripe_lock); | |
4558 | return 0; | |
4559 | } | |
4560 | ||
4561 | /* | |
4562 | * this stripe could be added to a batch list before we check | |
4563 | * BATCH_READY, skips it | |
4564 | */ | |
4565 | if (sh->batch_head != sh) { | |
4566 | spin_unlock(&sh->stripe_lock); | |
4567 | return 1; | |
4568 | } | |
4569 | spin_lock(&sh->batch_lock); | |
4570 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4571 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4572 | spin_unlock(&sh->batch_lock); | |
4573 | spin_unlock(&sh->stripe_lock); | |
4574 | ||
4575 | /* | |
4576 | * BATCH_READY is cleared, no new stripes can be added. | |
4577 | * batch_list can be accessed without lock | |
4578 | */ | |
4579 | return 0; | |
4580 | } | |
4581 | ||
3960ce79 N |
4582 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4583 | unsigned long handle_flags) | |
72ac7330 | 4584 | { |
4e3d62ff | 4585 | struct stripe_head *sh, *next; |
72ac7330 | 4586 | int i; |
fb642b92 | 4587 | int do_wakeup = 0; |
72ac7330 | 4588 | |
bb27051f N |
4589 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4590 | ||
72ac7330 | 4591 | list_del_init(&sh->batch_list); |
4592 | ||
fb3229d5 | 4593 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4594 | (1 << STRIPE_SYNCING) | |
4595 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4596 | (1 << STRIPE_DELAYED) | |
4597 | (1 << STRIPE_BIT_DELAY) | | |
4598 | (1 << STRIPE_FULL_WRITE) | | |
4599 | (1 << STRIPE_BIOFILL_RUN) | | |
4600 | (1 << STRIPE_COMPUTE_RUN) | | |
4601 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4602 | (1 << STRIPE_DISCARD) | | |
4603 | (1 << STRIPE_BATCH_READY) | | |
4604 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4605 | (1 << STRIPE_BITMAP_PENDING)), |
4606 | "stripe state: %lx\n", sh->state); | |
4607 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4608 | (1 << STRIPE_REPLACED)), | |
4609 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4610 | |
4611 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4612 | (1 << STRIPE_PREREAD_ACTIVE) | |
1b956f7a N |
4613 | (1 << STRIPE_DEGRADED)), |
4614 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4615 | ||
72ac7330 | 4616 | sh->check_state = head_sh->check_state; |
4617 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4618 | for (i = 0; i < sh->disks; i++) { |
4619 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4620 | do_wakeup = 1; | |
72ac7330 | 4621 | sh->dev[i].flags = head_sh->dev[i].flags & |
4622 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4623 | } |
72ac7330 | 4624 | spin_lock_irq(&sh->stripe_lock); |
4625 | sh->batch_head = NULL; | |
4626 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4627 | if (handle_flags == 0 || |
4628 | sh->state & handle_flags) | |
4629 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4630 | raid5_release_stripe(sh); |
72ac7330 | 4631 | } |
fb642b92 N |
4632 | spin_lock_irq(&head_sh->stripe_lock); |
4633 | head_sh->batch_head = NULL; | |
4634 | spin_unlock_irq(&head_sh->stripe_lock); | |
4635 | for (i = 0; i < head_sh->disks; i++) | |
4636 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4637 | do_wakeup = 1; | |
3960ce79 N |
4638 | if (head_sh->state & handle_flags) |
4639 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4640 | |
4641 | if (do_wakeup) | |
4642 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4643 | } |
4644 | ||
cc94015a N |
4645 | static void handle_stripe(struct stripe_head *sh) |
4646 | { | |
4647 | struct stripe_head_state s; | |
d1688a6d | 4648 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4649 | int i; |
84789554 N |
4650 | int prexor; |
4651 | int disks = sh->disks; | |
474af965 | 4652 | struct r5dev *pdev, *qdev; |
cc94015a N |
4653 | |
4654 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4655 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4656 | /* already being handled, ensure it gets handled |
4657 | * again when current action finishes */ | |
4658 | set_bit(STRIPE_HANDLE, &sh->state); | |
4659 | return; | |
4660 | } | |
4661 | ||
59fc630b | 4662 | if (clear_batch_ready(sh) ) { |
4663 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4664 | return; | |
4665 | } | |
4666 | ||
4e3d62ff | 4667 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4668 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4669 | |
dabc4ec6 | 4670 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4671 | spin_lock(&sh->stripe_lock); |
4672 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4673 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4674 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4675 | set_bit(STRIPE_SYNCING, &sh->state); | |
4676 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4677 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4678 | } |
4679 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4680 | } |
4681 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4682 | ||
4683 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4684 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4685 | (unsigned long long)sh->sector, sh->state, | |
4686 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4687 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4688 | |
acfe726b | 4689 | analyse_stripe(sh, &s); |
c5a31000 | 4690 | |
b70abcb2 SL |
4691 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4692 | goto finish; | |
4693 | ||
bc2607f3 N |
4694 | if (s.handle_bad_blocks) { |
4695 | set_bit(STRIPE_HANDLE, &sh->state); | |
4696 | goto finish; | |
4697 | } | |
4698 | ||
474af965 N |
4699 | if (unlikely(s.blocked_rdev)) { |
4700 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4701 | s.replacing || s.to_write || s.written) { |
474af965 N |
4702 | set_bit(STRIPE_HANDLE, &sh->state); |
4703 | goto finish; | |
4704 | } | |
4705 | /* There is nothing for the blocked_rdev to block */ | |
4706 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4707 | s.blocked_rdev = NULL; | |
4708 | } | |
4709 | ||
4710 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4711 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4712 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4713 | } | |
4714 | ||
4715 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4716 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4717 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4718 | s.failed_num[0], s.failed_num[1]); | |
4719 | /* check if the array has lost more than max_degraded devices and, | |
4720 | * if so, some requests might need to be failed. | |
4721 | */ | |
6e74a9cf | 4722 | if (s.failed > conf->max_degraded || s.log_failed) { |
9a3f530f N |
4723 | sh->check_state = 0; |
4724 | sh->reconstruct_state = 0; | |
626f2092 | 4725 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4726 | if (s.to_read+s.to_write+s.written) |
4727 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4728 | if (s.syncing + s.replacing) |
9a3f530f N |
4729 | handle_failed_sync(conf, sh, &s); |
4730 | } | |
474af965 | 4731 | |
84789554 N |
4732 | /* Now we check to see if any write operations have recently |
4733 | * completed | |
4734 | */ | |
4735 | prexor = 0; | |
4736 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4737 | prexor = 1; | |
4738 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4739 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4740 | sh->reconstruct_state = reconstruct_state_idle; | |
4741 | ||
4742 | /* All the 'written' buffers and the parity block are ready to | |
4743 | * be written back to disk | |
4744 | */ | |
9e444768 SL |
4745 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4746 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4747 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4748 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4749 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4750 | for (i = disks; i--; ) { |
4751 | struct r5dev *dev = &sh->dev[i]; | |
4752 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4753 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4754 | dev->written || test_bit(R5_InJournal, |
4755 | &dev->flags))) { | |
84789554 N |
4756 | pr_debug("Writing block %d\n", i); |
4757 | set_bit(R5_Wantwrite, &dev->flags); | |
4758 | if (prexor) | |
4759 | continue; | |
9c4bdf69 N |
4760 | if (s.failed > 1) |
4761 | continue; | |
84789554 N |
4762 | if (!test_bit(R5_Insync, &dev->flags) || |
4763 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4764 | s.failed == 0)) | |
4765 | set_bit(STRIPE_INSYNC, &sh->state); | |
4766 | } | |
4767 | } | |
4768 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4769 | s.dec_preread_active = 1; | |
4770 | } | |
4771 | ||
ef5b7c69 N |
4772 | /* |
4773 | * might be able to return some write requests if the parity blocks | |
4774 | * are safe, or on a failed drive | |
4775 | */ | |
4776 | pdev = &sh->dev[sh->pd_idx]; | |
4777 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4778 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4779 | qdev = &sh->dev[sh->qd_idx]; | |
4780 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4781 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4782 | || conf->level < 6; | |
4783 | ||
4784 | if (s.written && | |
4785 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4786 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4787 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4788 | test_bit(R5_Discard, &pdev->flags))))) && | |
4789 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4790 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4791 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4792 | test_bit(R5_Discard, &qdev->flags)))))) | |
4793 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4794 | ||
1e6d690b SL |
4795 | if (s.just_cached) |
4796 | r5c_handle_cached_data_endio(conf, sh, disks, &s.return_bi); | |
ff875738 | 4797 | log_stripe_write_finished(sh); |
1e6d690b | 4798 | |
ef5b7c69 N |
4799 | /* Now we might consider reading some blocks, either to check/generate |
4800 | * parity, or to satisfy requests | |
4801 | * or to load a block that is being partially written. | |
4802 | */ | |
4803 | if (s.to_read || s.non_overwrite | |
4804 | || (conf->level == 6 && s.to_write && s.failed) | |
4805 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4806 | || s.replacing | |
4807 | || s.expanding) | |
4808 | handle_stripe_fill(sh, &s, disks); | |
4809 | ||
2ded3703 SL |
4810 | /* |
4811 | * When the stripe finishes full journal write cycle (write to journal | |
4812 | * and raid disk), this is the clean up procedure so it is ready for | |
4813 | * next operation. | |
4814 | */ | |
4815 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4816 | ||
4817 | /* | |
4818 | * Now to consider new write requests, cache write back and what else, | |
4819 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4820 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4821 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4822 | * block. | |
2ded3703 | 4823 | * 3/ A r5c cache log write is in flight. |
84789554 | 4824 | */ |
2ded3703 SL |
4825 | |
4826 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4827 | if (!r5c_is_writeback(conf->log)) { | |
4828 | if (s.to_write) | |
4829 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4830 | } else { /* write back cache */ | |
4831 | int ret = 0; | |
4832 | ||
4833 | /* First, try handle writes in caching phase */ | |
4834 | if (s.to_write) | |
4835 | ret = r5c_try_caching_write(conf, sh, &s, | |
4836 | disks); | |
4837 | /* | |
4838 | * If caching phase failed: ret == -EAGAIN | |
4839 | * OR | |
4840 | * stripe under reclaim: !caching && injournal | |
4841 | * | |
4842 | * fall back to handle_stripe_dirtying() | |
4843 | */ | |
4844 | if (ret == -EAGAIN || | |
4845 | /* stripe under reclaim: !caching && injournal */ | |
4846 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4847 | s.injournal > 0)) { |
4848 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4849 | disks); | |
4850 | if (ret == -EAGAIN) | |
4851 | goto finish; | |
4852 | } | |
2ded3703 SL |
4853 | } |
4854 | } | |
84789554 N |
4855 | |
4856 | /* maybe we need to check and possibly fix the parity for this stripe | |
4857 | * Any reads will already have been scheduled, so we just see if enough | |
4858 | * data is available. The parity check is held off while parity | |
4859 | * dependent operations are in flight. | |
4860 | */ | |
4861 | if (sh->check_state || | |
4862 | (s.syncing && s.locked == 0 && | |
4863 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4864 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4865 | if (conf->level == 6) | |
4866 | handle_parity_checks6(conf, sh, &s, disks); | |
4867 | else | |
4868 | handle_parity_checks5(conf, sh, &s, disks); | |
4869 | } | |
c5a31000 | 4870 | |
f94c0b66 N |
4871 | if ((s.replacing || s.syncing) && s.locked == 0 |
4872 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4873 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4874 | /* Write out to replacement devices where possible */ |
4875 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4876 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4877 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4878 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4879 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4880 | s.locked++; | |
4881 | } | |
f94c0b66 N |
4882 | if (s.replacing) |
4883 | set_bit(STRIPE_INSYNC, &sh->state); | |
4884 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4885 | } |
4886 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4887 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4888 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4889 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4890 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4891 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4892 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4893 | } |
4894 | ||
4895 | /* If the failed drives are just a ReadError, then we might need | |
4896 | * to progress the repair/check process | |
4897 | */ | |
4898 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4899 | for (i = 0; i < s.failed; i++) { | |
4900 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4901 | if (test_bit(R5_ReadError, &dev->flags) | |
4902 | && !test_bit(R5_LOCKED, &dev->flags) | |
4903 | && test_bit(R5_UPTODATE, &dev->flags) | |
4904 | ) { | |
4905 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4906 | set_bit(R5_Wantwrite, &dev->flags); | |
4907 | set_bit(R5_ReWrite, &dev->flags); | |
4908 | set_bit(R5_LOCKED, &dev->flags); | |
4909 | s.locked++; | |
4910 | } else { | |
4911 | /* let's read it back */ | |
4912 | set_bit(R5_Wantread, &dev->flags); | |
4913 | set_bit(R5_LOCKED, &dev->flags); | |
4914 | s.locked++; | |
4915 | } | |
4916 | } | |
4917 | } | |
4918 | ||
3687c061 N |
4919 | /* Finish reconstruct operations initiated by the expansion process */ |
4920 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4921 | struct stripe_head *sh_src | |
6d036f7d | 4922 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4923 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4924 | /* sh cannot be written until sh_src has been read. | |
4925 | * so arrange for sh to be delayed a little | |
4926 | */ | |
4927 | set_bit(STRIPE_DELAYED, &sh->state); | |
4928 | set_bit(STRIPE_HANDLE, &sh->state); | |
4929 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4930 | &sh_src->state)) | |
4931 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4932 | raid5_release_stripe(sh_src); |
3687c061 N |
4933 | goto finish; |
4934 | } | |
4935 | if (sh_src) | |
6d036f7d | 4936 | raid5_release_stripe(sh_src); |
3687c061 N |
4937 | |
4938 | sh->reconstruct_state = reconstruct_state_idle; | |
4939 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4940 | for (i = conf->raid_disks; i--; ) { | |
4941 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4942 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4943 | s.locked++; | |
4944 | } | |
4945 | } | |
f416885e | 4946 | |
3687c061 N |
4947 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4948 | !sh->reconstruct_state) { | |
4949 | /* Need to write out all blocks after computing parity */ | |
4950 | sh->disks = conf->raid_disks; | |
4951 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4952 | schedule_reconstruction(sh, &s, 1, 1); | |
4953 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4954 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4955 | atomic_dec(&conf->reshape_stripes); | |
4956 | wake_up(&conf->wait_for_overlap); | |
4957 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4958 | } | |
4959 | ||
4960 | if (s.expanding && s.locked == 0 && | |
4961 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4962 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4963 | |
3687c061 | 4964 | finish: |
6bfe0b49 | 4965 | /* wait for this device to become unblocked */ |
5f066c63 N |
4966 | if (unlikely(s.blocked_rdev)) { |
4967 | if (conf->mddev->external) | |
4968 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4969 | conf->mddev); | |
4970 | else | |
4971 | /* Internal metadata will immediately | |
4972 | * be written by raid5d, so we don't | |
4973 | * need to wait here. | |
4974 | */ | |
4975 | rdev_dec_pending(s.blocked_rdev, | |
4976 | conf->mddev); | |
4977 | } | |
6bfe0b49 | 4978 | |
bc2607f3 N |
4979 | if (s.handle_bad_blocks) |
4980 | for (i = disks; i--; ) { | |
3cb03002 | 4981 | struct md_rdev *rdev; |
bc2607f3 N |
4982 | struct r5dev *dev = &sh->dev[i]; |
4983 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4984 | /* We own a safe reference to the rdev */ | |
4985 | rdev = conf->disks[i].rdev; | |
4986 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4987 | STRIPE_SECTORS, 0)) | |
4988 | md_error(conf->mddev, rdev); | |
4989 | rdev_dec_pending(rdev, conf->mddev); | |
4990 | } | |
b84db560 N |
4991 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4992 | rdev = conf->disks[i].rdev; | |
4993 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4994 | STRIPE_SECTORS, 0); |
b84db560 N |
4995 | rdev_dec_pending(rdev, conf->mddev); |
4996 | } | |
977df362 N |
4997 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4998 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4999 | if (!rdev) |
5000 | /* rdev have been moved down */ | |
5001 | rdev = conf->disks[i].rdev; | |
977df362 | 5002 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 5003 | STRIPE_SECTORS, 0); |
977df362 N |
5004 | rdev_dec_pending(rdev, conf->mddev); |
5005 | } | |
bc2607f3 N |
5006 | } |
5007 | ||
6c0069c0 YT |
5008 | if (s.ops_request) |
5009 | raid_run_ops(sh, s.ops_request); | |
5010 | ||
f0e43bcd | 5011 | ops_run_io(sh, &s); |
16a53ecc | 5012 | |
c5709ef6 | 5013 | if (s.dec_preread_active) { |
729a1866 | 5014 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 5015 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
5016 | * have actually been submitted. |
5017 | */ | |
5018 | atomic_dec(&conf->preread_active_stripes); | |
5019 | if (atomic_read(&conf->preread_active_stripes) < | |
5020 | IO_THRESHOLD) | |
5021 | md_wakeup_thread(conf->mddev->thread); | |
5022 | } | |
5023 | ||
c3cce6cd | 5024 | if (!bio_list_empty(&s.return_bi)) { |
2953079c | 5025 | if (test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { |
c3cce6cd N |
5026 | spin_lock_irq(&conf->device_lock); |
5027 | bio_list_merge(&conf->return_bi, &s.return_bi); | |
5028 | spin_unlock_irq(&conf->device_lock); | |
5029 | md_wakeup_thread(conf->mddev->thread); | |
5030 | } else | |
5031 | return_io(&s.return_bi); | |
5032 | } | |
16a53ecc | 5033 | |
257a4b42 | 5034 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
5035 | } |
5036 | ||
d1688a6d | 5037 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
5038 | { |
5039 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
5040 | while (!list_empty(&conf->delayed_list)) { | |
5041 | struct list_head *l = conf->delayed_list.next; | |
5042 | struct stripe_head *sh; | |
5043 | sh = list_entry(l, struct stripe_head, lru); | |
5044 | list_del_init(l); | |
5045 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5046 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5047 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 5048 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 5049 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 5050 | } |
482c0834 | 5051 | } |
16a53ecc N |
5052 | } |
5053 | ||
566c09c5 SL |
5054 | static void activate_bit_delay(struct r5conf *conf, |
5055 | struct list_head *temp_inactive_list) | |
16a53ecc N |
5056 | { |
5057 | /* device_lock is held */ | |
5058 | struct list_head head; | |
5059 | list_add(&head, &conf->bitmap_list); | |
5060 | list_del_init(&conf->bitmap_list); | |
5061 | while (!list_empty(&head)) { | |
5062 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 5063 | int hash; |
16a53ecc N |
5064 | list_del_init(&sh->lru); |
5065 | atomic_inc(&sh->count); | |
566c09c5 SL |
5066 | hash = sh->hash_lock_index; |
5067 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
5068 | } |
5069 | } | |
5070 | ||
5c675f83 | 5071 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 5072 | { |
d1688a6d | 5073 | struct r5conf *conf = mddev->private; |
f022b2fd N |
5074 | |
5075 | /* No difference between reads and writes. Just check | |
5076 | * how busy the stripe_cache is | |
5077 | */ | |
3fa841d7 | 5078 | |
5423399a | 5079 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd | 5080 | return 1; |
a39f7afd SL |
5081 | |
5082 | /* Also checks whether there is pressure on r5cache log space */ | |
5083 | if (test_bit(R5C_LOG_TIGHT, &conf->cache_state)) | |
5084 | return 1; | |
f022b2fd N |
5085 | if (conf->quiesce) |
5086 | return 1; | |
4bda556a | 5087 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
5088 | return 1; |
5089 | ||
5090 | return 0; | |
5091 | } | |
5092 | ||
fd01b88c | 5093 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 5094 | { |
3cb5edf4 | 5095 | struct r5conf *conf = mddev->private; |
4f024f37 | 5096 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 5097 | unsigned int chunk_sectors; |
aa8b57aa | 5098 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 5099 | |
3cb5edf4 | 5100 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
5101 | return chunk_sectors >= |
5102 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
5103 | } | |
5104 | ||
46031f9a RBJ |
5105 | /* |
5106 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
5107 | * later sampled by raid5d. | |
5108 | */ | |
d1688a6d | 5109 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
5110 | { |
5111 | unsigned long flags; | |
5112 | ||
5113 | spin_lock_irqsave(&conf->device_lock, flags); | |
5114 | ||
5115 | bi->bi_next = conf->retry_read_aligned_list; | |
5116 | conf->retry_read_aligned_list = bi; | |
5117 | ||
5118 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5119 | md_wakeup_thread(conf->mddev->thread); | |
5120 | } | |
5121 | ||
d1688a6d | 5122 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
5123 | { |
5124 | struct bio *bi; | |
5125 | ||
5126 | bi = conf->retry_read_aligned; | |
5127 | if (bi) { | |
5128 | conf->retry_read_aligned = NULL; | |
5129 | return bi; | |
5130 | } | |
5131 | bi = conf->retry_read_aligned_list; | |
5132 | if(bi) { | |
387bb173 | 5133 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5134 | bi->bi_next = NULL; |
960e739d JA |
5135 | /* |
5136 | * this sets the active strip count to 1 and the processed | |
5137 | * strip count to zero (upper 8 bits) | |
5138 | */ | |
e7836bd6 | 5139 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
5140 | } |
5141 | ||
5142 | return bi; | |
5143 | } | |
5144 | ||
f679623f RBJ |
5145 | /* |
5146 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5147 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5148 | * first). | |
5149 | * If the read failed.. | |
5150 | */ | |
4246a0b6 | 5151 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5152 | { |
5153 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5154 | struct mddev *mddev; |
d1688a6d | 5155 | struct r5conf *conf; |
3cb03002 | 5156 | struct md_rdev *rdev; |
9b81c842 | 5157 | int error = bi->bi_error; |
46031f9a | 5158 | |
f679623f | 5159 | bio_put(bi); |
46031f9a | 5160 | |
46031f9a RBJ |
5161 | rdev = (void*)raid_bi->bi_next; |
5162 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5163 | mddev = rdev->mddev; |
5164 | conf = mddev->private; | |
46031f9a RBJ |
5165 | |
5166 | rdev_dec_pending(rdev, conf->mddev); | |
5167 | ||
9b81c842 | 5168 | if (!error) { |
0a82a8d1 LT |
5169 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
5170 | raid_bi, 0); | |
4246a0b6 | 5171 | bio_endio(raid_bi); |
46031f9a | 5172 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5173 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5174 | return; |
46031f9a RBJ |
5175 | } |
5176 | ||
45b4233c | 5177 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5178 | |
5179 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5180 | } |
5181 | ||
7ef6b12a | 5182 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5183 | { |
d1688a6d | 5184 | struct r5conf *conf = mddev->private; |
8553fe7e | 5185 | int dd_idx; |
f679623f | 5186 | struct bio* align_bi; |
3cb03002 | 5187 | struct md_rdev *rdev; |
671488cc | 5188 | sector_t end_sector; |
f679623f RBJ |
5189 | |
5190 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5191 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5192 | return 0; |
5193 | } | |
5194 | /* | |
d7a10308 | 5195 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5196 | */ |
d7a10308 | 5197 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, mddev->bio_set); |
f679623f RBJ |
5198 | if (!align_bi) |
5199 | return 0; | |
5200 | /* | |
5201 | * set bi_end_io to a new function, and set bi_private to the | |
5202 | * original bio. | |
5203 | */ | |
5204 | align_bi->bi_end_io = raid5_align_endio; | |
5205 | align_bi->bi_private = raid_bio; | |
5206 | /* | |
5207 | * compute position | |
5208 | */ | |
4f024f37 KO |
5209 | align_bi->bi_iter.bi_sector = |
5210 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5211 | 0, &dd_idx, NULL); | |
f679623f | 5212 | |
f73a1c7d | 5213 | end_sector = bio_end_sector(align_bi); |
f679623f | 5214 | rcu_read_lock(); |
671488cc N |
5215 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5216 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5217 | rdev->recovery_offset < end_sector) { | |
5218 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5219 | if (rdev && | |
5220 | (test_bit(Faulty, &rdev->flags) || | |
5221 | !(test_bit(In_sync, &rdev->flags) || | |
5222 | rdev->recovery_offset >= end_sector))) | |
5223 | rdev = NULL; | |
5224 | } | |
03b047f4 SL |
5225 | |
5226 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5227 | rcu_read_unlock(); | |
5228 | bio_put(align_bi); | |
5229 | return 0; | |
5230 | } | |
5231 | ||
671488cc | 5232 | if (rdev) { |
31c176ec N |
5233 | sector_t first_bad; |
5234 | int bad_sectors; | |
5235 | ||
f679623f RBJ |
5236 | atomic_inc(&rdev->nr_pending); |
5237 | rcu_read_unlock(); | |
46031f9a RBJ |
5238 | raid_bio->bi_next = (void*)rdev; |
5239 | align_bi->bi_bdev = rdev->bdev; | |
b7c44ed9 | 5240 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 5241 | |
7140aafc | 5242 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5243 | bio_sectors(align_bi), |
31c176ec | 5244 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5245 | bio_put(align_bi); |
5246 | rdev_dec_pending(rdev, mddev); | |
5247 | return 0; | |
5248 | } | |
5249 | ||
6c0544e2 | 5250 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5251 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5252 | |
46031f9a | 5253 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5254 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5255 | conf->quiesce == 0, |
eed8c02e | 5256 | conf->device_lock); |
46031f9a RBJ |
5257 | atomic_inc(&conf->active_aligned_reads); |
5258 | spin_unlock_irq(&conf->device_lock); | |
5259 | ||
e3620a3a JB |
5260 | if (mddev->gendisk) |
5261 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
5262 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 5263 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
5264 | generic_make_request(align_bi); |
5265 | return 1; | |
5266 | } else { | |
5267 | rcu_read_unlock(); | |
46031f9a | 5268 | bio_put(align_bi); |
f679623f RBJ |
5269 | return 0; |
5270 | } | |
5271 | } | |
5272 | ||
7ef6b12a ML |
5273 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5274 | { | |
5275 | struct bio *split; | |
5276 | ||
5277 | do { | |
5278 | sector_t sector = raid_bio->bi_iter.bi_sector; | |
5279 | unsigned chunk_sects = mddev->chunk_sectors; | |
5280 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
5281 | ||
5282 | if (sectors < bio_sectors(raid_bio)) { | |
5283 | split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set); | |
5284 | bio_chain(split, raid_bio); | |
5285 | } else | |
5286 | split = raid_bio; | |
5287 | ||
5288 | if (!raid5_read_one_chunk(mddev, split)) { | |
5289 | if (split != raid_bio) | |
5290 | generic_make_request(raid_bio); | |
5291 | return split; | |
5292 | } | |
5293 | } while (split != raid_bio); | |
5294 | ||
5295 | return NULL; | |
5296 | } | |
5297 | ||
8b3e6cdc DW |
5298 | /* __get_priority_stripe - get the next stripe to process |
5299 | * | |
5300 | * Full stripe writes are allowed to pass preread active stripes up until | |
5301 | * the bypass_threshold is exceeded. In general the bypass_count | |
5302 | * increments when the handle_list is handled before the hold_list; however, it | |
5303 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5304 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5305 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5306 | * handle_list. | |
5307 | */ | |
851c30c9 | 5308 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5309 | { |
535ae4eb | 5310 | struct stripe_head *sh, *tmp; |
851c30c9 | 5311 | struct list_head *handle_list = NULL; |
535ae4eb SL |
5312 | struct r5worker_group *wg; |
5313 | bool second_try = !r5c_is_writeback(conf->log); | |
5314 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state); | |
851c30c9 | 5315 | |
535ae4eb SL |
5316 | again: |
5317 | wg = NULL; | |
5318 | sh = NULL; | |
851c30c9 | 5319 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5320 | handle_list = try_loprio ? &conf->loprio_list : |
5321 | &conf->handle_list; | |
851c30c9 | 5322 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5323 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5324 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5325 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5326 | } else { |
5327 | int i; | |
5328 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5329 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5330 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5331 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5332 | if (!list_empty(handle_list)) |
5333 | break; | |
5334 | } | |
5335 | } | |
8b3e6cdc DW |
5336 | |
5337 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5338 | __func__, | |
851c30c9 | 5339 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5340 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5341 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5342 | ||
851c30c9 SL |
5343 | if (!list_empty(handle_list)) { |
5344 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5345 | |
5346 | if (list_empty(&conf->hold_list)) | |
5347 | conf->bypass_count = 0; | |
5348 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5349 | if (conf->hold_list.next == conf->last_hold) | |
5350 | conf->bypass_count++; | |
5351 | else { | |
5352 | conf->last_hold = conf->hold_list.next; | |
5353 | conf->bypass_count -= conf->bypass_threshold; | |
5354 | if (conf->bypass_count < 0) | |
5355 | conf->bypass_count = 0; | |
5356 | } | |
5357 | } | |
5358 | } else if (!list_empty(&conf->hold_list) && | |
5359 | ((conf->bypass_threshold && | |
5360 | conf->bypass_count > conf->bypass_threshold) || | |
5361 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5362 | |
5363 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5364 | if (conf->worker_cnt_per_group == 0 || | |
5365 | group == ANY_GROUP || | |
5366 | !cpu_online(tmp->cpu) || | |
5367 | cpu_to_group(tmp->cpu) == group) { | |
5368 | sh = tmp; | |
5369 | break; | |
5370 | } | |
5371 | } | |
5372 | ||
5373 | if (sh) { | |
5374 | conf->bypass_count -= conf->bypass_threshold; | |
5375 | if (conf->bypass_count < 0) | |
5376 | conf->bypass_count = 0; | |
5377 | } | |
bfc90cb0 | 5378 | wg = NULL; |
851c30c9 SL |
5379 | } |
5380 | ||
535ae4eb SL |
5381 | if (!sh) { |
5382 | if (second_try) | |
5383 | return NULL; | |
5384 | second_try = true; | |
5385 | try_loprio = !try_loprio; | |
5386 | goto again; | |
5387 | } | |
8b3e6cdc | 5388 | |
bfc90cb0 SL |
5389 | if (wg) { |
5390 | wg->stripes_cnt--; | |
5391 | sh->group = NULL; | |
5392 | } | |
8b3e6cdc | 5393 | list_del_init(&sh->lru); |
c7a6d35e | 5394 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5395 | return sh; |
5396 | } | |
f679623f | 5397 | |
8811b596 SL |
5398 | struct raid5_plug_cb { |
5399 | struct blk_plug_cb cb; | |
5400 | struct list_head list; | |
566c09c5 | 5401 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5402 | }; |
5403 | ||
5404 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5405 | { | |
5406 | struct raid5_plug_cb *cb = container_of( | |
5407 | blk_cb, struct raid5_plug_cb, cb); | |
5408 | struct stripe_head *sh; | |
5409 | struct mddev *mddev = cb->cb.data; | |
5410 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5411 | int cnt = 0; |
566c09c5 | 5412 | int hash; |
8811b596 SL |
5413 | |
5414 | if (cb->list.next && !list_empty(&cb->list)) { | |
5415 | spin_lock_irq(&conf->device_lock); | |
5416 | while (!list_empty(&cb->list)) { | |
5417 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5418 | list_del_init(&sh->lru); | |
5419 | /* | |
5420 | * avoid race release_stripe_plug() sees | |
5421 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5422 | * is still in our list | |
5423 | */ | |
4e857c58 | 5424 | smp_mb__before_atomic(); |
8811b596 | 5425 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5426 | /* |
5427 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5428 | * case, the count is always > 1 here | |
5429 | */ | |
566c09c5 SL |
5430 | hash = sh->hash_lock_index; |
5431 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5432 | cnt++; |
8811b596 SL |
5433 | } |
5434 | spin_unlock_irq(&conf->device_lock); | |
5435 | } | |
566c09c5 SL |
5436 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5437 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5438 | if (mddev->queue) |
5439 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5440 | kfree(cb); |
5441 | } | |
5442 | ||
5443 | static void release_stripe_plug(struct mddev *mddev, | |
5444 | struct stripe_head *sh) | |
5445 | { | |
5446 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5447 | raid5_unplug, mddev, | |
5448 | sizeof(struct raid5_plug_cb)); | |
5449 | struct raid5_plug_cb *cb; | |
5450 | ||
5451 | if (!blk_cb) { | |
6d036f7d | 5452 | raid5_release_stripe(sh); |
8811b596 SL |
5453 | return; |
5454 | } | |
5455 | ||
5456 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5457 | ||
566c09c5 SL |
5458 | if (cb->list.next == NULL) { |
5459 | int i; | |
8811b596 | 5460 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5461 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5462 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5463 | } | |
8811b596 SL |
5464 | |
5465 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5466 | list_add_tail(&sh->lru, &cb->list); | |
5467 | else | |
6d036f7d | 5468 | raid5_release_stripe(sh); |
8811b596 SL |
5469 | } |
5470 | ||
620125f2 SL |
5471 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5472 | { | |
5473 | struct r5conf *conf = mddev->private; | |
5474 | sector_t logical_sector, last_sector; | |
5475 | struct stripe_head *sh; | |
5476 | int remaining; | |
5477 | int stripe_sectors; | |
5478 | ||
5479 | if (mddev->reshape_position != MaxSector) | |
5480 | /* Skip discard while reshape is happening */ | |
5481 | return; | |
5482 | ||
4f024f37 KO |
5483 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5484 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5485 | |
5486 | bi->bi_next = NULL; | |
5487 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
49728050 | 5488 | md_write_start(mddev, bi); |
620125f2 SL |
5489 | |
5490 | stripe_sectors = conf->chunk_sectors * | |
5491 | (conf->raid_disks - conf->max_degraded); | |
5492 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5493 | stripe_sectors); | |
5494 | sector_div(last_sector, stripe_sectors); | |
5495 | ||
5496 | logical_sector *= conf->chunk_sectors; | |
5497 | last_sector *= conf->chunk_sectors; | |
5498 | ||
5499 | for (; logical_sector < last_sector; | |
5500 | logical_sector += STRIPE_SECTORS) { | |
5501 | DEFINE_WAIT(w); | |
5502 | int d; | |
5503 | again: | |
6d036f7d | 5504 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5505 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5506 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5507 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5508 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5509 | raid5_release_stripe(sh); |
f8dfcffd N |
5510 | schedule(); |
5511 | goto again; | |
5512 | } | |
5513 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5514 | spin_lock_irq(&sh->stripe_lock); |
5515 | for (d = 0; d < conf->raid_disks; d++) { | |
5516 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5517 | continue; | |
5518 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5519 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5520 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5521 | raid5_release_stripe(sh); |
620125f2 SL |
5522 | schedule(); |
5523 | goto again; | |
5524 | } | |
5525 | } | |
f8dfcffd | 5526 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5527 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5528 | sh->overwrite_disks = 0; |
620125f2 SL |
5529 | for (d = 0; d < conf->raid_disks; d++) { |
5530 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5531 | continue; | |
5532 | sh->dev[d].towrite = bi; | |
5533 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5534 | raid5_inc_bi_active_stripes(bi); | |
49728050 | 5535 | md_write_inc(mddev, bi); |
7a87f434 | 5536 | sh->overwrite_disks++; |
620125f2 SL |
5537 | } |
5538 | spin_unlock_irq(&sh->stripe_lock); | |
5539 | if (conf->mddev->bitmap) { | |
5540 | for (d = 0; | |
5541 | d < conf->raid_disks - conf->max_degraded; | |
5542 | d++) | |
5543 | bitmap_startwrite(mddev->bitmap, | |
5544 | sh->sector, | |
5545 | STRIPE_SECTORS, | |
5546 | 0); | |
5547 | sh->bm_seq = conf->seq_flush + 1; | |
5548 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5549 | } | |
5550 | ||
5551 | set_bit(STRIPE_HANDLE, &sh->state); | |
5552 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5553 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5554 | atomic_inc(&conf->preread_active_stripes); | |
5555 | release_stripe_plug(mddev, sh); | |
5556 | } | |
5557 | ||
49728050 | 5558 | md_write_end(mddev); |
620125f2 SL |
5559 | remaining = raid5_dec_bi_active_stripes(bi); |
5560 | if (remaining == 0) { | |
4246a0b6 | 5561 | bio_endio(bi); |
620125f2 SL |
5562 | } |
5563 | } | |
5564 | ||
849674e4 | 5565 | static void raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5566 | { |
d1688a6d | 5567 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5568 | int dd_idx; |
1da177e4 LT |
5569 | sector_t new_sector; |
5570 | sector_t logical_sector, last_sector; | |
5571 | struct stripe_head *sh; | |
a362357b | 5572 | const int rw = bio_data_dir(bi); |
49077326 | 5573 | int remaining; |
27c0f68f SL |
5574 | DEFINE_WAIT(w); |
5575 | bool do_prepare; | |
3bddb7f8 | 5576 | bool do_flush = false; |
1da177e4 | 5577 | |
1eff9d32 | 5578 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
828cbe98 SL |
5579 | int ret = r5l_handle_flush_request(conf->log, bi); |
5580 | ||
5581 | if (ret == 0) | |
5582 | return; | |
5583 | if (ret == -ENODEV) { | |
5584 | md_flush_request(mddev, bi); | |
5585 | return; | |
5586 | } | |
5587 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5588 | /* |
5589 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5590 | * we need to flush journal device | |
5591 | */ | |
5592 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5593 | } |
5594 | ||
9ffc8f7c EM |
5595 | /* |
5596 | * If array is degraded, better not do chunk aligned read because | |
5597 | * later we might have to read it again in order to reconstruct | |
5598 | * data on failed drives. | |
5599 | */ | |
5600 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5601 | mddev->reshape_position == MaxSector) { |
5602 | bi = chunk_aligned_read(mddev, bi); | |
5603 | if (!bi) | |
5604 | return; | |
5605 | } | |
52488615 | 5606 | |
796a5cf0 | 5607 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 SL |
5608 | make_discard_request(mddev, bi); |
5609 | return; | |
5610 | } | |
5611 | ||
4f024f37 | 5612 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5613 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5614 | bi->bi_next = NULL; |
5615 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
49728050 | 5616 | md_write_start(mddev, bi); |
06d91a5f | 5617 | |
27c0f68f | 5618 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5619 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5620 | int previous; |
c46501b2 | 5621 | int seq; |
b578d55f | 5622 | |
27c0f68f | 5623 | do_prepare = false; |
7ecaa1e6 | 5624 | retry: |
c46501b2 | 5625 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5626 | previous = 0; |
27c0f68f SL |
5627 | if (do_prepare) |
5628 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5629 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5630 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5631 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5632 | * 64bit on a 32bit platform, and so it might be |
5633 | * possible to see a half-updated value | |
aeb878b0 | 5634 | * Of course reshape_progress could change after |
df8e7f76 N |
5635 | * the lock is dropped, so once we get a reference |
5636 | * to the stripe that we think it is, we will have | |
5637 | * to check again. | |
5638 | */ | |
7ecaa1e6 | 5639 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5640 | if (mddev->reshape_backwards |
fef9c61f N |
5641 | ? logical_sector < conf->reshape_progress |
5642 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5643 | previous = 1; |
5644 | } else { | |
2c810cdd | 5645 | if (mddev->reshape_backwards |
fef9c61f N |
5646 | ? logical_sector < conf->reshape_safe |
5647 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5648 | spin_unlock_irq(&conf->device_lock); |
5649 | schedule(); | |
27c0f68f | 5650 | do_prepare = true; |
b578d55f N |
5651 | goto retry; |
5652 | } | |
5653 | } | |
7ecaa1e6 N |
5654 | spin_unlock_irq(&conf->device_lock); |
5655 | } | |
16a53ecc | 5656 | |
112bf897 N |
5657 | new_sector = raid5_compute_sector(conf, logical_sector, |
5658 | previous, | |
911d4ee8 | 5659 | &dd_idx, NULL); |
849674e4 | 5660 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5661 | (unsigned long long)new_sector, |
1da177e4 LT |
5662 | (unsigned long long)logical_sector); |
5663 | ||
6d036f7d | 5664 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5665 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5666 | if (sh) { |
b0f9ec04 | 5667 | if (unlikely(previous)) { |
7ecaa1e6 | 5668 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5669 | * stripe, so we must do the range check again. |
5670 | * Expansion could still move past after this | |
5671 | * test, but as we are holding a reference to | |
5672 | * 'sh', we know that if that happens, | |
5673 | * STRIPE_EXPANDING will get set and the expansion | |
5674 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5675 | */ |
5676 | int must_retry = 0; | |
5677 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5678 | if (mddev->reshape_backwards |
b0f9ec04 N |
5679 | ? logical_sector >= conf->reshape_progress |
5680 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5681 | /* mismatch, need to try again */ |
5682 | must_retry = 1; | |
5683 | spin_unlock_irq(&conf->device_lock); | |
5684 | if (must_retry) { | |
6d036f7d | 5685 | raid5_release_stripe(sh); |
7a3ab908 | 5686 | schedule(); |
27c0f68f | 5687 | do_prepare = true; |
7ecaa1e6 N |
5688 | goto retry; |
5689 | } | |
5690 | } | |
c46501b2 N |
5691 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5692 | /* Might have got the wrong stripe_head | |
5693 | * by accident | |
5694 | */ | |
6d036f7d | 5695 | raid5_release_stripe(sh); |
c46501b2 N |
5696 | goto retry; |
5697 | } | |
e62e58a5 | 5698 | |
ffd96e35 | 5699 | if (rw == WRITE && |
a5c308d4 | 5700 | logical_sector >= mddev->suspend_lo && |
e464eafd | 5701 | logical_sector < mddev->suspend_hi) { |
6d036f7d | 5702 | raid5_release_stripe(sh); |
e62e58a5 N |
5703 | /* As the suspend_* range is controlled by |
5704 | * userspace, we want an interruptible | |
5705 | * wait. | |
5706 | */ | |
5707 | flush_signals(current); | |
5708 | prepare_to_wait(&conf->wait_for_overlap, | |
5709 | &w, TASK_INTERRUPTIBLE); | |
5710 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5711 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5712 | schedule(); |
27c0f68f SL |
5713 | do_prepare = true; |
5714 | } | |
e464eafd N |
5715 | goto retry; |
5716 | } | |
7ecaa1e6 N |
5717 | |
5718 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5719 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5720 | /* Stripe is busy expanding or |
5721 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5722 | * and wait a while |
5723 | */ | |
482c0834 | 5724 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5725 | raid5_release_stripe(sh); |
1da177e4 | 5726 | schedule(); |
27c0f68f | 5727 | do_prepare = true; |
1da177e4 LT |
5728 | goto retry; |
5729 | } | |
3bddb7f8 SL |
5730 | if (do_flush) { |
5731 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5732 | /* we only need flush for one stripe */ | |
5733 | do_flush = false; | |
5734 | } | |
5735 | ||
6ed3003c N |
5736 | set_bit(STRIPE_HANDLE, &sh->state); |
5737 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5738 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5739 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5740 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5741 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5742 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5743 | } else { |
5744 | /* cannot get stripe for read-ahead, just give-up */ | |
4246a0b6 | 5745 | bi->bi_error = -EIO; |
1da177e4 LT |
5746 | break; |
5747 | } | |
1da177e4 | 5748 | } |
27c0f68f | 5749 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5750 | |
49728050 N |
5751 | if (rw == WRITE) |
5752 | md_write_end(mddev); | |
e7836bd6 | 5753 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5754 | if (remaining == 0) { |
1da177e4 | 5755 | |
6712ecf8 | 5756 | |
0a82a8d1 LT |
5757 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5758 | bi, 0); | |
4246a0b6 | 5759 | bio_endio(bi); |
1da177e4 | 5760 | } |
1da177e4 LT |
5761 | } |
5762 | ||
fd01b88c | 5763 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5764 | |
fd01b88c | 5765 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5766 | { |
52c03291 N |
5767 | /* reshaping is quite different to recovery/resync so it is |
5768 | * handled quite separately ... here. | |
5769 | * | |
5770 | * On each call to sync_request, we gather one chunk worth of | |
5771 | * destination stripes and flag them as expanding. | |
5772 | * Then we find all the source stripes and request reads. | |
5773 | * As the reads complete, handle_stripe will copy the data | |
5774 | * into the destination stripe and release that stripe. | |
5775 | */ | |
d1688a6d | 5776 | struct r5conf *conf = mddev->private; |
1da177e4 | 5777 | struct stripe_head *sh; |
ccfcc3c1 | 5778 | sector_t first_sector, last_sector; |
f416885e N |
5779 | int raid_disks = conf->previous_raid_disks; |
5780 | int data_disks = raid_disks - conf->max_degraded; | |
5781 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5782 | int i; |
5783 | int dd_idx; | |
c8f517c4 | 5784 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5785 | sector_t stripe_addr; |
7a661381 | 5786 | int reshape_sectors; |
ab69ae12 | 5787 | struct list_head stripes; |
92140480 | 5788 | sector_t retn; |
52c03291 | 5789 | |
fef9c61f N |
5790 | if (sector_nr == 0) { |
5791 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5792 | if (mddev->reshape_backwards && |
fef9c61f N |
5793 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5794 | sector_nr = raid5_size(mddev, 0, 0) | |
5795 | - conf->reshape_progress; | |
6cbd8148 N |
5796 | } else if (mddev->reshape_backwards && |
5797 | conf->reshape_progress == MaxSector) { | |
5798 | /* shouldn't happen, but just in case, finish up.*/ | |
5799 | sector_nr = MaxSector; | |
2c810cdd | 5800 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5801 | conf->reshape_progress > 0) |
5802 | sector_nr = conf->reshape_progress; | |
f416885e | 5803 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5804 | if (sector_nr) { |
8dee7211 N |
5805 | mddev->curr_resync_completed = sector_nr; |
5806 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5807 | *skipped = 1; |
92140480 N |
5808 | retn = sector_nr; |
5809 | goto finish; | |
fef9c61f | 5810 | } |
52c03291 N |
5811 | } |
5812 | ||
7a661381 N |
5813 | /* We need to process a full chunk at a time. |
5814 | * If old and new chunk sizes differ, we need to process the | |
5815 | * largest of these | |
5816 | */ | |
3cb5edf4 N |
5817 | |
5818 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5819 | |
b5254dd5 N |
5820 | /* We update the metadata at least every 10 seconds, or when |
5821 | * the data about to be copied would over-write the source of | |
5822 | * the data at the front of the range. i.e. one new_stripe | |
5823 | * along from reshape_progress new_maps to after where | |
5824 | * reshape_safe old_maps to | |
52c03291 | 5825 | */ |
fef9c61f | 5826 | writepos = conf->reshape_progress; |
f416885e | 5827 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5828 | readpos = conf->reshape_progress; |
5829 | sector_div(readpos, data_disks); | |
fef9c61f | 5830 | safepos = conf->reshape_safe; |
f416885e | 5831 | sector_div(safepos, data_disks); |
2c810cdd | 5832 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5833 | BUG_ON(writepos < reshape_sectors); |
5834 | writepos -= reshape_sectors; | |
c8f517c4 | 5835 | readpos += reshape_sectors; |
7a661381 | 5836 | safepos += reshape_sectors; |
fef9c61f | 5837 | } else { |
7a661381 | 5838 | writepos += reshape_sectors; |
c74c0d76 N |
5839 | /* readpos and safepos are worst-case calculations. |
5840 | * A negative number is overly pessimistic, and causes | |
5841 | * obvious problems for unsigned storage. So clip to 0. | |
5842 | */ | |
ed37d83e N |
5843 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5844 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5845 | } |
52c03291 | 5846 | |
b5254dd5 N |
5847 | /* Having calculated the 'writepos' possibly use it |
5848 | * to set 'stripe_addr' which is where we will write to. | |
5849 | */ | |
5850 | if (mddev->reshape_backwards) { | |
5851 | BUG_ON(conf->reshape_progress == 0); | |
5852 | stripe_addr = writepos; | |
5853 | BUG_ON((mddev->dev_sectors & | |
5854 | ~((sector_t)reshape_sectors - 1)) | |
5855 | - reshape_sectors - stripe_addr | |
5856 | != sector_nr); | |
5857 | } else { | |
5858 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5859 | stripe_addr = sector_nr; | |
5860 | } | |
5861 | ||
c8f517c4 N |
5862 | /* 'writepos' is the most advanced device address we might write. |
5863 | * 'readpos' is the least advanced device address we might read. | |
5864 | * 'safepos' is the least address recorded in the metadata as having | |
5865 | * been reshaped. | |
b5254dd5 N |
5866 | * If there is a min_offset_diff, these are adjusted either by |
5867 | * increasing the safepos/readpos if diff is negative, or | |
5868 | * increasing writepos if diff is positive. | |
5869 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5870 | * ensure safety in the face of a crash - that must be done by userspace |
5871 | * making a backup of the data. So in that case there is no particular | |
5872 | * rush to update metadata. | |
5873 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5874 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5875 | * we can be safe in the event of a crash. | |
5876 | * So we insist on updating metadata if safepos is behind writepos and | |
5877 | * readpos is beyond writepos. | |
5878 | * In any case, update the metadata every 10 seconds. | |
5879 | * Maybe that number should be configurable, but I'm not sure it is | |
5880 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5881 | */ | |
b5254dd5 N |
5882 | if (conf->min_offset_diff < 0) { |
5883 | safepos += -conf->min_offset_diff; | |
5884 | readpos += -conf->min_offset_diff; | |
5885 | } else | |
5886 | writepos += conf->min_offset_diff; | |
5887 | ||
2c810cdd | 5888 | if ((mddev->reshape_backwards |
c8f517c4 N |
5889 | ? (safepos > writepos && readpos < writepos) |
5890 | : (safepos < writepos && readpos > writepos)) || | |
5891 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5892 | /* Cannot proceed until we've updated the superblock... */ |
5893 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5894 | atomic_read(&conf->reshape_stripes)==0 |
5895 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5896 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5897 | return 0; | |
fef9c61f | 5898 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5899 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5900 | conf->reshape_checkpoint = jiffies; |
2953079c | 5901 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5902 | md_wakeup_thread(mddev->thread); |
2953079c | 5903 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5904 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5905 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5906 | return 0; | |
52c03291 | 5907 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5908 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5909 | spin_unlock_irq(&conf->device_lock); |
5910 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5911 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5912 | } |
5913 | ||
ab69ae12 | 5914 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5915 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5916 | int j; |
a9f326eb | 5917 | int skipped_disk = 0; |
6d036f7d | 5918 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5919 | set_bit(STRIPE_EXPANDING, &sh->state); |
5920 | atomic_inc(&conf->reshape_stripes); | |
5921 | /* If any of this stripe is beyond the end of the old | |
5922 | * array, then we need to zero those blocks | |
5923 | */ | |
5924 | for (j=sh->disks; j--;) { | |
5925 | sector_t s; | |
5926 | if (j == sh->pd_idx) | |
5927 | continue; | |
f416885e | 5928 | if (conf->level == 6 && |
d0dabf7e | 5929 | j == sh->qd_idx) |
f416885e | 5930 | continue; |
6d036f7d | 5931 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5932 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5933 | skipped_disk = 1; |
52c03291 N |
5934 | continue; |
5935 | } | |
5936 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5937 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5938 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5939 | } | |
a9f326eb | 5940 | if (!skipped_disk) { |
52c03291 N |
5941 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5942 | set_bit(STRIPE_HANDLE, &sh->state); | |
5943 | } | |
ab69ae12 | 5944 | list_add(&sh->lru, &stripes); |
52c03291 N |
5945 | } |
5946 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5947 | if (mddev->reshape_backwards) |
7a661381 | 5948 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5949 | else |
7a661381 | 5950 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5951 | spin_unlock_irq(&conf->device_lock); |
5952 | /* Ok, those stripe are ready. We can start scheduling | |
5953 | * reads on the source stripes. | |
5954 | * The source stripes are determined by mapping the first and last | |
5955 | * block on the destination stripes. | |
5956 | */ | |
52c03291 | 5957 | first_sector = |
ec32a2bd | 5958 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5959 | 1, &dd_idx, NULL); |
52c03291 | 5960 | last_sector = |
0e6e0271 | 5961 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5962 | * new_data_disks - 1), |
911d4ee8 | 5963 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5964 | if (last_sector >= mddev->dev_sectors) |
5965 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5966 | while (first_sector <= last_sector) { |
6d036f7d | 5967 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5968 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5969 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5970 | raid5_release_stripe(sh); |
52c03291 N |
5971 | first_sector += STRIPE_SECTORS; |
5972 | } | |
ab69ae12 N |
5973 | /* Now that the sources are clearly marked, we can release |
5974 | * the destination stripes | |
5975 | */ | |
5976 | while (!list_empty(&stripes)) { | |
5977 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5978 | list_del_init(&sh->lru); | |
6d036f7d | 5979 | raid5_release_stripe(sh); |
ab69ae12 | 5980 | } |
c6207277 N |
5981 | /* If this takes us to the resync_max point where we have to pause, |
5982 | * then we need to write out the superblock. | |
5983 | */ | |
7a661381 | 5984 | sector_nr += reshape_sectors; |
92140480 N |
5985 | retn = reshape_sectors; |
5986 | finish: | |
c5e19d90 N |
5987 | if (mddev->curr_resync_completed > mddev->resync_max || |
5988 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5989 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5990 | /* Cannot proceed until we've updated the superblock... */ |
5991 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5992 | atomic_read(&conf->reshape_stripes) == 0 |
5993 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5994 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5995 | goto ret; | |
fef9c61f | 5996 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5997 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5998 | conf->reshape_checkpoint = jiffies; |
2953079c | 5999 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
6000 | md_wakeup_thread(mddev->thread); |
6001 | wait_event(mddev->sb_wait, | |
2953079c | 6002 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
6003 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
6004 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6005 | goto ret; | |
c6207277 | 6006 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 6007 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
6008 | spin_unlock_irq(&conf->device_lock); |
6009 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 6010 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 6011 | } |
c91abf5a | 6012 | ret: |
92140480 | 6013 | return retn; |
52c03291 N |
6014 | } |
6015 | ||
849674e4 SL |
6016 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
6017 | int *skipped) | |
52c03291 | 6018 | { |
d1688a6d | 6019 | struct r5conf *conf = mddev->private; |
52c03291 | 6020 | struct stripe_head *sh; |
58c0fed4 | 6021 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 6022 | sector_t sync_blocks; |
16a53ecc N |
6023 | int still_degraded = 0; |
6024 | int i; | |
1da177e4 | 6025 | |
72626685 | 6026 | if (sector_nr >= max_sector) { |
1da177e4 | 6027 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 6028 | |
29269553 N |
6029 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
6030 | end_reshape(conf); | |
6031 | return 0; | |
6032 | } | |
72626685 N |
6033 | |
6034 | if (mddev->curr_resync < max_sector) /* aborted */ | |
6035 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
6036 | &sync_blocks, 1); | |
16a53ecc | 6037 | else /* completed sync */ |
72626685 N |
6038 | conf->fullsync = 0; |
6039 | bitmap_close_sync(mddev->bitmap); | |
6040 | ||
1da177e4 LT |
6041 | return 0; |
6042 | } | |
ccfcc3c1 | 6043 | |
64bd660b N |
6044 | /* Allow raid5_quiesce to complete */ |
6045 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
6046 | ||
52c03291 N |
6047 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
6048 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 6049 | |
c6207277 N |
6050 | /* No need to check resync_max as we never do more than one |
6051 | * stripe, and as resync_max will always be on a chunk boundary, | |
6052 | * if the check in md_do_sync didn't fire, there is no chance | |
6053 | * of overstepping resync_max here | |
6054 | */ | |
6055 | ||
16a53ecc | 6056 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
6057 | * to resync, then assert that we are finished, because there is |
6058 | * nothing we can do. | |
6059 | */ | |
3285edf1 | 6060 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 6061 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 6062 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 6063 | *skipped = 1; |
1da177e4 LT |
6064 | return rv; |
6065 | } | |
6f608040 | 6066 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
6067 | !conf->fullsync && | |
6068 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
6069 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
6070 | /* we can skip this block, and probably more */ |
6071 | sync_blocks /= STRIPE_SECTORS; | |
6072 | *skipped = 1; | |
6073 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
6074 | } | |
1da177e4 | 6075 | |
c40f341f | 6076 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 6077 | |
6d036f7d | 6078 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 6079 | if (sh == NULL) { |
6d036f7d | 6080 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 6081 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 6082 | * is trying to get access |
1da177e4 | 6083 | */ |
66c006a5 | 6084 | schedule_timeout_uninterruptible(1); |
1da177e4 | 6085 | } |
16a53ecc | 6086 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
6087 | * Note in case of > 1 drive failures it's possible we're rebuilding |
6088 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 6089 | */ |
16d9cfab EM |
6090 | rcu_read_lock(); |
6091 | for (i = 0; i < conf->raid_disks; i++) { | |
6092 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
6093 | ||
6094 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 6095 | still_degraded = 1; |
16d9cfab EM |
6096 | } |
6097 | rcu_read_unlock(); | |
16a53ecc N |
6098 | |
6099 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
6100 | ||
83206d66 | 6101 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 6102 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 6103 | |
6d036f7d | 6104 | raid5_release_stripe(sh); |
1da177e4 LT |
6105 | |
6106 | return STRIPE_SECTORS; | |
6107 | } | |
6108 | ||
d1688a6d | 6109 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
6110 | { |
6111 | /* We may not be able to submit a whole bio at once as there | |
6112 | * may not be enough stripe_heads available. | |
6113 | * We cannot pre-allocate enough stripe_heads as we may need | |
6114 | * more than exist in the cache (if we allow ever large chunks). | |
6115 | * So we do one stripe head at a time and record in | |
6116 | * ->bi_hw_segments how many have been done. | |
6117 | * | |
6118 | * We *know* that this entire raid_bio is in one chunk, so | |
6119 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
6120 | */ | |
6121 | struct stripe_head *sh; | |
911d4ee8 | 6122 | int dd_idx; |
46031f9a RBJ |
6123 | sector_t sector, logical_sector, last_sector; |
6124 | int scnt = 0; | |
6125 | int remaining; | |
6126 | int handled = 0; | |
6127 | ||
4f024f37 KO |
6128 | logical_sector = raid_bio->bi_iter.bi_sector & |
6129 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 6130 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6131 | 0, &dd_idx, NULL); |
f73a1c7d | 6132 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6133 | |
6134 | for (; logical_sector < last_sector; | |
387bb173 NB |
6135 | logical_sector += STRIPE_SECTORS, |
6136 | sector += STRIPE_SECTORS, | |
6137 | scnt++) { | |
46031f9a | 6138 | |
e7836bd6 | 6139 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
6140 | /* already done this stripe */ |
6141 | continue; | |
6142 | ||
6d036f7d | 6143 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6144 | |
6145 | if (!sh) { | |
6146 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 6147 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
6148 | conf->retry_read_aligned = raid_bio; |
6149 | return handled; | |
6150 | } | |
6151 | ||
da41ba65 | 6152 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6153 | raid5_release_stripe(sh); |
e7836bd6 | 6154 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
6155 | conf->retry_read_aligned = raid_bio; |
6156 | return handled; | |
6157 | } | |
6158 | ||
3f9e7c14 | 6159 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6160 | handle_stripe(sh); |
6d036f7d | 6161 | raid5_release_stripe(sh); |
46031f9a RBJ |
6162 | handled++; |
6163 | } | |
e7836bd6 | 6164 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
6165 | if (remaining == 0) { |
6166 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
6167 | raid_bio, 0); | |
4246a0b6 | 6168 | bio_endio(raid_bio); |
0a82a8d1 | 6169 | } |
46031f9a | 6170 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6171 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6172 | return handled; |
6173 | } | |
6174 | ||
bfc90cb0 | 6175 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6176 | struct r5worker *worker, |
6177 | struct list_head *temp_inactive_list) | |
46a06401 SL |
6178 | { |
6179 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6180 | int i, batch_size = 0, hash; |
6181 | bool release_inactive = false; | |
46a06401 SL |
6182 | |
6183 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6184 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6185 | batch[batch_size++] = sh; |
6186 | ||
566c09c5 SL |
6187 | if (batch_size == 0) { |
6188 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6189 | if (!list_empty(temp_inactive_list + i)) | |
6190 | break; | |
a8c34f91 SL |
6191 | if (i == NR_STRIPE_HASH_LOCKS) { |
6192 | spin_unlock_irq(&conf->device_lock); | |
6193 | r5l_flush_stripe_to_raid(conf->log); | |
6194 | spin_lock_irq(&conf->device_lock); | |
566c09c5 | 6195 | return batch_size; |
a8c34f91 | 6196 | } |
566c09c5 SL |
6197 | release_inactive = true; |
6198 | } | |
46a06401 SL |
6199 | spin_unlock_irq(&conf->device_lock); |
6200 | ||
566c09c5 SL |
6201 | release_inactive_stripe_list(conf, temp_inactive_list, |
6202 | NR_STRIPE_HASH_LOCKS); | |
6203 | ||
a8c34f91 | 6204 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6205 | if (release_inactive) { |
6206 | spin_lock_irq(&conf->device_lock); | |
6207 | return 0; | |
6208 | } | |
6209 | ||
46a06401 SL |
6210 | for (i = 0; i < batch_size; i++) |
6211 | handle_stripe(batch[i]); | |
ff875738 | 6212 | log_write_stripe_run(conf); |
46a06401 SL |
6213 | |
6214 | cond_resched(); | |
6215 | ||
6216 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6217 | for (i = 0; i < batch_size; i++) { |
6218 | hash = batch[i]->hash_lock_index; | |
6219 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6220 | } | |
46a06401 SL |
6221 | return batch_size; |
6222 | } | |
46031f9a | 6223 | |
851c30c9 SL |
6224 | static void raid5_do_work(struct work_struct *work) |
6225 | { | |
6226 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6227 | struct r5worker_group *group = worker->group; | |
6228 | struct r5conf *conf = group->conf; | |
6229 | int group_id = group - conf->worker_groups; | |
6230 | int handled; | |
6231 | struct blk_plug plug; | |
6232 | ||
6233 | pr_debug("+++ raid5worker active\n"); | |
6234 | ||
6235 | blk_start_plug(&plug); | |
6236 | handled = 0; | |
6237 | spin_lock_irq(&conf->device_lock); | |
6238 | while (1) { | |
6239 | int batch_size, released; | |
6240 | ||
566c09c5 | 6241 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6242 | |
566c09c5 SL |
6243 | batch_size = handle_active_stripes(conf, group_id, worker, |
6244 | worker->temp_inactive_list); | |
bfc90cb0 | 6245 | worker->working = false; |
851c30c9 SL |
6246 | if (!batch_size && !released) |
6247 | break; | |
6248 | handled += batch_size; | |
6249 | } | |
6250 | pr_debug("%d stripes handled\n", handled); | |
6251 | ||
6252 | spin_unlock_irq(&conf->device_lock); | |
6253 | blk_finish_plug(&plug); | |
6254 | ||
6255 | pr_debug("--- raid5worker inactive\n"); | |
6256 | } | |
6257 | ||
1da177e4 LT |
6258 | /* |
6259 | * This is our raid5 kernel thread. | |
6260 | * | |
6261 | * We scan the hash table for stripes which can be handled now. | |
6262 | * During the scan, completed stripes are saved for us by the interrupt | |
6263 | * handler, so that they will not have to wait for our next wakeup. | |
6264 | */ | |
4ed8731d | 6265 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6266 | { |
4ed8731d | 6267 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6268 | struct r5conf *conf = mddev->private; |
1da177e4 | 6269 | int handled; |
e1dfa0a2 | 6270 | struct blk_plug plug; |
1da177e4 | 6271 | |
45b4233c | 6272 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6273 | |
6274 | md_check_recovery(mddev); | |
1da177e4 | 6275 | |
c3cce6cd | 6276 | if (!bio_list_empty(&conf->return_bi) && |
2953079c | 6277 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6278 | struct bio_list tmp = BIO_EMPTY_LIST; |
6279 | spin_lock_irq(&conf->device_lock); | |
2953079c | 6280 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6281 | bio_list_merge(&tmp, &conf->return_bi); |
6282 | bio_list_init(&conf->return_bi); | |
6283 | } | |
6284 | spin_unlock_irq(&conf->device_lock); | |
6285 | return_io(&tmp); | |
6286 | } | |
6287 | ||
e1dfa0a2 | 6288 | blk_start_plug(&plug); |
1da177e4 LT |
6289 | handled = 0; |
6290 | spin_lock_irq(&conf->device_lock); | |
6291 | while (1) { | |
46031f9a | 6292 | struct bio *bio; |
773ca82f SL |
6293 | int batch_size, released; |
6294 | ||
566c09c5 | 6295 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6296 | if (released) |
6297 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6298 | |
0021b7bc | 6299 | if ( |
7c13edc8 N |
6300 | !list_empty(&conf->bitmap_list)) { |
6301 | /* Now is a good time to flush some bitmap updates */ | |
6302 | conf->seq_flush++; | |
700e432d | 6303 | spin_unlock_irq(&conf->device_lock); |
72626685 | 6304 | bitmap_unplug(mddev->bitmap); |
700e432d | 6305 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6306 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6307 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6308 | } |
0021b7bc | 6309 | raid5_activate_delayed(conf); |
72626685 | 6310 | |
46031f9a RBJ |
6311 | while ((bio = remove_bio_from_retry(conf))) { |
6312 | int ok; | |
6313 | spin_unlock_irq(&conf->device_lock); | |
6314 | ok = retry_aligned_read(conf, bio); | |
6315 | spin_lock_irq(&conf->device_lock); | |
6316 | if (!ok) | |
6317 | break; | |
6318 | handled++; | |
6319 | } | |
6320 | ||
566c09c5 SL |
6321 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6322 | conf->temp_inactive_list); | |
773ca82f | 6323 | if (!batch_size && !released) |
1da177e4 | 6324 | break; |
46a06401 | 6325 | handled += batch_size; |
1da177e4 | 6326 | |
2953079c | 6327 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6328 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6329 | md_check_recovery(mddev); |
46a06401 SL |
6330 | spin_lock_irq(&conf->device_lock); |
6331 | } | |
1da177e4 | 6332 | } |
45b4233c | 6333 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6334 | |
6335 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6336 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6337 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6338 | grow_one_stripe(conf, __GFP_NOWARN); |
6339 | /* Set flag even if allocation failed. This helps | |
6340 | * slow down allocation requests when mem is short | |
6341 | */ | |
6342 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6343 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6344 | } |
1da177e4 | 6345 | |
765d704d SL |
6346 | flush_deferred_bios(conf); |
6347 | ||
0576b1c6 SL |
6348 | r5l_flush_stripe_to_raid(conf->log); |
6349 | ||
c9f21aaf | 6350 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6351 | blk_finish_plug(&plug); |
1da177e4 | 6352 | |
45b4233c | 6353 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6354 | } |
6355 | ||
3f294f4f | 6356 | static ssize_t |
fd01b88c | 6357 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6358 | { |
7b1485ba N |
6359 | struct r5conf *conf; |
6360 | int ret = 0; | |
6361 | spin_lock(&mddev->lock); | |
6362 | conf = mddev->private; | |
96de1e66 | 6363 | if (conf) |
edbe83ab | 6364 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6365 | spin_unlock(&mddev->lock); |
6366 | return ret; | |
3f294f4f N |
6367 | } |
6368 | ||
c41d4ac4 | 6369 | int |
fd01b88c | 6370 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6371 | { |
d1688a6d | 6372 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
6373 | int err; |
6374 | ||
c41d4ac4 | 6375 | if (size <= 16 || size > 32768) |
3f294f4f | 6376 | return -EINVAL; |
486f0644 | 6377 | |
edbe83ab | 6378 | conf->min_nr_stripes = size; |
2d5b569b | 6379 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6380 | while (size < conf->max_nr_stripes && |
6381 | drop_one_stripe(conf)) | |
6382 | ; | |
2d5b569b | 6383 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6384 | |
edbe83ab | 6385 | |
b5470dc5 DW |
6386 | err = md_allow_write(mddev); |
6387 | if (err) | |
6388 | return err; | |
486f0644 | 6389 | |
2d5b569b | 6390 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6391 | while (size > conf->max_nr_stripes) |
6392 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
6393 | break; | |
2d5b569b | 6394 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6395 | |
c41d4ac4 N |
6396 | return 0; |
6397 | } | |
6398 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6399 | ||
6400 | static ssize_t | |
fd01b88c | 6401 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6402 | { |
6791875e | 6403 | struct r5conf *conf; |
c41d4ac4 N |
6404 | unsigned long new; |
6405 | int err; | |
6406 | ||
6407 | if (len >= PAGE_SIZE) | |
6408 | return -EINVAL; | |
b29bebd6 | 6409 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6410 | return -EINVAL; |
6791875e | 6411 | err = mddev_lock(mddev); |
c41d4ac4 N |
6412 | if (err) |
6413 | return err; | |
6791875e N |
6414 | conf = mddev->private; |
6415 | if (!conf) | |
6416 | err = -ENODEV; | |
6417 | else | |
6418 | err = raid5_set_cache_size(mddev, new); | |
6419 | mddev_unlock(mddev); | |
6420 | ||
6421 | return err ?: len; | |
3f294f4f | 6422 | } |
007583c9 | 6423 | |
96de1e66 N |
6424 | static struct md_sysfs_entry |
6425 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6426 | raid5_show_stripe_cache_size, | |
6427 | raid5_store_stripe_cache_size); | |
3f294f4f | 6428 | |
d06f191f MS |
6429 | static ssize_t |
6430 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6431 | { | |
6432 | struct r5conf *conf = mddev->private; | |
6433 | if (conf) | |
6434 | return sprintf(page, "%d\n", conf->rmw_level); | |
6435 | else | |
6436 | return 0; | |
6437 | } | |
6438 | ||
6439 | static ssize_t | |
6440 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6441 | { | |
6442 | struct r5conf *conf = mddev->private; | |
6443 | unsigned long new; | |
6444 | ||
6445 | if (!conf) | |
6446 | return -ENODEV; | |
6447 | ||
6448 | if (len >= PAGE_SIZE) | |
6449 | return -EINVAL; | |
6450 | ||
6451 | if (kstrtoul(page, 10, &new)) | |
6452 | return -EINVAL; | |
6453 | ||
6454 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6455 | return -EINVAL; | |
6456 | ||
6457 | if (new != PARITY_DISABLE_RMW && | |
6458 | new != PARITY_ENABLE_RMW && | |
6459 | new != PARITY_PREFER_RMW) | |
6460 | return -EINVAL; | |
6461 | ||
6462 | conf->rmw_level = new; | |
6463 | return len; | |
6464 | } | |
6465 | ||
6466 | static struct md_sysfs_entry | |
6467 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6468 | raid5_show_rmw_level, | |
6469 | raid5_store_rmw_level); | |
6470 | ||
6471 | ||
8b3e6cdc | 6472 | static ssize_t |
fd01b88c | 6473 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6474 | { |
7b1485ba N |
6475 | struct r5conf *conf; |
6476 | int ret = 0; | |
6477 | spin_lock(&mddev->lock); | |
6478 | conf = mddev->private; | |
8b3e6cdc | 6479 | if (conf) |
7b1485ba N |
6480 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6481 | spin_unlock(&mddev->lock); | |
6482 | return ret; | |
8b3e6cdc DW |
6483 | } |
6484 | ||
6485 | static ssize_t | |
fd01b88c | 6486 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6487 | { |
6791875e | 6488 | struct r5conf *conf; |
4ef197d8 | 6489 | unsigned long new; |
6791875e N |
6490 | int err; |
6491 | ||
8b3e6cdc DW |
6492 | if (len >= PAGE_SIZE) |
6493 | return -EINVAL; | |
b29bebd6 | 6494 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6495 | return -EINVAL; |
6791875e N |
6496 | |
6497 | err = mddev_lock(mddev); | |
6498 | if (err) | |
6499 | return err; | |
6500 | conf = mddev->private; | |
6501 | if (!conf) | |
6502 | err = -ENODEV; | |
edbe83ab | 6503 | else if (new > conf->min_nr_stripes) |
6791875e N |
6504 | err = -EINVAL; |
6505 | else | |
6506 | conf->bypass_threshold = new; | |
6507 | mddev_unlock(mddev); | |
6508 | return err ?: len; | |
8b3e6cdc DW |
6509 | } |
6510 | ||
6511 | static struct md_sysfs_entry | |
6512 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6513 | S_IRUGO | S_IWUSR, | |
6514 | raid5_show_preread_threshold, | |
6515 | raid5_store_preread_threshold); | |
6516 | ||
d592a996 SL |
6517 | static ssize_t |
6518 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6519 | { | |
7b1485ba N |
6520 | struct r5conf *conf; |
6521 | int ret = 0; | |
6522 | spin_lock(&mddev->lock); | |
6523 | conf = mddev->private; | |
d592a996 | 6524 | if (conf) |
7b1485ba N |
6525 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6526 | spin_unlock(&mddev->lock); | |
6527 | return ret; | |
d592a996 SL |
6528 | } |
6529 | ||
6530 | static ssize_t | |
6531 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6532 | { | |
6791875e | 6533 | struct r5conf *conf; |
d592a996 | 6534 | unsigned long new; |
6791875e N |
6535 | int err; |
6536 | ||
d592a996 SL |
6537 | if (len >= PAGE_SIZE) |
6538 | return -EINVAL; | |
d592a996 SL |
6539 | if (kstrtoul(page, 10, &new)) |
6540 | return -EINVAL; | |
6541 | new = !!new; | |
6791875e N |
6542 | |
6543 | err = mddev_lock(mddev); | |
6544 | if (err) | |
6545 | return err; | |
6546 | conf = mddev->private; | |
6547 | if (!conf) | |
6548 | err = -ENODEV; | |
6549 | else if (new != conf->skip_copy) { | |
6550 | mddev_suspend(mddev); | |
6551 | conf->skip_copy = new; | |
6552 | if (new) | |
dc3b17cc | 6553 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6554 | BDI_CAP_STABLE_WRITES; |
6555 | else | |
dc3b17cc | 6556 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6557 | ~BDI_CAP_STABLE_WRITES; |
6558 | mddev_resume(mddev); | |
6559 | } | |
6560 | mddev_unlock(mddev); | |
6561 | return err ?: len; | |
d592a996 SL |
6562 | } |
6563 | ||
6564 | static struct md_sysfs_entry | |
6565 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6566 | raid5_show_skip_copy, | |
6567 | raid5_store_skip_copy); | |
6568 | ||
3f294f4f | 6569 | static ssize_t |
fd01b88c | 6570 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6571 | { |
d1688a6d | 6572 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6573 | if (conf) |
6574 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6575 | else | |
6576 | return 0; | |
3f294f4f N |
6577 | } |
6578 | ||
96de1e66 N |
6579 | static struct md_sysfs_entry |
6580 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6581 | |
b721420e SL |
6582 | static ssize_t |
6583 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6584 | { | |
7b1485ba N |
6585 | struct r5conf *conf; |
6586 | int ret = 0; | |
6587 | spin_lock(&mddev->lock); | |
6588 | conf = mddev->private; | |
b721420e | 6589 | if (conf) |
7b1485ba N |
6590 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6591 | spin_unlock(&mddev->lock); | |
6592 | return ret; | |
b721420e SL |
6593 | } |
6594 | ||
60aaf933 | 6595 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6596 | int *group_cnt, | |
6597 | int *worker_cnt_per_group, | |
6598 | struct r5worker_group **worker_groups); | |
b721420e SL |
6599 | static ssize_t |
6600 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6601 | { | |
6791875e | 6602 | struct r5conf *conf; |
b721420e SL |
6603 | unsigned long new; |
6604 | int err; | |
60aaf933 | 6605 | struct r5worker_group *new_groups, *old_groups; |
6606 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6607 | |
6608 | if (len >= PAGE_SIZE) | |
6609 | return -EINVAL; | |
b721420e SL |
6610 | if (kstrtoul(page, 10, &new)) |
6611 | return -EINVAL; | |
6612 | ||
6791875e N |
6613 | err = mddev_lock(mddev); |
6614 | if (err) | |
6615 | return err; | |
6616 | conf = mddev->private; | |
6617 | if (!conf) | |
6618 | err = -ENODEV; | |
6619 | else if (new != conf->worker_cnt_per_group) { | |
6620 | mddev_suspend(mddev); | |
b721420e | 6621 | |
6791875e N |
6622 | old_groups = conf->worker_groups; |
6623 | if (old_groups) | |
6624 | flush_workqueue(raid5_wq); | |
d206dcfa | 6625 | |
6791875e N |
6626 | err = alloc_thread_groups(conf, new, |
6627 | &group_cnt, &worker_cnt_per_group, | |
6628 | &new_groups); | |
6629 | if (!err) { | |
6630 | spin_lock_irq(&conf->device_lock); | |
6631 | conf->group_cnt = group_cnt; | |
6632 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6633 | conf->worker_groups = new_groups; | |
6634 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6635 | |
6791875e N |
6636 | if (old_groups) |
6637 | kfree(old_groups[0].workers); | |
6638 | kfree(old_groups); | |
6639 | } | |
6640 | mddev_resume(mddev); | |
b721420e | 6641 | } |
6791875e | 6642 | mddev_unlock(mddev); |
b721420e | 6643 | |
6791875e | 6644 | return err ?: len; |
b721420e SL |
6645 | } |
6646 | ||
6647 | static struct md_sysfs_entry | |
6648 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6649 | raid5_show_group_thread_cnt, | |
6650 | raid5_store_group_thread_cnt); | |
6651 | ||
007583c9 | 6652 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6653 | &raid5_stripecache_size.attr, |
6654 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6655 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6656 | &raid5_group_thread_cnt.attr, |
d592a996 | 6657 | &raid5_skip_copy.attr, |
d06f191f | 6658 | &raid5_rmw_level.attr, |
2c7da14b | 6659 | &r5c_journal_mode.attr, |
3f294f4f N |
6660 | NULL, |
6661 | }; | |
007583c9 N |
6662 | static struct attribute_group raid5_attrs_group = { |
6663 | .name = NULL, | |
6664 | .attrs = raid5_attrs, | |
3f294f4f N |
6665 | }; |
6666 | ||
60aaf933 | 6667 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6668 | int *group_cnt, | |
6669 | int *worker_cnt_per_group, | |
6670 | struct r5worker_group **worker_groups) | |
851c30c9 | 6671 | { |
566c09c5 | 6672 | int i, j, k; |
851c30c9 SL |
6673 | ssize_t size; |
6674 | struct r5worker *workers; | |
6675 | ||
60aaf933 | 6676 | *worker_cnt_per_group = cnt; |
851c30c9 | 6677 | if (cnt == 0) { |
60aaf933 | 6678 | *group_cnt = 0; |
6679 | *worker_groups = NULL; | |
851c30c9 SL |
6680 | return 0; |
6681 | } | |
60aaf933 | 6682 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6683 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6684 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6685 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6686 | *group_cnt, GFP_NOIO); | |
6687 | if (!*worker_groups || !workers) { | |
851c30c9 | 6688 | kfree(workers); |
60aaf933 | 6689 | kfree(*worker_groups); |
851c30c9 SL |
6690 | return -ENOMEM; |
6691 | } | |
6692 | ||
60aaf933 | 6693 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6694 | struct r5worker_group *group; |
6695 | ||
0c775d52 | 6696 | group = &(*worker_groups)[i]; |
851c30c9 | 6697 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6698 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6699 | group->conf = conf; |
6700 | group->workers = workers + i * cnt; | |
6701 | ||
6702 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6703 | struct r5worker *worker = group->workers + j; |
6704 | worker->group = group; | |
6705 | INIT_WORK(&worker->work, raid5_do_work); | |
6706 | ||
6707 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6708 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6709 | } |
6710 | } | |
6711 | ||
6712 | return 0; | |
6713 | } | |
6714 | ||
6715 | static void free_thread_groups(struct r5conf *conf) | |
6716 | { | |
6717 | if (conf->worker_groups) | |
6718 | kfree(conf->worker_groups[0].workers); | |
6719 | kfree(conf->worker_groups); | |
6720 | conf->worker_groups = NULL; | |
6721 | } | |
6722 | ||
80c3a6ce | 6723 | static sector_t |
fd01b88c | 6724 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6725 | { |
d1688a6d | 6726 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6727 | |
6728 | if (!sectors) | |
6729 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6730 | if (!raid_disks) |
7ec05478 | 6731 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6732 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6733 | |
3cb5edf4 N |
6734 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6735 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6736 | return sectors * (raid_disks - conf->max_degraded); |
6737 | } | |
6738 | ||
789b5e03 ON |
6739 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6740 | { | |
6741 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6742 | if (percpu->scribble) |
6743 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6744 | percpu->spare_page = NULL; |
6745 | percpu->scribble = NULL; | |
6746 | } | |
6747 | ||
6748 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6749 | { | |
6750 | if (conf->level == 6 && !percpu->spare_page) | |
6751 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6752 | if (!percpu->scribble) | |
46d5b785 | 6753 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6754 | conf->previous_raid_disks), |
6755 | max(conf->chunk_sectors, | |
6756 | conf->prev_chunk_sectors) | |
6757 | / STRIPE_SECTORS, | |
6758 | GFP_KERNEL); | |
789b5e03 ON |
6759 | |
6760 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6761 | free_scratch_buffer(conf, percpu); | |
6762 | return -ENOMEM; | |
6763 | } | |
6764 | ||
6765 | return 0; | |
6766 | } | |
6767 | ||
29c6d1bb | 6768 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6769 | { |
29c6d1bb SAS |
6770 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6771 | ||
6772 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6773 | return 0; | |
6774 | } | |
36d1c647 | 6775 | |
29c6d1bb SAS |
6776 | static void raid5_free_percpu(struct r5conf *conf) |
6777 | { | |
36d1c647 DW |
6778 | if (!conf->percpu) |
6779 | return; | |
6780 | ||
29c6d1bb | 6781 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6782 | free_percpu(conf->percpu); |
6783 | } | |
6784 | ||
d1688a6d | 6785 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6786 | { |
d7bd398e SL |
6787 | int i; |
6788 | ||
ff875738 AP |
6789 | log_exit(conf); |
6790 | ||
30c89465 | 6791 | if (conf->shrinker.nr_deferred) |
edbe83ab | 6792 | unregister_shrinker(&conf->shrinker); |
5c7e81c3 | 6793 | |
851c30c9 | 6794 | free_thread_groups(conf); |
95fc17aa | 6795 | shrink_stripes(conf); |
36d1c647 | 6796 | raid5_free_percpu(conf); |
d7bd398e SL |
6797 | for (i = 0; i < conf->pool_size; i++) |
6798 | if (conf->disks[i].extra_page) | |
6799 | put_page(conf->disks[i].extra_page); | |
95fc17aa DW |
6800 | kfree(conf->disks); |
6801 | kfree(conf->stripe_hashtbl); | |
aaf9f12e | 6802 | kfree(conf->pending_data); |
95fc17aa DW |
6803 | kfree(conf); |
6804 | } | |
6805 | ||
29c6d1bb | 6806 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6807 | { |
29c6d1bb | 6808 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6809 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6810 | ||
29c6d1bb | 6811 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6812 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6813 | __func__, cpu); | |
29c6d1bb | 6814 | return -ENOMEM; |
36d1c647 | 6815 | } |
29c6d1bb | 6816 | return 0; |
36d1c647 | 6817 | } |
36d1c647 | 6818 | |
d1688a6d | 6819 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6820 | { |
789b5e03 | 6821 | int err = 0; |
36d1c647 | 6822 | |
789b5e03 ON |
6823 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6824 | if (!conf->percpu) | |
36d1c647 | 6825 | return -ENOMEM; |
789b5e03 | 6826 | |
29c6d1bb | 6827 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6828 | if (!err) { |
6829 | conf->scribble_disks = max(conf->raid_disks, | |
6830 | conf->previous_raid_disks); | |
6831 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6832 | conf->prev_chunk_sectors); | |
6833 | } | |
36d1c647 DW |
6834 | return err; |
6835 | } | |
6836 | ||
edbe83ab N |
6837 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6838 | struct shrink_control *sc) | |
6839 | { | |
6840 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6841 | unsigned long ret = SHRINK_STOP; |
6842 | ||
6843 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6844 | ret= 0; | |
49895bcc N |
6845 | while (ret < sc->nr_to_scan && |
6846 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6847 | if (drop_one_stripe(conf) == 0) { |
6848 | ret = SHRINK_STOP; | |
6849 | break; | |
6850 | } | |
6851 | ret++; | |
6852 | } | |
6853 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6854 | } |
6855 | return ret; | |
6856 | } | |
6857 | ||
6858 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6859 | struct shrink_control *sc) | |
6860 | { | |
6861 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6862 | ||
6863 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6864 | /* unlikely, but not impossible */ | |
6865 | return 0; | |
6866 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6867 | } | |
6868 | ||
d1688a6d | 6869 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6870 | { |
d1688a6d | 6871 | struct r5conf *conf; |
5e5e3e78 | 6872 | int raid_disk, memory, max_disks; |
3cb03002 | 6873 | struct md_rdev *rdev; |
1da177e4 | 6874 | struct disk_info *disk; |
0232605d | 6875 | char pers_name[6]; |
566c09c5 | 6876 | int i; |
60aaf933 | 6877 | int group_cnt, worker_cnt_per_group; |
6878 | struct r5worker_group *new_group; | |
1da177e4 | 6879 | |
91adb564 N |
6880 | if (mddev->new_level != 5 |
6881 | && mddev->new_level != 4 | |
6882 | && mddev->new_level != 6) { | |
cc6167b4 N |
6883 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6884 | mdname(mddev), mddev->new_level); | |
91adb564 | 6885 | return ERR_PTR(-EIO); |
1da177e4 | 6886 | } |
91adb564 N |
6887 | if ((mddev->new_level == 5 |
6888 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6889 | (mddev->new_level == 6 | |
6890 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6891 | pr_warn("md/raid:%s: layout %d not supported\n", |
6892 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6893 | return ERR_PTR(-EIO); |
99c0fb5f | 6894 | } |
91adb564 | 6895 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6896 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6897 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6898 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6899 | } |
6900 | ||
664e7c41 AN |
6901 | if (!mddev->new_chunk_sectors || |
6902 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6903 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6904 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6905 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6906 | return ERR_PTR(-EINVAL); |
f6705578 N |
6907 | } |
6908 | ||
d1688a6d | 6909 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6910 | if (conf == NULL) |
1da177e4 | 6911 | goto abort; |
aaf9f12e SL |
6912 | INIT_LIST_HEAD(&conf->free_list); |
6913 | INIT_LIST_HEAD(&conf->pending_list); | |
6914 | conf->pending_data = kzalloc(sizeof(struct r5pending_data) * | |
6915 | PENDING_IO_MAX, GFP_KERNEL); | |
6916 | if (!conf->pending_data) | |
6917 | goto abort; | |
6918 | for (i = 0; i < PENDING_IO_MAX; i++) | |
6919 | list_add(&conf->pending_data[i].sibling, &conf->free_list); | |
851c30c9 | 6920 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6921 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6922 | &new_group)) { | |
6923 | conf->group_cnt = group_cnt; | |
6924 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6925 | conf->worker_groups = new_group; | |
6926 | } else | |
851c30c9 | 6927 | goto abort; |
f5efd45a | 6928 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6929 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6930 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6931 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6932 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6933 | init_waitqueue_head(&conf->wait_for_overlap); |
6934 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 6935 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
6936 | INIT_LIST_HEAD(&conf->hold_list); |
6937 | INIT_LIST_HEAD(&conf->delayed_list); | |
6938 | INIT_LIST_HEAD(&conf->bitmap_list); | |
c3cce6cd | 6939 | bio_list_init(&conf->return_bi); |
773ca82f | 6940 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6941 | atomic_set(&conf->active_stripes, 0); |
6942 | atomic_set(&conf->preread_active_stripes, 0); | |
6943 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
6944 | spin_lock_init(&conf->pending_bios_lock); |
6945 | conf->batch_bio_dispatch = true; | |
6946 | rdev_for_each(rdev, mddev) { | |
6947 | if (test_bit(Journal, &rdev->flags)) | |
6948 | continue; | |
6949 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
6950 | conf->batch_bio_dispatch = false; | |
6951 | break; | |
6952 | } | |
6953 | } | |
6954 | ||
f5efd45a | 6955 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 6956 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6957 | |
6958 | conf->raid_disks = mddev->raid_disks; | |
6959 | if (mddev->reshape_position == MaxSector) | |
6960 | conf->previous_raid_disks = mddev->raid_disks; | |
6961 | else | |
f6705578 | 6962 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6963 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6964 | |
5e5e3e78 | 6965 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc | 6966 | GFP_KERNEL); |
d7bd398e | 6967 | |
b55e6bfc N |
6968 | if (!conf->disks) |
6969 | goto abort; | |
9ffae0cf | 6970 | |
d7bd398e SL |
6971 | for (i = 0; i < max_disks; i++) { |
6972 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
6973 | if (!conf->disks[i].extra_page) | |
6974 | goto abort; | |
6975 | } | |
6976 | ||
1da177e4 LT |
6977 | conf->mddev = mddev; |
6978 | ||
fccddba0 | 6979 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6980 | goto abort; |
1da177e4 | 6981 | |
566c09c5 SL |
6982 | /* We init hash_locks[0] separately to that it can be used |
6983 | * as the reference lock in the spin_lock_nest_lock() call | |
6984 | * in lock_all_device_hash_locks_irq in order to convince | |
6985 | * lockdep that we know what we are doing. | |
6986 | */ | |
6987 | spin_lock_init(conf->hash_locks); | |
6988 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6989 | spin_lock_init(conf->hash_locks + i); | |
6990 | ||
6991 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6992 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6993 | ||
6994 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6995 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6996 | ||
1e6d690b SL |
6997 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
6998 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
6999 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
7000 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
7001 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
7002 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 7003 | |
36d1c647 | 7004 | conf->level = mddev->new_level; |
46d5b785 | 7005 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
7006 | if (raid5_alloc_percpu(conf) != 0) |
7007 | goto abort; | |
7008 | ||
0c55e022 | 7009 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 7010 | |
dafb20fa | 7011 | rdev_for_each(rdev, mddev) { |
1da177e4 | 7012 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 7013 | if (raid_disk >= max_disks |
f2076e7d | 7014 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
7015 | continue; |
7016 | disk = conf->disks + raid_disk; | |
7017 | ||
17045f52 N |
7018 | if (test_bit(Replacement, &rdev->flags)) { |
7019 | if (disk->replacement) | |
7020 | goto abort; | |
7021 | disk->replacement = rdev; | |
7022 | } else { | |
7023 | if (disk->rdev) | |
7024 | goto abort; | |
7025 | disk->rdev = rdev; | |
7026 | } | |
1da177e4 | 7027 | |
b2d444d7 | 7028 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 7029 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
7030 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
7031 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 7032 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
7033 | /* Cannot rely on bitmap to complete recovery */ |
7034 | conf->fullsync = 1; | |
1da177e4 LT |
7035 | } |
7036 | ||
91adb564 | 7037 | conf->level = mddev->new_level; |
584acdd4 | 7038 | if (conf->level == 6) { |
16a53ecc | 7039 | conf->max_degraded = 2; |
584acdd4 MS |
7040 | if (raid6_call.xor_syndrome) |
7041 | conf->rmw_level = PARITY_ENABLE_RMW; | |
7042 | else | |
7043 | conf->rmw_level = PARITY_DISABLE_RMW; | |
7044 | } else { | |
16a53ecc | 7045 | conf->max_degraded = 1; |
584acdd4 MS |
7046 | conf->rmw_level = PARITY_ENABLE_RMW; |
7047 | } | |
91adb564 | 7048 | conf->algorithm = mddev->new_layout; |
fef9c61f | 7049 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 7050 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 7051 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 7052 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
7053 | } else { |
7054 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
7055 | conf->prev_algo = conf->algorithm; | |
e183eaed | 7056 | } |
1da177e4 | 7057 | |
edbe83ab | 7058 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
7059 | if (mddev->reshape_position != MaxSector) { |
7060 | int stripes = max_t(int, | |
7061 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
7062 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
7063 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
7064 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 7065 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
7066 | mdname(mddev), conf->min_nr_stripes); |
7067 | } | |
edbe83ab | 7068 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 7069 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 7070 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 7071 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
7072 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
7073 | mdname(mddev), memory); | |
91adb564 N |
7074 | goto abort; |
7075 | } else | |
cc6167b4 | 7076 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
7077 | /* |
7078 | * Losing a stripe head costs more than the time to refill it, | |
7079 | * it reduces the queue depth and so can hurt throughput. | |
7080 | * So set it rather large, scaled by number of devices. | |
7081 | */ | |
7082 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
7083 | conf->shrinker.scan_objects = raid5_cache_scan; | |
7084 | conf->shrinker.count_objects = raid5_cache_count; | |
7085 | conf->shrinker.batch = 128; | |
7086 | conf->shrinker.flags = 0; | |
6a0f53ff | 7087 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
7088 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
7089 | mdname(mddev)); | |
6a0f53ff CY |
7090 | goto abort; |
7091 | } | |
1da177e4 | 7092 | |
0232605d N |
7093 | sprintf(pers_name, "raid%d", mddev->new_level); |
7094 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 7095 | if (!conf->thread) { |
cc6167b4 N |
7096 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
7097 | mdname(mddev)); | |
16a53ecc N |
7098 | goto abort; |
7099 | } | |
91adb564 N |
7100 | |
7101 | return conf; | |
7102 | ||
7103 | abort: | |
7104 | if (conf) { | |
95fc17aa | 7105 | free_conf(conf); |
91adb564 N |
7106 | return ERR_PTR(-EIO); |
7107 | } else | |
7108 | return ERR_PTR(-ENOMEM); | |
7109 | } | |
7110 | ||
c148ffdc N |
7111 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
7112 | { | |
7113 | switch (algo) { | |
7114 | case ALGORITHM_PARITY_0: | |
7115 | if (raid_disk < max_degraded) | |
7116 | return 1; | |
7117 | break; | |
7118 | case ALGORITHM_PARITY_N: | |
7119 | if (raid_disk >= raid_disks - max_degraded) | |
7120 | return 1; | |
7121 | break; | |
7122 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 7123 | if (raid_disk == 0 || |
c148ffdc N |
7124 | raid_disk == raid_disks - 1) |
7125 | return 1; | |
7126 | break; | |
7127 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7128 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7129 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7130 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7131 | if (raid_disk == raid_disks - 1) | |
7132 | return 1; | |
7133 | } | |
7134 | return 0; | |
7135 | } | |
7136 | ||
849674e4 | 7137 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7138 | { |
d1688a6d | 7139 | struct r5conf *conf; |
9f7c2220 | 7140 | int working_disks = 0; |
c148ffdc | 7141 | int dirty_parity_disks = 0; |
3cb03002 | 7142 | struct md_rdev *rdev; |
713cf5a6 | 7143 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7144 | sector_t reshape_offset = 0; |
17045f52 | 7145 | int i; |
b5254dd5 N |
7146 | long long min_offset_diff = 0; |
7147 | int first = 1; | |
91adb564 | 7148 | |
8c6ac868 | 7149 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7150 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7151 | mdname(mddev)); | |
b5254dd5 N |
7152 | |
7153 | rdev_for_each(rdev, mddev) { | |
7154 | long long diff; | |
713cf5a6 | 7155 | |
f2076e7d | 7156 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7157 | journal_dev = rdev; |
f2076e7d SL |
7158 | continue; |
7159 | } | |
b5254dd5 N |
7160 | if (rdev->raid_disk < 0) |
7161 | continue; | |
7162 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7163 | if (first) { | |
7164 | min_offset_diff = diff; | |
7165 | first = 0; | |
7166 | } else if (mddev->reshape_backwards && | |
7167 | diff < min_offset_diff) | |
7168 | min_offset_diff = diff; | |
7169 | else if (!mddev->reshape_backwards && | |
7170 | diff > min_offset_diff) | |
7171 | min_offset_diff = diff; | |
7172 | } | |
7173 | ||
91adb564 N |
7174 | if (mddev->reshape_position != MaxSector) { |
7175 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7176 | * Difficulties arise if the stripe we would write to |
7177 | * next is at or after the stripe we would read from next. | |
7178 | * For a reshape that changes the number of devices, this | |
7179 | * is only possible for a very short time, and mdadm makes | |
7180 | * sure that time appears to have past before assembling | |
7181 | * the array. So we fail if that time hasn't passed. | |
7182 | * For a reshape that keeps the number of devices the same | |
7183 | * mdadm must be monitoring the reshape can keeping the | |
7184 | * critical areas read-only and backed up. It will start | |
7185 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7186 | */ |
7187 | sector_t here_new, here_old; | |
7188 | int old_disks; | |
18b00334 | 7189 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7190 | int chunk_sectors; |
7191 | int new_data_disks; | |
91adb564 | 7192 | |
713cf5a6 | 7193 | if (journal_dev) { |
cc6167b4 N |
7194 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7195 | mdname(mddev)); | |
713cf5a6 SL |
7196 | return -EINVAL; |
7197 | } | |
7198 | ||
88ce4930 | 7199 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7200 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7201 | mdname(mddev)); | |
91adb564 N |
7202 | return -EINVAL; |
7203 | } | |
91adb564 N |
7204 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7205 | /* reshape_position must be on a new-stripe boundary, and one | |
7206 | * further up in new geometry must map after here in old | |
7207 | * geometry. | |
05256d98 N |
7208 | * If the chunk sizes are different, then as we perform reshape |
7209 | * in units of the largest of the two, reshape_position needs | |
7210 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7211 | */ |
7212 | here_new = mddev->reshape_position; | |
05256d98 N |
7213 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7214 | new_data_disks = mddev->raid_disks - max_degraded; | |
7215 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7216 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7217 | mdname(mddev)); | |
91adb564 N |
7218 | return -EINVAL; |
7219 | } | |
05256d98 | 7220 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7221 | /* here_new is the stripe we will write to */ |
7222 | here_old = mddev->reshape_position; | |
05256d98 | 7223 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7224 | /* here_old is the first stripe that we might need to read |
7225 | * from */ | |
67ac6011 N |
7226 | if (mddev->delta_disks == 0) { |
7227 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7228 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7229 | * and taking constant backups. |
7230 | * mdadm always starts a situation like this in | |
7231 | * readonly mode so it can take control before | |
7232 | * allowing any writes. So just check for that. | |
7233 | */ | |
b5254dd5 N |
7234 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7235 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7236 | /* not really in-place - so OK */; | |
7237 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7238 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7239 | mdname(mddev)); | |
67ac6011 N |
7240 | return -EINVAL; |
7241 | } | |
2c810cdd | 7242 | } else if (mddev->reshape_backwards |
05256d98 N |
7243 | ? (here_new * chunk_sectors + min_offset_diff <= |
7244 | here_old * chunk_sectors) | |
7245 | : (here_new * chunk_sectors >= | |
7246 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7247 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7248 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7249 | mdname(mddev)); | |
91adb564 N |
7250 | return -EINVAL; |
7251 | } | |
cc6167b4 | 7252 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7253 | /* OK, we should be able to continue; */ |
7254 | } else { | |
7255 | BUG_ON(mddev->level != mddev->new_level); | |
7256 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7257 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7258 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7259 | } |
91adb564 | 7260 | |
3418d036 AP |
7261 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && |
7262 | test_bit(MD_HAS_PPL, &mddev->flags)) { | |
7263 | pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n", | |
7264 | mdname(mddev)); | |
7265 | clear_bit(MD_HAS_PPL, &mddev->flags); | |
7266 | } | |
7267 | ||
245f46c2 N |
7268 | if (mddev->private == NULL) |
7269 | conf = setup_conf(mddev); | |
7270 | else | |
7271 | conf = mddev->private; | |
7272 | ||
91adb564 N |
7273 | if (IS_ERR(conf)) |
7274 | return PTR_ERR(conf); | |
7275 | ||
486b0f7b SL |
7276 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7277 | if (!journal_dev) { | |
cc6167b4 N |
7278 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7279 | mdname(mddev)); | |
486b0f7b SL |
7280 | mddev->ro = 1; |
7281 | set_disk_ro(mddev->gendisk, 1); | |
7282 | } else if (mddev->recovery_cp == MaxSector) | |
7283 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7284 | } |
7285 | ||
b5254dd5 | 7286 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7287 | mddev->thread = conf->thread; |
7288 | conf->thread = NULL; | |
7289 | mddev->private = conf; | |
7290 | ||
17045f52 N |
7291 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7292 | i++) { | |
7293 | rdev = conf->disks[i].rdev; | |
7294 | if (!rdev && conf->disks[i].replacement) { | |
7295 | /* The replacement is all we have yet */ | |
7296 | rdev = conf->disks[i].replacement; | |
7297 | conf->disks[i].replacement = NULL; | |
7298 | clear_bit(Replacement, &rdev->flags); | |
7299 | conf->disks[i].rdev = rdev; | |
7300 | } | |
7301 | if (!rdev) | |
c148ffdc | 7302 | continue; |
17045f52 N |
7303 | if (conf->disks[i].replacement && |
7304 | conf->reshape_progress != MaxSector) { | |
7305 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7306 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7307 | goto abort; |
7308 | } | |
2f115882 | 7309 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7310 | working_disks++; |
2f115882 N |
7311 | continue; |
7312 | } | |
c148ffdc N |
7313 | /* This disc is not fully in-sync. However if it |
7314 | * just stored parity (beyond the recovery_offset), | |
7315 | * when we don't need to be concerned about the | |
7316 | * array being dirty. | |
7317 | * When reshape goes 'backwards', we never have | |
7318 | * partially completed devices, so we only need | |
7319 | * to worry about reshape going forwards. | |
7320 | */ | |
7321 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7322 | if (mddev->major_version == 0 && | |
7323 | mddev->minor_version > 90) | |
7324 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7325 | |
c148ffdc N |
7326 | if (rdev->recovery_offset < reshape_offset) { |
7327 | /* We need to check old and new layout */ | |
7328 | if (!only_parity(rdev->raid_disk, | |
7329 | conf->algorithm, | |
7330 | conf->raid_disks, | |
7331 | conf->max_degraded)) | |
7332 | continue; | |
7333 | } | |
7334 | if (!only_parity(rdev->raid_disk, | |
7335 | conf->prev_algo, | |
7336 | conf->previous_raid_disks, | |
7337 | conf->max_degraded)) | |
7338 | continue; | |
7339 | dirty_parity_disks++; | |
7340 | } | |
91adb564 | 7341 | |
17045f52 N |
7342 | /* |
7343 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7344 | */ | |
2e38a37f | 7345 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7346 | |
674806d6 | 7347 | if (has_failed(conf)) { |
cc6167b4 | 7348 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7349 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7350 | goto abort; |
7351 | } | |
7352 | ||
91adb564 | 7353 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7354 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7355 | mddev->resync_max_sectors = mddev->dev_sectors; |
7356 | ||
c148ffdc | 7357 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7358 | mddev->recovery_cp != MaxSector) { |
4536bf9b AP |
7359 | if (test_bit(MD_HAS_PPL, &mddev->flags)) |
7360 | pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n", | |
7361 | mdname(mddev)); | |
7362 | else if (mddev->ok_start_degraded) | |
cc6167b4 N |
7363 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7364 | mdname(mddev)); | |
6ff8d8ec | 7365 | else { |
cc6167b4 N |
7366 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7367 | mdname(mddev)); | |
6ff8d8ec N |
7368 | goto abort; |
7369 | } | |
1da177e4 LT |
7370 | } |
7371 | ||
cc6167b4 N |
7372 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7373 | mdname(mddev), conf->level, | |
7374 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7375 | mddev->new_layout); | |
1da177e4 LT |
7376 | |
7377 | print_raid5_conf(conf); | |
7378 | ||
fef9c61f | 7379 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7380 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7381 | atomic_set(&conf->reshape_stripes, 0); |
7382 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7383 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7384 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7385 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7386 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7387 | "reshape"); |
f6705578 N |
7388 | } |
7389 | ||
1da177e4 | 7390 | /* Ok, everything is just fine now */ |
a64c876f N |
7391 | if (mddev->to_remove == &raid5_attrs_group) |
7392 | mddev->to_remove = NULL; | |
00bcb4ac N |
7393 | else if (mddev->kobj.sd && |
7394 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7395 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7396 | mdname(mddev)); | |
4a5add49 | 7397 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7398 | |
4a5add49 | 7399 | if (mddev->queue) { |
9f7c2220 | 7400 | int chunk_size; |
620125f2 | 7401 | bool discard_supported = true; |
4a5add49 N |
7402 | /* read-ahead size must cover two whole stripes, which |
7403 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7404 | * number of raid devices | |
7405 | */ | |
7406 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7407 | int stripe = data_disks * | |
7408 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7409 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7410 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7411 | |
9f7c2220 N |
7412 | chunk_size = mddev->chunk_sectors << 9; |
7413 | blk_queue_io_min(mddev->queue, chunk_size); | |
7414 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7415 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7416 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7417 | /* |
7418 | * We can only discard a whole stripe. It doesn't make sense to | |
7419 | * discard data disk but write parity disk | |
7420 | */ | |
7421 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7422 | /* Round up to power of 2, as discard handling |
7423 | * currently assumes that */ | |
7424 | while ((stripe-1) & stripe) | |
7425 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7426 | mddev->queue->limits.discard_alignment = stripe; |
7427 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 KK |
7428 | |
7429 | /* | |
7430 | * We use 16-bit counter of active stripes in bi_phys_segments | |
7431 | * (minus one for over-loaded initialization) | |
7432 | */ | |
7433 | blk_queue_max_hw_sectors(mddev->queue, 0xfffe * STRIPE_SECTORS); | |
7434 | blk_queue_max_discard_sectors(mddev->queue, | |
7435 | 0xfffe * STRIPE_SECTORS); | |
7436 | ||
620125f2 SL |
7437 | /* |
7438 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 7439 | * guarantee discard_zeroes_data |
620125f2 SL |
7440 | */ |
7441 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 7442 | |
5026d7a9 PA |
7443 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
7444 | ||
05616be5 | 7445 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7446 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7447 | rdev->data_offset << 9); | |
05616be5 N |
7448 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7449 | rdev->new_data_offset << 9); | |
620125f2 SL |
7450 | /* |
7451 | * discard_zeroes_data is required, otherwise data | |
7452 | * could be lost. Consider a scenario: discard a stripe | |
7453 | * (the stripe could be inconsistent if | |
7454 | * discard_zeroes_data is 0); write one disk of the | |
7455 | * stripe (the stripe could be inconsistent again | |
7456 | * depending on which disks are used to calculate | |
7457 | * parity); the disk is broken; The stripe data of this | |
7458 | * disk is lost. | |
7459 | */ | |
7460 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
7461 | !bdev_get_queue(rdev->bdev)-> | |
7462 | limits.discard_zeroes_data) | |
7463 | discard_supported = false; | |
8e0e99ba N |
7464 | /* Unfortunately, discard_zeroes_data is not currently |
7465 | * a guarantee - just a hint. So we only allow DISCARD | |
7466 | * if the sysadmin has confirmed that only safe devices | |
7467 | * are in use by setting a module parameter. | |
7468 | */ | |
7469 | if (!devices_handle_discard_safely) { | |
7470 | if (discard_supported) { | |
7471 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
7472 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
7473 | } | |
7474 | discard_supported = false; | |
7475 | } | |
05616be5 | 7476 | } |
620125f2 SL |
7477 | |
7478 | if (discard_supported && | |
e7597e69 JS |
7479 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7480 | mddev->queue->limits.discard_granularity >= stripe) | |
620125f2 SL |
7481 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
7482 | mddev->queue); | |
7483 | else | |
7484 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
7485 | mddev->queue); | |
1dffdddd SL |
7486 | |
7487 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7488 | } |
23032a0e | 7489 | |
ff875738 AP |
7490 | if (log_init(conf, journal_dev)) |
7491 | goto abort; | |
5c7e81c3 | 7492 | |
1da177e4 LT |
7493 | return 0; |
7494 | abort: | |
01f96c0a | 7495 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7496 | print_raid5_conf(conf); |
7497 | free_conf(conf); | |
1da177e4 | 7498 | mddev->private = NULL; |
cc6167b4 | 7499 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7500 | return -EIO; |
7501 | } | |
7502 | ||
afa0f557 | 7503 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7504 | { |
afa0f557 | 7505 | struct r5conf *conf = priv; |
1da177e4 | 7506 | |
95fc17aa | 7507 | free_conf(conf); |
a64c876f | 7508 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7509 | } |
7510 | ||
849674e4 | 7511 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7512 | { |
d1688a6d | 7513 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7514 | int i; |
7515 | ||
9d8f0363 | 7516 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7517 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7518 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7519 | rcu_read_lock(); |
7520 | for (i = 0; i < conf->raid_disks; i++) { | |
7521 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7522 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7523 | } | |
7524 | rcu_read_unlock(); | |
1da177e4 | 7525 | seq_printf (seq, "]"); |
1da177e4 LT |
7526 | } |
7527 | ||
d1688a6d | 7528 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7529 | { |
7530 | int i; | |
7531 | struct disk_info *tmp; | |
7532 | ||
cc6167b4 | 7533 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7534 | if (!conf) { |
cc6167b4 | 7535 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7536 | return; |
7537 | } | |
cc6167b4 | 7538 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7539 | conf->raid_disks, |
7540 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7541 | |
7542 | for (i = 0; i < conf->raid_disks; i++) { | |
7543 | char b[BDEVNAME_SIZE]; | |
7544 | tmp = conf->disks + i; | |
7545 | if (tmp->rdev) | |
cc6167b4 | 7546 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7547 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7548 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7549 | } |
7550 | } | |
7551 | ||
fd01b88c | 7552 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7553 | { |
7554 | int i; | |
d1688a6d | 7555 | struct r5conf *conf = mddev->private; |
1da177e4 | 7556 | struct disk_info *tmp; |
6b965620 N |
7557 | int count = 0; |
7558 | unsigned long flags; | |
1da177e4 LT |
7559 | |
7560 | for (i = 0; i < conf->raid_disks; i++) { | |
7561 | tmp = conf->disks + i; | |
dd054fce N |
7562 | if (tmp->replacement |
7563 | && tmp->replacement->recovery_offset == MaxSector | |
7564 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7565 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7566 | /* Replacement has just become active. */ | |
7567 | if (!tmp->rdev | |
7568 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7569 | count++; | |
7570 | if (tmp->rdev) { | |
7571 | /* Replaced device not technically faulty, | |
7572 | * but we need to be sure it gets removed | |
7573 | * and never re-added. | |
7574 | */ | |
7575 | set_bit(Faulty, &tmp->rdev->flags); | |
7576 | sysfs_notify_dirent_safe( | |
7577 | tmp->rdev->sysfs_state); | |
7578 | } | |
7579 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7580 | } else if (tmp->rdev | |
70fffd0b | 7581 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7582 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7583 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7584 | count++; |
43c73ca4 | 7585 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7586 | } |
7587 | } | |
6b965620 | 7588 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7589 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7590 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7591 | print_raid5_conf(conf); |
6b965620 | 7592 | return count; |
1da177e4 LT |
7593 | } |
7594 | ||
b8321b68 | 7595 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7596 | { |
d1688a6d | 7597 | struct r5conf *conf = mddev->private; |
1da177e4 | 7598 | int err = 0; |
b8321b68 | 7599 | int number = rdev->raid_disk; |
657e3e4d | 7600 | struct md_rdev **rdevp; |
1da177e4 LT |
7601 | struct disk_info *p = conf->disks + number; |
7602 | ||
7603 | print_raid5_conf(conf); | |
f6b6ec5c | 7604 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
c2bb6242 | 7605 | /* |
f6b6ec5c SL |
7606 | * we can't wait pending write here, as this is called in |
7607 | * raid5d, wait will deadlock. | |
c2bb6242 | 7608 | */ |
f6b6ec5c SL |
7609 | if (atomic_read(&mddev->writes_pending)) |
7610 | return -EBUSY; | |
ff875738 | 7611 | log_exit(conf); |
f6b6ec5c | 7612 | return 0; |
c2bb6242 | 7613 | } |
657e3e4d N |
7614 | if (rdev == p->rdev) |
7615 | rdevp = &p->rdev; | |
7616 | else if (rdev == p->replacement) | |
7617 | rdevp = &p->replacement; | |
7618 | else | |
7619 | return 0; | |
7620 | ||
7621 | if (number >= conf->raid_disks && | |
7622 | conf->reshape_progress == MaxSector) | |
7623 | clear_bit(In_sync, &rdev->flags); | |
7624 | ||
7625 | if (test_bit(In_sync, &rdev->flags) || | |
7626 | atomic_read(&rdev->nr_pending)) { | |
7627 | err = -EBUSY; | |
7628 | goto abort; | |
7629 | } | |
7630 | /* Only remove non-faulty devices if recovery | |
7631 | * isn't possible. | |
7632 | */ | |
7633 | if (!test_bit(Faulty, &rdev->flags) && | |
7634 | mddev->recovery_disabled != conf->recovery_disabled && | |
7635 | !has_failed(conf) && | |
dd054fce | 7636 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7637 | number < conf->raid_disks) { |
7638 | err = -EBUSY; | |
7639 | goto abort; | |
7640 | } | |
7641 | *rdevp = NULL; | |
d787be40 N |
7642 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7643 | synchronize_rcu(); | |
7644 | if (atomic_read(&rdev->nr_pending)) { | |
7645 | /* lost the race, try later */ | |
7646 | err = -EBUSY; | |
7647 | *rdevp = rdev; | |
7648 | } | |
7649 | } | |
6358c239 AP |
7650 | if (!err) { |
7651 | err = log_modify(conf, rdev, false); | |
7652 | if (err) | |
7653 | goto abort; | |
7654 | } | |
d787be40 | 7655 | if (p->replacement) { |
dd054fce N |
7656 | /* We must have just cleared 'rdev' */ |
7657 | p->rdev = p->replacement; | |
7658 | clear_bit(Replacement, &p->replacement->flags); | |
7659 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7660 | * but will never see neither - if they are careful | |
7661 | */ | |
7662 | p->replacement = NULL; | |
7663 | clear_bit(WantReplacement, &rdev->flags); | |
6358c239 AP |
7664 | |
7665 | if (!err) | |
7666 | err = log_modify(conf, p->rdev, true); | |
dd054fce N |
7667 | } else |
7668 | /* We might have just removed the Replacement as faulty- | |
7669 | * clear the bit just in case | |
7670 | */ | |
7671 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7672 | abort: |
7673 | ||
7674 | print_raid5_conf(conf); | |
7675 | return err; | |
7676 | } | |
7677 | ||
fd01b88c | 7678 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7679 | { |
d1688a6d | 7680 | struct r5conf *conf = mddev->private; |
199050ea | 7681 | int err = -EEXIST; |
1da177e4 LT |
7682 | int disk; |
7683 | struct disk_info *p; | |
6c2fce2e NB |
7684 | int first = 0; |
7685 | int last = conf->raid_disks - 1; | |
1da177e4 | 7686 | |
f6b6ec5c | 7687 | if (test_bit(Journal, &rdev->flags)) { |
f6b6ec5c SL |
7688 | if (conf->log) |
7689 | return -EBUSY; | |
7690 | ||
7691 | rdev->raid_disk = 0; | |
7692 | /* | |
7693 | * The array is in readonly mode if journal is missing, so no | |
7694 | * write requests running. We should be safe | |
7695 | */ | |
ff875738 | 7696 | log_init(conf, rdev); |
f6b6ec5c SL |
7697 | return 0; |
7698 | } | |
7f0da59b N |
7699 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7700 | return -EBUSY; | |
7701 | ||
dc10c643 | 7702 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7703 | /* no point adding a device */ |
199050ea | 7704 | return -EINVAL; |
1da177e4 | 7705 | |
6c2fce2e NB |
7706 | if (rdev->raid_disk >= 0) |
7707 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7708 | |
7709 | /* | |
16a53ecc N |
7710 | * find the disk ... but prefer rdev->saved_raid_disk |
7711 | * if possible. | |
1da177e4 | 7712 | */ |
16a53ecc | 7713 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7714 | rdev->saved_raid_disk >= first && |
16a53ecc | 7715 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7716 | first = rdev->saved_raid_disk; |
7717 | ||
7718 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7719 | p = conf->disks + disk; |
7720 | if (p->rdev == NULL) { | |
b2d444d7 | 7721 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7722 | rdev->raid_disk = disk; |
72626685 N |
7723 | if (rdev->saved_raid_disk != disk) |
7724 | conf->fullsync = 1; | |
d6065f7b | 7725 | rcu_assign_pointer(p->rdev, rdev); |
6358c239 AP |
7726 | |
7727 | err = log_modify(conf, rdev, true); | |
7728 | ||
5cfb22a1 | 7729 | goto out; |
1da177e4 | 7730 | } |
5cfb22a1 N |
7731 | } |
7732 | for (disk = first; disk <= last; disk++) { | |
7733 | p = conf->disks + disk; | |
7bfec5f3 N |
7734 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7735 | p->replacement == NULL) { | |
7736 | clear_bit(In_sync, &rdev->flags); | |
7737 | set_bit(Replacement, &rdev->flags); | |
7738 | rdev->raid_disk = disk; | |
7739 | err = 0; | |
7740 | conf->fullsync = 1; | |
7741 | rcu_assign_pointer(p->replacement, rdev); | |
7742 | break; | |
7743 | } | |
7744 | } | |
5cfb22a1 | 7745 | out: |
1da177e4 | 7746 | print_raid5_conf(conf); |
199050ea | 7747 | return err; |
1da177e4 LT |
7748 | } |
7749 | ||
fd01b88c | 7750 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7751 | { |
7752 | /* no resync is happening, and there is enough space | |
7753 | * on all devices, so we can resize. | |
7754 | * We need to make sure resync covers any new space. | |
7755 | * If the array is shrinking we should possibly wait until | |
7756 | * any io in the removed space completes, but it hardly seems | |
7757 | * worth it. | |
7758 | */ | |
a4a6125a | 7759 | sector_t newsize; |
3cb5edf4 N |
7760 | struct r5conf *conf = mddev->private; |
7761 | ||
3418d036 | 7762 | if (conf->log || raid5_has_ppl(conf)) |
713cf5a6 | 7763 | return -EINVAL; |
3cb5edf4 | 7764 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7765 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7766 | if (mddev->external_size && | |
7767 | mddev->array_sectors > newsize) | |
b522adcd | 7768 | return -EINVAL; |
a4a6125a N |
7769 | if (mddev->bitmap) { |
7770 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7771 | if (ret) | |
7772 | return ret; | |
7773 | } | |
7774 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7775 | if (sectors > mddev->dev_sectors && |
7776 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7777 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7778 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7779 | } | |
58c0fed4 | 7780 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7781 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7782 | return 0; |
7783 | } | |
7784 | ||
fd01b88c | 7785 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7786 | { |
7787 | /* Can only proceed if there are plenty of stripe_heads. | |
7788 | * We need a minimum of one full stripe,, and for sensible progress | |
7789 | * it is best to have about 4 times that. | |
7790 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7791 | * allow for chunk sizes up to 256K, which is probably OK. | |
7792 | * If the chunk size is greater, user-space should request more | |
7793 | * stripe_heads first. | |
7794 | */ | |
d1688a6d | 7795 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7796 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7797 | > conf->min_nr_stripes || |
01ee22b4 | 7798 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7799 | > conf->min_nr_stripes) { |
cc6167b4 N |
7800 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7801 | mdname(mddev), | |
7802 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
7803 | / STRIPE_SIZE)*4); | |
01ee22b4 N |
7804 | return 0; |
7805 | } | |
7806 | return 1; | |
7807 | } | |
7808 | ||
fd01b88c | 7809 | static int check_reshape(struct mddev *mddev) |
29269553 | 7810 | { |
d1688a6d | 7811 | struct r5conf *conf = mddev->private; |
29269553 | 7812 | |
3418d036 | 7813 | if (conf->log || raid5_has_ppl(conf)) |
713cf5a6 | 7814 | return -EINVAL; |
88ce4930 N |
7815 | if (mddev->delta_disks == 0 && |
7816 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7817 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7818 | return 0; /* nothing to do */ |
674806d6 | 7819 | if (has_failed(conf)) |
ec32a2bd | 7820 | return -EINVAL; |
fdcfbbb6 | 7821 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7822 | /* We might be able to shrink, but the devices must |
7823 | * be made bigger first. | |
7824 | * For raid6, 4 is the minimum size. | |
7825 | * Otherwise 2 is the minimum | |
7826 | */ | |
7827 | int min = 2; | |
7828 | if (mddev->level == 6) | |
7829 | min = 4; | |
7830 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7831 | return -EINVAL; | |
7832 | } | |
29269553 | 7833 | |
01ee22b4 | 7834 | if (!check_stripe_cache(mddev)) |
29269553 | 7835 | return -ENOSPC; |
29269553 | 7836 | |
738a2738 N |
7837 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7838 | mddev->delta_disks > 0) | |
7839 | if (resize_chunks(conf, | |
7840 | conf->previous_raid_disks | |
7841 | + max(0, mddev->delta_disks), | |
7842 | max(mddev->new_chunk_sectors, | |
7843 | mddev->chunk_sectors) | |
7844 | ) < 0) | |
7845 | return -ENOMEM; | |
e56108d6 N |
7846 | return resize_stripes(conf, (conf->previous_raid_disks |
7847 | + mddev->delta_disks)); | |
63c70c4f N |
7848 | } |
7849 | ||
fd01b88c | 7850 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7851 | { |
d1688a6d | 7852 | struct r5conf *conf = mddev->private; |
3cb03002 | 7853 | struct md_rdev *rdev; |
63c70c4f | 7854 | int spares = 0; |
c04be0aa | 7855 | unsigned long flags; |
63c70c4f | 7856 | |
f416885e | 7857 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7858 | return -EBUSY; |
7859 | ||
01ee22b4 N |
7860 | if (!check_stripe_cache(mddev)) |
7861 | return -ENOSPC; | |
7862 | ||
30b67645 N |
7863 | if (has_failed(conf)) |
7864 | return -EINVAL; | |
7865 | ||
c6563a8c | 7866 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7867 | if (!test_bit(In_sync, &rdev->flags) |
7868 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7869 | spares++; |
c6563a8c | 7870 | } |
63c70c4f | 7871 | |
f416885e | 7872 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7873 | /* Not enough devices even to make a degraded array |
7874 | * of that size | |
7875 | */ | |
7876 | return -EINVAL; | |
7877 | ||
ec32a2bd N |
7878 | /* Refuse to reduce size of the array. Any reductions in |
7879 | * array size must be through explicit setting of array_size | |
7880 | * attribute. | |
7881 | */ | |
7882 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7883 | < mddev->array_sectors) { | |
cc6167b4 N |
7884 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7885 | mdname(mddev)); | |
ec32a2bd N |
7886 | return -EINVAL; |
7887 | } | |
7888 | ||
f6705578 | 7889 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7890 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7891 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7892 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7893 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7894 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7895 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7896 | conf->prev_algo = conf->algorithm; |
7897 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7898 | conf->generation++; |
7899 | /* Code that selects data_offset needs to see the generation update | |
7900 | * if reshape_progress has been set - so a memory barrier needed. | |
7901 | */ | |
7902 | smp_mb(); | |
2c810cdd | 7903 | if (mddev->reshape_backwards) |
fef9c61f N |
7904 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7905 | else | |
7906 | conf->reshape_progress = 0; | |
7907 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7908 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7909 | spin_unlock_irq(&conf->device_lock); |
7910 | ||
4d77e3ba N |
7911 | /* Now make sure any requests that proceeded on the assumption |
7912 | * the reshape wasn't running - like Discard or Read - have | |
7913 | * completed. | |
7914 | */ | |
7915 | mddev_suspend(mddev); | |
7916 | mddev_resume(mddev); | |
7917 | ||
29269553 N |
7918 | /* Add some new drives, as many as will fit. |
7919 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7920 | * Don't add devices if we are reducing the number of |
7921 | * devices in the array. This is because it is not possible | |
7922 | * to correctly record the "partially reconstructed" state of | |
7923 | * such devices during the reshape and confusion could result. | |
29269553 | 7924 | */ |
87a8dec9 | 7925 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7926 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7927 | if (rdev->raid_disk < 0 && |
7928 | !test_bit(Faulty, &rdev->flags)) { | |
7929 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7930 | if (rdev->raid_disk |
9d4c7d87 | 7931 | >= conf->previous_raid_disks) |
87a8dec9 | 7932 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7933 | else |
87a8dec9 | 7934 | rdev->recovery_offset = 0; |
36fad858 NK |
7935 | |
7936 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7937 | /* Failure here is OK */; |
50da0840 | 7938 | } |
87a8dec9 N |
7939 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7940 | && !test_bit(Faulty, &rdev->flags)) { | |
7941 | /* This is a spare that was manually added */ | |
7942 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7943 | } |
29269553 | 7944 | |
87a8dec9 N |
7945 | /* When a reshape changes the number of devices, |
7946 | * ->degraded is measured against the larger of the | |
7947 | * pre and post number of devices. | |
7948 | */ | |
ec32a2bd | 7949 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7950 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
7951 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7952 | } | |
63c70c4f | 7953 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7954 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 7955 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 7956 | |
29269553 N |
7957 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7958 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7959 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7960 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7961 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7962 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7963 | "reshape"); |
29269553 N |
7964 | if (!mddev->sync_thread) { |
7965 | mddev->recovery = 0; | |
7966 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7967 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7968 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7969 | mddev->new_chunk_sectors = |
7970 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7971 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7972 | rdev_for_each(rdev, mddev) |
7973 | rdev->new_data_offset = rdev->data_offset; | |
7974 | smp_wmb(); | |
ba8805b9 | 7975 | conf->generation --; |
fef9c61f | 7976 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7977 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7978 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7979 | spin_unlock_irq(&conf->device_lock); |
7980 | return -EAGAIN; | |
7981 | } | |
c8f517c4 | 7982 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7983 | md_wakeup_thread(mddev->sync_thread); |
7984 | md_new_event(mddev); | |
7985 | return 0; | |
7986 | } | |
29269553 | 7987 | |
ec32a2bd N |
7988 | /* This is called from the reshape thread and should make any |
7989 | * changes needed in 'conf' | |
7990 | */ | |
d1688a6d | 7991 | static void end_reshape(struct r5conf *conf) |
29269553 | 7992 | { |
29269553 | 7993 | |
f6705578 | 7994 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7995 | struct md_rdev *rdev; |
f6705578 | 7996 | |
f6705578 | 7997 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7998 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7999 | rdev_for_each(rdev, conf->mddev) |
8000 | rdev->data_offset = rdev->new_data_offset; | |
8001 | smp_wmb(); | |
fef9c61f | 8002 | conf->reshape_progress = MaxSector; |
6cbd8148 | 8003 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 8004 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 8005 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
8006 | |
8007 | /* read-ahead size must cover two whole stripes, which is | |
8008 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
8009 | */ | |
4a5add49 | 8010 | if (conf->mddev->queue) { |
cea9c228 | 8011 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 8012 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 8013 | / PAGE_SIZE); |
dc3b17cc JK |
8014 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
8015 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 8016 | } |
29269553 | 8017 | } |
29269553 N |
8018 | } |
8019 | ||
ec32a2bd N |
8020 | /* This is called from the raid5d thread with mddev_lock held. |
8021 | * It makes config changes to the device. | |
8022 | */ | |
fd01b88c | 8023 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 8024 | { |
d1688a6d | 8025 | struct r5conf *conf = mddev->private; |
cea9c228 N |
8026 | |
8027 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8028 | ||
ec32a2bd N |
8029 | if (mddev->delta_disks > 0) { |
8030 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
fe67d19a HM |
8031 | if (mddev->queue) { |
8032 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
8033 | revalidate_disk(mddev->gendisk); | |
8034 | } | |
ec32a2bd N |
8035 | } else { |
8036 | int d; | |
908f4fbd | 8037 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 8038 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 8039 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
8040 | for (d = conf->raid_disks ; |
8041 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 8042 | d++) { |
3cb03002 | 8043 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
8044 | if (rdev) |
8045 | clear_bit(In_sync, &rdev->flags); | |
8046 | rdev = conf->disks[d].replacement; | |
8047 | if (rdev) | |
8048 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 8049 | } |
cea9c228 | 8050 | } |
88ce4930 | 8051 | mddev->layout = conf->algorithm; |
09c9e5fa | 8052 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
8053 | mddev->reshape_position = MaxSector; |
8054 | mddev->delta_disks = 0; | |
2c810cdd | 8055 | mddev->reshape_backwards = 0; |
cea9c228 N |
8056 | } |
8057 | } | |
8058 | ||
fd01b88c | 8059 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 8060 | { |
d1688a6d | 8061 | struct r5conf *conf = mddev->private; |
72626685 N |
8062 | |
8063 | switch(state) { | |
e464eafd N |
8064 | case 2: /* resume for a suspend */ |
8065 | wake_up(&conf->wait_for_overlap); | |
8066 | break; | |
8067 | ||
72626685 | 8068 | case 1: /* stop all writes */ |
566c09c5 | 8069 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
8070 | /* '2' tells resync/reshape to pause so that all |
8071 | * active stripes can drain | |
8072 | */ | |
a39f7afd | 8073 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 8074 | conf->quiesce = 2; |
b1b46486 | 8075 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
8076 | atomic_read(&conf->active_stripes) == 0 && |
8077 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
8078 | unlock_all_device_hash_locks_irq(conf), |
8079 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 8080 | conf->quiesce = 1; |
566c09c5 | 8081 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
8082 | /* allow reshape to continue */ |
8083 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
8084 | break; |
8085 | ||
8086 | case 0: /* re-enable writes */ | |
566c09c5 | 8087 | lock_all_device_hash_locks_irq(conf); |
72626685 | 8088 | conf->quiesce = 0; |
b1b46486 | 8089 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 8090 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 8091 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
8092 | break; |
8093 | } | |
e6c033f7 | 8094 | r5l_quiesce(conf->log, state); |
72626685 | 8095 | } |
b15c2e57 | 8096 | |
fd01b88c | 8097 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 8098 | { |
e373ab10 | 8099 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 8100 | sector_t sectors; |
54071b38 | 8101 | |
f1b29bca | 8102 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 8103 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
8104 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
8105 | mdname(mddev)); | |
f1b29bca DW |
8106 | return ERR_PTR(-EINVAL); |
8107 | } | |
8108 | ||
e373ab10 N |
8109 | sectors = raid0_conf->strip_zone[0].zone_end; |
8110 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 8111 | mddev->dev_sectors = sectors; |
f1b29bca | 8112 | mddev->new_level = level; |
54071b38 TM |
8113 | mddev->new_layout = ALGORITHM_PARITY_N; |
8114 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
8115 | mddev->raid_disks += 1; | |
8116 | mddev->delta_disks = 1; | |
8117 | /* make sure it will be not marked as dirty */ | |
8118 | mddev->recovery_cp = MaxSector; | |
8119 | ||
8120 | return setup_conf(mddev); | |
8121 | } | |
8122 | ||
fd01b88c | 8123 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
8124 | { |
8125 | int chunksect; | |
6995f0b2 | 8126 | void *ret; |
d562b0c4 N |
8127 | |
8128 | if (mddev->raid_disks != 2 || | |
8129 | mddev->degraded > 1) | |
8130 | return ERR_PTR(-EINVAL); | |
8131 | ||
8132 | /* Should check if there are write-behind devices? */ | |
8133 | ||
8134 | chunksect = 64*2; /* 64K by default */ | |
8135 | ||
8136 | /* The array must be an exact multiple of chunksize */ | |
8137 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8138 | chunksect >>= 1; | |
8139 | ||
8140 | if ((chunksect<<9) < STRIPE_SIZE) | |
8141 | /* array size does not allow a suitable chunk size */ | |
8142 | return ERR_PTR(-EINVAL); | |
8143 | ||
8144 | mddev->new_level = 5; | |
8145 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8146 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8147 | |
6995f0b2 | 8148 | ret = setup_conf(mddev); |
32cd7cbb | 8149 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8150 | mddev_clear_unsupported_flags(mddev, |
8151 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8152 | return ret; |
d562b0c4 N |
8153 | } |
8154 | ||
fd01b88c | 8155 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8156 | { |
8157 | int new_layout; | |
8158 | ||
8159 | switch (mddev->layout) { | |
8160 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8161 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8162 | break; | |
8163 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8164 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8165 | break; | |
8166 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8167 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8168 | break; | |
8169 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8170 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8171 | break; | |
8172 | case ALGORITHM_PARITY_0_6: | |
8173 | new_layout = ALGORITHM_PARITY_0; | |
8174 | break; | |
8175 | case ALGORITHM_PARITY_N: | |
8176 | new_layout = ALGORITHM_PARITY_N; | |
8177 | break; | |
8178 | default: | |
8179 | return ERR_PTR(-EINVAL); | |
8180 | } | |
8181 | mddev->new_level = 5; | |
8182 | mddev->new_layout = new_layout; | |
8183 | mddev->delta_disks = -1; | |
8184 | mddev->raid_disks -= 1; | |
8185 | return setup_conf(mddev); | |
8186 | } | |
8187 | ||
fd01b88c | 8188 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8189 | { |
88ce4930 N |
8190 | /* For a 2-drive array, the layout and chunk size can be changed |
8191 | * immediately as not restriping is needed. | |
8192 | * For larger arrays we record the new value - after validation | |
8193 | * to be used by a reshape pass. | |
b3546035 | 8194 | */ |
d1688a6d | 8195 | struct r5conf *conf = mddev->private; |
597a711b | 8196 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8197 | |
597a711b | 8198 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8199 | return -EINVAL; |
8200 | if (new_chunk > 0) { | |
0ba459d2 | 8201 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8202 | return -EINVAL; |
597a711b | 8203 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8204 | return -EINVAL; |
597a711b | 8205 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8206 | /* not factor of array size */ |
8207 | return -EINVAL; | |
8208 | } | |
8209 | ||
8210 | /* They look valid */ | |
8211 | ||
88ce4930 | 8212 | if (mddev->raid_disks == 2) { |
597a711b N |
8213 | /* can make the change immediately */ |
8214 | if (mddev->new_layout >= 0) { | |
8215 | conf->algorithm = mddev->new_layout; | |
8216 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8217 | } |
8218 | if (new_chunk > 0) { | |
597a711b N |
8219 | conf->chunk_sectors = new_chunk ; |
8220 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8221 | } |
2953079c | 8222 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8223 | md_wakeup_thread(mddev->thread); |
b3546035 | 8224 | } |
50ac168a | 8225 | return check_reshape(mddev); |
88ce4930 N |
8226 | } |
8227 | ||
fd01b88c | 8228 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8229 | { |
597a711b | 8230 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8231 | |
597a711b | 8232 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8233 | return -EINVAL; |
b3546035 | 8234 | if (new_chunk > 0) { |
0ba459d2 | 8235 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8236 | return -EINVAL; |
597a711b | 8237 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8238 | return -EINVAL; |
597a711b | 8239 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8240 | /* not factor of array size */ |
8241 | return -EINVAL; | |
b3546035 | 8242 | } |
88ce4930 N |
8243 | |
8244 | /* They look valid */ | |
50ac168a | 8245 | return check_reshape(mddev); |
b3546035 N |
8246 | } |
8247 | ||
fd01b88c | 8248 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8249 | { |
8250 | /* raid5 can take over: | |
f1b29bca | 8251 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8252 | * raid1 - if there are two drives. We need to know the chunk size |
8253 | * raid4 - trivial - just use a raid4 layout. | |
8254 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8255 | */ |
f1b29bca DW |
8256 | if (mddev->level == 0) |
8257 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8258 | if (mddev->level == 1) |
8259 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8260 | if (mddev->level == 4) { |
8261 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8262 | mddev->new_level = 5; | |
8263 | return setup_conf(mddev); | |
8264 | } | |
fc9739c6 N |
8265 | if (mddev->level == 6) |
8266 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8267 | |
8268 | return ERR_PTR(-EINVAL); | |
8269 | } | |
8270 | ||
fd01b88c | 8271 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8272 | { |
f1b29bca DW |
8273 | /* raid4 can take over: |
8274 | * raid0 - if there is only one strip zone | |
8275 | * raid5 - if layout is right | |
a78d38a1 | 8276 | */ |
f1b29bca DW |
8277 | if (mddev->level == 0) |
8278 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8279 | if (mddev->level == 5 && |
8280 | mddev->layout == ALGORITHM_PARITY_N) { | |
8281 | mddev->new_layout = 0; | |
8282 | mddev->new_level = 4; | |
8283 | return setup_conf(mddev); | |
8284 | } | |
8285 | return ERR_PTR(-EINVAL); | |
8286 | } | |
d562b0c4 | 8287 | |
84fc4b56 | 8288 | static struct md_personality raid5_personality; |
245f46c2 | 8289 | |
fd01b88c | 8290 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8291 | { |
8292 | /* Currently can only take over a raid5. We map the | |
8293 | * personality to an equivalent raid6 personality | |
8294 | * with the Q block at the end. | |
8295 | */ | |
8296 | int new_layout; | |
8297 | ||
8298 | if (mddev->pers != &raid5_personality) | |
8299 | return ERR_PTR(-EINVAL); | |
8300 | if (mddev->degraded > 1) | |
8301 | return ERR_PTR(-EINVAL); | |
8302 | if (mddev->raid_disks > 253) | |
8303 | return ERR_PTR(-EINVAL); | |
8304 | if (mddev->raid_disks < 3) | |
8305 | return ERR_PTR(-EINVAL); | |
8306 | ||
8307 | switch (mddev->layout) { | |
8308 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8309 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8310 | break; | |
8311 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8312 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8313 | break; | |
8314 | case ALGORITHM_LEFT_SYMMETRIC: | |
8315 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8316 | break; | |
8317 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8318 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8319 | break; | |
8320 | case ALGORITHM_PARITY_0: | |
8321 | new_layout = ALGORITHM_PARITY_0_6; | |
8322 | break; | |
8323 | case ALGORITHM_PARITY_N: | |
8324 | new_layout = ALGORITHM_PARITY_N; | |
8325 | break; | |
8326 | default: | |
8327 | return ERR_PTR(-EINVAL); | |
8328 | } | |
8329 | mddev->new_level = 6; | |
8330 | mddev->new_layout = new_layout; | |
8331 | mddev->delta_disks = 1; | |
8332 | mddev->raid_disks += 1; | |
8333 | return setup_conf(mddev); | |
8334 | } | |
8335 | ||
ba903a3e AP |
8336 | static void raid5_reset_stripe_cache(struct mddev *mddev) |
8337 | { | |
8338 | struct r5conf *conf = mddev->private; | |
8339 | ||
8340 | mutex_lock(&conf->cache_size_mutex); | |
8341 | while (conf->max_nr_stripes && | |
8342 | drop_one_stripe(conf)) | |
8343 | ; | |
8344 | while (conf->min_nr_stripes > conf->max_nr_stripes && | |
8345 | grow_one_stripe(conf, GFP_KERNEL)) | |
8346 | ; | |
8347 | mutex_unlock(&conf->cache_size_mutex); | |
8348 | } | |
8349 | ||
8350 | static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf) | |
8351 | { | |
8352 | struct r5conf *conf; | |
8353 | int err; | |
8354 | ||
8355 | err = mddev_lock(mddev); | |
8356 | if (err) | |
8357 | return err; | |
8358 | conf = mddev->private; | |
8359 | if (!conf) { | |
8360 | mddev_unlock(mddev); | |
8361 | return -ENODEV; | |
8362 | } | |
8363 | ||
8364 | if (strncmp(buf, "ppl", 3) == 0 && !raid5_has_ppl(conf)) { | |
8365 | mddev_suspend(mddev); | |
8366 | set_bit(MD_HAS_PPL, &mddev->flags); | |
8367 | err = log_init(conf, NULL); | |
8368 | if (!err) | |
8369 | raid5_reset_stripe_cache(mddev); | |
8370 | mddev_resume(mddev); | |
8371 | } else if (strncmp(buf, "resync", 6) == 0 && raid5_has_ppl(conf)) { | |
8372 | mddev_suspend(mddev); | |
8373 | log_exit(conf); | |
8374 | raid5_reset_stripe_cache(mddev); | |
8375 | mddev_resume(mddev); | |
8376 | } else { | |
8377 | err = -EINVAL; | |
8378 | } | |
8379 | ||
8380 | if (!err) | |
8381 | md_update_sb(mddev, 1); | |
8382 | ||
8383 | mddev_unlock(mddev); | |
8384 | ||
8385 | return err; | |
8386 | } | |
8387 | ||
84fc4b56 | 8388 | static struct md_personality raid6_personality = |
16a53ecc N |
8389 | { |
8390 | .name = "raid6", | |
8391 | .level = 6, | |
8392 | .owner = THIS_MODULE, | |
849674e4 SL |
8393 | .make_request = raid5_make_request, |
8394 | .run = raid5_run, | |
afa0f557 | 8395 | .free = raid5_free, |
849674e4 SL |
8396 | .status = raid5_status, |
8397 | .error_handler = raid5_error, | |
16a53ecc N |
8398 | .hot_add_disk = raid5_add_disk, |
8399 | .hot_remove_disk= raid5_remove_disk, | |
8400 | .spare_active = raid5_spare_active, | |
849674e4 | 8401 | .sync_request = raid5_sync_request, |
16a53ecc | 8402 | .resize = raid5_resize, |
80c3a6ce | 8403 | .size = raid5_size, |
50ac168a | 8404 | .check_reshape = raid6_check_reshape, |
f416885e | 8405 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8406 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8407 | .quiesce = raid5_quiesce, |
245f46c2 | 8408 | .takeover = raid6_takeover, |
5c675f83 | 8409 | .congested = raid5_congested, |
16a53ecc | 8410 | }; |
84fc4b56 | 8411 | static struct md_personality raid5_personality = |
1da177e4 LT |
8412 | { |
8413 | .name = "raid5", | |
2604b703 | 8414 | .level = 5, |
1da177e4 | 8415 | .owner = THIS_MODULE, |
849674e4 SL |
8416 | .make_request = raid5_make_request, |
8417 | .run = raid5_run, | |
afa0f557 | 8418 | .free = raid5_free, |
849674e4 SL |
8419 | .status = raid5_status, |
8420 | .error_handler = raid5_error, | |
1da177e4 LT |
8421 | .hot_add_disk = raid5_add_disk, |
8422 | .hot_remove_disk= raid5_remove_disk, | |
8423 | .spare_active = raid5_spare_active, | |
849674e4 | 8424 | .sync_request = raid5_sync_request, |
1da177e4 | 8425 | .resize = raid5_resize, |
80c3a6ce | 8426 | .size = raid5_size, |
63c70c4f N |
8427 | .check_reshape = raid5_check_reshape, |
8428 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8429 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8430 | .quiesce = raid5_quiesce, |
d562b0c4 | 8431 | .takeover = raid5_takeover, |
5c675f83 | 8432 | .congested = raid5_congested, |
ba903a3e | 8433 | .change_consistency_policy = raid5_change_consistency_policy, |
1da177e4 LT |
8434 | }; |
8435 | ||
84fc4b56 | 8436 | static struct md_personality raid4_personality = |
1da177e4 | 8437 | { |
2604b703 N |
8438 | .name = "raid4", |
8439 | .level = 4, | |
8440 | .owner = THIS_MODULE, | |
849674e4 SL |
8441 | .make_request = raid5_make_request, |
8442 | .run = raid5_run, | |
afa0f557 | 8443 | .free = raid5_free, |
849674e4 SL |
8444 | .status = raid5_status, |
8445 | .error_handler = raid5_error, | |
2604b703 N |
8446 | .hot_add_disk = raid5_add_disk, |
8447 | .hot_remove_disk= raid5_remove_disk, | |
8448 | .spare_active = raid5_spare_active, | |
849674e4 | 8449 | .sync_request = raid5_sync_request, |
2604b703 | 8450 | .resize = raid5_resize, |
80c3a6ce | 8451 | .size = raid5_size, |
3d37890b N |
8452 | .check_reshape = raid5_check_reshape, |
8453 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8454 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8455 | .quiesce = raid5_quiesce, |
a78d38a1 | 8456 | .takeover = raid4_takeover, |
5c675f83 | 8457 | .congested = raid5_congested, |
2604b703 N |
8458 | }; |
8459 | ||
8460 | static int __init raid5_init(void) | |
8461 | { | |
29c6d1bb SAS |
8462 | int ret; |
8463 | ||
851c30c9 SL |
8464 | raid5_wq = alloc_workqueue("raid5wq", |
8465 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8466 | if (!raid5_wq) | |
8467 | return -ENOMEM; | |
29c6d1bb SAS |
8468 | |
8469 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8470 | "md/raid5:prepare", | |
8471 | raid456_cpu_up_prepare, | |
8472 | raid456_cpu_dead); | |
8473 | if (ret) { | |
8474 | destroy_workqueue(raid5_wq); | |
8475 | return ret; | |
8476 | } | |
16a53ecc | 8477 | register_md_personality(&raid6_personality); |
2604b703 N |
8478 | register_md_personality(&raid5_personality); |
8479 | register_md_personality(&raid4_personality); | |
8480 | return 0; | |
1da177e4 LT |
8481 | } |
8482 | ||
2604b703 | 8483 | static void raid5_exit(void) |
1da177e4 | 8484 | { |
16a53ecc | 8485 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8486 | unregister_md_personality(&raid5_personality); |
8487 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8488 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8489 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8490 | } |
8491 | ||
8492 | module_init(raid5_init); | |
8493 | module_exit(raid5_exit); | |
8494 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8495 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8496 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8497 | MODULE_ALIAS("md-raid5"); |
8498 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8499 | MODULE_ALIAS("md-level-5"); |
8500 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8501 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8502 | MODULE_ALIAS("md-raid6"); | |
8503 | MODULE_ALIAS("md-level-6"); | |
8504 | ||
8505 | /* This used to be two separate modules, they were: */ | |
8506 | MODULE_ALIAS("raid5"); | |
8507 | MODULE_ALIAS("raid6"); |