]>
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 N |
60 | #include <trace/events/block.h> |
61 | ||
43b2e5d8 | 62 | #include "md.h" |
bff61975 | 63 | #include "raid5.h" |
54071b38 | 64 | #include "raid0.h" |
ef740c37 | 65 | #include "bitmap.h" |
72626685 | 66 | |
394ed8e4 SL |
67 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
68 | ||
851c30c9 SL |
69 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
70 | #define ANY_GROUP NUMA_NO_NODE | |
71 | ||
8e0e99ba N |
72 | static bool devices_handle_discard_safely = false; |
73 | module_param(devices_handle_discard_safely, bool, 0644); | |
74 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
75 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 76 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 77 | |
d1688a6d | 78 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
79 | { |
80 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
81 | return &conf->stripe_hashtbl[hash]; | |
82 | } | |
1da177e4 | 83 | |
566c09c5 SL |
84 | static inline int stripe_hash_locks_hash(sector_t sect) |
85 | { | |
86 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
87 | } | |
88 | ||
89 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
90 | { | |
91 | spin_lock_irq(conf->hash_locks + hash); | |
92 | spin_lock(&conf->device_lock); | |
93 | } | |
94 | ||
95 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
96 | { | |
97 | spin_unlock(&conf->device_lock); | |
98 | spin_unlock_irq(conf->hash_locks + hash); | |
99 | } | |
100 | ||
101 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
102 | { | |
103 | int i; | |
104 | local_irq_disable(); | |
105 | spin_lock(conf->hash_locks); | |
106 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
107 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
108 | spin_lock(&conf->device_lock); | |
109 | } | |
110 | ||
111 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
112 | { | |
113 | int i; | |
114 | spin_unlock(&conf->device_lock); | |
115 | for (i = NR_STRIPE_HASH_LOCKS; i; i--) | |
116 | spin_unlock(conf->hash_locks + i - 1); | |
117 | local_irq_enable(); | |
118 | } | |
119 | ||
d0dabf7e N |
120 | /* Find first data disk in a raid6 stripe */ |
121 | static inline int raid6_d0(struct stripe_head *sh) | |
122 | { | |
67cc2b81 N |
123 | if (sh->ddf_layout) |
124 | /* ddf always start from first device */ | |
125 | return 0; | |
126 | /* md starts just after Q block */ | |
d0dabf7e N |
127 | if (sh->qd_idx == sh->disks - 1) |
128 | return 0; | |
129 | else | |
130 | return sh->qd_idx + 1; | |
131 | } | |
16a53ecc N |
132 | static inline int raid6_next_disk(int disk, int raid_disks) |
133 | { | |
134 | disk++; | |
135 | return (disk < raid_disks) ? disk : 0; | |
136 | } | |
a4456856 | 137 | |
d0dabf7e N |
138 | /* When walking through the disks in a raid5, starting at raid6_d0, |
139 | * We need to map each disk to a 'slot', where the data disks are slot | |
140 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
141 | * is raid_disks-1. This help does that mapping. | |
142 | */ | |
67cc2b81 N |
143 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
144 | int *count, int syndrome_disks) | |
d0dabf7e | 145 | { |
6629542e | 146 | int slot = *count; |
67cc2b81 | 147 | |
e4424fee | 148 | if (sh->ddf_layout) |
6629542e | 149 | (*count)++; |
d0dabf7e | 150 | if (idx == sh->pd_idx) |
67cc2b81 | 151 | return syndrome_disks; |
d0dabf7e | 152 | if (idx == sh->qd_idx) |
67cc2b81 | 153 | return syndrome_disks + 1; |
e4424fee | 154 | if (!sh->ddf_layout) |
6629542e | 155 | (*count)++; |
d0dabf7e N |
156 | return slot; |
157 | } | |
158 | ||
34a6f80e | 159 | static void return_io(struct bio_list *return_bi) |
a4456856 | 160 | { |
34a6f80e N |
161 | struct bio *bi; |
162 | while ((bi = bio_list_pop(return_bi)) != NULL) { | |
4f024f37 | 163 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
164 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
165 | bi, 0); | |
4246a0b6 | 166 | bio_endio(bi); |
a4456856 DW |
167 | } |
168 | } | |
169 | ||
d1688a6d | 170 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 171 | |
600aa109 DW |
172 | static int stripe_operations_active(struct stripe_head *sh) |
173 | { | |
174 | return sh->check_state || sh->reconstruct_state || | |
175 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
176 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
177 | } | |
178 | ||
851c30c9 SL |
179 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
180 | { | |
181 | struct r5conf *conf = sh->raid_conf; | |
182 | struct r5worker_group *group; | |
bfc90cb0 | 183 | int thread_cnt; |
851c30c9 SL |
184 | int i, cpu = sh->cpu; |
185 | ||
186 | if (!cpu_online(cpu)) { | |
187 | cpu = cpumask_any(cpu_online_mask); | |
188 | sh->cpu = cpu; | |
189 | } | |
190 | ||
191 | if (list_empty(&sh->lru)) { | |
192 | struct r5worker_group *group; | |
193 | group = conf->worker_groups + cpu_to_group(cpu); | |
194 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
195 | group->stripes_cnt++; |
196 | sh->group = group; | |
851c30c9 SL |
197 | } |
198 | ||
199 | if (conf->worker_cnt_per_group == 0) { | |
200 | md_wakeup_thread(conf->mddev->thread); | |
201 | return; | |
202 | } | |
203 | ||
204 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
205 | ||
bfc90cb0 SL |
206 | group->workers[0].working = true; |
207 | /* at least one worker should run to avoid race */ | |
208 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
209 | ||
210 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
211 | /* wakeup more workers */ | |
212 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
213 | if (group->workers[i].working == false) { | |
214 | group->workers[i].working = true; | |
215 | queue_work_on(sh->cpu, raid5_wq, | |
216 | &group->workers[i].work); | |
217 | thread_cnt--; | |
218 | } | |
219 | } | |
851c30c9 SL |
220 | } |
221 | ||
566c09c5 SL |
222 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
223 | struct list_head *temp_inactive_list) | |
1da177e4 | 224 | { |
1e6d690b SL |
225 | int i; |
226 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
227 | ||
4eb788df SL |
228 | BUG_ON(!list_empty(&sh->lru)); |
229 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
230 | |
231 | if (r5c_is_writeback(conf->log)) | |
232 | for (i = sh->disks; i--; ) | |
233 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
234 | injournal++; | |
a39f7afd SL |
235 | /* |
236 | * When quiesce in r5c write back, set STRIPE_HANDLE for stripes with | |
237 | * data in journal, so they are not released to cached lists | |
238 | */ | |
239 | if (conf->quiesce && r5c_is_writeback(conf->log) && | |
240 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0) { | |
241 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
242 | r5c_make_stripe_write_out(sh); | |
243 | set_bit(STRIPE_HANDLE, &sh->state); | |
244 | } | |
1e6d690b | 245 | |
4eb788df SL |
246 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
247 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 248 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 249 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 250 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
251 | sh->bm_seq - conf->seq_write > 0) |
252 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
253 | else { | |
254 | clear_bit(STRIPE_DELAYED, &sh->state); | |
255 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 SL |
256 | if (conf->worker_cnt_per_group == 0) { |
257 | list_add_tail(&sh->lru, &conf->handle_list); | |
258 | } else { | |
259 | raid5_wakeup_stripe_thread(sh); | |
260 | return; | |
261 | } | |
4eb788df SL |
262 | } |
263 | md_wakeup_thread(conf->mddev->thread); | |
264 | } else { | |
265 | BUG_ON(stripe_operations_active(sh)); | |
266 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
267 | if (atomic_dec_return(&conf->preread_active_stripes) | |
268 | < IO_THRESHOLD) | |
269 | md_wakeup_thread(conf->mddev->thread); | |
270 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
271 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
272 | if (!r5c_is_writeback(conf->log)) | |
273 | list_add_tail(&sh->lru, temp_inactive_list); | |
274 | else { | |
275 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
276 | if (injournal == 0) | |
277 | list_add_tail(&sh->lru, temp_inactive_list); | |
278 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
279 | /* full stripe */ | |
280 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
281 | atomic_inc(&conf->r5c_cached_full_stripes); | |
282 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
283 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
284 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 285 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
286 | } else |
287 | /* | |
288 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
289 | * r5c_try_caching_write(). No need to | |
290 | * set it again. | |
291 | */ | |
1e6d690b | 292 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
293 | } |
294 | } | |
1da177e4 LT |
295 | } |
296 | } | |
d0dabf7e | 297 | |
566c09c5 SL |
298 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
299 | struct list_head *temp_inactive_list) | |
4eb788df SL |
300 | { |
301 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
302 | do_release_stripe(conf, sh, temp_inactive_list); |
303 | } | |
304 | ||
305 | /* | |
306 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
307 | * | |
308 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
309 | * given time. Adding stripes only takes device lock, while deleting stripes | |
310 | * only takes hash lock. | |
311 | */ | |
312 | static void release_inactive_stripe_list(struct r5conf *conf, | |
313 | struct list_head *temp_inactive_list, | |
314 | int hash) | |
315 | { | |
316 | int size; | |
6ab2a4b8 | 317 | bool do_wakeup = false; |
566c09c5 SL |
318 | unsigned long flags; |
319 | ||
320 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
321 | size = NR_STRIPE_HASH_LOCKS; | |
322 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
323 | } else | |
324 | size = 1; | |
325 | while (size) { | |
326 | struct list_head *list = &temp_inactive_list[size - 1]; | |
327 | ||
328 | /* | |
6d036f7d | 329 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
330 | * remove stripes from the list |
331 | */ | |
332 | if (!list_empty_careful(list)) { | |
333 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
334 | if (list_empty(conf->inactive_list + hash) && |
335 | !list_empty(list)) | |
336 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 337 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 338 | do_wakeup = true; |
566c09c5 SL |
339 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
340 | } | |
341 | size--; | |
342 | hash--; | |
343 | } | |
344 | ||
345 | if (do_wakeup) { | |
6ab2a4b8 | 346 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
347 | if (atomic_read(&conf->active_stripes) == 0) |
348 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
349 | if (conf->retry_read_aligned) |
350 | md_wakeup_thread(conf->mddev->thread); | |
351 | } | |
4eb788df SL |
352 | } |
353 | ||
773ca82f | 354 | /* should hold conf->device_lock already */ |
566c09c5 SL |
355 | static int release_stripe_list(struct r5conf *conf, |
356 | struct list_head *temp_inactive_list) | |
773ca82f | 357 | { |
eae8263f | 358 | struct stripe_head *sh, *t; |
773ca82f SL |
359 | int count = 0; |
360 | struct llist_node *head; | |
361 | ||
362 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 363 | head = llist_reverse_order(head); |
eae8263f | 364 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
365 | int hash; |
366 | ||
773ca82f SL |
367 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
368 | smp_mb(); | |
369 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
370 | /* | |
371 | * Don't worry the bit is set here, because if the bit is set | |
372 | * again, the count is always > 1. This is true for | |
373 | * STRIPE_ON_UNPLUG_LIST bit too. | |
374 | */ | |
566c09c5 SL |
375 | hash = sh->hash_lock_index; |
376 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
377 | count++; |
378 | } | |
379 | ||
380 | return count; | |
381 | } | |
382 | ||
6d036f7d | 383 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 384 | { |
d1688a6d | 385 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 386 | unsigned long flags; |
566c09c5 SL |
387 | struct list_head list; |
388 | int hash; | |
773ca82f | 389 | bool wakeup; |
16a53ecc | 390 | |
cf170f3f ES |
391 | /* Avoid release_list until the last reference. |
392 | */ | |
393 | if (atomic_add_unless(&sh->count, -1, 1)) | |
394 | return; | |
395 | ||
ad4068de | 396 | if (unlikely(!conf->mddev->thread) || |
397 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
398 | goto slow_path; |
399 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
400 | if (wakeup) | |
401 | md_wakeup_thread(conf->mddev->thread); | |
402 | return; | |
403 | slow_path: | |
4eb788df | 404 | local_irq_save(flags); |
773ca82f | 405 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 406 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
407 | INIT_LIST_HEAD(&list); |
408 | hash = sh->hash_lock_index; | |
409 | do_release_stripe(conf, sh, &list); | |
4eb788df | 410 | spin_unlock(&conf->device_lock); |
566c09c5 | 411 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
412 | } |
413 | local_irq_restore(flags); | |
1da177e4 LT |
414 | } |
415 | ||
fccddba0 | 416 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 417 | { |
45b4233c DW |
418 | pr_debug("remove_hash(), stripe %llu\n", |
419 | (unsigned long long)sh->sector); | |
1da177e4 | 420 | |
fccddba0 | 421 | hlist_del_init(&sh->hash); |
1da177e4 LT |
422 | } |
423 | ||
d1688a6d | 424 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 425 | { |
fccddba0 | 426 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 427 | |
45b4233c DW |
428 | pr_debug("insert_hash(), stripe %llu\n", |
429 | (unsigned long long)sh->sector); | |
1da177e4 | 430 | |
fccddba0 | 431 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
432 | } |
433 | ||
1da177e4 | 434 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 435 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
436 | { |
437 | struct stripe_head *sh = NULL; | |
438 | struct list_head *first; | |
439 | ||
566c09c5 | 440 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 441 | goto out; |
566c09c5 | 442 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
443 | sh = list_entry(first, struct stripe_head, lru); |
444 | list_del_init(first); | |
445 | remove_hash(sh); | |
446 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 447 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
448 | if (list_empty(conf->inactive_list + hash)) |
449 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
450 | out: |
451 | return sh; | |
452 | } | |
453 | ||
e4e11e38 | 454 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
455 | { |
456 | struct page *p; | |
457 | int i; | |
e4e11e38 | 458 | int num = sh->raid_conf->pool_size; |
1da177e4 | 459 | |
e4e11e38 | 460 | for (i = 0; i < num ; i++) { |
d592a996 | 461 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
462 | p = sh->dev[i].page; |
463 | if (!p) | |
464 | continue; | |
465 | sh->dev[i].page = NULL; | |
2d1f3b5d | 466 | put_page(p); |
1da177e4 LT |
467 | } |
468 | } | |
469 | ||
a9683a79 | 470 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
471 | { |
472 | int i; | |
e4e11e38 | 473 | int num = sh->raid_conf->pool_size; |
1da177e4 | 474 | |
e4e11e38 | 475 | for (i = 0; i < num; i++) { |
1da177e4 LT |
476 | struct page *page; |
477 | ||
a9683a79 | 478 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
479 | return 1; |
480 | } | |
481 | sh->dev[i].page = page; | |
d592a996 | 482 | sh->dev[i].orig_page = page; |
1da177e4 LT |
483 | } |
484 | return 0; | |
485 | } | |
486 | ||
784052ec | 487 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 488 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 489 | struct stripe_head *sh); |
1da177e4 | 490 | |
b5663ba4 | 491 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 492 | { |
d1688a6d | 493 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 494 | int i, seq; |
1da177e4 | 495 | |
78bafebd ES |
496 | BUG_ON(atomic_read(&sh->count) != 0); |
497 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 498 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 499 | BUG_ON(sh->batch_head); |
d84e0f10 | 500 | |
45b4233c | 501 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 502 | (unsigned long long)sector); |
566c09c5 SL |
503 | retry: |
504 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 505 | sh->generation = conf->generation - previous; |
b5663ba4 | 506 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 507 | sh->sector = sector; |
911d4ee8 | 508 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
509 | sh->state = 0; |
510 | ||
7ecaa1e6 | 511 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
512 | struct r5dev *dev = &sh->dev[i]; |
513 | ||
d84e0f10 | 514 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 515 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 516 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 517 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 518 | dev->read, dev->towrite, dev->written, |
1da177e4 | 519 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 520 | WARN_ON(1); |
1da177e4 LT |
521 | } |
522 | dev->flags = 0; | |
784052ec | 523 | raid5_build_block(sh, i, previous); |
1da177e4 | 524 | } |
566c09c5 SL |
525 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
526 | goto retry; | |
7a87f434 | 527 | sh->overwrite_disks = 0; |
1da177e4 | 528 | insert_hash(conf, sh); |
851c30c9 | 529 | sh->cpu = smp_processor_id(); |
da41ba65 | 530 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
531 | } |
532 | ||
d1688a6d | 533 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 534 | short generation) |
1da177e4 LT |
535 | { |
536 | struct stripe_head *sh; | |
537 | ||
45b4233c | 538 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 539 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 540 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 541 | return sh; |
45b4233c | 542 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
543 | return NULL; |
544 | } | |
545 | ||
674806d6 N |
546 | /* |
547 | * Need to check if array has failed when deciding whether to: | |
548 | * - start an array | |
549 | * - remove non-faulty devices | |
550 | * - add a spare | |
551 | * - allow a reshape | |
552 | * This determination is simple when no reshape is happening. | |
553 | * However if there is a reshape, we need to carefully check | |
554 | * both the before and after sections. | |
555 | * This is because some failed devices may only affect one | |
556 | * of the two sections, and some non-in_sync devices may | |
557 | * be insync in the section most affected by failed devices. | |
558 | */ | |
2e38a37f | 559 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 560 | { |
908f4fbd | 561 | int degraded, degraded2; |
674806d6 | 562 | int i; |
674806d6 N |
563 | |
564 | rcu_read_lock(); | |
565 | degraded = 0; | |
566 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 567 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
568 | if (rdev && test_bit(Faulty, &rdev->flags)) |
569 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
570 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
571 | degraded++; | |
572 | else if (test_bit(In_sync, &rdev->flags)) | |
573 | ; | |
574 | else | |
575 | /* not in-sync or faulty. | |
576 | * If the reshape increases the number of devices, | |
577 | * this is being recovered by the reshape, so | |
578 | * this 'previous' section is not in_sync. | |
579 | * If the number of devices is being reduced however, | |
580 | * the device can only be part of the array if | |
581 | * we are reverting a reshape, so this section will | |
582 | * be in-sync. | |
583 | */ | |
584 | if (conf->raid_disks >= conf->previous_raid_disks) | |
585 | degraded++; | |
586 | } | |
587 | rcu_read_unlock(); | |
908f4fbd N |
588 | if (conf->raid_disks == conf->previous_raid_disks) |
589 | return degraded; | |
674806d6 | 590 | rcu_read_lock(); |
908f4fbd | 591 | degraded2 = 0; |
674806d6 | 592 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 593 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
594 | if (rdev && test_bit(Faulty, &rdev->flags)) |
595 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 596 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 597 | degraded2++; |
674806d6 N |
598 | else if (test_bit(In_sync, &rdev->flags)) |
599 | ; | |
600 | else | |
601 | /* not in-sync or faulty. | |
602 | * If reshape increases the number of devices, this | |
603 | * section has already been recovered, else it | |
604 | * almost certainly hasn't. | |
605 | */ | |
606 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 607 | degraded2++; |
674806d6 N |
608 | } |
609 | rcu_read_unlock(); | |
908f4fbd N |
610 | if (degraded2 > degraded) |
611 | return degraded2; | |
612 | return degraded; | |
613 | } | |
614 | ||
615 | static int has_failed(struct r5conf *conf) | |
616 | { | |
617 | int degraded; | |
618 | ||
619 | if (conf->mddev->reshape_position == MaxSector) | |
620 | return conf->mddev->degraded > conf->max_degraded; | |
621 | ||
2e38a37f | 622 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
623 | if (degraded > conf->max_degraded) |
624 | return 1; | |
625 | return 0; | |
626 | } | |
627 | ||
6d036f7d SL |
628 | struct stripe_head * |
629 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
630 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
631 | { |
632 | struct stripe_head *sh; | |
566c09c5 | 633 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 634 | int inc_empty_inactive_list_flag; |
1da177e4 | 635 | |
45b4233c | 636 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 637 | |
566c09c5 | 638 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
639 | |
640 | do { | |
b1b46486 | 641 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 642 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 643 | *(conf->hash_locks + hash)); |
86b42c71 | 644 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 645 | if (!sh) { |
edbe83ab | 646 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 647 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
648 | if (!sh && !test_bit(R5_DID_ALLOC, |
649 | &conf->cache_state)) | |
edbe83ab N |
650 | set_bit(R5_ALLOC_MORE, |
651 | &conf->cache_state); | |
652 | } | |
1da177e4 LT |
653 | if (noblock && sh == NULL) |
654 | break; | |
a39f7afd SL |
655 | |
656 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 657 | if (!sh) { |
5423399a N |
658 | set_bit(R5_INACTIVE_BLOCKED, |
659 | &conf->cache_state); | |
a39f7afd | 660 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
661 | wait_event_lock_irq( |
662 | conf->wait_for_stripe, | |
566c09c5 SL |
663 | !list_empty(conf->inactive_list + hash) && |
664 | (atomic_read(&conf->active_stripes) | |
665 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
666 | || !test_bit(R5_INACTIVE_BLOCKED, |
667 | &conf->cache_state)), | |
6ab2a4b8 | 668 | *(conf->hash_locks + hash)); |
5423399a N |
669 | clear_bit(R5_INACTIVE_BLOCKED, |
670 | &conf->cache_state); | |
7da9d450 | 671 | } else { |
b5663ba4 | 672 | init_stripe(sh, sector, previous); |
7da9d450 N |
673 | atomic_inc(&sh->count); |
674 | } | |
e240c183 | 675 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 676 | spin_lock(&conf->device_lock); |
e240c183 | 677 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
678 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
679 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
680 | BUG_ON(list_empty(&sh->lru) && |
681 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
682 | inc_empty_inactive_list_flag = 0; |
683 | if (!list_empty(conf->inactive_list + hash)) | |
684 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 685 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
686 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
687 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
688 | if (sh->group) { |
689 | sh->group->stripes_cnt--; | |
690 | sh->group = NULL; | |
691 | } | |
1da177e4 | 692 | } |
7da9d450 | 693 | atomic_inc(&sh->count); |
6d183de4 | 694 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
695 | } |
696 | } while (sh == NULL); | |
697 | ||
566c09c5 | 698 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
699 | return sh; |
700 | } | |
701 | ||
7a87f434 | 702 | static bool is_full_stripe_write(struct stripe_head *sh) |
703 | { | |
704 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
705 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
706 | } | |
707 | ||
59fc630b | 708 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
709 | { | |
710 | local_irq_disable(); | |
711 | if (sh1 > sh2) { | |
712 | spin_lock(&sh2->stripe_lock); | |
713 | spin_lock_nested(&sh1->stripe_lock, 1); | |
714 | } else { | |
715 | spin_lock(&sh1->stripe_lock); | |
716 | spin_lock_nested(&sh2->stripe_lock, 1); | |
717 | } | |
718 | } | |
719 | ||
720 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
721 | { | |
722 | spin_unlock(&sh1->stripe_lock); | |
723 | spin_unlock(&sh2->stripe_lock); | |
724 | local_irq_enable(); | |
725 | } | |
726 | ||
727 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
728 | static bool stripe_can_batch(struct stripe_head *sh) | |
729 | { | |
9c3e333d SL |
730 | struct r5conf *conf = sh->raid_conf; |
731 | ||
732 | if (conf->log) | |
733 | return false; | |
59fc630b | 734 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 735 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 736 | is_full_stripe_write(sh); |
737 | } | |
738 | ||
739 | /* we only do back search */ | |
740 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
741 | { | |
742 | struct stripe_head *head; | |
743 | sector_t head_sector, tmp_sec; | |
744 | int hash; | |
745 | int dd_idx; | |
ff00d3b4 | 746 | int inc_empty_inactive_list_flag; |
59fc630b | 747 | |
59fc630b | 748 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
749 | tmp_sec = sh->sector; | |
750 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
751 | return; | |
752 | head_sector = sh->sector - STRIPE_SECTORS; | |
753 | ||
754 | hash = stripe_hash_locks_hash(head_sector); | |
755 | spin_lock_irq(conf->hash_locks + hash); | |
756 | head = __find_stripe(conf, head_sector, conf->generation); | |
757 | if (head && !atomic_inc_not_zero(&head->count)) { | |
758 | spin_lock(&conf->device_lock); | |
759 | if (!atomic_read(&head->count)) { | |
760 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
761 | atomic_inc(&conf->active_stripes); | |
762 | BUG_ON(list_empty(&head->lru) && | |
763 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
764 | inc_empty_inactive_list_flag = 0; |
765 | if (!list_empty(conf->inactive_list + hash)) | |
766 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 767 | list_del_init(&head->lru); |
ff00d3b4 ZL |
768 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
769 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 770 | if (head->group) { |
771 | head->group->stripes_cnt--; | |
772 | head->group = NULL; | |
773 | } | |
774 | } | |
775 | atomic_inc(&head->count); | |
776 | spin_unlock(&conf->device_lock); | |
777 | } | |
778 | spin_unlock_irq(conf->hash_locks + hash); | |
779 | ||
780 | if (!head) | |
781 | return; | |
782 | if (!stripe_can_batch(head)) | |
783 | goto out; | |
784 | ||
785 | lock_two_stripes(head, sh); | |
786 | /* clear_batch_ready clear the flag */ | |
787 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
788 | goto unlock_out; | |
789 | ||
790 | if (sh->batch_head) | |
791 | goto unlock_out; | |
792 | ||
793 | dd_idx = 0; | |
794 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
795 | dd_idx++; | |
1eff9d32 | 796 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 797 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 798 | goto unlock_out; |
799 | ||
800 | if (head->batch_head) { | |
801 | spin_lock(&head->batch_head->batch_lock); | |
802 | /* This batch list is already running */ | |
803 | if (!stripe_can_batch(head)) { | |
804 | spin_unlock(&head->batch_head->batch_lock); | |
805 | goto unlock_out; | |
806 | } | |
807 | ||
808 | /* | |
809 | * at this point, head's BATCH_READY could be cleared, but we | |
810 | * can still add the stripe to batch list | |
811 | */ | |
812 | list_add(&sh->batch_list, &head->batch_list); | |
813 | spin_unlock(&head->batch_head->batch_lock); | |
814 | ||
815 | sh->batch_head = head->batch_head; | |
816 | } else { | |
817 | head->batch_head = head; | |
818 | sh->batch_head = head->batch_head; | |
819 | spin_lock(&head->batch_lock); | |
820 | list_add_tail(&sh->batch_list, &head->batch_list); | |
821 | spin_unlock(&head->batch_lock); | |
822 | } | |
823 | ||
824 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
825 | if (atomic_dec_return(&conf->preread_active_stripes) | |
826 | < IO_THRESHOLD) | |
827 | md_wakeup_thread(conf->mddev->thread); | |
828 | ||
2b6b2457 N |
829 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
830 | int seq = sh->bm_seq; | |
831 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
832 | sh->batch_head->bm_seq > seq) | |
833 | seq = sh->batch_head->bm_seq; | |
834 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
835 | sh->batch_head->bm_seq = seq; | |
836 | } | |
837 | ||
59fc630b | 838 | atomic_inc(&sh->count); |
839 | unlock_out: | |
840 | unlock_two_stripes(head, sh); | |
841 | out: | |
6d036f7d | 842 | raid5_release_stripe(head); |
59fc630b | 843 | } |
844 | ||
05616be5 N |
845 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
846 | * in this stripe_head. | |
847 | */ | |
848 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
849 | { | |
850 | sector_t progress = conf->reshape_progress; | |
851 | /* Need a memory barrier to make sure we see the value | |
852 | * of conf->generation, or ->data_offset that was set before | |
853 | * reshape_progress was updated. | |
854 | */ | |
855 | smp_rmb(); | |
856 | if (progress == MaxSector) | |
857 | return 0; | |
858 | if (sh->generation == conf->generation - 1) | |
859 | return 0; | |
860 | /* We are in a reshape, and this is a new-generation stripe, | |
861 | * so use new_data_offset. | |
862 | */ | |
863 | return 1; | |
864 | } | |
865 | ||
765d704d SL |
866 | static void flush_deferred_bios(struct r5conf *conf) |
867 | { | |
868 | struct bio_list tmp; | |
869 | struct bio *bio; | |
870 | ||
871 | if (!conf->batch_bio_dispatch || !conf->group_cnt) | |
872 | return; | |
873 | ||
874 | bio_list_init(&tmp); | |
875 | spin_lock(&conf->pending_bios_lock); | |
876 | bio_list_merge(&tmp, &conf->pending_bios); | |
877 | bio_list_init(&conf->pending_bios); | |
878 | spin_unlock(&conf->pending_bios_lock); | |
879 | ||
880 | while ((bio = bio_list_pop(&tmp))) | |
881 | generic_make_request(bio); | |
882 | } | |
883 | ||
884 | static void defer_bio_issue(struct r5conf *conf, struct bio *bio) | |
885 | { | |
886 | /* | |
887 | * change group_cnt will drain all bios, so this is safe | |
888 | * | |
889 | * A read generally means a read-modify-write, which usually means a | |
890 | * randwrite, so we don't delay it | |
891 | */ | |
892 | if (!conf->batch_bio_dispatch || !conf->group_cnt || | |
893 | bio_op(bio) == REQ_OP_READ) { | |
894 | generic_make_request(bio); | |
895 | return; | |
896 | } | |
897 | spin_lock(&conf->pending_bios_lock); | |
898 | bio_list_add(&conf->pending_bios, bio); | |
899 | spin_unlock(&conf->pending_bios_lock); | |
900 | md_wakeup_thread(conf->mddev->thread); | |
901 | } | |
902 | ||
6712ecf8 | 903 | static void |
4246a0b6 | 904 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 905 | static void |
4246a0b6 | 906 | raid5_end_write_request(struct bio *bi); |
91c00924 | 907 | |
c4e5ac0a | 908 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 909 | { |
d1688a6d | 910 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 911 | int i, disks = sh->disks; |
59fc630b | 912 | struct stripe_head *head_sh = sh; |
91c00924 DW |
913 | |
914 | might_sleep(); | |
915 | ||
1e6d690b SL |
916 | if (!test_bit(STRIPE_R5C_CACHING, &sh->state)) { |
917 | /* writing out phase */ | |
d7bd398e SL |
918 | if (s->waiting_extra_page) |
919 | return; | |
1e6d690b SL |
920 | if (r5l_write_stripe(conf->log, sh) == 0) |
921 | return; | |
922 | } else { /* caching phase */ | |
923 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) { | |
924 | r5c_cache_data(conf->log, sh, s); | |
925 | return; | |
926 | } | |
927 | } | |
928 | ||
91c00924 | 929 | for (i = disks; i--; ) { |
796a5cf0 | 930 | int op, op_flags = 0; |
9a3e1101 | 931 | int replace_only = 0; |
977df362 N |
932 | struct bio *bi, *rbi; |
933 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 934 | |
935 | sh = head_sh; | |
e9c7469b | 936 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 937 | op = REQ_OP_WRITE; |
e9c7469b | 938 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 939 | op_flags = REQ_FUA; |
9e444768 | 940 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 941 | op = REQ_OP_DISCARD; |
e9c7469b | 942 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 943 | op = REQ_OP_READ; |
9a3e1101 N |
944 | else if (test_and_clear_bit(R5_WantReplace, |
945 | &sh->dev[i].flags)) { | |
796a5cf0 | 946 | op = REQ_OP_WRITE; |
9a3e1101 N |
947 | replace_only = 1; |
948 | } else | |
91c00924 | 949 | continue; |
bc0934f0 | 950 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 951 | op_flags |= REQ_SYNC; |
91c00924 | 952 | |
59fc630b | 953 | again: |
91c00924 | 954 | bi = &sh->dev[i].req; |
977df362 | 955 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 956 | |
91c00924 | 957 | rcu_read_lock(); |
9a3e1101 | 958 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
959 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
960 | rdev = rcu_dereference(conf->disks[i].rdev); | |
961 | if (!rdev) { | |
962 | rdev = rrdev; | |
963 | rrdev = NULL; | |
964 | } | |
796a5cf0 | 965 | if (op_is_write(op)) { |
9a3e1101 N |
966 | if (replace_only) |
967 | rdev = NULL; | |
dd054fce N |
968 | if (rdev == rrdev) |
969 | /* We raced and saw duplicates */ | |
970 | rrdev = NULL; | |
9a3e1101 | 971 | } else { |
59fc630b | 972 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
973 | rdev = rrdev; |
974 | rrdev = NULL; | |
975 | } | |
977df362 | 976 | |
91c00924 DW |
977 | if (rdev && test_bit(Faulty, &rdev->flags)) |
978 | rdev = NULL; | |
979 | if (rdev) | |
980 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
981 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
982 | rrdev = NULL; | |
983 | if (rrdev) | |
984 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
985 | rcu_read_unlock(); |
986 | ||
73e92e51 | 987 | /* We have already checked bad blocks for reads. Now |
977df362 N |
988 | * need to check for writes. We never accept write errors |
989 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 990 | */ |
796a5cf0 | 991 | while (op_is_write(op) && rdev && |
73e92e51 N |
992 | test_bit(WriteErrorSeen, &rdev->flags)) { |
993 | sector_t first_bad; | |
994 | int bad_sectors; | |
995 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
996 | &first_bad, &bad_sectors); | |
997 | if (!bad) | |
998 | break; | |
999 | ||
1000 | if (bad < 0) { | |
1001 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1002 | if (!conf->mddev->external && | |
2953079c | 1003 | conf->mddev->sb_flags) { |
73e92e51 N |
1004 | /* It is very unlikely, but we might |
1005 | * still need to write out the | |
1006 | * bad block log - better give it | |
1007 | * a chance*/ | |
1008 | md_check_recovery(conf->mddev); | |
1009 | } | |
1850753d | 1010 | /* |
1011 | * Because md_wait_for_blocked_rdev | |
1012 | * will dec nr_pending, we must | |
1013 | * increment it first. | |
1014 | */ | |
1015 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1016 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1017 | } else { | |
1018 | /* Acknowledged bad block - skip the write */ | |
1019 | rdev_dec_pending(rdev, conf->mddev); | |
1020 | rdev = NULL; | |
1021 | } | |
1022 | } | |
1023 | ||
91c00924 | 1024 | if (rdev) { |
9a3e1101 N |
1025 | if (s->syncing || s->expanding || s->expanded |
1026 | || s->replacing) | |
91c00924 DW |
1027 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1028 | ||
2b7497f0 DW |
1029 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1030 | ||
91c00924 | 1031 | bi->bi_bdev = rdev->bdev; |
796a5cf0 MC |
1032 | bio_set_op_attrs(bi, op, op_flags); |
1033 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1034 | ? raid5_end_write_request |
1035 | : raid5_end_read_request; | |
1036 | bi->bi_private = sh; | |
1037 | ||
6296b960 | 1038 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1039 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1040 | bi->bi_opf, i); |
91c00924 | 1041 | atomic_inc(&sh->count); |
59fc630b | 1042 | if (sh != head_sh) |
1043 | atomic_inc(&head_sh->count); | |
05616be5 | 1044 | if (use_new_offset(conf, sh)) |
4f024f37 | 1045 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1046 | + rdev->new_data_offset); |
1047 | else | |
4f024f37 | 1048 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1049 | + rdev->data_offset); |
59fc630b | 1050 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1051 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1052 | |
d592a996 SL |
1053 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1054 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1055 | |
1056 | if (!op_is_write(op) && | |
1057 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1058 | /* | |
1059 | * issuing read for a page in journal, this | |
1060 | * must be preparing for prexor in rmw; read | |
1061 | * the data into orig_page | |
1062 | */ | |
1063 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1064 | else | |
1065 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1066 | bi->bi_vcnt = 1; |
91c00924 DW |
1067 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1068 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1069 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1070 | /* |
1071 | * If this is discard request, set bi_vcnt 0. We don't | |
1072 | * want to confuse SCSI because SCSI will replace payload | |
1073 | */ | |
796a5cf0 | 1074 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1075 | bi->bi_vcnt = 0; |
977df362 N |
1076 | if (rrdev) |
1077 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1078 | |
1079 | if (conf->mddev->gendisk) | |
1080 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1081 | bi, disk_devt(conf->mddev->gendisk), | |
1082 | sh->dev[i].sector); | |
765d704d | 1083 | defer_bio_issue(conf, bi); |
977df362 N |
1084 | } |
1085 | if (rrdev) { | |
9a3e1101 N |
1086 | if (s->syncing || s->expanding || s->expanded |
1087 | || s->replacing) | |
977df362 N |
1088 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1089 | ||
1090 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1091 | ||
1092 | rbi->bi_bdev = rrdev->bdev; | |
796a5cf0 MC |
1093 | bio_set_op_attrs(rbi, op, op_flags); |
1094 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1095 | rbi->bi_end_io = raid5_end_write_request; |
1096 | rbi->bi_private = sh; | |
1097 | ||
6296b960 | 1098 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1099 | "replacement disc %d\n", |
1100 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1101 | rbi->bi_opf, i); |
977df362 | 1102 | atomic_inc(&sh->count); |
59fc630b | 1103 | if (sh != head_sh) |
1104 | atomic_inc(&head_sh->count); | |
05616be5 | 1105 | if (use_new_offset(conf, sh)) |
4f024f37 | 1106 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1107 | + rrdev->new_data_offset); |
1108 | else | |
4f024f37 | 1109 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1110 | + rrdev->data_offset); |
d592a996 SL |
1111 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1112 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1113 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1114 | rbi->bi_vcnt = 1; |
977df362 N |
1115 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1116 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1117 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1118 | /* |
1119 | * If this is discard request, set bi_vcnt 0. We don't | |
1120 | * want to confuse SCSI because SCSI will replace payload | |
1121 | */ | |
796a5cf0 | 1122 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1123 | rbi->bi_vcnt = 0; |
e3620a3a JB |
1124 | if (conf->mddev->gendisk) |
1125 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1126 | rbi, disk_devt(conf->mddev->gendisk), | |
1127 | sh->dev[i].sector); | |
765d704d | 1128 | defer_bio_issue(conf, rbi); |
977df362 N |
1129 | } |
1130 | if (!rdev && !rrdev) { | |
796a5cf0 | 1131 | if (op_is_write(op)) |
91c00924 | 1132 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1133 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1134 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1135 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1136 | set_bit(STRIPE_HANDLE, &sh->state); | |
1137 | } | |
59fc630b | 1138 | |
1139 | if (!head_sh->batch_head) | |
1140 | continue; | |
1141 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1142 | batch_list); | |
1143 | if (sh != head_sh) | |
1144 | goto again; | |
91c00924 DW |
1145 | } |
1146 | } | |
1147 | ||
1148 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1149 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1150 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1151 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1152 | { |
7988613b KO |
1153 | struct bio_vec bvl; |
1154 | struct bvec_iter iter; | |
91c00924 | 1155 | struct page *bio_page; |
91c00924 | 1156 | int page_offset; |
a08abd8c | 1157 | struct async_submit_ctl submit; |
0403e382 | 1158 | enum async_tx_flags flags = 0; |
91c00924 | 1159 | |
4f024f37 KO |
1160 | if (bio->bi_iter.bi_sector >= sector) |
1161 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1162 | else |
4f024f37 | 1163 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1164 | |
0403e382 DW |
1165 | if (frombio) |
1166 | flags |= ASYNC_TX_FENCE; | |
1167 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1168 | ||
7988613b KO |
1169 | bio_for_each_segment(bvl, bio, iter) { |
1170 | int len = bvl.bv_len; | |
91c00924 DW |
1171 | int clen; |
1172 | int b_offset = 0; | |
1173 | ||
1174 | if (page_offset < 0) { | |
1175 | b_offset = -page_offset; | |
1176 | page_offset += b_offset; | |
1177 | len -= b_offset; | |
1178 | } | |
1179 | ||
1180 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1181 | clen = STRIPE_SIZE - page_offset; | |
1182 | else | |
1183 | clen = len; | |
1184 | ||
1185 | if (clen > 0) { | |
7988613b KO |
1186 | b_offset += bvl.bv_offset; |
1187 | bio_page = bvl.bv_page; | |
d592a996 SL |
1188 | if (frombio) { |
1189 | if (sh->raid_conf->skip_copy && | |
1190 | b_offset == 0 && page_offset == 0 && | |
1e6d690b SL |
1191 | clen == STRIPE_SIZE && |
1192 | !no_skipcopy) | |
d592a996 SL |
1193 | *page = bio_page; |
1194 | else | |
1195 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1196 | b_offset, clen, &submit); |
d592a996 SL |
1197 | } else |
1198 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1199 | page_offset, clen, &submit); |
91c00924 | 1200 | } |
a08abd8c DW |
1201 | /* chain the operations */ |
1202 | submit.depend_tx = tx; | |
1203 | ||
91c00924 DW |
1204 | if (clen < len) /* hit end of page */ |
1205 | break; | |
1206 | page_offset += len; | |
1207 | } | |
1208 | ||
1209 | return tx; | |
1210 | } | |
1211 | ||
1212 | static void ops_complete_biofill(void *stripe_head_ref) | |
1213 | { | |
1214 | struct stripe_head *sh = stripe_head_ref; | |
34a6f80e | 1215 | struct bio_list return_bi = BIO_EMPTY_LIST; |
e4d84909 | 1216 | int i; |
91c00924 | 1217 | |
e46b272b | 1218 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1219 | (unsigned long long)sh->sector); |
1220 | ||
1221 | /* clear completed biofills */ | |
1222 | for (i = sh->disks; i--; ) { | |
1223 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1224 | |
1225 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1226 | /* and check if we need to reply to a read request, |
1227 | * new R5_Wantfill requests are held off until | |
83de75cc | 1228 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1229 | */ |
1230 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1231 | struct bio *rbi, *rbi2; |
91c00924 | 1232 | |
91c00924 DW |
1233 | BUG_ON(!dev->read); |
1234 | rbi = dev->read; | |
1235 | dev->read = NULL; | |
4f024f37 | 1236 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1237 | dev->sector + STRIPE_SECTORS) { |
1238 | rbi2 = r5_next_bio(rbi, dev->sector); | |
34a6f80e N |
1239 | if (!raid5_dec_bi_active_stripes(rbi)) |
1240 | bio_list_add(&return_bi, rbi); | |
91c00924 DW |
1241 | rbi = rbi2; |
1242 | } | |
1243 | } | |
1244 | } | |
83de75cc | 1245 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1246 | |
34a6f80e | 1247 | return_io(&return_bi); |
91c00924 | 1248 | |
e4d84909 | 1249 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1250 | raid5_release_stripe(sh); |
91c00924 DW |
1251 | } |
1252 | ||
1253 | static void ops_run_biofill(struct stripe_head *sh) | |
1254 | { | |
1255 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1256 | struct async_submit_ctl submit; |
91c00924 DW |
1257 | int i; |
1258 | ||
59fc630b | 1259 | BUG_ON(sh->batch_head); |
e46b272b | 1260 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1261 | (unsigned long long)sh->sector); |
1262 | ||
1263 | for (i = sh->disks; i--; ) { | |
1264 | struct r5dev *dev = &sh->dev[i]; | |
1265 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1266 | struct bio *rbi; | |
b17459c0 | 1267 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1268 | dev->read = rbi = dev->toread; |
1269 | dev->toread = NULL; | |
b17459c0 | 1270 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1271 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1272 | dev->sector + STRIPE_SECTORS) { |
d592a996 | 1273 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1274 | dev->sector, tx, sh, 0); |
91c00924 DW |
1275 | rbi = r5_next_bio(rbi, dev->sector); |
1276 | } | |
1277 | } | |
1278 | } | |
1279 | ||
1280 | atomic_inc(&sh->count); | |
a08abd8c DW |
1281 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1282 | async_trigger_callback(&submit); | |
91c00924 DW |
1283 | } |
1284 | ||
4e7d2c0a | 1285 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1286 | { |
4e7d2c0a | 1287 | struct r5dev *tgt; |
91c00924 | 1288 | |
4e7d2c0a DW |
1289 | if (target < 0) |
1290 | return; | |
91c00924 | 1291 | |
4e7d2c0a | 1292 | tgt = &sh->dev[target]; |
91c00924 DW |
1293 | set_bit(R5_UPTODATE, &tgt->flags); |
1294 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1295 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1296 | } |
1297 | ||
ac6b53b6 | 1298 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1299 | { |
1300 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1301 | |
e46b272b | 1302 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1303 | (unsigned long long)sh->sector); |
1304 | ||
ac6b53b6 | 1305 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1306 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1307 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1308 | |
ecc65c9b DW |
1309 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1310 | if (sh->check_state == check_state_compute_run) | |
1311 | sh->check_state = check_state_compute_result; | |
91c00924 | 1312 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1313 | raid5_release_stripe(sh); |
91c00924 DW |
1314 | } |
1315 | ||
d6f38f31 DW |
1316 | /* return a pointer to the address conversion region of the scribble buffer */ |
1317 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1318 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1319 | { |
46d5b785 | 1320 | void *addr; |
1321 | ||
1322 | addr = flex_array_get(percpu->scribble, i); | |
1323 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1324 | } | |
1325 | ||
1326 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1327 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1328 | { | |
1329 | void *addr; | |
1330 | ||
1331 | addr = flex_array_get(percpu->scribble, i); | |
1332 | return addr; | |
d6f38f31 DW |
1333 | } |
1334 | ||
1335 | static struct dma_async_tx_descriptor * | |
1336 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1337 | { |
91c00924 | 1338 | int disks = sh->disks; |
46d5b785 | 1339 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1340 | int target = sh->ops.target; |
1341 | struct r5dev *tgt = &sh->dev[target]; | |
1342 | struct page *xor_dest = tgt->page; | |
1343 | int count = 0; | |
1344 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1345 | struct async_submit_ctl submit; |
91c00924 DW |
1346 | int i; |
1347 | ||
59fc630b | 1348 | BUG_ON(sh->batch_head); |
1349 | ||
91c00924 | 1350 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1351 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1352 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1353 | ||
1354 | for (i = disks; i--; ) | |
1355 | if (i != target) | |
1356 | xor_srcs[count++] = sh->dev[i].page; | |
1357 | ||
1358 | atomic_inc(&sh->count); | |
1359 | ||
0403e382 | 1360 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1361 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1362 | if (unlikely(count == 1)) |
a08abd8c | 1363 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1364 | else |
a08abd8c | 1365 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1366 | |
91c00924 DW |
1367 | return tx; |
1368 | } | |
1369 | ||
ac6b53b6 DW |
1370 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1371 | * @srcs - (struct page *) array of size sh->disks | |
1372 | * @sh - stripe_head to parse | |
1373 | * | |
1374 | * Populates srcs in proper layout order for the stripe and returns the | |
1375 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1376 | * destination buffer is recorded in srcs[count] and the Q destination | |
1377 | * is recorded in srcs[count+1]]. | |
1378 | */ | |
584acdd4 MS |
1379 | static int set_syndrome_sources(struct page **srcs, |
1380 | struct stripe_head *sh, | |
1381 | int srctype) | |
ac6b53b6 DW |
1382 | { |
1383 | int disks = sh->disks; | |
1384 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1385 | int d0_idx = raid6_d0(sh); | |
1386 | int count; | |
1387 | int i; | |
1388 | ||
1389 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1390 | srcs[i] = NULL; |
ac6b53b6 DW |
1391 | |
1392 | count = 0; | |
1393 | i = d0_idx; | |
1394 | do { | |
1395 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1396 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1397 | |
584acdd4 MS |
1398 | if (i == sh->qd_idx || i == sh->pd_idx || |
1399 | (srctype == SYNDROME_SRC_ALL) || | |
1400 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1401 | (test_bit(R5_Wantdrain, &dev->flags) || |
1402 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1403 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1404 | (dev->written || |
1405 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1406 | if (test_bit(R5_InJournal, &dev->flags)) |
1407 | srcs[slot] = sh->dev[i].orig_page; | |
1408 | else | |
1409 | srcs[slot] = sh->dev[i].page; | |
1410 | } | |
ac6b53b6 DW |
1411 | i = raid6_next_disk(i, disks); |
1412 | } while (i != d0_idx); | |
ac6b53b6 | 1413 | |
e4424fee | 1414 | return syndrome_disks; |
ac6b53b6 DW |
1415 | } |
1416 | ||
1417 | static struct dma_async_tx_descriptor * | |
1418 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1419 | { | |
1420 | int disks = sh->disks; | |
46d5b785 | 1421 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1422 | int target; |
1423 | int qd_idx = sh->qd_idx; | |
1424 | struct dma_async_tx_descriptor *tx; | |
1425 | struct async_submit_ctl submit; | |
1426 | struct r5dev *tgt; | |
1427 | struct page *dest; | |
1428 | int i; | |
1429 | int count; | |
1430 | ||
59fc630b | 1431 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1432 | if (sh->ops.target < 0) |
1433 | target = sh->ops.target2; | |
1434 | else if (sh->ops.target2 < 0) | |
1435 | target = sh->ops.target; | |
91c00924 | 1436 | else |
ac6b53b6 DW |
1437 | /* we should only have one valid target */ |
1438 | BUG(); | |
1439 | BUG_ON(target < 0); | |
1440 | pr_debug("%s: stripe %llu block: %d\n", | |
1441 | __func__, (unsigned long long)sh->sector, target); | |
1442 | ||
1443 | tgt = &sh->dev[target]; | |
1444 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1445 | dest = tgt->page; | |
1446 | ||
1447 | atomic_inc(&sh->count); | |
1448 | ||
1449 | if (target == qd_idx) { | |
584acdd4 | 1450 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1451 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1452 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1453 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1454 | ops_complete_compute, sh, | |
46d5b785 | 1455 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1456 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1457 | } else { | |
1458 | /* Compute any data- or p-drive using XOR */ | |
1459 | count = 0; | |
1460 | for (i = disks; i-- ; ) { | |
1461 | if (i == target || i == qd_idx) | |
1462 | continue; | |
1463 | blocks[count++] = sh->dev[i].page; | |
1464 | } | |
1465 | ||
0403e382 DW |
1466 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1467 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1468 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1469 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1470 | } | |
91c00924 | 1471 | |
91c00924 DW |
1472 | return tx; |
1473 | } | |
1474 | ||
ac6b53b6 DW |
1475 | static struct dma_async_tx_descriptor * |
1476 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1477 | { | |
1478 | int i, count, disks = sh->disks; | |
1479 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1480 | int d0_idx = raid6_d0(sh); | |
1481 | int faila = -1, failb = -1; | |
1482 | int target = sh->ops.target; | |
1483 | int target2 = sh->ops.target2; | |
1484 | struct r5dev *tgt = &sh->dev[target]; | |
1485 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1486 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1487 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1488 | struct async_submit_ctl submit; |
1489 | ||
59fc630b | 1490 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1491 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1492 | __func__, (unsigned long long)sh->sector, target, target2); | |
1493 | BUG_ON(target < 0 || target2 < 0); | |
1494 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1495 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1496 | ||
6c910a78 | 1497 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1498 | * slot number conversion for 'faila' and 'failb' |
1499 | */ | |
1500 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1501 | blocks[i] = NULL; |
ac6b53b6 DW |
1502 | count = 0; |
1503 | i = d0_idx; | |
1504 | do { | |
1505 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1506 | ||
1507 | blocks[slot] = sh->dev[i].page; | |
1508 | ||
1509 | if (i == target) | |
1510 | faila = slot; | |
1511 | if (i == target2) | |
1512 | failb = slot; | |
1513 | i = raid6_next_disk(i, disks); | |
1514 | } while (i != d0_idx); | |
ac6b53b6 DW |
1515 | |
1516 | BUG_ON(faila == failb); | |
1517 | if (failb < faila) | |
1518 | swap(faila, failb); | |
1519 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1520 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1521 | ||
1522 | atomic_inc(&sh->count); | |
1523 | ||
1524 | if (failb == syndrome_disks+1) { | |
1525 | /* Q disk is one of the missing disks */ | |
1526 | if (faila == syndrome_disks) { | |
1527 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1528 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1529 | ops_complete_compute, sh, | |
46d5b785 | 1530 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1531 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1532 | STRIPE_SIZE, &submit); |
1533 | } else { | |
1534 | struct page *dest; | |
1535 | int data_target; | |
1536 | int qd_idx = sh->qd_idx; | |
1537 | ||
1538 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1539 | if (target == qd_idx) | |
1540 | data_target = target2; | |
1541 | else | |
1542 | data_target = target; | |
1543 | ||
1544 | count = 0; | |
1545 | for (i = disks; i-- ; ) { | |
1546 | if (i == data_target || i == qd_idx) | |
1547 | continue; | |
1548 | blocks[count++] = sh->dev[i].page; | |
1549 | } | |
1550 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1551 | init_async_submit(&submit, |
1552 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1553 | NULL, NULL, NULL, | |
46d5b785 | 1554 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1555 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1556 | &submit); | |
1557 | ||
584acdd4 | 1558 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1559 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1560 | ops_complete_compute, sh, | |
46d5b785 | 1561 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1562 | return async_gen_syndrome(blocks, 0, count+2, |
1563 | STRIPE_SIZE, &submit); | |
1564 | } | |
ac6b53b6 | 1565 | } else { |
6c910a78 DW |
1566 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1567 | ops_complete_compute, sh, | |
46d5b785 | 1568 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1569 | if (failb == syndrome_disks) { |
1570 | /* We're missing D+P. */ | |
1571 | return async_raid6_datap_recov(syndrome_disks+2, | |
1572 | STRIPE_SIZE, faila, | |
1573 | blocks, &submit); | |
1574 | } else { | |
1575 | /* We're missing D+D. */ | |
1576 | return async_raid6_2data_recov(syndrome_disks+2, | |
1577 | STRIPE_SIZE, faila, failb, | |
1578 | blocks, &submit); | |
1579 | } | |
ac6b53b6 DW |
1580 | } |
1581 | } | |
1582 | ||
91c00924 DW |
1583 | static void ops_complete_prexor(void *stripe_head_ref) |
1584 | { | |
1585 | struct stripe_head *sh = stripe_head_ref; | |
1586 | ||
e46b272b | 1587 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1588 | (unsigned long long)sh->sector); |
1e6d690b SL |
1589 | |
1590 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1591 | /* | |
1592 | * raid5-cache write back uses orig_page during prexor. | |
1593 | * After prexor, it is time to free orig_page | |
1594 | */ | |
1595 | r5c_release_extra_page(sh); | |
91c00924 DW |
1596 | } |
1597 | ||
1598 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1599 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1600 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1601 | { |
91c00924 | 1602 | int disks = sh->disks; |
46d5b785 | 1603 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1604 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1605 | struct async_submit_ctl submit; |
91c00924 DW |
1606 | |
1607 | /* existing parity data subtracted */ | |
1608 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1609 | ||
59fc630b | 1610 | BUG_ON(sh->batch_head); |
e46b272b | 1611 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1612 | (unsigned long long)sh->sector); |
1613 | ||
1614 | for (i = disks; i--; ) { | |
1615 | struct r5dev *dev = &sh->dev[i]; | |
1616 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1617 | if (test_bit(R5_InJournal, &dev->flags)) |
1618 | xor_srcs[count++] = dev->orig_page; | |
1619 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1620 | xor_srcs[count++] = dev->page; |
1621 | } | |
1622 | ||
0403e382 | 1623 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1624 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1625 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1626 | |
1627 | return tx; | |
1628 | } | |
1629 | ||
584acdd4 MS |
1630 | static struct dma_async_tx_descriptor * |
1631 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1632 | struct dma_async_tx_descriptor *tx) | |
1633 | { | |
1634 | struct page **blocks = to_addr_page(percpu, 0); | |
1635 | int count; | |
1636 | struct async_submit_ctl submit; | |
1637 | ||
1638 | pr_debug("%s: stripe %llu\n", __func__, | |
1639 | (unsigned long long)sh->sector); | |
1640 | ||
1641 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1642 | ||
1643 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1644 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1645 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1646 | ||
1647 | return tx; | |
1648 | } | |
1649 | ||
91c00924 | 1650 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1651 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1652 | { |
1e6d690b | 1653 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1654 | int disks = sh->disks; |
d8ee0728 | 1655 | int i; |
59fc630b | 1656 | struct stripe_head *head_sh = sh; |
91c00924 | 1657 | |
e46b272b | 1658 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1659 | (unsigned long long)sh->sector); |
1660 | ||
1661 | for (i = disks; i--; ) { | |
59fc630b | 1662 | struct r5dev *dev; |
91c00924 | 1663 | struct bio *chosen; |
91c00924 | 1664 | |
59fc630b | 1665 | sh = head_sh; |
1666 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1667 | struct bio *wbi; |
1668 | ||
59fc630b | 1669 | again: |
1670 | dev = &sh->dev[i]; | |
1e6d690b SL |
1671 | /* |
1672 | * clear R5_InJournal, so when rewriting a page in | |
1673 | * journal, it is not skipped by r5l_log_stripe() | |
1674 | */ | |
1675 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1676 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1677 | chosen = dev->towrite; |
1678 | dev->towrite = NULL; | |
7a87f434 | 1679 | sh->overwrite_disks = 0; |
91c00924 DW |
1680 | BUG_ON(dev->written); |
1681 | wbi = dev->written = chosen; | |
b17459c0 | 1682 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1683 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1684 | |
4f024f37 | 1685 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1686 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1687 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1688 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1689 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1690 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1691 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1692 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1693 | else { |
1694 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1695 | dev->sector, tx, sh, |
1696 | r5c_is_writeback(conf->log)); | |
1697 | if (dev->page != dev->orig_page && | |
1698 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1699 | set_bit(R5_SkipCopy, &dev->flags); |
1700 | clear_bit(R5_UPTODATE, &dev->flags); | |
1701 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1702 | } | |
1703 | } | |
91c00924 DW |
1704 | wbi = r5_next_bio(wbi, dev->sector); |
1705 | } | |
59fc630b | 1706 | |
1707 | if (head_sh->batch_head) { | |
1708 | sh = list_first_entry(&sh->batch_list, | |
1709 | struct stripe_head, | |
1710 | batch_list); | |
1711 | if (sh == head_sh) | |
1712 | continue; | |
1713 | goto again; | |
1714 | } | |
91c00924 DW |
1715 | } |
1716 | } | |
1717 | ||
1718 | return tx; | |
1719 | } | |
1720 | ||
ac6b53b6 | 1721 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1722 | { |
1723 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1724 | int disks = sh->disks; |
1725 | int pd_idx = sh->pd_idx; | |
1726 | int qd_idx = sh->qd_idx; | |
1727 | int i; | |
9e444768 | 1728 | bool fua = false, sync = false, discard = false; |
91c00924 | 1729 | |
e46b272b | 1730 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1731 | (unsigned long long)sh->sector); |
1732 | ||
bc0934f0 | 1733 | for (i = disks; i--; ) { |
e9c7469b | 1734 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1735 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1736 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1737 | } |
e9c7469b | 1738 | |
91c00924 DW |
1739 | for (i = disks; i--; ) { |
1740 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1741 | |
e9c7469b | 1742 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1743 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1744 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1745 | if (fua) |
1746 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1747 | if (sync) |
1748 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1749 | } |
91c00924 DW |
1750 | } |
1751 | ||
d8ee0728 DW |
1752 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1753 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1754 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1755 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1756 | else { | |
1757 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1758 | sh->reconstruct_state = reconstruct_state_result; | |
1759 | } | |
91c00924 DW |
1760 | |
1761 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1762 | raid5_release_stripe(sh); |
91c00924 DW |
1763 | } |
1764 | ||
1765 | static void | |
ac6b53b6 DW |
1766 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1767 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1768 | { |
91c00924 | 1769 | int disks = sh->disks; |
59fc630b | 1770 | struct page **xor_srcs; |
a08abd8c | 1771 | struct async_submit_ctl submit; |
59fc630b | 1772 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1773 | struct page *xor_dest; |
d8ee0728 | 1774 | int prexor = 0; |
91c00924 | 1775 | unsigned long flags; |
59fc630b | 1776 | int j = 0; |
1777 | struct stripe_head *head_sh = sh; | |
1778 | int last_stripe; | |
91c00924 | 1779 | |
e46b272b | 1780 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1781 | (unsigned long long)sh->sector); |
1782 | ||
620125f2 SL |
1783 | for (i = 0; i < sh->disks; i++) { |
1784 | if (pd_idx == i) | |
1785 | continue; | |
1786 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1787 | break; | |
1788 | } | |
1789 | if (i >= sh->disks) { | |
1790 | atomic_inc(&sh->count); | |
620125f2 SL |
1791 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1792 | ops_complete_reconstruct(sh); | |
1793 | return; | |
1794 | } | |
59fc630b | 1795 | again: |
1796 | count = 0; | |
1797 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1798 | /* check if prexor is active which means only process blocks |
1799 | * that are part of a read-modify-write (written) | |
1800 | */ | |
59fc630b | 1801 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1802 | prexor = 1; |
91c00924 DW |
1803 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1804 | for (i = disks; i--; ) { | |
1805 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1806 | if (head_sh->dev[i].written || |
1807 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1808 | xor_srcs[count++] = dev->page; |
1809 | } | |
1810 | } else { | |
1811 | xor_dest = sh->dev[pd_idx].page; | |
1812 | for (i = disks; i--; ) { | |
1813 | struct r5dev *dev = &sh->dev[i]; | |
1814 | if (i != pd_idx) | |
1815 | xor_srcs[count++] = dev->page; | |
1816 | } | |
1817 | } | |
1818 | ||
91c00924 DW |
1819 | /* 1/ if we prexor'd then the dest is reused as a source |
1820 | * 2/ if we did not prexor then we are redoing the parity | |
1821 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1822 | * for the synchronous xor case | |
1823 | */ | |
59fc630b | 1824 | last_stripe = !head_sh->batch_head || |
1825 | list_first_entry(&sh->batch_list, | |
1826 | struct stripe_head, batch_list) == head_sh; | |
1827 | if (last_stripe) { | |
1828 | flags = ASYNC_TX_ACK | | |
1829 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1830 | ||
1831 | atomic_inc(&head_sh->count); | |
1832 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1833 | to_addr_conv(sh, percpu, j)); | |
1834 | } else { | |
1835 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1836 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1837 | to_addr_conv(sh, percpu, j)); | |
1838 | } | |
91c00924 | 1839 | |
a08abd8c DW |
1840 | if (unlikely(count == 1)) |
1841 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1842 | else | |
1843 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1844 | if (!last_stripe) { |
1845 | j++; | |
1846 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1847 | batch_list); | |
1848 | goto again; | |
1849 | } | |
91c00924 DW |
1850 | } |
1851 | ||
ac6b53b6 DW |
1852 | static void |
1853 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1854 | struct dma_async_tx_descriptor *tx) | |
1855 | { | |
1856 | struct async_submit_ctl submit; | |
59fc630b | 1857 | struct page **blocks; |
1858 | int count, i, j = 0; | |
1859 | struct stripe_head *head_sh = sh; | |
1860 | int last_stripe; | |
584acdd4 MS |
1861 | int synflags; |
1862 | unsigned long txflags; | |
ac6b53b6 DW |
1863 | |
1864 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1865 | ||
620125f2 SL |
1866 | for (i = 0; i < sh->disks; i++) { |
1867 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1868 | continue; | |
1869 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1870 | break; | |
1871 | } | |
1872 | if (i >= sh->disks) { | |
1873 | atomic_inc(&sh->count); | |
620125f2 SL |
1874 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1875 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1876 | ops_complete_reconstruct(sh); | |
1877 | return; | |
1878 | } | |
1879 | ||
59fc630b | 1880 | again: |
1881 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1882 | |
1883 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1884 | synflags = SYNDROME_SRC_WRITTEN; | |
1885 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1886 | } else { | |
1887 | synflags = SYNDROME_SRC_ALL; | |
1888 | txflags = ASYNC_TX_ACK; | |
1889 | } | |
1890 | ||
1891 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1892 | last_stripe = !head_sh->batch_head || |
1893 | list_first_entry(&sh->batch_list, | |
1894 | struct stripe_head, batch_list) == head_sh; | |
1895 | ||
1896 | if (last_stripe) { | |
1897 | atomic_inc(&head_sh->count); | |
584acdd4 | 1898 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1899 | head_sh, to_addr_conv(sh, percpu, j)); |
1900 | } else | |
1901 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1902 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1903 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1904 | if (!last_stripe) { |
1905 | j++; | |
1906 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1907 | batch_list); | |
1908 | goto again; | |
1909 | } | |
91c00924 DW |
1910 | } |
1911 | ||
1912 | static void ops_complete_check(void *stripe_head_ref) | |
1913 | { | |
1914 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1915 | |
e46b272b | 1916 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1917 | (unsigned long long)sh->sector); |
1918 | ||
ecc65c9b | 1919 | sh->check_state = check_state_check_result; |
91c00924 | 1920 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1921 | raid5_release_stripe(sh); |
91c00924 DW |
1922 | } |
1923 | ||
ac6b53b6 | 1924 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1925 | { |
91c00924 | 1926 | int disks = sh->disks; |
ac6b53b6 DW |
1927 | int pd_idx = sh->pd_idx; |
1928 | int qd_idx = sh->qd_idx; | |
1929 | struct page *xor_dest; | |
46d5b785 | 1930 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1931 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1932 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1933 | int count; |
1934 | int i; | |
91c00924 | 1935 | |
e46b272b | 1936 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1937 | (unsigned long long)sh->sector); |
1938 | ||
59fc630b | 1939 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1940 | count = 0; |
1941 | xor_dest = sh->dev[pd_idx].page; | |
1942 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1943 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1944 | if (i == pd_idx || i == qd_idx) |
1945 | continue; | |
1946 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1947 | } |
1948 | ||
d6f38f31 | 1949 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1950 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1951 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1952 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1953 | |
91c00924 | 1954 | atomic_inc(&sh->count); |
a08abd8c DW |
1955 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1956 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1957 | } |
1958 | ||
ac6b53b6 DW |
1959 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1960 | { | |
46d5b785 | 1961 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1962 | struct async_submit_ctl submit; |
1963 | int count; | |
1964 | ||
1965 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1966 | (unsigned long long)sh->sector, checkp); | |
1967 | ||
59fc630b | 1968 | BUG_ON(sh->batch_head); |
584acdd4 | 1969 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1970 | if (!checkp) |
1971 | srcs[count] = NULL; | |
91c00924 | 1972 | |
91c00924 | 1973 | atomic_inc(&sh->count); |
ac6b53b6 | 1974 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1975 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1976 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1977 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1978 | } |
1979 | ||
51acbcec | 1980 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1981 | { |
1982 | int overlap_clear = 0, i, disks = sh->disks; | |
1983 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1984 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 1985 | int level = conf->level; |
d6f38f31 DW |
1986 | struct raid5_percpu *percpu; |
1987 | unsigned long cpu; | |
91c00924 | 1988 | |
d6f38f31 DW |
1989 | cpu = get_cpu(); |
1990 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 1991 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
1992 | ops_run_biofill(sh); |
1993 | overlap_clear++; | |
1994 | } | |
1995 | ||
7b3a871e | 1996 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
1997 | if (level < 6) |
1998 | tx = ops_run_compute5(sh, percpu); | |
1999 | else { | |
2000 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2001 | tx = ops_run_compute6_1(sh, percpu); | |
2002 | else | |
2003 | tx = ops_run_compute6_2(sh, percpu); | |
2004 | } | |
2005 | /* terminate the chain if reconstruct is not set to be run */ | |
2006 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2007 | async_tx_ack(tx); |
2008 | } | |
91c00924 | 2009 | |
584acdd4 MS |
2010 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2011 | if (level < 6) | |
2012 | tx = ops_run_prexor5(sh, percpu, tx); | |
2013 | else | |
2014 | tx = ops_run_prexor6(sh, percpu, tx); | |
2015 | } | |
91c00924 | 2016 | |
600aa109 | 2017 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2018 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2019 | overlap_clear++; |
2020 | } | |
2021 | ||
ac6b53b6 DW |
2022 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2023 | if (level < 6) | |
2024 | ops_run_reconstruct5(sh, percpu, tx); | |
2025 | else | |
2026 | ops_run_reconstruct6(sh, percpu, tx); | |
2027 | } | |
91c00924 | 2028 | |
ac6b53b6 DW |
2029 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2030 | if (sh->check_state == check_state_run) | |
2031 | ops_run_check_p(sh, percpu); | |
2032 | else if (sh->check_state == check_state_run_q) | |
2033 | ops_run_check_pq(sh, percpu, 0); | |
2034 | else if (sh->check_state == check_state_run_pq) | |
2035 | ops_run_check_pq(sh, percpu, 1); | |
2036 | else | |
2037 | BUG(); | |
2038 | } | |
91c00924 | 2039 | |
59fc630b | 2040 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2041 | for (i = disks; i--; ) { |
2042 | struct r5dev *dev = &sh->dev[i]; | |
2043 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2044 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2045 | } | |
d6f38f31 | 2046 | put_cpu(); |
91c00924 DW |
2047 | } |
2048 | ||
5f9d1fde SL |
2049 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
2050 | int disks) | |
f18c1a35 N |
2051 | { |
2052 | struct stripe_head *sh; | |
5f9d1fde | 2053 | int i; |
f18c1a35 N |
2054 | |
2055 | sh = kmem_cache_zalloc(sc, gfp); | |
2056 | if (sh) { | |
2057 | spin_lock_init(&sh->stripe_lock); | |
2058 | spin_lock_init(&sh->batch_lock); | |
2059 | INIT_LIST_HEAD(&sh->batch_list); | |
2060 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2061 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2062 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2063 | atomic_set(&sh->count, 1); |
a39f7afd | 2064 | sh->log_start = MaxSector; |
5f9d1fde SL |
2065 | for (i = 0; i < disks; i++) { |
2066 | struct r5dev *dev = &sh->dev[i]; | |
2067 | ||
3a83f467 ML |
2068 | bio_init(&dev->req, &dev->vec, 1); |
2069 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2070 | } |
f18c1a35 N |
2071 | } |
2072 | return sh; | |
2073 | } | |
486f0644 | 2074 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2075 | { |
2076 | struct stripe_head *sh; | |
f18c1a35 | 2077 | |
5f9d1fde | 2078 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size); |
3f294f4f N |
2079 | if (!sh) |
2080 | return 0; | |
6ce32846 | 2081 | |
3f294f4f | 2082 | sh->raid_conf = conf; |
3f294f4f | 2083 | |
a9683a79 | 2084 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2085 | shrink_buffers(sh); |
3f294f4f N |
2086 | kmem_cache_free(conf->slab_cache, sh); |
2087 | return 0; | |
2088 | } | |
486f0644 N |
2089 | sh->hash_lock_index = |
2090 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2091 | /* we just created an active stripe so... */ |
3f294f4f | 2092 | atomic_inc(&conf->active_stripes); |
59fc630b | 2093 | |
6d036f7d | 2094 | raid5_release_stripe(sh); |
486f0644 | 2095 | conf->max_nr_stripes++; |
3f294f4f N |
2096 | return 1; |
2097 | } | |
2098 | ||
d1688a6d | 2099 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2100 | { |
e18b890b | 2101 | struct kmem_cache *sc; |
5e5e3e78 | 2102 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2103 | |
f4be6b43 N |
2104 | if (conf->mddev->gendisk) |
2105 | sprintf(conf->cache_name[0], | |
2106 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2107 | else | |
2108 | sprintf(conf->cache_name[0], | |
2109 | "raid%d-%p", conf->level, conf->mddev); | |
2110 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2111 | ||
ad01c9e3 N |
2112 | conf->active_name = 0; |
2113 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2114 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2115 | 0, 0, NULL); |
1da177e4 LT |
2116 | if (!sc) |
2117 | return 1; | |
2118 | conf->slab_cache = sc; | |
ad01c9e3 | 2119 | conf->pool_size = devs; |
486f0644 N |
2120 | while (num--) |
2121 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2122 | return 1; |
486f0644 | 2123 | |
1da177e4 LT |
2124 | return 0; |
2125 | } | |
29269553 | 2126 | |
d6f38f31 DW |
2127 | /** |
2128 | * scribble_len - return the required size of the scribble region | |
2129 | * @num - total number of disks in the array | |
2130 | * | |
2131 | * The size must be enough to contain: | |
2132 | * 1/ a struct page pointer for each device in the array +2 | |
2133 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2134 | * (dma_map_page()) or page (page_address()) address. | |
2135 | * | |
2136 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2137 | * calculate over all devices (not just the data blocks), using zeros in place | |
2138 | * of the P and Q blocks. | |
2139 | */ | |
46d5b785 | 2140 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2141 | { |
46d5b785 | 2142 | struct flex_array *ret; |
d6f38f31 DW |
2143 | size_t len; |
2144 | ||
2145 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2146 | ret = flex_array_alloc(len, cnt, flags); |
2147 | if (!ret) | |
2148 | return NULL; | |
2149 | /* always prealloc all elements, so no locking is required */ | |
2150 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2151 | flex_array_free(ret); | |
2152 | return NULL; | |
2153 | } | |
2154 | return ret; | |
d6f38f31 DW |
2155 | } |
2156 | ||
738a2738 N |
2157 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2158 | { | |
2159 | unsigned long cpu; | |
2160 | int err = 0; | |
2161 | ||
27a353c0 SL |
2162 | /* |
2163 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2164 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2165 | * should equal to new_disks and new_sectors | |
2166 | */ | |
2167 | if (conf->scribble_disks >= new_disks && | |
2168 | conf->scribble_sectors >= new_sectors) | |
2169 | return 0; | |
738a2738 N |
2170 | mddev_suspend(conf->mddev); |
2171 | get_online_cpus(); | |
2172 | for_each_present_cpu(cpu) { | |
2173 | struct raid5_percpu *percpu; | |
2174 | struct flex_array *scribble; | |
2175 | ||
2176 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2177 | scribble = scribble_alloc(new_disks, | |
2178 | new_sectors / STRIPE_SECTORS, | |
2179 | GFP_NOIO); | |
2180 | ||
2181 | if (scribble) { | |
2182 | flex_array_free(percpu->scribble); | |
2183 | percpu->scribble = scribble; | |
2184 | } else { | |
2185 | err = -ENOMEM; | |
2186 | break; | |
2187 | } | |
2188 | } | |
2189 | put_online_cpus(); | |
2190 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2191 | if (!err) { |
2192 | conf->scribble_disks = new_disks; | |
2193 | conf->scribble_sectors = new_sectors; | |
2194 | } | |
738a2738 N |
2195 | return err; |
2196 | } | |
2197 | ||
d1688a6d | 2198 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2199 | { |
2200 | /* Make all the stripes able to hold 'newsize' devices. | |
2201 | * New slots in each stripe get 'page' set to a new page. | |
2202 | * | |
2203 | * This happens in stages: | |
2204 | * 1/ create a new kmem_cache and allocate the required number of | |
2205 | * stripe_heads. | |
83f0d77a | 2206 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2207 | * to the new stripe_heads. This will have the side effect of |
2208 | * freezing the array as once all stripe_heads have been collected, | |
2209 | * no IO will be possible. Old stripe heads are freed once their | |
2210 | * pages have been transferred over, and the old kmem_cache is | |
2211 | * freed when all stripes are done. | |
2212 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2213 | * we simple return a failre status - no need to clean anything up. | |
2214 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2215 | * If this fails, we don't bother trying the shrink the | |
2216 | * stripe_heads down again, we just leave them as they are. | |
2217 | * As each stripe_head is processed the new one is released into | |
2218 | * active service. | |
2219 | * | |
2220 | * Once step2 is started, we cannot afford to wait for a write, | |
2221 | * so we use GFP_NOIO allocations. | |
2222 | */ | |
2223 | struct stripe_head *osh, *nsh; | |
2224 | LIST_HEAD(newstripes); | |
2225 | struct disk_info *ndisks; | |
b5470dc5 | 2226 | int err; |
e18b890b | 2227 | struct kmem_cache *sc; |
ad01c9e3 | 2228 | int i; |
566c09c5 | 2229 | int hash, cnt; |
ad01c9e3 N |
2230 | |
2231 | if (newsize <= conf->pool_size) | |
2232 | return 0; /* never bother to shrink */ | |
2233 | ||
b5470dc5 DW |
2234 | err = md_allow_write(conf->mddev); |
2235 | if (err) | |
2236 | return err; | |
2a2275d6 | 2237 | |
ad01c9e3 N |
2238 | /* Step 1 */ |
2239 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2240 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2241 | 0, 0, NULL); |
ad01c9e3 N |
2242 | if (!sc) |
2243 | return -ENOMEM; | |
2244 | ||
2d5b569b N |
2245 | /* Need to ensure auto-resizing doesn't interfere */ |
2246 | mutex_lock(&conf->cache_size_mutex); | |
2247 | ||
ad01c9e3 | 2248 | for (i = conf->max_nr_stripes; i; i--) { |
5f9d1fde | 2249 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize); |
ad01c9e3 N |
2250 | if (!nsh) |
2251 | break; | |
2252 | ||
ad01c9e3 | 2253 | nsh->raid_conf = conf; |
ad01c9e3 N |
2254 | list_add(&nsh->lru, &newstripes); |
2255 | } | |
2256 | if (i) { | |
2257 | /* didn't get enough, give up */ | |
2258 | while (!list_empty(&newstripes)) { | |
2259 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2260 | list_del(&nsh->lru); | |
2261 | kmem_cache_free(sc, nsh); | |
2262 | } | |
2263 | kmem_cache_destroy(sc); | |
2d5b569b | 2264 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2265 | return -ENOMEM; |
2266 | } | |
2267 | /* Step 2 - Must use GFP_NOIO now. | |
2268 | * OK, we have enough stripes, start collecting inactive | |
2269 | * stripes and copying them over | |
2270 | */ | |
566c09c5 SL |
2271 | hash = 0; |
2272 | cnt = 0; | |
ad01c9e3 | 2273 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2274 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2275 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2276 | !list_empty(conf->inactive_list + hash), |
2277 | unlock_device_hash_lock(conf, hash), | |
2278 | lock_device_hash_lock(conf, hash)); | |
2279 | osh = get_free_stripe(conf, hash); | |
2280 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2281 | |
d592a996 | 2282 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2283 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2284 | nsh->dev[i].orig_page = osh->dev[i].page; |
2285 | } | |
566c09c5 | 2286 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2287 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2288 | cnt++; |
2289 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2290 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2291 | hash++; | |
2292 | cnt = 0; | |
2293 | } | |
ad01c9e3 N |
2294 | } |
2295 | kmem_cache_destroy(conf->slab_cache); | |
2296 | ||
2297 | /* Step 3. | |
2298 | * At this point, we are holding all the stripes so the array | |
2299 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2300 | * conf->disks and the scribble region |
ad01c9e3 N |
2301 | */ |
2302 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2303 | if (ndisks) { | |
d7bd398e | 2304 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2305 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2306 | |
2307 | for (i = conf->pool_size; i < newsize; i++) { | |
2308 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2309 | if (!ndisks[i].extra_page) | |
2310 | err = -ENOMEM; | |
2311 | } | |
2312 | ||
2313 | if (err) { | |
2314 | for (i = conf->pool_size; i < newsize; i++) | |
2315 | if (ndisks[i].extra_page) | |
2316 | put_page(ndisks[i].extra_page); | |
2317 | kfree(ndisks); | |
2318 | } else { | |
2319 | kfree(conf->disks); | |
2320 | conf->disks = ndisks; | |
2321 | } | |
ad01c9e3 N |
2322 | } else |
2323 | err = -ENOMEM; | |
2324 | ||
2d5b569b | 2325 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2326 | /* Step 4, return new stripes to service */ |
2327 | while(!list_empty(&newstripes)) { | |
2328 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2329 | list_del_init(&nsh->lru); | |
d6f38f31 | 2330 | |
ad01c9e3 N |
2331 | for (i=conf->raid_disks; i < newsize; i++) |
2332 | if (nsh->dev[i].page == NULL) { | |
2333 | struct page *p = alloc_page(GFP_NOIO); | |
2334 | nsh->dev[i].page = p; | |
d592a996 | 2335 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2336 | if (!p) |
2337 | err = -ENOMEM; | |
2338 | } | |
6d036f7d | 2339 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2340 | } |
2341 | /* critical section pass, GFP_NOIO no longer needed */ | |
2342 | ||
2343 | conf->slab_cache = sc; | |
2344 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2345 | if (!err) |
2346 | conf->pool_size = newsize; | |
ad01c9e3 N |
2347 | return err; |
2348 | } | |
1da177e4 | 2349 | |
486f0644 | 2350 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2351 | { |
2352 | struct stripe_head *sh; | |
49895bcc | 2353 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2354 | |
566c09c5 SL |
2355 | spin_lock_irq(conf->hash_locks + hash); |
2356 | sh = get_free_stripe(conf, hash); | |
2357 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2358 | if (!sh) |
2359 | return 0; | |
78bafebd | 2360 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2361 | shrink_buffers(sh); |
3f294f4f N |
2362 | kmem_cache_free(conf->slab_cache, sh); |
2363 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2364 | conf->max_nr_stripes--; |
3f294f4f N |
2365 | return 1; |
2366 | } | |
2367 | ||
d1688a6d | 2368 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2369 | { |
486f0644 N |
2370 | while (conf->max_nr_stripes && |
2371 | drop_one_stripe(conf)) | |
2372 | ; | |
3f294f4f | 2373 | |
644df1a8 | 2374 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2375 | conf->slab_cache = NULL; |
2376 | } | |
2377 | ||
4246a0b6 | 2378 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2379 | { |
99c0fb5f | 2380 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2381 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2382 | int disks = sh->disks, i; |
d6950432 | 2383 | char b[BDEVNAME_SIZE]; |
dd054fce | 2384 | struct md_rdev *rdev = NULL; |
05616be5 | 2385 | sector_t s; |
1da177e4 LT |
2386 | |
2387 | for (i=0 ; i<disks; i++) | |
2388 | if (bi == &sh->dev[i].req) | |
2389 | break; | |
2390 | ||
4246a0b6 | 2391 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2392 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2393 | bi->bi_error); |
1da177e4 | 2394 | if (i == disks) { |
5f9d1fde | 2395 | bio_reset(bi); |
1da177e4 | 2396 | BUG(); |
6712ecf8 | 2397 | return; |
1da177e4 | 2398 | } |
14a75d3e | 2399 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2400 | /* If replacement finished while this request was outstanding, |
2401 | * 'replacement' might be NULL already. | |
2402 | * In that case it moved down to 'rdev'. | |
2403 | * rdev is not removed until all requests are finished. | |
2404 | */ | |
14a75d3e | 2405 | rdev = conf->disks[i].replacement; |
dd054fce | 2406 | if (!rdev) |
14a75d3e | 2407 | rdev = conf->disks[i].rdev; |
1da177e4 | 2408 | |
05616be5 N |
2409 | if (use_new_offset(conf, sh)) |
2410 | s = sh->sector + rdev->new_data_offset; | |
2411 | else | |
2412 | s = sh->sector + rdev->data_offset; | |
4246a0b6 | 2413 | if (!bi->bi_error) { |
1da177e4 | 2414 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2415 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2416 | /* Note that this cannot happen on a |
2417 | * replacement device. We just fail those on | |
2418 | * any error | |
2419 | */ | |
cc6167b4 N |
2420 | pr_info_ratelimited( |
2421 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
8bda470e | 2422 | mdname(conf->mddev), STRIPE_SECTORS, |
05616be5 | 2423 | (unsigned long long)s, |
8bda470e | 2424 | bdevname(rdev->bdev, b)); |
ddd5115f | 2425 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2426 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2427 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2428 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2429 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2430 | ||
86aa1397 SL |
2431 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2432 | /* | |
2433 | * end read for a page in journal, this | |
2434 | * must be preparing for prexor in rmw | |
2435 | */ | |
2436 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2437 | ||
14a75d3e N |
2438 | if (atomic_read(&rdev->read_errors)) |
2439 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2440 | } else { |
14a75d3e | 2441 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2442 | int retry = 0; |
2e8ac303 | 2443 | int set_bad = 0; |
d6950432 | 2444 | |
1da177e4 | 2445 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2446 | atomic_inc(&rdev->read_errors); |
14a75d3e | 2447 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2448 | pr_warn_ratelimited( |
2449 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2450 | mdname(conf->mddev), |
05616be5 | 2451 | (unsigned long long)s, |
14a75d3e | 2452 | bdn); |
2e8ac303 | 2453 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2454 | set_bad = 1; | |
cc6167b4 N |
2455 | pr_warn_ratelimited( |
2456 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2457 | mdname(conf->mddev), |
05616be5 | 2458 | (unsigned long long)s, |
8bda470e | 2459 | bdn); |
2e8ac303 | 2460 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2461 | /* Oh, no!!! */ |
2e8ac303 | 2462 | set_bad = 1; |
cc6167b4 N |
2463 | pr_warn_ratelimited( |
2464 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2465 | mdname(conf->mddev), |
05616be5 | 2466 | (unsigned long long)s, |
8bda470e | 2467 | bdn); |
2e8ac303 | 2468 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2469 | > conf->max_nr_stripes) |
cc6167b4 | 2470 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2471 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2472 | else |
2473 | retry = 1; | |
edfa1f65 BY |
2474 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2475 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2476 | retry = 1; | |
ba22dcbf | 2477 | if (retry) |
3f9e7c14 | 2478 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2479 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2480 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2481 | } else | |
2482 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2483 | else { |
4e5314b5 N |
2484 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2485 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2486 | if (!(set_bad |
2487 | && test_bit(In_sync, &rdev->flags) | |
2488 | && rdev_set_badblocks( | |
2489 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2490 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2491 | } |
1da177e4 | 2492 | } |
14a75d3e | 2493 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2494 | bio_reset(bi); |
1da177e4 LT |
2495 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2496 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2497 | raid5_release_stripe(sh); |
1da177e4 LT |
2498 | } |
2499 | ||
4246a0b6 | 2500 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2501 | { |
99c0fb5f | 2502 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2503 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2504 | int disks = sh->disks, i; |
977df362 | 2505 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2506 | sector_t first_bad; |
2507 | int bad_sectors; | |
977df362 | 2508 | int replacement = 0; |
1da177e4 | 2509 | |
977df362 N |
2510 | for (i = 0 ; i < disks; i++) { |
2511 | if (bi == &sh->dev[i].req) { | |
2512 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2513 | break; |
977df362 N |
2514 | } |
2515 | if (bi == &sh->dev[i].rreq) { | |
2516 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2517 | if (rdev) |
2518 | replacement = 1; | |
2519 | else | |
2520 | /* rdev was removed and 'replacement' | |
2521 | * replaced it. rdev is not removed | |
2522 | * until all requests are finished. | |
2523 | */ | |
2524 | rdev = conf->disks[i].rdev; | |
977df362 N |
2525 | break; |
2526 | } | |
2527 | } | |
4246a0b6 | 2528 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2529 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2530 | bi->bi_error); |
1da177e4 | 2531 | if (i == disks) { |
5f9d1fde | 2532 | bio_reset(bi); |
1da177e4 | 2533 | BUG(); |
6712ecf8 | 2534 | return; |
1da177e4 LT |
2535 | } |
2536 | ||
977df362 | 2537 | if (replacement) { |
4246a0b6 | 2538 | if (bi->bi_error) |
977df362 N |
2539 | md_error(conf->mddev, rdev); |
2540 | else if (is_badblock(rdev, sh->sector, | |
2541 | STRIPE_SECTORS, | |
2542 | &first_bad, &bad_sectors)) | |
2543 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2544 | } else { | |
4246a0b6 | 2545 | if (bi->bi_error) { |
9f97e4b1 | 2546 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2547 | set_bit(WriteErrorSeen, &rdev->flags); |
2548 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2549 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2550 | set_bit(MD_RECOVERY_NEEDED, | |
2551 | &rdev->mddev->recovery); | |
977df362 N |
2552 | } else if (is_badblock(rdev, sh->sector, |
2553 | STRIPE_SECTORS, | |
c0b32972 | 2554 | &first_bad, &bad_sectors)) { |
977df362 | 2555 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2556 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2557 | /* That was a successful write so make | |
2558 | * sure it looks like we already did | |
2559 | * a re-write. | |
2560 | */ | |
2561 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2562 | } | |
977df362 N |
2563 | } |
2564 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2565 | |
4246a0b6 | 2566 | if (sh->batch_head && bi->bi_error && !replacement) |
72ac7330 | 2567 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2568 | ||
c9445555 | 2569 | bio_reset(bi); |
977df362 N |
2570 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2571 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2572 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2573 | raid5_release_stripe(sh); |
59fc630b | 2574 | |
2575 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2576 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2577 | } |
2578 | ||
784052ec | 2579 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2580 | { |
2581 | struct r5dev *dev = &sh->dev[i]; | |
2582 | ||
1da177e4 | 2583 | dev->flags = 0; |
6d036f7d | 2584 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 LT |
2585 | } |
2586 | ||
849674e4 | 2587 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2588 | { |
2589 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2590 | struct r5conf *conf = mddev->private; |
908f4fbd | 2591 | unsigned long flags; |
0c55e022 | 2592 | pr_debug("raid456: error called\n"); |
1da177e4 | 2593 | |
908f4fbd N |
2594 | spin_lock_irqsave(&conf->device_lock, flags); |
2595 | clear_bit(In_sync, &rdev->flags); | |
2e38a37f | 2596 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2597 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2598 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2599 | ||
de393cde | 2600 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 | 2601 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
2602 | set_mask_bits(&mddev->sb_flags, 0, |
2603 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2604 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2605 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2606 | mdname(mddev), | |
2607 | bdevname(rdev->bdev, b), | |
2608 | mdname(mddev), | |
2609 | conf->raid_disks - mddev->degraded); | |
2e38a37f | 2610 | r5c_update_on_rdev_error(mddev); |
16a53ecc | 2611 | } |
1da177e4 LT |
2612 | |
2613 | /* | |
2614 | * Input: a 'big' sector number, | |
2615 | * Output: index of the data and parity disk, and the sector # in them. | |
2616 | */ | |
6d036f7d SL |
2617 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2618 | int previous, int *dd_idx, | |
2619 | struct stripe_head *sh) | |
1da177e4 | 2620 | { |
6e3b96ed | 2621 | sector_t stripe, stripe2; |
35f2a591 | 2622 | sector_t chunk_number; |
1da177e4 | 2623 | unsigned int chunk_offset; |
911d4ee8 | 2624 | int pd_idx, qd_idx; |
67cc2b81 | 2625 | int ddf_layout = 0; |
1da177e4 | 2626 | sector_t new_sector; |
e183eaed N |
2627 | int algorithm = previous ? conf->prev_algo |
2628 | : conf->algorithm; | |
09c9e5fa AN |
2629 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2630 | : conf->chunk_sectors; | |
112bf897 N |
2631 | int raid_disks = previous ? conf->previous_raid_disks |
2632 | : conf->raid_disks; | |
2633 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2634 | |
2635 | /* First compute the information on this sector */ | |
2636 | ||
2637 | /* | |
2638 | * Compute the chunk number and the sector offset inside the chunk | |
2639 | */ | |
2640 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2641 | chunk_number = r_sector; | |
1da177e4 LT |
2642 | |
2643 | /* | |
2644 | * Compute the stripe number | |
2645 | */ | |
35f2a591 N |
2646 | stripe = chunk_number; |
2647 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2648 | stripe2 = stripe; |
1da177e4 LT |
2649 | /* |
2650 | * Select the parity disk based on the user selected algorithm. | |
2651 | */ | |
84789554 | 2652 | pd_idx = qd_idx = -1; |
16a53ecc N |
2653 | switch(conf->level) { |
2654 | case 4: | |
911d4ee8 | 2655 | pd_idx = data_disks; |
16a53ecc N |
2656 | break; |
2657 | case 5: | |
e183eaed | 2658 | switch (algorithm) { |
1da177e4 | 2659 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2660 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2661 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2662 | (*dd_idx)++; |
2663 | break; | |
2664 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2665 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2666 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2667 | (*dd_idx)++; |
2668 | break; | |
2669 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2670 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2671 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2672 | break; |
2673 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2674 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2675 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2676 | break; |
99c0fb5f N |
2677 | case ALGORITHM_PARITY_0: |
2678 | pd_idx = 0; | |
2679 | (*dd_idx)++; | |
2680 | break; | |
2681 | case ALGORITHM_PARITY_N: | |
2682 | pd_idx = data_disks; | |
2683 | break; | |
1da177e4 | 2684 | default: |
99c0fb5f | 2685 | BUG(); |
16a53ecc N |
2686 | } |
2687 | break; | |
2688 | case 6: | |
2689 | ||
e183eaed | 2690 | switch (algorithm) { |
16a53ecc | 2691 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2692 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2693 | qd_idx = pd_idx + 1; |
2694 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2695 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2696 | qd_idx = 0; |
2697 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2698 | (*dd_idx) += 2; /* D D P Q D */ |
2699 | break; | |
2700 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2701 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2702 | qd_idx = pd_idx + 1; |
2703 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2704 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2705 | qd_idx = 0; |
2706 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2707 | (*dd_idx) += 2; /* D D P Q D */ |
2708 | break; | |
2709 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2710 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2711 | qd_idx = (pd_idx + 1) % raid_disks; |
2712 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2713 | break; |
2714 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2715 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2716 | qd_idx = (pd_idx + 1) % raid_disks; |
2717 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2718 | break; |
99c0fb5f N |
2719 | |
2720 | case ALGORITHM_PARITY_0: | |
2721 | pd_idx = 0; | |
2722 | qd_idx = 1; | |
2723 | (*dd_idx) += 2; | |
2724 | break; | |
2725 | case ALGORITHM_PARITY_N: | |
2726 | pd_idx = data_disks; | |
2727 | qd_idx = data_disks + 1; | |
2728 | break; | |
2729 | ||
2730 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2731 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2732 | * of blocks for computing Q is different. | |
2733 | */ | |
6e3b96ed | 2734 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2735 | qd_idx = pd_idx + 1; |
2736 | if (pd_idx == raid_disks-1) { | |
2737 | (*dd_idx)++; /* Q D D D P */ | |
2738 | qd_idx = 0; | |
2739 | } else if (*dd_idx >= pd_idx) | |
2740 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2741 | ddf_layout = 1; |
99c0fb5f N |
2742 | break; |
2743 | ||
2744 | case ALGORITHM_ROTATING_N_RESTART: | |
2745 | /* Same a left_asymmetric, by first stripe is | |
2746 | * D D D P Q rather than | |
2747 | * Q D D D P | |
2748 | */ | |
6e3b96ed N |
2749 | stripe2 += 1; |
2750 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2751 | qd_idx = pd_idx + 1; |
2752 | if (pd_idx == raid_disks-1) { | |
2753 | (*dd_idx)++; /* Q D D D P */ | |
2754 | qd_idx = 0; | |
2755 | } else if (*dd_idx >= pd_idx) | |
2756 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2757 | ddf_layout = 1; |
99c0fb5f N |
2758 | break; |
2759 | ||
2760 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2761 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2762 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2763 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2764 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2765 | ddf_layout = 1; |
99c0fb5f N |
2766 | break; |
2767 | ||
2768 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2769 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2770 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2771 | if (*dd_idx >= pd_idx) |
2772 | (*dd_idx)++; | |
2773 | qd_idx = raid_disks - 1; | |
2774 | break; | |
2775 | ||
2776 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2777 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2778 | if (*dd_idx >= pd_idx) |
2779 | (*dd_idx)++; | |
2780 | qd_idx = raid_disks - 1; | |
2781 | break; | |
2782 | ||
2783 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2784 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2785 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2786 | qd_idx = raid_disks - 1; | |
2787 | break; | |
2788 | ||
2789 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2790 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2791 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2792 | qd_idx = raid_disks - 1; | |
2793 | break; | |
2794 | ||
2795 | case ALGORITHM_PARITY_0_6: | |
2796 | pd_idx = 0; | |
2797 | (*dd_idx)++; | |
2798 | qd_idx = raid_disks - 1; | |
2799 | break; | |
2800 | ||
16a53ecc | 2801 | default: |
99c0fb5f | 2802 | BUG(); |
16a53ecc N |
2803 | } |
2804 | break; | |
1da177e4 LT |
2805 | } |
2806 | ||
911d4ee8 N |
2807 | if (sh) { |
2808 | sh->pd_idx = pd_idx; | |
2809 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2810 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2811 | } |
1da177e4 LT |
2812 | /* |
2813 | * Finally, compute the new sector number | |
2814 | */ | |
2815 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2816 | return new_sector; | |
2817 | } | |
2818 | ||
6d036f7d | 2819 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2820 | { |
d1688a6d | 2821 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2822 | int raid_disks = sh->disks; |
2823 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2824 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2825 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2826 | : conf->chunk_sectors; | |
e183eaed N |
2827 | int algorithm = previous ? conf->prev_algo |
2828 | : conf->algorithm; | |
1da177e4 LT |
2829 | sector_t stripe; |
2830 | int chunk_offset; | |
35f2a591 N |
2831 | sector_t chunk_number; |
2832 | int dummy1, dd_idx = i; | |
1da177e4 | 2833 | sector_t r_sector; |
911d4ee8 | 2834 | struct stripe_head sh2; |
1da177e4 LT |
2835 | |
2836 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2837 | stripe = new_sector; | |
1da177e4 | 2838 | |
16a53ecc N |
2839 | if (i == sh->pd_idx) |
2840 | return 0; | |
2841 | switch(conf->level) { | |
2842 | case 4: break; | |
2843 | case 5: | |
e183eaed | 2844 | switch (algorithm) { |
1da177e4 LT |
2845 | case ALGORITHM_LEFT_ASYMMETRIC: |
2846 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2847 | if (i > sh->pd_idx) | |
2848 | i--; | |
2849 | break; | |
2850 | case ALGORITHM_LEFT_SYMMETRIC: | |
2851 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2852 | if (i < sh->pd_idx) | |
2853 | i += raid_disks; | |
2854 | i -= (sh->pd_idx + 1); | |
2855 | break; | |
99c0fb5f N |
2856 | case ALGORITHM_PARITY_0: |
2857 | i -= 1; | |
2858 | break; | |
2859 | case ALGORITHM_PARITY_N: | |
2860 | break; | |
1da177e4 | 2861 | default: |
99c0fb5f | 2862 | BUG(); |
16a53ecc N |
2863 | } |
2864 | break; | |
2865 | case 6: | |
d0dabf7e | 2866 | if (i == sh->qd_idx) |
16a53ecc | 2867 | return 0; /* It is the Q disk */ |
e183eaed | 2868 | switch (algorithm) { |
16a53ecc N |
2869 | case ALGORITHM_LEFT_ASYMMETRIC: |
2870 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2871 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2872 | case ALGORITHM_ROTATING_N_RESTART: | |
2873 | if (sh->pd_idx == raid_disks-1) | |
2874 | i--; /* Q D D D P */ | |
16a53ecc N |
2875 | else if (i > sh->pd_idx) |
2876 | i -= 2; /* D D P Q D */ | |
2877 | break; | |
2878 | case ALGORITHM_LEFT_SYMMETRIC: | |
2879 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2880 | if (sh->pd_idx == raid_disks-1) | |
2881 | i--; /* Q D D D P */ | |
2882 | else { | |
2883 | /* D D P Q D */ | |
2884 | if (i < sh->pd_idx) | |
2885 | i += raid_disks; | |
2886 | i -= (sh->pd_idx + 2); | |
2887 | } | |
2888 | break; | |
99c0fb5f N |
2889 | case ALGORITHM_PARITY_0: |
2890 | i -= 2; | |
2891 | break; | |
2892 | case ALGORITHM_PARITY_N: | |
2893 | break; | |
2894 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2895 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2896 | if (sh->pd_idx == 0) |
2897 | i--; /* P D D D Q */ | |
e4424fee N |
2898 | else { |
2899 | /* D D Q P D */ | |
2900 | if (i < sh->pd_idx) | |
2901 | i += raid_disks; | |
2902 | i -= (sh->pd_idx + 1); | |
2903 | } | |
99c0fb5f N |
2904 | break; |
2905 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2906 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2907 | if (i > sh->pd_idx) | |
2908 | i--; | |
2909 | break; | |
2910 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2911 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2912 | if (i < sh->pd_idx) | |
2913 | i += data_disks + 1; | |
2914 | i -= (sh->pd_idx + 1); | |
2915 | break; | |
2916 | case ALGORITHM_PARITY_0_6: | |
2917 | i -= 1; | |
2918 | break; | |
16a53ecc | 2919 | default: |
99c0fb5f | 2920 | BUG(); |
16a53ecc N |
2921 | } |
2922 | break; | |
1da177e4 LT |
2923 | } |
2924 | ||
2925 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2926 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2927 | |
112bf897 | 2928 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2929 | previous, &dummy1, &sh2); |
911d4ee8 N |
2930 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2931 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
2932 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
2933 | mdname(conf->mddev)); | |
1da177e4 LT |
2934 | return 0; |
2935 | } | |
2936 | return r_sector; | |
2937 | } | |
2938 | ||
07e83364 SL |
2939 | /* |
2940 | * There are cases where we want handle_stripe_dirtying() and | |
2941 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
2942 | * | |
2943 | * This function checks whether we want to delay the towrite. Specifically, | |
2944 | * we delay the towrite when: | |
2945 | * | |
2946 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
2947 | * stripe has data in journal (for other devices). | |
2948 | * | |
2949 | * In this case, when reading data for the non-overwrite dev, it is | |
2950 | * necessary to handle complex rmw of write back cache (prexor with | |
2951 | * orig_page, and xor with page). To keep read path simple, we would | |
2952 | * like to flush data in journal to RAID disks first, so complex rmw | |
2953 | * is handled in the write patch (handle_stripe_dirtying). | |
2954 | * | |
39b99586 SL |
2955 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
2956 | * | |
2957 | * It is important to be able to flush all stripes in raid5-cache. | |
2958 | * Therefore, we need reserve some space on the journal device for | |
2959 | * these flushes. If flush operation includes pending writes to the | |
2960 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
2961 | * for the flush out. If we exclude these pending writes from flush | |
2962 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
2963 | * Therefore, excluding pending writes in these cases enables more | |
2964 | * efficient use of the journal device. | |
2965 | * | |
2966 | * Note: To make sure the stripe makes progress, we only delay | |
2967 | * towrite for stripes with data already in journal (injournal > 0). | |
2968 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
2969 | * no_space_stripes list. | |
2970 | * | |
07e83364 | 2971 | */ |
39b99586 SL |
2972 | static inline bool delay_towrite(struct r5conf *conf, |
2973 | struct r5dev *dev, | |
2974 | struct stripe_head_state *s) | |
07e83364 | 2975 | { |
39b99586 SL |
2976 | /* case 1 above */ |
2977 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2978 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
2979 | return true; | |
2980 | /* case 2 above */ | |
2981 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
2982 | s->injournal > 0) | |
2983 | return true; | |
2984 | return false; | |
07e83364 SL |
2985 | } |
2986 | ||
600aa109 | 2987 | static void |
c0f7bddb | 2988 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 2989 | int rcw, int expand) |
e33129d8 | 2990 | { |
584acdd4 | 2991 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 2992 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 2993 | int level = conf->level; |
e33129d8 DW |
2994 | |
2995 | if (rcw) { | |
1e6d690b SL |
2996 | /* |
2997 | * In some cases, handle_stripe_dirtying initially decided to | |
2998 | * run rmw and allocates extra page for prexor. However, rcw is | |
2999 | * cheaper later on. We need to free the extra page now, | |
3000 | * because we won't be able to do that in ops_complete_prexor(). | |
3001 | */ | |
3002 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3003 | |
3004 | for (i = disks; i--; ) { | |
3005 | struct r5dev *dev = &sh->dev[i]; | |
3006 | ||
39b99586 | 3007 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3008 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3009 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3010 | if (!expand) |
3011 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3012 | s->locked++; |
1e6d690b SL |
3013 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3014 | set_bit(R5_LOCKED, &dev->flags); | |
3015 | s->locked++; | |
e33129d8 DW |
3016 | } |
3017 | } | |
ce7d363a N |
3018 | /* if we are not expanding this is a proper write request, and |
3019 | * there will be bios with new data to be drained into the | |
3020 | * stripe cache | |
3021 | */ | |
3022 | if (!expand) { | |
3023 | if (!s->locked) | |
3024 | /* False alarm, nothing to do */ | |
3025 | return; | |
3026 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3027 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3028 | } else | |
3029 | sh->reconstruct_state = reconstruct_state_run; | |
3030 | ||
3031 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3032 | ||
c0f7bddb | 3033 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3034 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3035 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3036 | } else { |
3037 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3038 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3039 | BUG_ON(level == 6 && |
3040 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3041 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3042 | |
e33129d8 DW |
3043 | for (i = disks; i--; ) { |
3044 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3045 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3046 | continue; |
3047 | ||
e33129d8 DW |
3048 | if (dev->towrite && |
3049 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3050 | test_bit(R5_Wantcompute, &dev->flags))) { |
3051 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3052 | set_bit(R5_LOCKED, &dev->flags); |
3053 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3054 | s->locked++; |
1e6d690b SL |
3055 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3056 | set_bit(R5_LOCKED, &dev->flags); | |
3057 | s->locked++; | |
e33129d8 DW |
3058 | } |
3059 | } | |
ce7d363a N |
3060 | if (!s->locked) |
3061 | /* False alarm - nothing to do */ | |
3062 | return; | |
3063 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3064 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3065 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3066 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3067 | } |
3068 | ||
c0f7bddb | 3069 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3070 | * are in flight |
3071 | */ | |
3072 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3073 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3074 | s->locked++; |
e33129d8 | 3075 | |
c0f7bddb YT |
3076 | if (level == 6) { |
3077 | int qd_idx = sh->qd_idx; | |
3078 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3079 | ||
3080 | set_bit(R5_LOCKED, &dev->flags); | |
3081 | clear_bit(R5_UPTODATE, &dev->flags); | |
3082 | s->locked++; | |
3083 | } | |
3084 | ||
600aa109 | 3085 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3086 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3087 | s->locked, s->ops_request); |
e33129d8 | 3088 | } |
16a53ecc | 3089 | |
1da177e4 LT |
3090 | /* |
3091 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3092 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3093 | * The bi_next chain must be in order. |
3094 | */ | |
da41ba65 | 3095 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3096 | int forwrite, int previous) | |
1da177e4 LT |
3097 | { |
3098 | struct bio **bip; | |
d1688a6d | 3099 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3100 | int firstwrite=0; |
1da177e4 | 3101 | |
cbe47ec5 | 3102 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3103 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3104 | (unsigned long long)sh->sector); |
3105 | ||
b17459c0 SL |
3106 | /* |
3107 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
3108 | * reference count to avoid race. The reference count should already be | |
3109 | * increased before this function is called (for example, in | |
849674e4 | 3110 | * raid5_make_request()), so other bio sharing this stripe will not free the |
b17459c0 SL |
3111 | * stripe. If a stripe is owned by one stripe, the stripe lock will |
3112 | * protect it. | |
3113 | */ | |
3114 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 3115 | /* Don't allow new IO added to stripes in batch list */ |
3116 | if (sh->batch_head) | |
3117 | goto overlap; | |
72626685 | 3118 | if (forwrite) { |
1da177e4 | 3119 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3120 | if (*bip == NULL) |
72626685 N |
3121 | firstwrite = 1; |
3122 | } else | |
1da177e4 | 3123 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3124 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3125 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3126 | goto overlap; |
3127 | bip = & (*bip)->bi_next; | |
3128 | } | |
4f024f37 | 3129 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3130 | goto overlap; |
3131 | ||
da41ba65 | 3132 | if (!forwrite || previous) |
3133 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3134 | ||
78bafebd | 3135 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3136 | if (*bip) |
3137 | bi->bi_next = *bip; | |
3138 | *bip = bi; | |
e7836bd6 | 3139 | raid5_inc_bi_active_stripes(bi); |
72626685 | 3140 | |
1da177e4 LT |
3141 | if (forwrite) { |
3142 | /* check if page is covered */ | |
3143 | sector_t sector = sh->dev[dd_idx].sector; | |
3144 | for (bi=sh->dev[dd_idx].towrite; | |
3145 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3146 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3147 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3148 | if (bio_end_sector(bi) >= sector) |
3149 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3150 | } |
3151 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3152 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3153 | sh->overwrite_disks++; | |
1da177e4 | 3154 | } |
cbe47ec5 N |
3155 | |
3156 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3157 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3158 | (unsigned long long)sh->sector, dd_idx); |
3159 | ||
3160 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3161 | /* Cannot hold spinlock over bitmap_startwrite, |
3162 | * but must ensure this isn't added to a batch until | |
3163 | * we have added to the bitmap and set bm_seq. | |
3164 | * So set STRIPE_BITMAP_PENDING to prevent | |
3165 | * batching. | |
3166 | * If multiple add_stripe_bio() calls race here they | |
3167 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3168 | * to complete "bitmap_startwrite" gets to set | |
3169 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3170 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3171 | * any more. | |
3172 | */ | |
3173 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3174 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3175 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3176 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3177 | spin_lock_irq(&sh->stripe_lock); |
3178 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3179 | if (!sh->batch_head) { | |
3180 | sh->bm_seq = conf->seq_flush+1; | |
3181 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3182 | } | |
cbe47ec5 | 3183 | } |
d0852df5 | 3184 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3185 | |
3186 | if (stripe_can_batch(sh)) | |
3187 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3188 | return 1; |
3189 | ||
3190 | overlap: | |
3191 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3192 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3193 | return 0; |
3194 | } | |
3195 | ||
d1688a6d | 3196 | static void end_reshape(struct r5conf *conf); |
29269553 | 3197 | |
d1688a6d | 3198 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3199 | struct stripe_head *sh) |
ccfcc3c1 | 3200 | { |
784052ec | 3201 | int sectors_per_chunk = |
09c9e5fa | 3202 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3203 | int dd_idx; |
2d2063ce | 3204 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3205 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3206 | |
112bf897 N |
3207 | raid5_compute_sector(conf, |
3208 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3209 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3210 | previous, |
911d4ee8 | 3211 | &dd_idx, sh); |
ccfcc3c1 N |
3212 | } |
3213 | ||
a4456856 | 3214 | static void |
d1688a6d | 3215 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 | 3216 | struct stripe_head_state *s, int disks, |
34a6f80e | 3217 | struct bio_list *return_bi) |
a4456856 DW |
3218 | { |
3219 | int i; | |
59fc630b | 3220 | BUG_ON(sh->batch_head); |
a4456856 DW |
3221 | for (i = disks; i--; ) { |
3222 | struct bio *bi; | |
3223 | int bitmap_end = 0; | |
3224 | ||
3225 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3226 | struct md_rdev *rdev; |
a4456856 DW |
3227 | rcu_read_lock(); |
3228 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3229 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3230 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3231 | atomic_inc(&rdev->nr_pending); |
3232 | else | |
3233 | rdev = NULL; | |
a4456856 | 3234 | rcu_read_unlock(); |
7f0da59b N |
3235 | if (rdev) { |
3236 | if (!rdev_set_badblocks( | |
3237 | rdev, | |
3238 | sh->sector, | |
3239 | STRIPE_SECTORS, 0)) | |
3240 | md_error(conf->mddev, rdev); | |
3241 | rdev_dec_pending(rdev, conf->mddev); | |
3242 | } | |
a4456856 | 3243 | } |
b17459c0 | 3244 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3245 | /* fail all writes first */ |
3246 | bi = sh->dev[i].towrite; | |
3247 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3248 | sh->overwrite_disks = 0; |
b17459c0 | 3249 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3250 | if (bi) |
a4456856 | 3251 | bitmap_end = 1; |
a4456856 | 3252 | |
0576b1c6 SL |
3253 | r5l_stripe_write_finished(sh); |
3254 | ||
a4456856 DW |
3255 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3256 | wake_up(&conf->wait_for_overlap); | |
3257 | ||
4f024f37 | 3258 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3259 | sh->dev[i].sector + STRIPE_SECTORS) { |
3260 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3261 | |
3262 | bi->bi_error = -EIO; | |
e7836bd6 | 3263 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3264 | md_write_end(conf->mddev); |
34a6f80e | 3265 | bio_list_add(return_bi, bi); |
a4456856 DW |
3266 | } |
3267 | bi = nextbi; | |
3268 | } | |
7eaf7e8e SL |
3269 | if (bitmap_end) |
3270 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3271 | STRIPE_SECTORS, 0, 0); | |
3272 | bitmap_end = 0; | |
a4456856 DW |
3273 | /* and fail all 'written' */ |
3274 | bi = sh->dev[i].written; | |
3275 | sh->dev[i].written = NULL; | |
d592a996 SL |
3276 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3277 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3278 | sh->dev[i].page = sh->dev[i].orig_page; | |
3279 | } | |
3280 | ||
a4456856 | 3281 | if (bi) bitmap_end = 1; |
4f024f37 | 3282 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3283 | sh->dev[i].sector + STRIPE_SECTORS) { |
3284 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3285 | |
3286 | bi->bi_error = -EIO; | |
e7836bd6 | 3287 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3288 | md_write_end(conf->mddev); |
34a6f80e | 3289 | bio_list_add(return_bi, bi); |
a4456856 DW |
3290 | } |
3291 | bi = bi2; | |
3292 | } | |
3293 | ||
b5e98d65 DW |
3294 | /* fail any reads if this device is non-operational and |
3295 | * the data has not reached the cache yet. | |
3296 | */ | |
3297 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3298 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3299 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3300 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3301 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3302 | bi = sh->dev[i].toread; |
3303 | sh->dev[i].toread = NULL; | |
143c4d05 | 3304 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3305 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3306 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3307 | if (bi) |
3308 | s->to_read--; | |
4f024f37 | 3309 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3310 | sh->dev[i].sector + STRIPE_SECTORS) { |
3311 | struct bio *nextbi = | |
3312 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3313 | |
3314 | bi->bi_error = -EIO; | |
34a6f80e N |
3315 | if (!raid5_dec_bi_active_stripes(bi)) |
3316 | bio_list_add(return_bi, bi); | |
a4456856 DW |
3317 | bi = nextbi; |
3318 | } | |
3319 | } | |
a4456856 DW |
3320 | if (bitmap_end) |
3321 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3322 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3323 | /* If we were in the middle of a write the parity block might |
3324 | * still be locked - so just clear all R5_LOCKED flags | |
3325 | */ | |
3326 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3327 | } |
ebda780b SL |
3328 | s->to_write = 0; |
3329 | s->written = 0; | |
a4456856 | 3330 | |
8b3e6cdc DW |
3331 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3332 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3333 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3334 | } |
3335 | ||
7f0da59b | 3336 | static void |
d1688a6d | 3337 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3338 | struct stripe_head_state *s) |
3339 | { | |
3340 | int abort = 0; | |
3341 | int i; | |
3342 | ||
59fc630b | 3343 | BUG_ON(sh->batch_head); |
7f0da59b | 3344 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3345 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3346 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3347 | s->syncing = 0; |
9a3e1101 | 3348 | s->replacing = 0; |
7f0da59b | 3349 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3350 | * Don't even need to abort as that is handled elsewhere |
3351 | * if needed, and not always wanted e.g. if there is a known | |
3352 | * bad block here. | |
9a3e1101 | 3353 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3354 | * non-sync devices, or abort the recovery |
3355 | */ | |
18b9837e N |
3356 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3357 | /* During recovery devices cannot be removed, so | |
3358 | * locking and refcounting of rdevs is not needed | |
3359 | */ | |
e50d3992 | 3360 | rcu_read_lock(); |
18b9837e | 3361 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3362 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3363 | if (rdev |
3364 | && !test_bit(Faulty, &rdev->flags) | |
3365 | && !test_bit(In_sync, &rdev->flags) | |
3366 | && !rdev_set_badblocks(rdev, sh->sector, | |
3367 | STRIPE_SECTORS, 0)) | |
3368 | abort = 1; | |
e50d3992 | 3369 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3370 | if (rdev |
3371 | && !test_bit(Faulty, &rdev->flags) | |
3372 | && !test_bit(In_sync, &rdev->flags) | |
3373 | && !rdev_set_badblocks(rdev, sh->sector, | |
3374 | STRIPE_SECTORS, 0)) | |
3375 | abort = 1; | |
3376 | } | |
e50d3992 | 3377 | rcu_read_unlock(); |
18b9837e N |
3378 | if (abort) |
3379 | conf->recovery_disabled = | |
3380 | conf->mddev->recovery_disabled; | |
7f0da59b | 3381 | } |
18b9837e | 3382 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3383 | } |
3384 | ||
9a3e1101 N |
3385 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3386 | { | |
3387 | struct md_rdev *rdev; | |
3388 | int rv = 0; | |
3f232d6a N |
3389 | |
3390 | rcu_read_lock(); | |
3391 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3392 | if (rdev |
3393 | && !test_bit(Faulty, &rdev->flags) | |
3394 | && !test_bit(In_sync, &rdev->flags) | |
3395 | && (rdev->recovery_offset <= sh->sector | |
3396 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3397 | rv = 1; | |
3f232d6a | 3398 | rcu_read_unlock(); |
9a3e1101 N |
3399 | return rv; |
3400 | } | |
3401 | ||
2c58f06e N |
3402 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3403 | int disk_idx, int disks) | |
a4456856 | 3404 | { |
5599becc | 3405 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3406 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3407 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3408 | int i; |
5599becc | 3409 | |
a79cfe12 N |
3410 | |
3411 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3412 | test_bit(R5_UPTODATE, &dev->flags)) | |
3413 | /* No point reading this as we already have it or have | |
3414 | * decided to get it. | |
3415 | */ | |
3416 | return 0; | |
3417 | ||
3418 | if (dev->toread || | |
3419 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3420 | /* We need this block to directly satisfy a request */ | |
3421 | return 1; | |
3422 | ||
3423 | if (s->syncing || s->expanding || | |
3424 | (s->replacing && want_replace(sh, disk_idx))) | |
3425 | /* When syncing, or expanding we read everything. | |
3426 | * When replacing, we need the replaced block. | |
3427 | */ | |
3428 | return 1; | |
3429 | ||
3430 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3431 | (s->failed >= 2 && fdev[1]->toread)) | |
3432 | /* If we want to read from a failed device, then | |
3433 | * we need to actually read every other device. | |
3434 | */ | |
3435 | return 1; | |
3436 | ||
a9d56950 N |
3437 | /* Sometimes neither read-modify-write nor reconstruct-write |
3438 | * cycles can work. In those cases we read every block we | |
3439 | * can. Then the parity-update is certain to have enough to | |
3440 | * work with. | |
3441 | * This can only be a problem when we need to write something, | |
3442 | * and some device has failed. If either of those tests | |
3443 | * fail we need look no further. | |
3444 | */ | |
3445 | if (!s->failed || !s->to_write) | |
3446 | return 0; | |
3447 | ||
3448 | if (test_bit(R5_Insync, &dev->flags) && | |
3449 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3450 | /* Pre-reads at not permitted until after short delay | |
3451 | * to gather multiple requests. However if this | |
3452 | * device is no Insync, the block could only be be computed | |
3453 | * and there is no need to delay that. | |
3454 | */ | |
3455 | return 0; | |
ea664c82 | 3456 | |
36707bb2 | 3457 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3458 | if (fdev[i]->towrite && |
3459 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3460 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3461 | /* If we have a partial write to a failed | |
3462 | * device, then we will need to reconstruct | |
3463 | * the content of that device, so all other | |
3464 | * devices must be read. | |
3465 | */ | |
3466 | return 1; | |
3467 | } | |
3468 | ||
3469 | /* If we are forced to do a reconstruct-write, either because | |
3470 | * the current RAID6 implementation only supports that, or | |
3471 | * or because parity cannot be trusted and we are currently | |
3472 | * recovering it, there is extra need to be careful. | |
3473 | * If one of the devices that we would need to read, because | |
3474 | * it is not being overwritten (and maybe not written at all) | |
3475 | * is missing/faulty, then we need to read everything we can. | |
3476 | */ | |
3477 | if (sh->raid_conf->level != 6 && | |
3478 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3479 | /* reconstruct-write isn't being forced */ | |
3480 | return 0; | |
36707bb2 | 3481 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3482 | if (s->failed_num[i] != sh->pd_idx && |
3483 | s->failed_num[i] != sh->qd_idx && | |
3484 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3485 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3486 | return 1; | |
3487 | } | |
3488 | ||
2c58f06e N |
3489 | return 0; |
3490 | } | |
3491 | ||
ba02684d SL |
3492 | /* fetch_block - checks the given member device to see if its data needs |
3493 | * to be read or computed to satisfy a request. | |
3494 | * | |
3495 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3496 | * 0 to tell the loop in handle_stripe_fill to continue | |
3497 | */ | |
2c58f06e N |
3498 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3499 | int disk_idx, int disks) | |
3500 | { | |
3501 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3502 | ||
3503 | /* is the data in this block needed, and can we get it? */ | |
3504 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3505 | /* we would like to get this block, possibly by computing it, |
3506 | * otherwise read it if the backing disk is insync | |
3507 | */ | |
3508 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3509 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3510 | BUG_ON(sh->batch_head); |
5599becc | 3511 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3512 | (s->failed && (disk_idx == s->failed_num[0] || |
3513 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3514 | /* have disk failed, and we're requested to fetch it; |
3515 | * do compute it | |
a4456856 | 3516 | */ |
5599becc YT |
3517 | pr_debug("Computing stripe %llu block %d\n", |
3518 | (unsigned long long)sh->sector, disk_idx); | |
3519 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3520 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3521 | set_bit(R5_Wantcompute, &dev->flags); | |
3522 | sh->ops.target = disk_idx; | |
3523 | sh->ops.target2 = -1; /* no 2nd target */ | |
3524 | s->req_compute = 1; | |
93b3dbce N |
3525 | /* Careful: from this point on 'uptodate' is in the eye |
3526 | * of raid_run_ops which services 'compute' operations | |
3527 | * before writes. R5_Wantcompute flags a block that will | |
3528 | * be R5_UPTODATE by the time it is needed for a | |
3529 | * subsequent operation. | |
3530 | */ | |
5599becc YT |
3531 | s->uptodate++; |
3532 | return 1; | |
3533 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3534 | /* Computing 2-failure is *very* expensive; only | |
3535 | * do it if failed >= 2 | |
3536 | */ | |
3537 | int other; | |
3538 | for (other = disks; other--; ) { | |
3539 | if (other == disk_idx) | |
3540 | continue; | |
3541 | if (!test_bit(R5_UPTODATE, | |
3542 | &sh->dev[other].flags)) | |
3543 | break; | |
a4456856 | 3544 | } |
5599becc YT |
3545 | BUG_ON(other < 0); |
3546 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3547 | (unsigned long long)sh->sector, | |
3548 | disk_idx, other); | |
3549 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3550 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3551 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3552 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3553 | sh->ops.target = disk_idx; | |
3554 | sh->ops.target2 = other; | |
3555 | s->uptodate += 2; | |
3556 | s->req_compute = 1; | |
3557 | return 1; | |
3558 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3559 | set_bit(R5_LOCKED, &dev->flags); | |
3560 | set_bit(R5_Wantread, &dev->flags); | |
3561 | s->locked++; | |
3562 | pr_debug("Reading block %d (sync=%d)\n", | |
3563 | disk_idx, s->syncing); | |
a4456856 DW |
3564 | } |
3565 | } | |
5599becc YT |
3566 | |
3567 | return 0; | |
3568 | } | |
3569 | ||
3570 | /** | |
93b3dbce | 3571 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3572 | */ |
93b3dbce N |
3573 | static void handle_stripe_fill(struct stripe_head *sh, |
3574 | struct stripe_head_state *s, | |
3575 | int disks) | |
5599becc YT |
3576 | { |
3577 | int i; | |
3578 | ||
3579 | /* look for blocks to read/compute, skip this if a compute | |
3580 | * is already in flight, or if the stripe contents are in the | |
3581 | * midst of changing due to a write | |
3582 | */ | |
3583 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3584 | !sh->reconstruct_state) { |
3585 | ||
3586 | /* | |
3587 | * For degraded stripe with data in journal, do not handle | |
3588 | * read requests yet, instead, flush the stripe to raid | |
3589 | * disks first, this avoids handling complex rmw of write | |
3590 | * back cache (prexor with orig_page, and then xor with | |
3591 | * page) in the read path | |
3592 | */ | |
3593 | if (s->injournal && s->failed) { | |
3594 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3595 | r5c_make_stripe_write_out(sh); | |
3596 | goto out; | |
3597 | } | |
3598 | ||
5599becc | 3599 | for (i = disks; i--; ) |
93b3dbce | 3600 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3601 | break; |
07e83364 SL |
3602 | } |
3603 | out: | |
a4456856 DW |
3604 | set_bit(STRIPE_HANDLE, &sh->state); |
3605 | } | |
3606 | ||
787b76fa N |
3607 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3608 | unsigned long handle_flags); | |
1fe797e6 | 3609 | /* handle_stripe_clean_event |
a4456856 DW |
3610 | * any written block on an uptodate or failed drive can be returned. |
3611 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3612 | * never LOCKED, so we don't need to test 'failed' directly. | |
3613 | */ | |
d1688a6d | 3614 | static void handle_stripe_clean_event(struct r5conf *conf, |
34a6f80e | 3615 | struct stripe_head *sh, int disks, struct bio_list *return_bi) |
a4456856 DW |
3616 | { |
3617 | int i; | |
3618 | struct r5dev *dev; | |
f8dfcffd | 3619 | int discard_pending = 0; |
59fc630b | 3620 | struct stripe_head *head_sh = sh; |
3621 | bool do_endio = false; | |
a4456856 DW |
3622 | |
3623 | for (i = disks; i--; ) | |
3624 | if (sh->dev[i].written) { | |
3625 | dev = &sh->dev[i]; | |
3626 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3627 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3628 | test_bit(R5_Discard, &dev->flags) || |
3629 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3630 | /* We can return any write requests */ |
3631 | struct bio *wbi, *wbi2; | |
45b4233c | 3632 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3633 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3634 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3635 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3636 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3637 | } |
59fc630b | 3638 | do_endio = true; |
3639 | ||
3640 | returnbi: | |
3641 | dev->page = dev->orig_page; | |
a4456856 DW |
3642 | wbi = dev->written; |
3643 | dev->written = NULL; | |
4f024f37 | 3644 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3645 | dev->sector + STRIPE_SECTORS) { |
3646 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3647 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 | 3648 | md_write_end(conf->mddev); |
34a6f80e | 3649 | bio_list_add(return_bi, wbi); |
a4456856 DW |
3650 | } |
3651 | wbi = wbi2; | |
3652 | } | |
7eaf7e8e SL |
3653 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3654 | STRIPE_SECTORS, | |
a4456856 | 3655 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3656 | 0); |
59fc630b | 3657 | if (head_sh->batch_head) { |
3658 | sh = list_first_entry(&sh->batch_list, | |
3659 | struct stripe_head, | |
3660 | batch_list); | |
3661 | if (sh != head_sh) { | |
3662 | dev = &sh->dev[i]; | |
3663 | goto returnbi; | |
3664 | } | |
3665 | } | |
3666 | sh = head_sh; | |
3667 | dev = &sh->dev[i]; | |
f8dfcffd N |
3668 | } else if (test_bit(R5_Discard, &dev->flags)) |
3669 | discard_pending = 1; | |
3670 | } | |
f6bed0ef | 3671 | |
0576b1c6 SL |
3672 | r5l_stripe_write_finished(sh); |
3673 | ||
f8dfcffd N |
3674 | if (!discard_pending && |
3675 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3676 | int hash; |
f8dfcffd N |
3677 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3678 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3679 | if (sh->qd_idx >= 0) { | |
3680 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3681 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3682 | } | |
3683 | /* now that discard is done we can proceed with any sync */ | |
3684 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3685 | /* |
3686 | * SCSI discard will change some bio fields and the stripe has | |
3687 | * no updated data, so remove it from hash list and the stripe | |
3688 | * will be reinitialized | |
3689 | */ | |
59fc630b | 3690 | unhash: |
b8a9d66d RG |
3691 | hash = sh->hash_lock_index; |
3692 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3693 | remove_hash(sh); |
b8a9d66d | 3694 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3695 | if (head_sh->batch_head) { |
3696 | sh = list_first_entry(&sh->batch_list, | |
3697 | struct stripe_head, batch_list); | |
3698 | if (sh != head_sh) | |
3699 | goto unhash; | |
3700 | } | |
59fc630b | 3701 | sh = head_sh; |
3702 | ||
f8dfcffd N |
3703 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3704 | set_bit(STRIPE_HANDLE, &sh->state); | |
3705 | ||
3706 | } | |
8b3e6cdc DW |
3707 | |
3708 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3709 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3710 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3711 | |
787b76fa N |
3712 | if (head_sh->batch_head && do_endio) |
3713 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3714 | } |
3715 | ||
86aa1397 SL |
3716 | /* |
3717 | * For RMW in write back cache, we need extra page in prexor to store the | |
3718 | * old data. This page is stored in dev->orig_page. | |
3719 | * | |
3720 | * This function checks whether we have data for prexor. The exact logic | |
3721 | * is: | |
3722 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3723 | */ | |
3724 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3725 | { | |
3726 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3727 | (!test_bit(R5_InJournal, &dev->flags) || | |
3728 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3729 | } | |
3730 | ||
d7bd398e SL |
3731 | static int handle_stripe_dirtying(struct r5conf *conf, |
3732 | struct stripe_head *sh, | |
3733 | struct stripe_head_state *s, | |
3734 | int disks) | |
a4456856 DW |
3735 | { |
3736 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3737 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3738 | ||
584acdd4 | 3739 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3740 | * If yes, then the array is dirty (after unclean shutdown or |
3741 | * initial creation), so parity in some stripes might be inconsistent. | |
3742 | * In this case, we need to always do reconstruct-write, to ensure | |
3743 | * that in case of drive failure or read-error correction, we | |
3744 | * generate correct data from the parity. | |
3745 | */ | |
584acdd4 | 3746 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3747 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3748 | s->failed == 0)) { | |
a7854487 | 3749 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3750 | * look like rcw is cheaper |
3751 | */ | |
3752 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3753 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3754 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3755 | (unsigned long long)sh->sector); |
c8ac1803 | 3756 | } else for (i = disks; i--; ) { |
a4456856 DW |
3757 | /* would I have to read this buffer for read_modify_write */ |
3758 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3759 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3760 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3761 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3762 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3763 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3764 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3765 | if (test_bit(R5_Insync, &dev->flags)) |
3766 | rmw++; | |
3767 | else | |
3768 | rmw += 2*disks; /* cannot read it */ | |
3769 | } | |
3770 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3771 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3772 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3773 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3774 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3775 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3776 | if (test_bit(R5_Insync, &dev->flags)) |
3777 | rcw++; | |
a4456856 DW |
3778 | else |
3779 | rcw += 2*disks; | |
3780 | } | |
3781 | } | |
1e6d690b | 3782 | |
39b99586 SL |
3783 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3784 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3785 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3786 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3787 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3788 | if (conf->mddev->queue) |
3789 | blk_add_trace_msg(conf->mddev->queue, | |
3790 | "raid5 rmw %llu %d", | |
3791 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3792 | for (i = disks; i--; ) { |
3793 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3794 | if (test_bit(R5_InJournal, &dev->flags) && |
3795 | dev->page == dev->orig_page && | |
3796 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3797 | /* alloc page for prexor */ | |
d7bd398e SL |
3798 | struct page *p = alloc_page(GFP_NOIO); |
3799 | ||
3800 | if (p) { | |
3801 | dev->orig_page = p; | |
3802 | continue; | |
3803 | } | |
3804 | ||
3805 | /* | |
3806 | * alloc_page() failed, try use | |
3807 | * disk_info->extra_page | |
3808 | */ | |
3809 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3810 | &conf->cache_state)) { | |
3811 | r5c_use_extra_page(sh); | |
3812 | break; | |
3813 | } | |
1e6d690b | 3814 | |
d7bd398e SL |
3815 | /* extra_page in use, add to delayed_list */ |
3816 | set_bit(STRIPE_DELAYED, &sh->state); | |
3817 | s->waiting_extra_page = 1; | |
3818 | return -EAGAIN; | |
1e6d690b | 3819 | } |
d7bd398e | 3820 | } |
1e6d690b | 3821 | |
d7bd398e SL |
3822 | for (i = disks; i--; ) { |
3823 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3824 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3825 | i == sh->pd_idx || i == sh->qd_idx || |
3826 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3827 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3828 | !(uptodate_for_rmw(dev) || |
1e6d690b | 3829 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 3830 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3831 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3832 | &sh->state)) { | |
3833 | pr_debug("Read_old block %d for r-m-w\n", | |
3834 | i); | |
a4456856 DW |
3835 | set_bit(R5_LOCKED, &dev->flags); |
3836 | set_bit(R5_Wantread, &dev->flags); | |
3837 | s->locked++; | |
3838 | } else { | |
3839 | set_bit(STRIPE_DELAYED, &sh->state); | |
3840 | set_bit(STRIPE_HANDLE, &sh->state); | |
3841 | } | |
3842 | } | |
3843 | } | |
a9add5d9 | 3844 | } |
41257580 | 3845 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3846 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3847 | int qread =0; |
c8ac1803 | 3848 | rcw = 0; |
a4456856 DW |
3849 | for (i = disks; i--; ) { |
3850 | struct r5dev *dev = &sh->dev[i]; | |
3851 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3852 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3853 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3854 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3855 | test_bit(R5_Wantcompute, &dev->flags))) { |
3856 | rcw++; | |
67f45548 N |
3857 | if (test_bit(R5_Insync, &dev->flags) && |
3858 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3859 | &sh->state)) { | |
45b4233c | 3860 | pr_debug("Read_old block " |
a4456856 DW |
3861 | "%d for Reconstruct\n", i); |
3862 | set_bit(R5_LOCKED, &dev->flags); | |
3863 | set_bit(R5_Wantread, &dev->flags); | |
3864 | s->locked++; | |
a9add5d9 | 3865 | qread++; |
a4456856 DW |
3866 | } else { |
3867 | set_bit(STRIPE_DELAYED, &sh->state); | |
3868 | set_bit(STRIPE_HANDLE, &sh->state); | |
3869 | } | |
3870 | } | |
3871 | } | |
e3620a3a | 3872 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3873 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3874 | (unsigned long long)sh->sector, | |
3875 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3876 | } |
b1b02fe9 N |
3877 | |
3878 | if (rcw > disks && rmw > disks && | |
3879 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3880 | set_bit(STRIPE_DELAYED, &sh->state); | |
3881 | ||
a4456856 DW |
3882 | /* now if nothing is locked, and if we have enough data, |
3883 | * we can start a write request | |
3884 | */ | |
f38e1219 DW |
3885 | /* since handle_stripe can be called at any time we need to handle the |
3886 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3887 | * subsequent call wants to start a write request. raid_run_ops only |
3888 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3889 | * simultaneously. If this is not the case then new writes need to be |
3890 | * held off until the compute completes. | |
3891 | */ | |
976ea8d4 DW |
3892 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3893 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 3894 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 3895 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 3896 | return 0; |
a4456856 DW |
3897 | } |
3898 | ||
d1688a6d | 3899 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3900 | struct stripe_head_state *s, int disks) |
3901 | { | |
ecc65c9b | 3902 | struct r5dev *dev = NULL; |
bd2ab670 | 3903 | |
59fc630b | 3904 | BUG_ON(sh->batch_head); |
a4456856 | 3905 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3906 | |
ecc65c9b DW |
3907 | switch (sh->check_state) { |
3908 | case check_state_idle: | |
3909 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3910 | if (s->failed == 0) { |
bd2ab670 | 3911 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3912 | sh->check_state = check_state_run; |
3913 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3914 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3915 | s->uptodate--; |
ecc65c9b | 3916 | break; |
bd2ab670 | 3917 | } |
f2b3b44d | 3918 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3919 | /* fall through */ |
3920 | case check_state_compute_result: | |
3921 | sh->check_state = check_state_idle; | |
3922 | if (!dev) | |
3923 | dev = &sh->dev[sh->pd_idx]; | |
3924 | ||
3925 | /* check that a write has not made the stripe insync */ | |
3926 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3927 | break; | |
c8894419 | 3928 | |
a4456856 | 3929 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3930 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3931 | BUG_ON(s->uptodate != disks); | |
3932 | ||
3933 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3934 | s->locked++; |
a4456856 | 3935 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3936 | |
a4456856 | 3937 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3938 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3939 | break; |
3940 | case check_state_run: | |
3941 | break; /* we will be called again upon completion */ | |
3942 | case check_state_check_result: | |
3943 | sh->check_state = check_state_idle; | |
3944 | ||
3945 | /* if a failure occurred during the check operation, leave | |
3946 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3947 | */ | |
3948 | if (s->failed) | |
3949 | break; | |
3950 | ||
3951 | /* handle a successful check operation, if parity is correct | |
3952 | * we are done. Otherwise update the mismatch count and repair | |
3953 | * parity if !MD_RECOVERY_CHECK | |
3954 | */ | |
ad283ea4 | 3955 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3956 | /* parity is correct (on disc, |
3957 | * not in buffer any more) | |
3958 | */ | |
3959 | set_bit(STRIPE_INSYNC, &sh->state); | |
3960 | else { | |
7f7583d4 | 3961 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3962 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3963 | /* don't try to repair!! */ | |
3964 | set_bit(STRIPE_INSYNC, &sh->state); | |
3965 | else { | |
3966 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3967 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3968 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3969 | set_bit(R5_Wantcompute, | |
3970 | &sh->dev[sh->pd_idx].flags); | |
3971 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3972 | sh->ops.target2 = -1; |
ecc65c9b DW |
3973 | s->uptodate++; |
3974 | } | |
3975 | } | |
3976 | break; | |
3977 | case check_state_compute_run: | |
3978 | break; | |
3979 | default: | |
cc6167b4 | 3980 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
3981 | __func__, sh->check_state, |
3982 | (unsigned long long) sh->sector); | |
3983 | BUG(); | |
a4456856 DW |
3984 | } |
3985 | } | |
3986 | ||
d1688a6d | 3987 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 3988 | struct stripe_head_state *s, |
f2b3b44d | 3989 | int disks) |
a4456856 | 3990 | { |
a4456856 | 3991 | int pd_idx = sh->pd_idx; |
34e04e87 | 3992 | int qd_idx = sh->qd_idx; |
d82dfee0 | 3993 | struct r5dev *dev; |
a4456856 | 3994 | |
59fc630b | 3995 | BUG_ON(sh->batch_head); |
a4456856 DW |
3996 | set_bit(STRIPE_HANDLE, &sh->state); |
3997 | ||
3998 | BUG_ON(s->failed > 2); | |
d82dfee0 | 3999 | |
a4456856 DW |
4000 | /* Want to check and possibly repair P and Q. |
4001 | * However there could be one 'failed' device, in which | |
4002 | * case we can only check one of them, possibly using the | |
4003 | * other to generate missing data | |
4004 | */ | |
4005 | ||
d82dfee0 DW |
4006 | switch (sh->check_state) { |
4007 | case check_state_idle: | |
4008 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4009 | if (s->failed == s->q_failed) { |
d82dfee0 | 4010 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4011 | * makes sense to check P (If anything else were failed, |
4012 | * we would have used P to recreate it). | |
4013 | */ | |
d82dfee0 | 4014 | sh->check_state = check_state_run; |
a4456856 | 4015 | } |
f2b3b44d | 4016 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4017 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4018 | * anything, so it makes sense to check it |
4019 | */ | |
d82dfee0 DW |
4020 | if (sh->check_state == check_state_run) |
4021 | sh->check_state = check_state_run_pq; | |
4022 | else | |
4023 | sh->check_state = check_state_run_q; | |
a4456856 | 4024 | } |
a4456856 | 4025 | |
d82dfee0 DW |
4026 | /* discard potentially stale zero_sum_result */ |
4027 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4028 | |
d82dfee0 DW |
4029 | if (sh->check_state == check_state_run) { |
4030 | /* async_xor_zero_sum destroys the contents of P */ | |
4031 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4032 | s->uptodate--; | |
a4456856 | 4033 | } |
d82dfee0 DW |
4034 | if (sh->check_state >= check_state_run && |
4035 | sh->check_state <= check_state_run_pq) { | |
4036 | /* async_syndrome_zero_sum preserves P and Q, so | |
4037 | * no need to mark them !uptodate here | |
4038 | */ | |
4039 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4040 | break; | |
a4456856 DW |
4041 | } |
4042 | ||
d82dfee0 DW |
4043 | /* we have 2-disk failure */ |
4044 | BUG_ON(s->failed != 2); | |
4045 | /* fall through */ | |
4046 | case check_state_compute_result: | |
4047 | sh->check_state = check_state_idle; | |
a4456856 | 4048 | |
d82dfee0 DW |
4049 | /* check that a write has not made the stripe insync */ |
4050 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4051 | break; | |
a4456856 DW |
4052 | |
4053 | /* now write out any block on a failed drive, | |
d82dfee0 | 4054 | * or P or Q if they were recomputed |
a4456856 | 4055 | */ |
d82dfee0 | 4056 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 4057 | if (s->failed == 2) { |
f2b3b44d | 4058 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4059 | s->locked++; |
4060 | set_bit(R5_LOCKED, &dev->flags); | |
4061 | set_bit(R5_Wantwrite, &dev->flags); | |
4062 | } | |
4063 | if (s->failed >= 1) { | |
f2b3b44d | 4064 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4065 | s->locked++; |
4066 | set_bit(R5_LOCKED, &dev->flags); | |
4067 | set_bit(R5_Wantwrite, &dev->flags); | |
4068 | } | |
d82dfee0 | 4069 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4070 | dev = &sh->dev[pd_idx]; |
4071 | s->locked++; | |
4072 | set_bit(R5_LOCKED, &dev->flags); | |
4073 | set_bit(R5_Wantwrite, &dev->flags); | |
4074 | } | |
d82dfee0 | 4075 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4076 | dev = &sh->dev[qd_idx]; |
4077 | s->locked++; | |
4078 | set_bit(R5_LOCKED, &dev->flags); | |
4079 | set_bit(R5_Wantwrite, &dev->flags); | |
4080 | } | |
4081 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
4082 | ||
4083 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4084 | break; |
4085 | case check_state_run: | |
4086 | case check_state_run_q: | |
4087 | case check_state_run_pq: | |
4088 | break; /* we will be called again upon completion */ | |
4089 | case check_state_check_result: | |
4090 | sh->check_state = check_state_idle; | |
4091 | ||
4092 | /* handle a successful check operation, if parity is correct | |
4093 | * we are done. Otherwise update the mismatch count and repair | |
4094 | * parity if !MD_RECOVERY_CHECK | |
4095 | */ | |
4096 | if (sh->ops.zero_sum_result == 0) { | |
4097 | /* both parities are correct */ | |
4098 | if (!s->failed) | |
4099 | set_bit(STRIPE_INSYNC, &sh->state); | |
4100 | else { | |
4101 | /* in contrast to the raid5 case we can validate | |
4102 | * parity, but still have a failure to write | |
4103 | * back | |
4104 | */ | |
4105 | sh->check_state = check_state_compute_result; | |
4106 | /* Returning at this point means that we may go | |
4107 | * off and bring p and/or q uptodate again so | |
4108 | * we make sure to check zero_sum_result again | |
4109 | * to verify if p or q need writeback | |
4110 | */ | |
4111 | } | |
4112 | } else { | |
7f7583d4 | 4113 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
4114 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
4115 | /* don't try to repair!! */ | |
4116 | set_bit(STRIPE_INSYNC, &sh->state); | |
4117 | else { | |
4118 | int *target = &sh->ops.target; | |
4119 | ||
4120 | sh->ops.target = -1; | |
4121 | sh->ops.target2 = -1; | |
4122 | sh->check_state = check_state_compute_run; | |
4123 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4124 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4125 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4126 | set_bit(R5_Wantcompute, | |
4127 | &sh->dev[pd_idx].flags); | |
4128 | *target = pd_idx; | |
4129 | target = &sh->ops.target2; | |
4130 | s->uptodate++; | |
4131 | } | |
4132 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4133 | set_bit(R5_Wantcompute, | |
4134 | &sh->dev[qd_idx].flags); | |
4135 | *target = qd_idx; | |
4136 | s->uptodate++; | |
4137 | } | |
4138 | } | |
4139 | } | |
4140 | break; | |
4141 | case check_state_compute_run: | |
4142 | break; | |
4143 | default: | |
cc6167b4 N |
4144 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4145 | __func__, sh->check_state, | |
4146 | (unsigned long long) sh->sector); | |
d82dfee0 | 4147 | BUG(); |
a4456856 DW |
4148 | } |
4149 | } | |
4150 | ||
d1688a6d | 4151 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4152 | { |
4153 | int i; | |
4154 | ||
4155 | /* We have read all the blocks in this stripe and now we need to | |
4156 | * copy some of them into a target stripe for expand. | |
4157 | */ | |
f0a50d37 | 4158 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4159 | BUG_ON(sh->batch_head); |
a4456856 DW |
4160 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4161 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4162 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4163 | int dd_idx, j; |
a4456856 | 4164 | struct stripe_head *sh2; |
a08abd8c | 4165 | struct async_submit_ctl submit; |
a4456856 | 4166 | |
6d036f7d | 4167 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4168 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4169 | &dd_idx, NULL); | |
6d036f7d | 4170 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4171 | if (sh2 == NULL) |
4172 | /* so far only the early blocks of this stripe | |
4173 | * have been requested. When later blocks | |
4174 | * get requested, we will try again | |
4175 | */ | |
4176 | continue; | |
4177 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4178 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4179 | /* must have already done this block */ | |
6d036f7d | 4180 | raid5_release_stripe(sh2); |
a4456856 DW |
4181 | continue; |
4182 | } | |
f0a50d37 DW |
4183 | |
4184 | /* place all the copies on one channel */ | |
a08abd8c | 4185 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4186 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 4187 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 4188 | &submit); |
f0a50d37 | 4189 | |
a4456856 DW |
4190 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4191 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4192 | for (j = 0; j < conf->raid_disks; j++) | |
4193 | if (j != sh2->pd_idx && | |
86c374ba | 4194 | j != sh2->qd_idx && |
a4456856 DW |
4195 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4196 | break; | |
4197 | if (j == conf->raid_disks) { | |
4198 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4199 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4200 | } | |
6d036f7d | 4201 | raid5_release_stripe(sh2); |
f0a50d37 | 4202 | |
a4456856 | 4203 | } |
a2e08551 | 4204 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4205 | async_tx_quiesce(&tx); |
a4456856 | 4206 | } |
1da177e4 LT |
4207 | |
4208 | /* | |
4209 | * handle_stripe - do things to a stripe. | |
4210 | * | |
9a3e1101 N |
4211 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4212 | * state of various bits to see what needs to be done. | |
1da177e4 | 4213 | * Possible results: |
9a3e1101 N |
4214 | * return some read requests which now have data |
4215 | * return some write requests which are safely on storage | |
1da177e4 LT |
4216 | * schedule a read on some buffers |
4217 | * schedule a write of some buffers | |
4218 | * return confirmation of parity correctness | |
4219 | * | |
1da177e4 | 4220 | */ |
a4456856 | 4221 | |
acfe726b | 4222 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4223 | { |
d1688a6d | 4224 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4225 | int disks = sh->disks; |
474af965 N |
4226 | struct r5dev *dev; |
4227 | int i; | |
9a3e1101 | 4228 | int do_recovery = 0; |
1da177e4 | 4229 | |
acfe726b N |
4230 | memset(s, 0, sizeof(*s)); |
4231 | ||
dabc4ec6 | 4232 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4233 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4234 | s->failed_num[0] = -1; |
4235 | s->failed_num[1] = -1; | |
6e74a9cf | 4236 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4237 | |
acfe726b | 4238 | /* Now to look around and see what can be done */ |
1da177e4 | 4239 | rcu_read_lock(); |
16a53ecc | 4240 | for (i=disks; i--; ) { |
3cb03002 | 4241 | struct md_rdev *rdev; |
31c176ec N |
4242 | sector_t first_bad; |
4243 | int bad_sectors; | |
4244 | int is_bad = 0; | |
acfe726b | 4245 | |
16a53ecc | 4246 | dev = &sh->dev[i]; |
1da177e4 | 4247 | |
45b4233c | 4248 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4249 | i, dev->flags, |
4250 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4251 | /* maybe we can reply to a read |
4252 | * | |
4253 | * new wantfill requests are only permitted while | |
4254 | * ops_complete_biofill is guaranteed to be inactive | |
4255 | */ | |
4256 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4257 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4258 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4259 | |
16a53ecc | 4260 | /* now count some things */ |
cc94015a N |
4261 | if (test_bit(R5_LOCKED, &dev->flags)) |
4262 | s->locked++; | |
4263 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4264 | s->uptodate++; | |
2d6e4ecc | 4265 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4266 | s->compute++; |
4267 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4268 | } |
1da177e4 | 4269 | |
acfe726b | 4270 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4271 | s->to_fill++; |
acfe726b | 4272 | else if (dev->toread) |
cc94015a | 4273 | s->to_read++; |
16a53ecc | 4274 | if (dev->towrite) { |
cc94015a | 4275 | s->to_write++; |
16a53ecc | 4276 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4277 | s->non_overwrite++; |
16a53ecc | 4278 | } |
a4456856 | 4279 | if (dev->written) |
cc94015a | 4280 | s->written++; |
14a75d3e N |
4281 | /* Prefer to use the replacement for reads, but only |
4282 | * if it is recovered enough and has no bad blocks. | |
4283 | */ | |
4284 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4285 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4286 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4287 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4288 | &first_bad, &bad_sectors)) | |
4289 | set_bit(R5_ReadRepl, &dev->flags); | |
4290 | else { | |
e6030cb0 | 4291 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4292 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4293 | else |
4294 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4295 | rdev = rcu_dereference(conf->disks[i].rdev); |
4296 | clear_bit(R5_ReadRepl, &dev->flags); | |
4297 | } | |
9283d8c5 N |
4298 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4299 | rdev = NULL; | |
31c176ec N |
4300 | if (rdev) { |
4301 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4302 | &first_bad, &bad_sectors); | |
4303 | if (s->blocked_rdev == NULL | |
4304 | && (test_bit(Blocked, &rdev->flags) | |
4305 | || is_bad < 0)) { | |
4306 | if (is_bad < 0) | |
4307 | set_bit(BlockedBadBlocks, | |
4308 | &rdev->flags); | |
4309 | s->blocked_rdev = rdev; | |
4310 | atomic_inc(&rdev->nr_pending); | |
4311 | } | |
6bfe0b49 | 4312 | } |
415e72d0 N |
4313 | clear_bit(R5_Insync, &dev->flags); |
4314 | if (!rdev) | |
4315 | /* Not in-sync */; | |
31c176ec N |
4316 | else if (is_bad) { |
4317 | /* also not in-sync */ | |
18b9837e N |
4318 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4319 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4320 | /* treat as in-sync, but with a read error |
4321 | * which we can now try to correct | |
4322 | */ | |
4323 | set_bit(R5_Insync, &dev->flags); | |
4324 | set_bit(R5_ReadError, &dev->flags); | |
4325 | } | |
4326 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4327 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4328 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4329 | /* in sync if before recovery_offset */ |
30d7a483 N |
4330 | set_bit(R5_Insync, &dev->flags); |
4331 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4332 | test_bit(R5_Expanded, &dev->flags)) | |
4333 | /* If we've reshaped into here, we assume it is Insync. | |
4334 | * We will shortly update recovery_offset to make | |
4335 | * it official. | |
4336 | */ | |
4337 | set_bit(R5_Insync, &dev->flags); | |
4338 | ||
1cc03eb9 | 4339 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4340 | /* This flag does not apply to '.replacement' |
4341 | * only to .rdev, so make sure to check that*/ | |
4342 | struct md_rdev *rdev2 = rcu_dereference( | |
4343 | conf->disks[i].rdev); | |
4344 | if (rdev2 == rdev) | |
4345 | clear_bit(R5_Insync, &dev->flags); | |
4346 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4347 | s->handle_bad_blocks = 1; |
14a75d3e | 4348 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4349 | } else |
4350 | clear_bit(R5_WriteError, &dev->flags); | |
4351 | } | |
1cc03eb9 | 4352 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4353 | /* This flag does not apply to '.replacement' |
4354 | * only to .rdev, so make sure to check that*/ | |
4355 | struct md_rdev *rdev2 = rcu_dereference( | |
4356 | conf->disks[i].rdev); | |
4357 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4358 | s->handle_bad_blocks = 1; |
14a75d3e | 4359 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4360 | } else |
4361 | clear_bit(R5_MadeGood, &dev->flags); | |
4362 | } | |
977df362 N |
4363 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4364 | struct md_rdev *rdev2 = rcu_dereference( | |
4365 | conf->disks[i].replacement); | |
4366 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4367 | s->handle_bad_blocks = 1; | |
4368 | atomic_inc(&rdev2->nr_pending); | |
4369 | } else | |
4370 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4371 | } | |
415e72d0 | 4372 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4373 | /* The ReadError flag will just be confusing now */ |
4374 | clear_bit(R5_ReadError, &dev->flags); | |
4375 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4376 | } |
415e72d0 N |
4377 | if (test_bit(R5_ReadError, &dev->flags)) |
4378 | clear_bit(R5_Insync, &dev->flags); | |
4379 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4380 | if (s->failed < 2) |
4381 | s->failed_num[s->failed] = i; | |
4382 | s->failed++; | |
9a3e1101 N |
4383 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4384 | do_recovery = 1; | |
415e72d0 | 4385 | } |
2ded3703 SL |
4386 | |
4387 | if (test_bit(R5_InJournal, &dev->flags)) | |
4388 | s->injournal++; | |
1e6d690b SL |
4389 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4390 | s->just_cached++; | |
1da177e4 | 4391 | } |
9a3e1101 N |
4392 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4393 | /* If there is a failed device being replaced, | |
4394 | * we must be recovering. | |
4395 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4396 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4397 | * else we can only be replacing |
4398 | * sync and recovery both need to read all devices, and so | |
4399 | * use the same flag. | |
4400 | */ | |
4401 | if (do_recovery || | |
c6d2e084 | 4402 | sh->sector >= conf->mddev->recovery_cp || |
4403 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4404 | s->syncing = 1; |
4405 | else | |
4406 | s->replacing = 1; | |
4407 | } | |
1da177e4 | 4408 | rcu_read_unlock(); |
cc94015a N |
4409 | } |
4410 | ||
59fc630b | 4411 | static int clear_batch_ready(struct stripe_head *sh) |
4412 | { | |
b15a9dbd N |
4413 | /* Return '1' if this is a member of batch, or |
4414 | * '0' if it is a lone stripe or a head which can now be | |
4415 | * handled. | |
4416 | */ | |
59fc630b | 4417 | struct stripe_head *tmp; |
4418 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4419 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4420 | spin_lock(&sh->stripe_lock); |
4421 | if (!sh->batch_head) { | |
4422 | spin_unlock(&sh->stripe_lock); | |
4423 | return 0; | |
4424 | } | |
4425 | ||
4426 | /* | |
4427 | * this stripe could be added to a batch list before we check | |
4428 | * BATCH_READY, skips it | |
4429 | */ | |
4430 | if (sh->batch_head != sh) { | |
4431 | spin_unlock(&sh->stripe_lock); | |
4432 | return 1; | |
4433 | } | |
4434 | spin_lock(&sh->batch_lock); | |
4435 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4436 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4437 | spin_unlock(&sh->batch_lock); | |
4438 | spin_unlock(&sh->stripe_lock); | |
4439 | ||
4440 | /* | |
4441 | * BATCH_READY is cleared, no new stripes can be added. | |
4442 | * batch_list can be accessed without lock | |
4443 | */ | |
4444 | return 0; | |
4445 | } | |
4446 | ||
3960ce79 N |
4447 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4448 | unsigned long handle_flags) | |
72ac7330 | 4449 | { |
4e3d62ff | 4450 | struct stripe_head *sh, *next; |
72ac7330 | 4451 | int i; |
fb642b92 | 4452 | int do_wakeup = 0; |
72ac7330 | 4453 | |
bb27051f N |
4454 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4455 | ||
72ac7330 | 4456 | list_del_init(&sh->batch_list); |
4457 | ||
fb3229d5 | 4458 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4459 | (1 << STRIPE_SYNCING) | |
4460 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4461 | (1 << STRIPE_DELAYED) | |
4462 | (1 << STRIPE_BIT_DELAY) | | |
4463 | (1 << STRIPE_FULL_WRITE) | | |
4464 | (1 << STRIPE_BIOFILL_RUN) | | |
4465 | (1 << STRIPE_COMPUTE_RUN) | | |
4466 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4467 | (1 << STRIPE_DISCARD) | | |
4468 | (1 << STRIPE_BATCH_READY) | | |
4469 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4470 | (1 << STRIPE_BITMAP_PENDING)), |
4471 | "stripe state: %lx\n", sh->state); | |
4472 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4473 | (1 << STRIPE_REPLACED)), | |
4474 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4475 | |
4476 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4477 | (1 << STRIPE_PREREAD_ACTIVE) | |
1b956f7a N |
4478 | (1 << STRIPE_DEGRADED)), |
4479 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4480 | ||
72ac7330 | 4481 | sh->check_state = head_sh->check_state; |
4482 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4483 | for (i = 0; i < sh->disks; i++) { |
4484 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4485 | do_wakeup = 1; | |
72ac7330 | 4486 | sh->dev[i].flags = head_sh->dev[i].flags & |
4487 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4488 | } |
72ac7330 | 4489 | spin_lock_irq(&sh->stripe_lock); |
4490 | sh->batch_head = NULL; | |
4491 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4492 | if (handle_flags == 0 || |
4493 | sh->state & handle_flags) | |
4494 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4495 | raid5_release_stripe(sh); |
72ac7330 | 4496 | } |
fb642b92 N |
4497 | spin_lock_irq(&head_sh->stripe_lock); |
4498 | head_sh->batch_head = NULL; | |
4499 | spin_unlock_irq(&head_sh->stripe_lock); | |
4500 | for (i = 0; i < head_sh->disks; i++) | |
4501 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4502 | do_wakeup = 1; | |
3960ce79 N |
4503 | if (head_sh->state & handle_flags) |
4504 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4505 | |
4506 | if (do_wakeup) | |
4507 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4508 | } |
4509 | ||
cc94015a N |
4510 | static void handle_stripe(struct stripe_head *sh) |
4511 | { | |
4512 | struct stripe_head_state s; | |
d1688a6d | 4513 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4514 | int i; |
84789554 N |
4515 | int prexor; |
4516 | int disks = sh->disks; | |
474af965 | 4517 | struct r5dev *pdev, *qdev; |
cc94015a N |
4518 | |
4519 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4520 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4521 | /* already being handled, ensure it gets handled |
4522 | * again when current action finishes */ | |
4523 | set_bit(STRIPE_HANDLE, &sh->state); | |
4524 | return; | |
4525 | } | |
4526 | ||
59fc630b | 4527 | if (clear_batch_ready(sh) ) { |
4528 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4529 | return; | |
4530 | } | |
4531 | ||
4e3d62ff | 4532 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4533 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4534 | |
dabc4ec6 | 4535 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4536 | spin_lock(&sh->stripe_lock); |
4537 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4538 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4539 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4540 | set_bit(STRIPE_SYNCING, &sh->state); | |
4541 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4542 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4543 | } |
4544 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4545 | } |
4546 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4547 | ||
4548 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4549 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4550 | (unsigned long long)sh->sector, sh->state, | |
4551 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4552 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4553 | |
acfe726b | 4554 | analyse_stripe(sh, &s); |
c5a31000 | 4555 | |
b70abcb2 SL |
4556 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4557 | goto finish; | |
4558 | ||
bc2607f3 N |
4559 | if (s.handle_bad_blocks) { |
4560 | set_bit(STRIPE_HANDLE, &sh->state); | |
4561 | goto finish; | |
4562 | } | |
4563 | ||
474af965 N |
4564 | if (unlikely(s.blocked_rdev)) { |
4565 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4566 | s.replacing || s.to_write || s.written) { |
474af965 N |
4567 | set_bit(STRIPE_HANDLE, &sh->state); |
4568 | goto finish; | |
4569 | } | |
4570 | /* There is nothing for the blocked_rdev to block */ | |
4571 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4572 | s.blocked_rdev = NULL; | |
4573 | } | |
4574 | ||
4575 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4576 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4577 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4578 | } | |
4579 | ||
4580 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4581 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4582 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4583 | s.failed_num[0], s.failed_num[1]); | |
4584 | /* check if the array has lost more than max_degraded devices and, | |
4585 | * if so, some requests might need to be failed. | |
4586 | */ | |
6e74a9cf | 4587 | if (s.failed > conf->max_degraded || s.log_failed) { |
9a3f530f N |
4588 | sh->check_state = 0; |
4589 | sh->reconstruct_state = 0; | |
626f2092 | 4590 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4591 | if (s.to_read+s.to_write+s.written) |
4592 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4593 | if (s.syncing + s.replacing) |
9a3f530f N |
4594 | handle_failed_sync(conf, sh, &s); |
4595 | } | |
474af965 | 4596 | |
84789554 N |
4597 | /* Now we check to see if any write operations have recently |
4598 | * completed | |
4599 | */ | |
4600 | prexor = 0; | |
4601 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4602 | prexor = 1; | |
4603 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4604 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4605 | sh->reconstruct_state = reconstruct_state_idle; | |
4606 | ||
4607 | /* All the 'written' buffers and the parity block are ready to | |
4608 | * be written back to disk | |
4609 | */ | |
9e444768 SL |
4610 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4611 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4612 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4613 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4614 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4615 | for (i = disks; i--; ) { |
4616 | struct r5dev *dev = &sh->dev[i]; | |
4617 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4618 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4619 | dev->written || test_bit(R5_InJournal, |
4620 | &dev->flags))) { | |
84789554 N |
4621 | pr_debug("Writing block %d\n", i); |
4622 | set_bit(R5_Wantwrite, &dev->flags); | |
4623 | if (prexor) | |
4624 | continue; | |
9c4bdf69 N |
4625 | if (s.failed > 1) |
4626 | continue; | |
84789554 N |
4627 | if (!test_bit(R5_Insync, &dev->flags) || |
4628 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4629 | s.failed == 0)) | |
4630 | set_bit(STRIPE_INSYNC, &sh->state); | |
4631 | } | |
4632 | } | |
4633 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4634 | s.dec_preread_active = 1; | |
4635 | } | |
4636 | ||
ef5b7c69 N |
4637 | /* |
4638 | * might be able to return some write requests if the parity blocks | |
4639 | * are safe, or on a failed drive | |
4640 | */ | |
4641 | pdev = &sh->dev[sh->pd_idx]; | |
4642 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4643 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4644 | qdev = &sh->dev[sh->qd_idx]; | |
4645 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4646 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4647 | || conf->level < 6; | |
4648 | ||
4649 | if (s.written && | |
4650 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4651 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4652 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4653 | test_bit(R5_Discard, &pdev->flags))))) && | |
4654 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4655 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4656 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4657 | test_bit(R5_Discard, &qdev->flags)))))) | |
4658 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4659 | ||
1e6d690b SL |
4660 | if (s.just_cached) |
4661 | r5c_handle_cached_data_endio(conf, sh, disks, &s.return_bi); | |
4662 | r5l_stripe_write_finished(sh); | |
4663 | ||
ef5b7c69 N |
4664 | /* Now we might consider reading some blocks, either to check/generate |
4665 | * parity, or to satisfy requests | |
4666 | * or to load a block that is being partially written. | |
4667 | */ | |
4668 | if (s.to_read || s.non_overwrite | |
4669 | || (conf->level == 6 && s.to_write && s.failed) | |
4670 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4671 | || s.replacing | |
4672 | || s.expanding) | |
4673 | handle_stripe_fill(sh, &s, disks); | |
4674 | ||
2ded3703 SL |
4675 | /* |
4676 | * When the stripe finishes full journal write cycle (write to journal | |
4677 | * and raid disk), this is the clean up procedure so it is ready for | |
4678 | * next operation. | |
4679 | */ | |
4680 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4681 | ||
4682 | /* | |
4683 | * Now to consider new write requests, cache write back and what else, | |
4684 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4685 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4686 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4687 | * block. | |
2ded3703 | 4688 | * 3/ A r5c cache log write is in flight. |
84789554 | 4689 | */ |
2ded3703 SL |
4690 | |
4691 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4692 | if (!r5c_is_writeback(conf->log)) { | |
4693 | if (s.to_write) | |
4694 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4695 | } else { /* write back cache */ | |
4696 | int ret = 0; | |
4697 | ||
4698 | /* First, try handle writes in caching phase */ | |
4699 | if (s.to_write) | |
4700 | ret = r5c_try_caching_write(conf, sh, &s, | |
4701 | disks); | |
4702 | /* | |
4703 | * If caching phase failed: ret == -EAGAIN | |
4704 | * OR | |
4705 | * stripe under reclaim: !caching && injournal | |
4706 | * | |
4707 | * fall back to handle_stripe_dirtying() | |
4708 | */ | |
4709 | if (ret == -EAGAIN || | |
4710 | /* stripe under reclaim: !caching && injournal */ | |
4711 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4712 | s.injournal > 0)) { |
4713 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4714 | disks); | |
4715 | if (ret == -EAGAIN) | |
4716 | goto finish; | |
4717 | } | |
2ded3703 SL |
4718 | } |
4719 | } | |
84789554 N |
4720 | |
4721 | /* maybe we need to check and possibly fix the parity for this stripe | |
4722 | * Any reads will already have been scheduled, so we just see if enough | |
4723 | * data is available. The parity check is held off while parity | |
4724 | * dependent operations are in flight. | |
4725 | */ | |
4726 | if (sh->check_state || | |
4727 | (s.syncing && s.locked == 0 && | |
4728 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4729 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4730 | if (conf->level == 6) | |
4731 | handle_parity_checks6(conf, sh, &s, disks); | |
4732 | else | |
4733 | handle_parity_checks5(conf, sh, &s, disks); | |
4734 | } | |
c5a31000 | 4735 | |
f94c0b66 N |
4736 | if ((s.replacing || s.syncing) && s.locked == 0 |
4737 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4738 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4739 | /* Write out to replacement devices where possible */ |
4740 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4741 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4742 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4743 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4744 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4745 | s.locked++; | |
4746 | } | |
f94c0b66 N |
4747 | if (s.replacing) |
4748 | set_bit(STRIPE_INSYNC, &sh->state); | |
4749 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4750 | } |
4751 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4752 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4753 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4754 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4755 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4756 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4757 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4758 | } |
4759 | ||
4760 | /* If the failed drives are just a ReadError, then we might need | |
4761 | * to progress the repair/check process | |
4762 | */ | |
4763 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4764 | for (i = 0; i < s.failed; i++) { | |
4765 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4766 | if (test_bit(R5_ReadError, &dev->flags) | |
4767 | && !test_bit(R5_LOCKED, &dev->flags) | |
4768 | && test_bit(R5_UPTODATE, &dev->flags) | |
4769 | ) { | |
4770 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4771 | set_bit(R5_Wantwrite, &dev->flags); | |
4772 | set_bit(R5_ReWrite, &dev->flags); | |
4773 | set_bit(R5_LOCKED, &dev->flags); | |
4774 | s.locked++; | |
4775 | } else { | |
4776 | /* let's read it back */ | |
4777 | set_bit(R5_Wantread, &dev->flags); | |
4778 | set_bit(R5_LOCKED, &dev->flags); | |
4779 | s.locked++; | |
4780 | } | |
4781 | } | |
4782 | } | |
4783 | ||
3687c061 N |
4784 | /* Finish reconstruct operations initiated by the expansion process */ |
4785 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4786 | struct stripe_head *sh_src | |
6d036f7d | 4787 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4788 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4789 | /* sh cannot be written until sh_src has been read. | |
4790 | * so arrange for sh to be delayed a little | |
4791 | */ | |
4792 | set_bit(STRIPE_DELAYED, &sh->state); | |
4793 | set_bit(STRIPE_HANDLE, &sh->state); | |
4794 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4795 | &sh_src->state)) | |
4796 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4797 | raid5_release_stripe(sh_src); |
3687c061 N |
4798 | goto finish; |
4799 | } | |
4800 | if (sh_src) | |
6d036f7d | 4801 | raid5_release_stripe(sh_src); |
3687c061 N |
4802 | |
4803 | sh->reconstruct_state = reconstruct_state_idle; | |
4804 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4805 | for (i = conf->raid_disks; i--; ) { | |
4806 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4807 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4808 | s.locked++; | |
4809 | } | |
4810 | } | |
f416885e | 4811 | |
3687c061 N |
4812 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4813 | !sh->reconstruct_state) { | |
4814 | /* Need to write out all blocks after computing parity */ | |
4815 | sh->disks = conf->raid_disks; | |
4816 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4817 | schedule_reconstruction(sh, &s, 1, 1); | |
4818 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4819 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4820 | atomic_dec(&conf->reshape_stripes); | |
4821 | wake_up(&conf->wait_for_overlap); | |
4822 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4823 | } | |
4824 | ||
4825 | if (s.expanding && s.locked == 0 && | |
4826 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4827 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4828 | |
3687c061 | 4829 | finish: |
6bfe0b49 | 4830 | /* wait for this device to become unblocked */ |
5f066c63 N |
4831 | if (unlikely(s.blocked_rdev)) { |
4832 | if (conf->mddev->external) | |
4833 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4834 | conf->mddev); | |
4835 | else | |
4836 | /* Internal metadata will immediately | |
4837 | * be written by raid5d, so we don't | |
4838 | * need to wait here. | |
4839 | */ | |
4840 | rdev_dec_pending(s.blocked_rdev, | |
4841 | conf->mddev); | |
4842 | } | |
6bfe0b49 | 4843 | |
bc2607f3 N |
4844 | if (s.handle_bad_blocks) |
4845 | for (i = disks; i--; ) { | |
3cb03002 | 4846 | struct md_rdev *rdev; |
bc2607f3 N |
4847 | struct r5dev *dev = &sh->dev[i]; |
4848 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4849 | /* We own a safe reference to the rdev */ | |
4850 | rdev = conf->disks[i].rdev; | |
4851 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4852 | STRIPE_SECTORS, 0)) | |
4853 | md_error(conf->mddev, rdev); | |
4854 | rdev_dec_pending(rdev, conf->mddev); | |
4855 | } | |
b84db560 N |
4856 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4857 | rdev = conf->disks[i].rdev; | |
4858 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4859 | STRIPE_SECTORS, 0); |
b84db560 N |
4860 | rdev_dec_pending(rdev, conf->mddev); |
4861 | } | |
977df362 N |
4862 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4863 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4864 | if (!rdev) |
4865 | /* rdev have been moved down */ | |
4866 | rdev = conf->disks[i].rdev; | |
977df362 | 4867 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4868 | STRIPE_SECTORS, 0); |
977df362 N |
4869 | rdev_dec_pending(rdev, conf->mddev); |
4870 | } | |
bc2607f3 N |
4871 | } |
4872 | ||
6c0069c0 YT |
4873 | if (s.ops_request) |
4874 | raid_run_ops(sh, s.ops_request); | |
4875 | ||
f0e43bcd | 4876 | ops_run_io(sh, &s); |
16a53ecc | 4877 | |
c5709ef6 | 4878 | if (s.dec_preread_active) { |
729a1866 | 4879 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4880 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4881 | * have actually been submitted. |
4882 | */ | |
4883 | atomic_dec(&conf->preread_active_stripes); | |
4884 | if (atomic_read(&conf->preread_active_stripes) < | |
4885 | IO_THRESHOLD) | |
4886 | md_wakeup_thread(conf->mddev->thread); | |
4887 | } | |
4888 | ||
c3cce6cd | 4889 | if (!bio_list_empty(&s.return_bi)) { |
2953079c | 4890 | if (test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { |
c3cce6cd N |
4891 | spin_lock_irq(&conf->device_lock); |
4892 | bio_list_merge(&conf->return_bi, &s.return_bi); | |
4893 | spin_unlock_irq(&conf->device_lock); | |
4894 | md_wakeup_thread(conf->mddev->thread); | |
4895 | } else | |
4896 | return_io(&s.return_bi); | |
4897 | } | |
16a53ecc | 4898 | |
257a4b42 | 4899 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4900 | } |
4901 | ||
d1688a6d | 4902 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4903 | { |
4904 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4905 | while (!list_empty(&conf->delayed_list)) { | |
4906 | struct list_head *l = conf->delayed_list.next; | |
4907 | struct stripe_head *sh; | |
4908 | sh = list_entry(l, struct stripe_head, lru); | |
4909 | list_del_init(l); | |
4910 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4911 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4912 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4913 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4914 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4915 | } |
482c0834 | 4916 | } |
16a53ecc N |
4917 | } |
4918 | ||
566c09c5 SL |
4919 | static void activate_bit_delay(struct r5conf *conf, |
4920 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4921 | { |
4922 | /* device_lock is held */ | |
4923 | struct list_head head; | |
4924 | list_add(&head, &conf->bitmap_list); | |
4925 | list_del_init(&conf->bitmap_list); | |
4926 | while (!list_empty(&head)) { | |
4927 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4928 | int hash; |
16a53ecc N |
4929 | list_del_init(&sh->lru); |
4930 | atomic_inc(&sh->count); | |
566c09c5 SL |
4931 | hash = sh->hash_lock_index; |
4932 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4933 | } |
4934 | } | |
4935 | ||
5c675f83 | 4936 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4937 | { |
d1688a6d | 4938 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4939 | |
4940 | /* No difference between reads and writes. Just check | |
4941 | * how busy the stripe_cache is | |
4942 | */ | |
3fa841d7 | 4943 | |
5423399a | 4944 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd | 4945 | return 1; |
a39f7afd SL |
4946 | |
4947 | /* Also checks whether there is pressure on r5cache log space */ | |
4948 | if (test_bit(R5C_LOG_TIGHT, &conf->cache_state)) | |
4949 | return 1; | |
f022b2fd N |
4950 | if (conf->quiesce) |
4951 | return 1; | |
4bda556a | 4952 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4953 | return 1; |
4954 | ||
4955 | return 0; | |
4956 | } | |
4957 | ||
fd01b88c | 4958 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4959 | { |
3cb5edf4 | 4960 | struct r5conf *conf = mddev->private; |
4f024f37 | 4961 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 4962 | unsigned int chunk_sectors; |
aa8b57aa | 4963 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4964 | |
3cb5edf4 | 4965 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
4966 | return chunk_sectors >= |
4967 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4968 | } | |
4969 | ||
46031f9a RBJ |
4970 | /* |
4971 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4972 | * later sampled by raid5d. | |
4973 | */ | |
d1688a6d | 4974 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4975 | { |
4976 | unsigned long flags; | |
4977 | ||
4978 | spin_lock_irqsave(&conf->device_lock, flags); | |
4979 | ||
4980 | bi->bi_next = conf->retry_read_aligned_list; | |
4981 | conf->retry_read_aligned_list = bi; | |
4982 | ||
4983 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4984 | md_wakeup_thread(conf->mddev->thread); | |
4985 | } | |
4986 | ||
d1688a6d | 4987 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
4988 | { |
4989 | struct bio *bi; | |
4990 | ||
4991 | bi = conf->retry_read_aligned; | |
4992 | if (bi) { | |
4993 | conf->retry_read_aligned = NULL; | |
4994 | return bi; | |
4995 | } | |
4996 | bi = conf->retry_read_aligned_list; | |
4997 | if(bi) { | |
387bb173 | 4998 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 4999 | bi->bi_next = NULL; |
960e739d JA |
5000 | /* |
5001 | * this sets the active strip count to 1 and the processed | |
5002 | * strip count to zero (upper 8 bits) | |
5003 | */ | |
e7836bd6 | 5004 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
5005 | } |
5006 | ||
5007 | return bi; | |
5008 | } | |
5009 | ||
f679623f RBJ |
5010 | /* |
5011 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5012 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5013 | * first). | |
5014 | * If the read failed.. | |
5015 | */ | |
4246a0b6 | 5016 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5017 | { |
5018 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5019 | struct mddev *mddev; |
d1688a6d | 5020 | struct r5conf *conf; |
3cb03002 | 5021 | struct md_rdev *rdev; |
9b81c842 | 5022 | int error = bi->bi_error; |
46031f9a | 5023 | |
f679623f | 5024 | bio_put(bi); |
46031f9a | 5025 | |
46031f9a RBJ |
5026 | rdev = (void*)raid_bi->bi_next; |
5027 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5028 | mddev = rdev->mddev; |
5029 | conf = mddev->private; | |
46031f9a RBJ |
5030 | |
5031 | rdev_dec_pending(rdev, conf->mddev); | |
5032 | ||
9b81c842 | 5033 | if (!error) { |
4246a0b6 | 5034 | bio_endio(raid_bi); |
46031f9a | 5035 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5036 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5037 | return; |
46031f9a RBJ |
5038 | } |
5039 | ||
45b4233c | 5040 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5041 | |
5042 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5043 | } |
5044 | ||
7ef6b12a | 5045 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5046 | { |
d1688a6d | 5047 | struct r5conf *conf = mddev->private; |
8553fe7e | 5048 | int dd_idx; |
f679623f | 5049 | struct bio* align_bi; |
3cb03002 | 5050 | struct md_rdev *rdev; |
671488cc | 5051 | sector_t end_sector; |
f679623f RBJ |
5052 | |
5053 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5054 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5055 | return 0; |
5056 | } | |
5057 | /* | |
d7a10308 | 5058 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5059 | */ |
d7a10308 | 5060 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, mddev->bio_set); |
f679623f RBJ |
5061 | if (!align_bi) |
5062 | return 0; | |
5063 | /* | |
5064 | * set bi_end_io to a new function, and set bi_private to the | |
5065 | * original bio. | |
5066 | */ | |
5067 | align_bi->bi_end_io = raid5_align_endio; | |
5068 | align_bi->bi_private = raid_bio; | |
5069 | /* | |
5070 | * compute position | |
5071 | */ | |
4f024f37 KO |
5072 | align_bi->bi_iter.bi_sector = |
5073 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5074 | 0, &dd_idx, NULL); | |
f679623f | 5075 | |
f73a1c7d | 5076 | end_sector = bio_end_sector(align_bi); |
f679623f | 5077 | rcu_read_lock(); |
671488cc N |
5078 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5079 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5080 | rdev->recovery_offset < end_sector) { | |
5081 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5082 | if (rdev && | |
5083 | (test_bit(Faulty, &rdev->flags) || | |
5084 | !(test_bit(In_sync, &rdev->flags) || | |
5085 | rdev->recovery_offset >= end_sector))) | |
5086 | rdev = NULL; | |
5087 | } | |
03b047f4 SL |
5088 | |
5089 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5090 | rcu_read_unlock(); | |
5091 | bio_put(align_bi); | |
5092 | return 0; | |
5093 | } | |
5094 | ||
671488cc | 5095 | if (rdev) { |
31c176ec N |
5096 | sector_t first_bad; |
5097 | int bad_sectors; | |
5098 | ||
f679623f RBJ |
5099 | atomic_inc(&rdev->nr_pending); |
5100 | rcu_read_unlock(); | |
46031f9a RBJ |
5101 | raid_bio->bi_next = (void*)rdev; |
5102 | align_bi->bi_bdev = rdev->bdev; | |
b7c44ed9 | 5103 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 5104 | |
7140aafc | 5105 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5106 | bio_sectors(align_bi), |
31c176ec | 5107 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5108 | bio_put(align_bi); |
5109 | rdev_dec_pending(rdev, mddev); | |
5110 | return 0; | |
5111 | } | |
5112 | ||
6c0544e2 | 5113 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5114 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5115 | |
46031f9a | 5116 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5117 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5118 | conf->quiesce == 0, |
eed8c02e | 5119 | conf->device_lock); |
46031f9a RBJ |
5120 | atomic_inc(&conf->active_aligned_reads); |
5121 | spin_unlock_irq(&conf->device_lock); | |
5122 | ||
e3620a3a JB |
5123 | if (mddev->gendisk) |
5124 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
5125 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 5126 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
5127 | generic_make_request(align_bi); |
5128 | return 1; | |
5129 | } else { | |
5130 | rcu_read_unlock(); | |
46031f9a | 5131 | bio_put(align_bi); |
f679623f RBJ |
5132 | return 0; |
5133 | } | |
5134 | } | |
5135 | ||
7ef6b12a ML |
5136 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5137 | { | |
5138 | struct bio *split; | |
5139 | ||
5140 | do { | |
5141 | sector_t sector = raid_bio->bi_iter.bi_sector; | |
5142 | unsigned chunk_sects = mddev->chunk_sectors; | |
5143 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
5144 | ||
5145 | if (sectors < bio_sectors(raid_bio)) { | |
5146 | split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set); | |
5147 | bio_chain(split, raid_bio); | |
5148 | } else | |
5149 | split = raid_bio; | |
5150 | ||
5151 | if (!raid5_read_one_chunk(mddev, split)) { | |
5152 | if (split != raid_bio) | |
5153 | generic_make_request(raid_bio); | |
5154 | return split; | |
5155 | } | |
5156 | } while (split != raid_bio); | |
5157 | ||
5158 | return NULL; | |
5159 | } | |
5160 | ||
8b3e6cdc DW |
5161 | /* __get_priority_stripe - get the next stripe to process |
5162 | * | |
5163 | * Full stripe writes are allowed to pass preread active stripes up until | |
5164 | * the bypass_threshold is exceeded. In general the bypass_count | |
5165 | * increments when the handle_list is handled before the hold_list; however, it | |
5166 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5167 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5168 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5169 | * handle_list. | |
5170 | */ | |
851c30c9 | 5171 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5172 | { |
851c30c9 SL |
5173 | struct stripe_head *sh = NULL, *tmp; |
5174 | struct list_head *handle_list = NULL; | |
bfc90cb0 | 5175 | struct r5worker_group *wg = NULL; |
851c30c9 SL |
5176 | |
5177 | if (conf->worker_cnt_per_group == 0) { | |
5178 | handle_list = &conf->handle_list; | |
5179 | } else if (group != ANY_GROUP) { | |
5180 | handle_list = &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5181 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5182 | } else { |
5183 | int i; | |
5184 | for (i = 0; i < conf->group_cnt; i++) { | |
5185 | handle_list = &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5186 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5187 | if (!list_empty(handle_list)) |
5188 | break; | |
5189 | } | |
5190 | } | |
8b3e6cdc DW |
5191 | |
5192 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5193 | __func__, | |
851c30c9 | 5194 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5195 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5196 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5197 | ||
851c30c9 SL |
5198 | if (!list_empty(handle_list)) { |
5199 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5200 | |
5201 | if (list_empty(&conf->hold_list)) | |
5202 | conf->bypass_count = 0; | |
5203 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5204 | if (conf->hold_list.next == conf->last_hold) | |
5205 | conf->bypass_count++; | |
5206 | else { | |
5207 | conf->last_hold = conf->hold_list.next; | |
5208 | conf->bypass_count -= conf->bypass_threshold; | |
5209 | if (conf->bypass_count < 0) | |
5210 | conf->bypass_count = 0; | |
5211 | } | |
5212 | } | |
5213 | } else if (!list_empty(&conf->hold_list) && | |
5214 | ((conf->bypass_threshold && | |
5215 | conf->bypass_count > conf->bypass_threshold) || | |
5216 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5217 | |
5218 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5219 | if (conf->worker_cnt_per_group == 0 || | |
5220 | group == ANY_GROUP || | |
5221 | !cpu_online(tmp->cpu) || | |
5222 | cpu_to_group(tmp->cpu) == group) { | |
5223 | sh = tmp; | |
5224 | break; | |
5225 | } | |
5226 | } | |
5227 | ||
5228 | if (sh) { | |
5229 | conf->bypass_count -= conf->bypass_threshold; | |
5230 | if (conf->bypass_count < 0) | |
5231 | conf->bypass_count = 0; | |
5232 | } | |
bfc90cb0 | 5233 | wg = NULL; |
851c30c9 SL |
5234 | } |
5235 | ||
5236 | if (!sh) | |
8b3e6cdc DW |
5237 | return NULL; |
5238 | ||
bfc90cb0 SL |
5239 | if (wg) { |
5240 | wg->stripes_cnt--; | |
5241 | sh->group = NULL; | |
5242 | } | |
8b3e6cdc | 5243 | list_del_init(&sh->lru); |
c7a6d35e | 5244 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5245 | return sh; |
5246 | } | |
f679623f | 5247 | |
8811b596 SL |
5248 | struct raid5_plug_cb { |
5249 | struct blk_plug_cb cb; | |
5250 | struct list_head list; | |
566c09c5 | 5251 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5252 | }; |
5253 | ||
5254 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5255 | { | |
5256 | struct raid5_plug_cb *cb = container_of( | |
5257 | blk_cb, struct raid5_plug_cb, cb); | |
5258 | struct stripe_head *sh; | |
5259 | struct mddev *mddev = cb->cb.data; | |
5260 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5261 | int cnt = 0; |
566c09c5 | 5262 | int hash; |
8811b596 SL |
5263 | |
5264 | if (cb->list.next && !list_empty(&cb->list)) { | |
5265 | spin_lock_irq(&conf->device_lock); | |
5266 | while (!list_empty(&cb->list)) { | |
5267 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5268 | list_del_init(&sh->lru); | |
5269 | /* | |
5270 | * avoid race release_stripe_plug() sees | |
5271 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5272 | * is still in our list | |
5273 | */ | |
4e857c58 | 5274 | smp_mb__before_atomic(); |
8811b596 | 5275 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5276 | /* |
5277 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5278 | * case, the count is always > 1 here | |
5279 | */ | |
566c09c5 SL |
5280 | hash = sh->hash_lock_index; |
5281 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5282 | cnt++; |
8811b596 SL |
5283 | } |
5284 | spin_unlock_irq(&conf->device_lock); | |
5285 | } | |
566c09c5 SL |
5286 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5287 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5288 | if (mddev->queue) |
5289 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5290 | kfree(cb); |
5291 | } | |
5292 | ||
5293 | static void release_stripe_plug(struct mddev *mddev, | |
5294 | struct stripe_head *sh) | |
5295 | { | |
5296 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5297 | raid5_unplug, mddev, | |
5298 | sizeof(struct raid5_plug_cb)); | |
5299 | struct raid5_plug_cb *cb; | |
5300 | ||
5301 | if (!blk_cb) { | |
6d036f7d | 5302 | raid5_release_stripe(sh); |
8811b596 SL |
5303 | return; |
5304 | } | |
5305 | ||
5306 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5307 | ||
566c09c5 SL |
5308 | if (cb->list.next == NULL) { |
5309 | int i; | |
8811b596 | 5310 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5311 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5312 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5313 | } | |
8811b596 SL |
5314 | |
5315 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5316 | list_add_tail(&sh->lru, &cb->list); | |
5317 | else | |
6d036f7d | 5318 | raid5_release_stripe(sh); |
8811b596 SL |
5319 | } |
5320 | ||
620125f2 SL |
5321 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5322 | { | |
5323 | struct r5conf *conf = mddev->private; | |
5324 | sector_t logical_sector, last_sector; | |
5325 | struct stripe_head *sh; | |
5326 | int remaining; | |
5327 | int stripe_sectors; | |
5328 | ||
5329 | if (mddev->reshape_position != MaxSector) | |
5330 | /* Skip discard while reshape is happening */ | |
5331 | return; | |
5332 | ||
4f024f37 KO |
5333 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5334 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5335 | |
5336 | bi->bi_next = NULL; | |
5337 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5338 | ||
5339 | stripe_sectors = conf->chunk_sectors * | |
5340 | (conf->raid_disks - conf->max_degraded); | |
5341 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5342 | stripe_sectors); | |
5343 | sector_div(last_sector, stripe_sectors); | |
5344 | ||
5345 | logical_sector *= conf->chunk_sectors; | |
5346 | last_sector *= conf->chunk_sectors; | |
5347 | ||
5348 | for (; logical_sector < last_sector; | |
5349 | logical_sector += STRIPE_SECTORS) { | |
5350 | DEFINE_WAIT(w); | |
5351 | int d; | |
5352 | again: | |
6d036f7d | 5353 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5354 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5355 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5356 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5357 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5358 | raid5_release_stripe(sh); |
f8dfcffd N |
5359 | schedule(); |
5360 | goto again; | |
5361 | } | |
5362 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5363 | spin_lock_irq(&sh->stripe_lock); |
5364 | for (d = 0; d < conf->raid_disks; d++) { | |
5365 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5366 | continue; | |
5367 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5368 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5369 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5370 | raid5_release_stripe(sh); |
620125f2 SL |
5371 | schedule(); |
5372 | goto again; | |
5373 | } | |
5374 | } | |
f8dfcffd | 5375 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5376 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5377 | sh->overwrite_disks = 0; |
620125f2 SL |
5378 | for (d = 0; d < conf->raid_disks; d++) { |
5379 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5380 | continue; | |
5381 | sh->dev[d].towrite = bi; | |
5382 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5383 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5384 | sh->overwrite_disks++; |
620125f2 SL |
5385 | } |
5386 | spin_unlock_irq(&sh->stripe_lock); | |
5387 | if (conf->mddev->bitmap) { | |
5388 | for (d = 0; | |
5389 | d < conf->raid_disks - conf->max_degraded; | |
5390 | d++) | |
5391 | bitmap_startwrite(mddev->bitmap, | |
5392 | sh->sector, | |
5393 | STRIPE_SECTORS, | |
5394 | 0); | |
5395 | sh->bm_seq = conf->seq_flush + 1; | |
5396 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5397 | } | |
5398 | ||
5399 | set_bit(STRIPE_HANDLE, &sh->state); | |
5400 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5401 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5402 | atomic_inc(&conf->preread_active_stripes); | |
5403 | release_stripe_plug(mddev, sh); | |
5404 | } | |
5405 | ||
5406 | remaining = raid5_dec_bi_active_stripes(bi); | |
5407 | if (remaining == 0) { | |
5408 | md_write_end(mddev); | |
4246a0b6 | 5409 | bio_endio(bi); |
620125f2 SL |
5410 | } |
5411 | } | |
5412 | ||
849674e4 | 5413 | static void raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5414 | { |
d1688a6d | 5415 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5416 | int dd_idx; |
1da177e4 LT |
5417 | sector_t new_sector; |
5418 | sector_t logical_sector, last_sector; | |
5419 | struct stripe_head *sh; | |
a362357b | 5420 | const int rw = bio_data_dir(bi); |
49077326 | 5421 | int remaining; |
27c0f68f SL |
5422 | DEFINE_WAIT(w); |
5423 | bool do_prepare; | |
3bddb7f8 | 5424 | bool do_flush = false; |
1da177e4 | 5425 | |
1eff9d32 | 5426 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
828cbe98 SL |
5427 | int ret = r5l_handle_flush_request(conf->log, bi); |
5428 | ||
5429 | if (ret == 0) | |
5430 | return; | |
5431 | if (ret == -ENODEV) { | |
5432 | md_flush_request(mddev, bi); | |
5433 | return; | |
5434 | } | |
5435 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5436 | /* |
5437 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5438 | * we need to flush journal device | |
5439 | */ | |
5440 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5441 | } |
5442 | ||
3d310eb7 | 5443 | md_write_start(mddev, bi); |
06d91a5f | 5444 | |
9ffc8f7c EM |
5445 | /* |
5446 | * If array is degraded, better not do chunk aligned read because | |
5447 | * later we might have to read it again in order to reconstruct | |
5448 | * data on failed drives. | |
5449 | */ | |
5450 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5451 | mddev->reshape_position == MaxSector) { |
5452 | bi = chunk_aligned_read(mddev, bi); | |
5453 | if (!bi) | |
5454 | return; | |
5455 | } | |
52488615 | 5456 | |
796a5cf0 | 5457 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 SL |
5458 | make_discard_request(mddev, bi); |
5459 | return; | |
5460 | } | |
5461 | ||
4f024f37 | 5462 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5463 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5464 | bi->bi_next = NULL; |
5465 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5466 | |
27c0f68f | 5467 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5468 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5469 | int previous; |
c46501b2 | 5470 | int seq; |
b578d55f | 5471 | |
27c0f68f | 5472 | do_prepare = false; |
7ecaa1e6 | 5473 | retry: |
c46501b2 | 5474 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5475 | previous = 0; |
27c0f68f SL |
5476 | if (do_prepare) |
5477 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5478 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5479 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5480 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5481 | * 64bit on a 32bit platform, and so it might be |
5482 | * possible to see a half-updated value | |
aeb878b0 | 5483 | * Of course reshape_progress could change after |
df8e7f76 N |
5484 | * the lock is dropped, so once we get a reference |
5485 | * to the stripe that we think it is, we will have | |
5486 | * to check again. | |
5487 | */ | |
7ecaa1e6 | 5488 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5489 | if (mddev->reshape_backwards |
fef9c61f N |
5490 | ? logical_sector < conf->reshape_progress |
5491 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5492 | previous = 1; |
5493 | } else { | |
2c810cdd | 5494 | if (mddev->reshape_backwards |
fef9c61f N |
5495 | ? logical_sector < conf->reshape_safe |
5496 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5497 | spin_unlock_irq(&conf->device_lock); |
5498 | schedule(); | |
27c0f68f | 5499 | do_prepare = true; |
b578d55f N |
5500 | goto retry; |
5501 | } | |
5502 | } | |
7ecaa1e6 N |
5503 | spin_unlock_irq(&conf->device_lock); |
5504 | } | |
16a53ecc | 5505 | |
112bf897 N |
5506 | new_sector = raid5_compute_sector(conf, logical_sector, |
5507 | previous, | |
911d4ee8 | 5508 | &dd_idx, NULL); |
849674e4 | 5509 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5510 | (unsigned long long)new_sector, |
1da177e4 LT |
5511 | (unsigned long long)logical_sector); |
5512 | ||
6d036f7d | 5513 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5514 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5515 | if (sh) { |
b0f9ec04 | 5516 | if (unlikely(previous)) { |
7ecaa1e6 | 5517 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5518 | * stripe, so we must do the range check again. |
5519 | * Expansion could still move past after this | |
5520 | * test, but as we are holding a reference to | |
5521 | * 'sh', we know that if that happens, | |
5522 | * STRIPE_EXPANDING will get set and the expansion | |
5523 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5524 | */ |
5525 | int must_retry = 0; | |
5526 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5527 | if (mddev->reshape_backwards |
b0f9ec04 N |
5528 | ? logical_sector >= conf->reshape_progress |
5529 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5530 | /* mismatch, need to try again */ |
5531 | must_retry = 1; | |
5532 | spin_unlock_irq(&conf->device_lock); | |
5533 | if (must_retry) { | |
6d036f7d | 5534 | raid5_release_stripe(sh); |
7a3ab908 | 5535 | schedule(); |
27c0f68f | 5536 | do_prepare = true; |
7ecaa1e6 N |
5537 | goto retry; |
5538 | } | |
5539 | } | |
c46501b2 N |
5540 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5541 | /* Might have got the wrong stripe_head | |
5542 | * by accident | |
5543 | */ | |
6d036f7d | 5544 | raid5_release_stripe(sh); |
c46501b2 N |
5545 | goto retry; |
5546 | } | |
e62e58a5 | 5547 | |
ffd96e35 | 5548 | if (rw == WRITE && |
a5c308d4 | 5549 | logical_sector >= mddev->suspend_lo && |
e464eafd | 5550 | logical_sector < mddev->suspend_hi) { |
6d036f7d | 5551 | raid5_release_stripe(sh); |
e62e58a5 N |
5552 | /* As the suspend_* range is controlled by |
5553 | * userspace, we want an interruptible | |
5554 | * wait. | |
5555 | */ | |
5556 | flush_signals(current); | |
5557 | prepare_to_wait(&conf->wait_for_overlap, | |
5558 | &w, TASK_INTERRUPTIBLE); | |
5559 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5560 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5561 | schedule(); |
27c0f68f SL |
5562 | do_prepare = true; |
5563 | } | |
e464eafd N |
5564 | goto retry; |
5565 | } | |
7ecaa1e6 N |
5566 | |
5567 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5568 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5569 | /* Stripe is busy expanding or |
5570 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5571 | * and wait a while |
5572 | */ | |
482c0834 | 5573 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5574 | raid5_release_stripe(sh); |
1da177e4 | 5575 | schedule(); |
27c0f68f | 5576 | do_prepare = true; |
1da177e4 LT |
5577 | goto retry; |
5578 | } | |
3bddb7f8 SL |
5579 | if (do_flush) { |
5580 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5581 | /* we only need flush for one stripe */ | |
5582 | do_flush = false; | |
5583 | } | |
5584 | ||
6ed3003c N |
5585 | set_bit(STRIPE_HANDLE, &sh->state); |
5586 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5587 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5588 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5589 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5590 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5591 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5592 | } else { |
5593 | /* cannot get stripe for read-ahead, just give-up */ | |
4246a0b6 | 5594 | bi->bi_error = -EIO; |
1da177e4 LT |
5595 | break; |
5596 | } | |
1da177e4 | 5597 | } |
27c0f68f | 5598 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5599 | |
e7836bd6 | 5600 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5601 | if (remaining == 0) { |
1da177e4 | 5602 | |
16a53ecc | 5603 | if ( rw == WRITE ) |
1da177e4 | 5604 | md_write_end(mddev); |
6712ecf8 | 5605 | |
0a82a8d1 LT |
5606 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5607 | bi, 0); | |
4246a0b6 | 5608 | bio_endio(bi); |
1da177e4 | 5609 | } |
1da177e4 LT |
5610 | } |
5611 | ||
fd01b88c | 5612 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5613 | |
fd01b88c | 5614 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5615 | { |
52c03291 N |
5616 | /* reshaping is quite different to recovery/resync so it is |
5617 | * handled quite separately ... here. | |
5618 | * | |
5619 | * On each call to sync_request, we gather one chunk worth of | |
5620 | * destination stripes and flag them as expanding. | |
5621 | * Then we find all the source stripes and request reads. | |
5622 | * As the reads complete, handle_stripe will copy the data | |
5623 | * into the destination stripe and release that stripe. | |
5624 | */ | |
d1688a6d | 5625 | struct r5conf *conf = mddev->private; |
1da177e4 | 5626 | struct stripe_head *sh; |
ccfcc3c1 | 5627 | sector_t first_sector, last_sector; |
f416885e N |
5628 | int raid_disks = conf->previous_raid_disks; |
5629 | int data_disks = raid_disks - conf->max_degraded; | |
5630 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5631 | int i; |
5632 | int dd_idx; | |
c8f517c4 | 5633 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5634 | sector_t stripe_addr; |
7a661381 | 5635 | int reshape_sectors; |
ab69ae12 | 5636 | struct list_head stripes; |
92140480 | 5637 | sector_t retn; |
52c03291 | 5638 | |
fef9c61f N |
5639 | if (sector_nr == 0) { |
5640 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5641 | if (mddev->reshape_backwards && |
fef9c61f N |
5642 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5643 | sector_nr = raid5_size(mddev, 0, 0) | |
5644 | - conf->reshape_progress; | |
6cbd8148 N |
5645 | } else if (mddev->reshape_backwards && |
5646 | conf->reshape_progress == MaxSector) { | |
5647 | /* shouldn't happen, but just in case, finish up.*/ | |
5648 | sector_nr = MaxSector; | |
2c810cdd | 5649 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5650 | conf->reshape_progress > 0) |
5651 | sector_nr = conf->reshape_progress; | |
f416885e | 5652 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5653 | if (sector_nr) { |
8dee7211 N |
5654 | mddev->curr_resync_completed = sector_nr; |
5655 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5656 | *skipped = 1; |
92140480 N |
5657 | retn = sector_nr; |
5658 | goto finish; | |
fef9c61f | 5659 | } |
52c03291 N |
5660 | } |
5661 | ||
7a661381 N |
5662 | /* We need to process a full chunk at a time. |
5663 | * If old and new chunk sizes differ, we need to process the | |
5664 | * largest of these | |
5665 | */ | |
3cb5edf4 N |
5666 | |
5667 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5668 | |
b5254dd5 N |
5669 | /* We update the metadata at least every 10 seconds, or when |
5670 | * the data about to be copied would over-write the source of | |
5671 | * the data at the front of the range. i.e. one new_stripe | |
5672 | * along from reshape_progress new_maps to after where | |
5673 | * reshape_safe old_maps to | |
52c03291 | 5674 | */ |
fef9c61f | 5675 | writepos = conf->reshape_progress; |
f416885e | 5676 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5677 | readpos = conf->reshape_progress; |
5678 | sector_div(readpos, data_disks); | |
fef9c61f | 5679 | safepos = conf->reshape_safe; |
f416885e | 5680 | sector_div(safepos, data_disks); |
2c810cdd | 5681 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5682 | BUG_ON(writepos < reshape_sectors); |
5683 | writepos -= reshape_sectors; | |
c8f517c4 | 5684 | readpos += reshape_sectors; |
7a661381 | 5685 | safepos += reshape_sectors; |
fef9c61f | 5686 | } else { |
7a661381 | 5687 | writepos += reshape_sectors; |
c74c0d76 N |
5688 | /* readpos and safepos are worst-case calculations. |
5689 | * A negative number is overly pessimistic, and causes | |
5690 | * obvious problems for unsigned storage. So clip to 0. | |
5691 | */ | |
ed37d83e N |
5692 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5693 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5694 | } |
52c03291 | 5695 | |
b5254dd5 N |
5696 | /* Having calculated the 'writepos' possibly use it |
5697 | * to set 'stripe_addr' which is where we will write to. | |
5698 | */ | |
5699 | if (mddev->reshape_backwards) { | |
5700 | BUG_ON(conf->reshape_progress == 0); | |
5701 | stripe_addr = writepos; | |
5702 | BUG_ON((mddev->dev_sectors & | |
5703 | ~((sector_t)reshape_sectors - 1)) | |
5704 | - reshape_sectors - stripe_addr | |
5705 | != sector_nr); | |
5706 | } else { | |
5707 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5708 | stripe_addr = sector_nr; | |
5709 | } | |
5710 | ||
c8f517c4 N |
5711 | /* 'writepos' is the most advanced device address we might write. |
5712 | * 'readpos' is the least advanced device address we might read. | |
5713 | * 'safepos' is the least address recorded in the metadata as having | |
5714 | * been reshaped. | |
b5254dd5 N |
5715 | * If there is a min_offset_diff, these are adjusted either by |
5716 | * increasing the safepos/readpos if diff is negative, or | |
5717 | * increasing writepos if diff is positive. | |
5718 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5719 | * ensure safety in the face of a crash - that must be done by userspace |
5720 | * making a backup of the data. So in that case there is no particular | |
5721 | * rush to update metadata. | |
5722 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5723 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5724 | * we can be safe in the event of a crash. | |
5725 | * So we insist on updating metadata if safepos is behind writepos and | |
5726 | * readpos is beyond writepos. | |
5727 | * In any case, update the metadata every 10 seconds. | |
5728 | * Maybe that number should be configurable, but I'm not sure it is | |
5729 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5730 | */ | |
b5254dd5 N |
5731 | if (conf->min_offset_diff < 0) { |
5732 | safepos += -conf->min_offset_diff; | |
5733 | readpos += -conf->min_offset_diff; | |
5734 | } else | |
5735 | writepos += conf->min_offset_diff; | |
5736 | ||
2c810cdd | 5737 | if ((mddev->reshape_backwards |
c8f517c4 N |
5738 | ? (safepos > writepos && readpos < writepos) |
5739 | : (safepos < writepos && readpos > writepos)) || | |
5740 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5741 | /* Cannot proceed until we've updated the superblock... */ |
5742 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5743 | atomic_read(&conf->reshape_stripes)==0 |
5744 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5745 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5746 | return 0; | |
fef9c61f | 5747 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5748 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5749 | conf->reshape_checkpoint = jiffies; |
2953079c | 5750 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5751 | md_wakeup_thread(mddev->thread); |
2953079c | 5752 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5753 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5754 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5755 | return 0; | |
52c03291 | 5756 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5757 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5758 | spin_unlock_irq(&conf->device_lock); |
5759 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5760 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5761 | } |
5762 | ||
ab69ae12 | 5763 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5764 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5765 | int j; |
a9f326eb | 5766 | int skipped_disk = 0; |
6d036f7d | 5767 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5768 | set_bit(STRIPE_EXPANDING, &sh->state); |
5769 | atomic_inc(&conf->reshape_stripes); | |
5770 | /* If any of this stripe is beyond the end of the old | |
5771 | * array, then we need to zero those blocks | |
5772 | */ | |
5773 | for (j=sh->disks; j--;) { | |
5774 | sector_t s; | |
5775 | if (j == sh->pd_idx) | |
5776 | continue; | |
f416885e | 5777 | if (conf->level == 6 && |
d0dabf7e | 5778 | j == sh->qd_idx) |
f416885e | 5779 | continue; |
6d036f7d | 5780 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5781 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5782 | skipped_disk = 1; |
52c03291 N |
5783 | continue; |
5784 | } | |
5785 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5786 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5787 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5788 | } | |
a9f326eb | 5789 | if (!skipped_disk) { |
52c03291 N |
5790 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5791 | set_bit(STRIPE_HANDLE, &sh->state); | |
5792 | } | |
ab69ae12 | 5793 | list_add(&sh->lru, &stripes); |
52c03291 N |
5794 | } |
5795 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5796 | if (mddev->reshape_backwards) |
7a661381 | 5797 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5798 | else |
7a661381 | 5799 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5800 | spin_unlock_irq(&conf->device_lock); |
5801 | /* Ok, those stripe are ready. We can start scheduling | |
5802 | * reads on the source stripes. | |
5803 | * The source stripes are determined by mapping the first and last | |
5804 | * block on the destination stripes. | |
5805 | */ | |
52c03291 | 5806 | first_sector = |
ec32a2bd | 5807 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5808 | 1, &dd_idx, NULL); |
52c03291 | 5809 | last_sector = |
0e6e0271 | 5810 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5811 | * new_data_disks - 1), |
911d4ee8 | 5812 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5813 | if (last_sector >= mddev->dev_sectors) |
5814 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5815 | while (first_sector <= last_sector) { |
6d036f7d | 5816 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5817 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5818 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5819 | raid5_release_stripe(sh); |
52c03291 N |
5820 | first_sector += STRIPE_SECTORS; |
5821 | } | |
ab69ae12 N |
5822 | /* Now that the sources are clearly marked, we can release |
5823 | * the destination stripes | |
5824 | */ | |
5825 | while (!list_empty(&stripes)) { | |
5826 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5827 | list_del_init(&sh->lru); | |
6d036f7d | 5828 | raid5_release_stripe(sh); |
ab69ae12 | 5829 | } |
c6207277 N |
5830 | /* If this takes us to the resync_max point where we have to pause, |
5831 | * then we need to write out the superblock. | |
5832 | */ | |
7a661381 | 5833 | sector_nr += reshape_sectors; |
92140480 N |
5834 | retn = reshape_sectors; |
5835 | finish: | |
c5e19d90 N |
5836 | if (mddev->curr_resync_completed > mddev->resync_max || |
5837 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5838 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5839 | /* Cannot proceed until we've updated the superblock... */ |
5840 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5841 | atomic_read(&conf->reshape_stripes) == 0 |
5842 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5843 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5844 | goto ret; | |
fef9c61f | 5845 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5846 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5847 | conf->reshape_checkpoint = jiffies; |
2953079c | 5848 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
5849 | md_wakeup_thread(mddev->thread); |
5850 | wait_event(mddev->sb_wait, | |
2953079c | 5851 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
5852 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5853 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5854 | goto ret; | |
c6207277 | 5855 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5856 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5857 | spin_unlock_irq(&conf->device_lock); |
5858 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5859 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5860 | } |
c91abf5a | 5861 | ret: |
92140480 | 5862 | return retn; |
52c03291 N |
5863 | } |
5864 | ||
849674e4 SL |
5865 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
5866 | int *skipped) | |
52c03291 | 5867 | { |
d1688a6d | 5868 | struct r5conf *conf = mddev->private; |
52c03291 | 5869 | struct stripe_head *sh; |
58c0fed4 | 5870 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5871 | sector_t sync_blocks; |
16a53ecc N |
5872 | int still_degraded = 0; |
5873 | int i; | |
1da177e4 | 5874 | |
72626685 | 5875 | if (sector_nr >= max_sector) { |
1da177e4 | 5876 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5877 | |
29269553 N |
5878 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5879 | end_reshape(conf); | |
5880 | return 0; | |
5881 | } | |
72626685 N |
5882 | |
5883 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5884 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5885 | &sync_blocks, 1); | |
16a53ecc | 5886 | else /* completed sync */ |
72626685 N |
5887 | conf->fullsync = 0; |
5888 | bitmap_close_sync(mddev->bitmap); | |
5889 | ||
1da177e4 LT |
5890 | return 0; |
5891 | } | |
ccfcc3c1 | 5892 | |
64bd660b N |
5893 | /* Allow raid5_quiesce to complete */ |
5894 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5895 | ||
52c03291 N |
5896 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5897 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5898 | |
c6207277 N |
5899 | /* No need to check resync_max as we never do more than one |
5900 | * stripe, and as resync_max will always be on a chunk boundary, | |
5901 | * if the check in md_do_sync didn't fire, there is no chance | |
5902 | * of overstepping resync_max here | |
5903 | */ | |
5904 | ||
16a53ecc | 5905 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5906 | * to resync, then assert that we are finished, because there is |
5907 | * nothing we can do. | |
5908 | */ | |
3285edf1 | 5909 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5910 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5911 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5912 | *skipped = 1; |
1da177e4 LT |
5913 | return rv; |
5914 | } | |
6f608040 | 5915 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5916 | !conf->fullsync && | |
5917 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5918 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5919 | /* we can skip this block, and probably more */ |
5920 | sync_blocks /= STRIPE_SECTORS; | |
5921 | *skipped = 1; | |
5922 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5923 | } | |
1da177e4 | 5924 | |
c40f341f | 5925 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 5926 | |
6d036f7d | 5927 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5928 | if (sh == NULL) { |
6d036f7d | 5929 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5930 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5931 | * is trying to get access |
1da177e4 | 5932 | */ |
66c006a5 | 5933 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5934 | } |
16a53ecc | 5935 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5936 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5937 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5938 | */ |
16d9cfab EM |
5939 | rcu_read_lock(); |
5940 | for (i = 0; i < conf->raid_disks; i++) { | |
5941 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5942 | ||
5943 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5944 | still_degraded = 1; |
16d9cfab EM |
5945 | } |
5946 | rcu_read_unlock(); | |
16a53ecc N |
5947 | |
5948 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5949 | ||
83206d66 | 5950 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5951 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5952 | |
6d036f7d | 5953 | raid5_release_stripe(sh); |
1da177e4 LT |
5954 | |
5955 | return STRIPE_SECTORS; | |
5956 | } | |
5957 | ||
d1688a6d | 5958 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5959 | { |
5960 | /* We may not be able to submit a whole bio at once as there | |
5961 | * may not be enough stripe_heads available. | |
5962 | * We cannot pre-allocate enough stripe_heads as we may need | |
5963 | * more than exist in the cache (if we allow ever large chunks). | |
5964 | * So we do one stripe head at a time and record in | |
5965 | * ->bi_hw_segments how many have been done. | |
5966 | * | |
5967 | * We *know* that this entire raid_bio is in one chunk, so | |
5968 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5969 | */ | |
5970 | struct stripe_head *sh; | |
911d4ee8 | 5971 | int dd_idx; |
46031f9a RBJ |
5972 | sector_t sector, logical_sector, last_sector; |
5973 | int scnt = 0; | |
5974 | int remaining; | |
5975 | int handled = 0; | |
5976 | ||
4f024f37 KO |
5977 | logical_sector = raid_bio->bi_iter.bi_sector & |
5978 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 5979 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 5980 | 0, &dd_idx, NULL); |
f73a1c7d | 5981 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
5982 | |
5983 | for (; logical_sector < last_sector; | |
387bb173 NB |
5984 | logical_sector += STRIPE_SECTORS, |
5985 | sector += STRIPE_SECTORS, | |
5986 | scnt++) { | |
46031f9a | 5987 | |
e7836bd6 | 5988 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
5989 | /* already done this stripe */ |
5990 | continue; | |
5991 | ||
6d036f7d | 5992 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
5993 | |
5994 | if (!sh) { | |
5995 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 5996 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
5997 | conf->retry_read_aligned = raid_bio; |
5998 | return handled; | |
5999 | } | |
6000 | ||
da41ba65 | 6001 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6002 | raid5_release_stripe(sh); |
e7836bd6 | 6003 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
6004 | conf->retry_read_aligned = raid_bio; |
6005 | return handled; | |
6006 | } | |
6007 | ||
3f9e7c14 | 6008 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6009 | handle_stripe(sh); |
6d036f7d | 6010 | raid5_release_stripe(sh); |
46031f9a RBJ |
6011 | handled++; |
6012 | } | |
e7836bd6 | 6013 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
6014 | if (remaining == 0) { |
6015 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
6016 | raid_bio, 0); | |
4246a0b6 | 6017 | bio_endio(raid_bio); |
0a82a8d1 | 6018 | } |
46031f9a | 6019 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6020 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6021 | return handled; |
6022 | } | |
6023 | ||
bfc90cb0 | 6024 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6025 | struct r5worker *worker, |
6026 | struct list_head *temp_inactive_list) | |
46a06401 SL |
6027 | { |
6028 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6029 | int i, batch_size = 0, hash; |
6030 | bool release_inactive = false; | |
46a06401 SL |
6031 | |
6032 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6033 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6034 | batch[batch_size++] = sh; |
6035 | ||
566c09c5 SL |
6036 | if (batch_size == 0) { |
6037 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6038 | if (!list_empty(temp_inactive_list + i)) | |
6039 | break; | |
a8c34f91 SL |
6040 | if (i == NR_STRIPE_HASH_LOCKS) { |
6041 | spin_unlock_irq(&conf->device_lock); | |
6042 | r5l_flush_stripe_to_raid(conf->log); | |
6043 | spin_lock_irq(&conf->device_lock); | |
566c09c5 | 6044 | return batch_size; |
a8c34f91 | 6045 | } |
566c09c5 SL |
6046 | release_inactive = true; |
6047 | } | |
46a06401 SL |
6048 | spin_unlock_irq(&conf->device_lock); |
6049 | ||
566c09c5 SL |
6050 | release_inactive_stripe_list(conf, temp_inactive_list, |
6051 | NR_STRIPE_HASH_LOCKS); | |
6052 | ||
a8c34f91 | 6053 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6054 | if (release_inactive) { |
6055 | spin_lock_irq(&conf->device_lock); | |
6056 | return 0; | |
6057 | } | |
6058 | ||
46a06401 SL |
6059 | for (i = 0; i < batch_size; i++) |
6060 | handle_stripe(batch[i]); | |
f6bed0ef | 6061 | r5l_write_stripe_run(conf->log); |
46a06401 SL |
6062 | |
6063 | cond_resched(); | |
6064 | ||
6065 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6066 | for (i = 0; i < batch_size; i++) { |
6067 | hash = batch[i]->hash_lock_index; | |
6068 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6069 | } | |
46a06401 SL |
6070 | return batch_size; |
6071 | } | |
46031f9a | 6072 | |
851c30c9 SL |
6073 | static void raid5_do_work(struct work_struct *work) |
6074 | { | |
6075 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6076 | struct r5worker_group *group = worker->group; | |
6077 | struct r5conf *conf = group->conf; | |
6078 | int group_id = group - conf->worker_groups; | |
6079 | int handled; | |
6080 | struct blk_plug plug; | |
6081 | ||
6082 | pr_debug("+++ raid5worker active\n"); | |
6083 | ||
6084 | blk_start_plug(&plug); | |
6085 | handled = 0; | |
6086 | spin_lock_irq(&conf->device_lock); | |
6087 | while (1) { | |
6088 | int batch_size, released; | |
6089 | ||
566c09c5 | 6090 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6091 | |
566c09c5 SL |
6092 | batch_size = handle_active_stripes(conf, group_id, worker, |
6093 | worker->temp_inactive_list); | |
bfc90cb0 | 6094 | worker->working = false; |
851c30c9 SL |
6095 | if (!batch_size && !released) |
6096 | break; | |
6097 | handled += batch_size; | |
6098 | } | |
6099 | pr_debug("%d stripes handled\n", handled); | |
6100 | ||
6101 | spin_unlock_irq(&conf->device_lock); | |
6102 | blk_finish_plug(&plug); | |
6103 | ||
6104 | pr_debug("--- raid5worker inactive\n"); | |
6105 | } | |
6106 | ||
1da177e4 LT |
6107 | /* |
6108 | * This is our raid5 kernel thread. | |
6109 | * | |
6110 | * We scan the hash table for stripes which can be handled now. | |
6111 | * During the scan, completed stripes are saved for us by the interrupt | |
6112 | * handler, so that they will not have to wait for our next wakeup. | |
6113 | */ | |
4ed8731d | 6114 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6115 | { |
4ed8731d | 6116 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6117 | struct r5conf *conf = mddev->private; |
1da177e4 | 6118 | int handled; |
e1dfa0a2 | 6119 | struct blk_plug plug; |
1da177e4 | 6120 | |
45b4233c | 6121 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6122 | |
6123 | md_check_recovery(mddev); | |
1da177e4 | 6124 | |
c3cce6cd | 6125 | if (!bio_list_empty(&conf->return_bi) && |
2953079c | 6126 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6127 | struct bio_list tmp = BIO_EMPTY_LIST; |
6128 | spin_lock_irq(&conf->device_lock); | |
2953079c | 6129 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6130 | bio_list_merge(&tmp, &conf->return_bi); |
6131 | bio_list_init(&conf->return_bi); | |
6132 | } | |
6133 | spin_unlock_irq(&conf->device_lock); | |
6134 | return_io(&tmp); | |
6135 | } | |
6136 | ||
e1dfa0a2 | 6137 | blk_start_plug(&plug); |
1da177e4 LT |
6138 | handled = 0; |
6139 | spin_lock_irq(&conf->device_lock); | |
6140 | while (1) { | |
46031f9a | 6141 | struct bio *bio; |
773ca82f SL |
6142 | int batch_size, released; |
6143 | ||
566c09c5 | 6144 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6145 | if (released) |
6146 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6147 | |
0021b7bc | 6148 | if ( |
7c13edc8 N |
6149 | !list_empty(&conf->bitmap_list)) { |
6150 | /* Now is a good time to flush some bitmap updates */ | |
6151 | conf->seq_flush++; | |
700e432d | 6152 | spin_unlock_irq(&conf->device_lock); |
72626685 | 6153 | bitmap_unplug(mddev->bitmap); |
700e432d | 6154 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6155 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6156 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6157 | } |
0021b7bc | 6158 | raid5_activate_delayed(conf); |
72626685 | 6159 | |
46031f9a RBJ |
6160 | while ((bio = remove_bio_from_retry(conf))) { |
6161 | int ok; | |
6162 | spin_unlock_irq(&conf->device_lock); | |
6163 | ok = retry_aligned_read(conf, bio); | |
6164 | spin_lock_irq(&conf->device_lock); | |
6165 | if (!ok) | |
6166 | break; | |
6167 | handled++; | |
6168 | } | |
6169 | ||
566c09c5 SL |
6170 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6171 | conf->temp_inactive_list); | |
773ca82f | 6172 | if (!batch_size && !released) |
1da177e4 | 6173 | break; |
46a06401 | 6174 | handled += batch_size; |
1da177e4 | 6175 | |
2953079c | 6176 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6177 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6178 | md_check_recovery(mddev); |
46a06401 SL |
6179 | spin_lock_irq(&conf->device_lock); |
6180 | } | |
1da177e4 | 6181 | } |
45b4233c | 6182 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6183 | |
6184 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6185 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6186 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6187 | grow_one_stripe(conf, __GFP_NOWARN); |
6188 | /* Set flag even if allocation failed. This helps | |
6189 | * slow down allocation requests when mem is short | |
6190 | */ | |
6191 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6192 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6193 | } |
1da177e4 | 6194 | |
765d704d SL |
6195 | flush_deferred_bios(conf); |
6196 | ||
0576b1c6 SL |
6197 | r5l_flush_stripe_to_raid(conf->log); |
6198 | ||
c9f21aaf | 6199 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6200 | blk_finish_plug(&plug); |
1da177e4 | 6201 | |
45b4233c | 6202 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6203 | } |
6204 | ||
3f294f4f | 6205 | static ssize_t |
fd01b88c | 6206 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6207 | { |
7b1485ba N |
6208 | struct r5conf *conf; |
6209 | int ret = 0; | |
6210 | spin_lock(&mddev->lock); | |
6211 | conf = mddev->private; | |
96de1e66 | 6212 | if (conf) |
edbe83ab | 6213 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6214 | spin_unlock(&mddev->lock); |
6215 | return ret; | |
3f294f4f N |
6216 | } |
6217 | ||
c41d4ac4 | 6218 | int |
fd01b88c | 6219 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6220 | { |
d1688a6d | 6221 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
6222 | int err; |
6223 | ||
c41d4ac4 | 6224 | if (size <= 16 || size > 32768) |
3f294f4f | 6225 | return -EINVAL; |
486f0644 | 6226 | |
edbe83ab | 6227 | conf->min_nr_stripes = size; |
2d5b569b | 6228 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6229 | while (size < conf->max_nr_stripes && |
6230 | drop_one_stripe(conf)) | |
6231 | ; | |
2d5b569b | 6232 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6233 | |
edbe83ab | 6234 | |
b5470dc5 DW |
6235 | err = md_allow_write(mddev); |
6236 | if (err) | |
6237 | return err; | |
486f0644 | 6238 | |
2d5b569b | 6239 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6240 | while (size > conf->max_nr_stripes) |
6241 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
6242 | break; | |
2d5b569b | 6243 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6244 | |
c41d4ac4 N |
6245 | return 0; |
6246 | } | |
6247 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6248 | ||
6249 | static ssize_t | |
fd01b88c | 6250 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6251 | { |
6791875e | 6252 | struct r5conf *conf; |
c41d4ac4 N |
6253 | unsigned long new; |
6254 | int err; | |
6255 | ||
6256 | if (len >= PAGE_SIZE) | |
6257 | return -EINVAL; | |
b29bebd6 | 6258 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6259 | return -EINVAL; |
6791875e | 6260 | err = mddev_lock(mddev); |
c41d4ac4 N |
6261 | if (err) |
6262 | return err; | |
6791875e N |
6263 | conf = mddev->private; |
6264 | if (!conf) | |
6265 | err = -ENODEV; | |
6266 | else | |
6267 | err = raid5_set_cache_size(mddev, new); | |
6268 | mddev_unlock(mddev); | |
6269 | ||
6270 | return err ?: len; | |
3f294f4f | 6271 | } |
007583c9 | 6272 | |
96de1e66 N |
6273 | static struct md_sysfs_entry |
6274 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6275 | raid5_show_stripe_cache_size, | |
6276 | raid5_store_stripe_cache_size); | |
3f294f4f | 6277 | |
d06f191f MS |
6278 | static ssize_t |
6279 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6280 | { | |
6281 | struct r5conf *conf = mddev->private; | |
6282 | if (conf) | |
6283 | return sprintf(page, "%d\n", conf->rmw_level); | |
6284 | else | |
6285 | return 0; | |
6286 | } | |
6287 | ||
6288 | static ssize_t | |
6289 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6290 | { | |
6291 | struct r5conf *conf = mddev->private; | |
6292 | unsigned long new; | |
6293 | ||
6294 | if (!conf) | |
6295 | return -ENODEV; | |
6296 | ||
6297 | if (len >= PAGE_SIZE) | |
6298 | return -EINVAL; | |
6299 | ||
6300 | if (kstrtoul(page, 10, &new)) | |
6301 | return -EINVAL; | |
6302 | ||
6303 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6304 | return -EINVAL; | |
6305 | ||
6306 | if (new != PARITY_DISABLE_RMW && | |
6307 | new != PARITY_ENABLE_RMW && | |
6308 | new != PARITY_PREFER_RMW) | |
6309 | return -EINVAL; | |
6310 | ||
6311 | conf->rmw_level = new; | |
6312 | return len; | |
6313 | } | |
6314 | ||
6315 | static struct md_sysfs_entry | |
6316 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6317 | raid5_show_rmw_level, | |
6318 | raid5_store_rmw_level); | |
6319 | ||
6320 | ||
8b3e6cdc | 6321 | static ssize_t |
fd01b88c | 6322 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6323 | { |
7b1485ba N |
6324 | struct r5conf *conf; |
6325 | int ret = 0; | |
6326 | spin_lock(&mddev->lock); | |
6327 | conf = mddev->private; | |
8b3e6cdc | 6328 | if (conf) |
7b1485ba N |
6329 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6330 | spin_unlock(&mddev->lock); | |
6331 | return ret; | |
8b3e6cdc DW |
6332 | } |
6333 | ||
6334 | static ssize_t | |
fd01b88c | 6335 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6336 | { |
6791875e | 6337 | struct r5conf *conf; |
4ef197d8 | 6338 | unsigned long new; |
6791875e N |
6339 | int err; |
6340 | ||
8b3e6cdc DW |
6341 | if (len >= PAGE_SIZE) |
6342 | return -EINVAL; | |
b29bebd6 | 6343 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6344 | return -EINVAL; |
6791875e N |
6345 | |
6346 | err = mddev_lock(mddev); | |
6347 | if (err) | |
6348 | return err; | |
6349 | conf = mddev->private; | |
6350 | if (!conf) | |
6351 | err = -ENODEV; | |
edbe83ab | 6352 | else if (new > conf->min_nr_stripes) |
6791875e N |
6353 | err = -EINVAL; |
6354 | else | |
6355 | conf->bypass_threshold = new; | |
6356 | mddev_unlock(mddev); | |
6357 | return err ?: len; | |
8b3e6cdc DW |
6358 | } |
6359 | ||
6360 | static struct md_sysfs_entry | |
6361 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6362 | S_IRUGO | S_IWUSR, | |
6363 | raid5_show_preread_threshold, | |
6364 | raid5_store_preread_threshold); | |
6365 | ||
d592a996 SL |
6366 | static ssize_t |
6367 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6368 | { | |
7b1485ba N |
6369 | struct r5conf *conf; |
6370 | int ret = 0; | |
6371 | spin_lock(&mddev->lock); | |
6372 | conf = mddev->private; | |
d592a996 | 6373 | if (conf) |
7b1485ba N |
6374 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6375 | spin_unlock(&mddev->lock); | |
6376 | return ret; | |
d592a996 SL |
6377 | } |
6378 | ||
6379 | static ssize_t | |
6380 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6381 | { | |
6791875e | 6382 | struct r5conf *conf; |
d592a996 | 6383 | unsigned long new; |
6791875e N |
6384 | int err; |
6385 | ||
d592a996 SL |
6386 | if (len >= PAGE_SIZE) |
6387 | return -EINVAL; | |
d592a996 SL |
6388 | if (kstrtoul(page, 10, &new)) |
6389 | return -EINVAL; | |
6390 | new = !!new; | |
6791875e N |
6391 | |
6392 | err = mddev_lock(mddev); | |
6393 | if (err) | |
6394 | return err; | |
6395 | conf = mddev->private; | |
6396 | if (!conf) | |
6397 | err = -ENODEV; | |
6398 | else if (new != conf->skip_copy) { | |
6399 | mddev_suspend(mddev); | |
6400 | conf->skip_copy = new; | |
6401 | if (new) | |
dc3b17cc | 6402 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6403 | BDI_CAP_STABLE_WRITES; |
6404 | else | |
dc3b17cc | 6405 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6406 | ~BDI_CAP_STABLE_WRITES; |
6407 | mddev_resume(mddev); | |
6408 | } | |
6409 | mddev_unlock(mddev); | |
6410 | return err ?: len; | |
d592a996 SL |
6411 | } |
6412 | ||
6413 | static struct md_sysfs_entry | |
6414 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6415 | raid5_show_skip_copy, | |
6416 | raid5_store_skip_copy); | |
6417 | ||
3f294f4f | 6418 | static ssize_t |
fd01b88c | 6419 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6420 | { |
d1688a6d | 6421 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6422 | if (conf) |
6423 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6424 | else | |
6425 | return 0; | |
3f294f4f N |
6426 | } |
6427 | ||
96de1e66 N |
6428 | static struct md_sysfs_entry |
6429 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6430 | |
b721420e SL |
6431 | static ssize_t |
6432 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6433 | { | |
7b1485ba N |
6434 | struct r5conf *conf; |
6435 | int ret = 0; | |
6436 | spin_lock(&mddev->lock); | |
6437 | conf = mddev->private; | |
b721420e | 6438 | if (conf) |
7b1485ba N |
6439 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6440 | spin_unlock(&mddev->lock); | |
6441 | return ret; | |
b721420e SL |
6442 | } |
6443 | ||
60aaf933 | 6444 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6445 | int *group_cnt, | |
6446 | int *worker_cnt_per_group, | |
6447 | struct r5worker_group **worker_groups); | |
b721420e SL |
6448 | static ssize_t |
6449 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6450 | { | |
6791875e | 6451 | struct r5conf *conf; |
b721420e SL |
6452 | unsigned long new; |
6453 | int err; | |
60aaf933 | 6454 | struct r5worker_group *new_groups, *old_groups; |
6455 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6456 | |
6457 | if (len >= PAGE_SIZE) | |
6458 | return -EINVAL; | |
b721420e SL |
6459 | if (kstrtoul(page, 10, &new)) |
6460 | return -EINVAL; | |
6461 | ||
6791875e N |
6462 | err = mddev_lock(mddev); |
6463 | if (err) | |
6464 | return err; | |
6465 | conf = mddev->private; | |
6466 | if (!conf) | |
6467 | err = -ENODEV; | |
6468 | else if (new != conf->worker_cnt_per_group) { | |
6469 | mddev_suspend(mddev); | |
b721420e | 6470 | |
6791875e N |
6471 | old_groups = conf->worker_groups; |
6472 | if (old_groups) | |
6473 | flush_workqueue(raid5_wq); | |
d206dcfa | 6474 | |
6791875e N |
6475 | err = alloc_thread_groups(conf, new, |
6476 | &group_cnt, &worker_cnt_per_group, | |
6477 | &new_groups); | |
6478 | if (!err) { | |
6479 | spin_lock_irq(&conf->device_lock); | |
6480 | conf->group_cnt = group_cnt; | |
6481 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6482 | conf->worker_groups = new_groups; | |
6483 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6484 | |
6791875e N |
6485 | if (old_groups) |
6486 | kfree(old_groups[0].workers); | |
6487 | kfree(old_groups); | |
6488 | } | |
6489 | mddev_resume(mddev); | |
b721420e | 6490 | } |
6791875e | 6491 | mddev_unlock(mddev); |
b721420e | 6492 | |
6791875e | 6493 | return err ?: len; |
b721420e SL |
6494 | } |
6495 | ||
6496 | static struct md_sysfs_entry | |
6497 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6498 | raid5_show_group_thread_cnt, | |
6499 | raid5_store_group_thread_cnt); | |
6500 | ||
007583c9 | 6501 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6502 | &raid5_stripecache_size.attr, |
6503 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6504 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6505 | &raid5_group_thread_cnt.attr, |
d592a996 | 6506 | &raid5_skip_copy.attr, |
d06f191f | 6507 | &raid5_rmw_level.attr, |
2c7da14b | 6508 | &r5c_journal_mode.attr, |
3f294f4f N |
6509 | NULL, |
6510 | }; | |
007583c9 N |
6511 | static struct attribute_group raid5_attrs_group = { |
6512 | .name = NULL, | |
6513 | .attrs = raid5_attrs, | |
3f294f4f N |
6514 | }; |
6515 | ||
60aaf933 | 6516 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6517 | int *group_cnt, | |
6518 | int *worker_cnt_per_group, | |
6519 | struct r5worker_group **worker_groups) | |
851c30c9 | 6520 | { |
566c09c5 | 6521 | int i, j, k; |
851c30c9 SL |
6522 | ssize_t size; |
6523 | struct r5worker *workers; | |
6524 | ||
60aaf933 | 6525 | *worker_cnt_per_group = cnt; |
851c30c9 | 6526 | if (cnt == 0) { |
60aaf933 | 6527 | *group_cnt = 0; |
6528 | *worker_groups = NULL; | |
851c30c9 SL |
6529 | return 0; |
6530 | } | |
60aaf933 | 6531 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6532 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6533 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6534 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6535 | *group_cnt, GFP_NOIO); | |
6536 | if (!*worker_groups || !workers) { | |
851c30c9 | 6537 | kfree(workers); |
60aaf933 | 6538 | kfree(*worker_groups); |
851c30c9 SL |
6539 | return -ENOMEM; |
6540 | } | |
6541 | ||
60aaf933 | 6542 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6543 | struct r5worker_group *group; |
6544 | ||
0c775d52 | 6545 | group = &(*worker_groups)[i]; |
851c30c9 SL |
6546 | INIT_LIST_HEAD(&group->handle_list); |
6547 | group->conf = conf; | |
6548 | group->workers = workers + i * cnt; | |
6549 | ||
6550 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6551 | struct r5worker *worker = group->workers + j; |
6552 | worker->group = group; | |
6553 | INIT_WORK(&worker->work, raid5_do_work); | |
6554 | ||
6555 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6556 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6557 | } |
6558 | } | |
6559 | ||
6560 | return 0; | |
6561 | } | |
6562 | ||
6563 | static void free_thread_groups(struct r5conf *conf) | |
6564 | { | |
6565 | if (conf->worker_groups) | |
6566 | kfree(conf->worker_groups[0].workers); | |
6567 | kfree(conf->worker_groups); | |
6568 | conf->worker_groups = NULL; | |
6569 | } | |
6570 | ||
80c3a6ce | 6571 | static sector_t |
fd01b88c | 6572 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6573 | { |
d1688a6d | 6574 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6575 | |
6576 | if (!sectors) | |
6577 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6578 | if (!raid_disks) |
7ec05478 | 6579 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6580 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6581 | |
3cb5edf4 N |
6582 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6583 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6584 | return sectors * (raid_disks - conf->max_degraded); |
6585 | } | |
6586 | ||
789b5e03 ON |
6587 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6588 | { | |
6589 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6590 | if (percpu->scribble) |
6591 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6592 | percpu->spare_page = NULL; |
6593 | percpu->scribble = NULL; | |
6594 | } | |
6595 | ||
6596 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6597 | { | |
6598 | if (conf->level == 6 && !percpu->spare_page) | |
6599 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6600 | if (!percpu->scribble) | |
46d5b785 | 6601 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6602 | conf->previous_raid_disks), |
6603 | max(conf->chunk_sectors, | |
6604 | conf->prev_chunk_sectors) | |
6605 | / STRIPE_SECTORS, | |
6606 | GFP_KERNEL); | |
789b5e03 ON |
6607 | |
6608 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6609 | free_scratch_buffer(conf, percpu); | |
6610 | return -ENOMEM; | |
6611 | } | |
6612 | ||
6613 | return 0; | |
6614 | } | |
6615 | ||
29c6d1bb | 6616 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6617 | { |
29c6d1bb SAS |
6618 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6619 | ||
6620 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6621 | return 0; | |
6622 | } | |
36d1c647 | 6623 | |
29c6d1bb SAS |
6624 | static void raid5_free_percpu(struct r5conf *conf) |
6625 | { | |
36d1c647 DW |
6626 | if (!conf->percpu) |
6627 | return; | |
6628 | ||
29c6d1bb | 6629 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6630 | free_percpu(conf->percpu); |
6631 | } | |
6632 | ||
d1688a6d | 6633 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6634 | { |
d7bd398e SL |
6635 | int i; |
6636 | ||
5c7e81c3 SL |
6637 | if (conf->log) |
6638 | r5l_exit_log(conf->log); | |
30c89465 | 6639 | if (conf->shrinker.nr_deferred) |
edbe83ab | 6640 | unregister_shrinker(&conf->shrinker); |
5c7e81c3 | 6641 | |
851c30c9 | 6642 | free_thread_groups(conf); |
95fc17aa | 6643 | shrink_stripes(conf); |
36d1c647 | 6644 | raid5_free_percpu(conf); |
d7bd398e SL |
6645 | for (i = 0; i < conf->pool_size; i++) |
6646 | if (conf->disks[i].extra_page) | |
6647 | put_page(conf->disks[i].extra_page); | |
95fc17aa DW |
6648 | kfree(conf->disks); |
6649 | kfree(conf->stripe_hashtbl); | |
6650 | kfree(conf); | |
6651 | } | |
6652 | ||
29c6d1bb | 6653 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6654 | { |
29c6d1bb | 6655 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6656 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6657 | ||
29c6d1bb | 6658 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6659 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6660 | __func__, cpu); | |
29c6d1bb | 6661 | return -ENOMEM; |
36d1c647 | 6662 | } |
29c6d1bb | 6663 | return 0; |
36d1c647 | 6664 | } |
36d1c647 | 6665 | |
d1688a6d | 6666 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6667 | { |
789b5e03 | 6668 | int err = 0; |
36d1c647 | 6669 | |
789b5e03 ON |
6670 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6671 | if (!conf->percpu) | |
36d1c647 | 6672 | return -ENOMEM; |
789b5e03 | 6673 | |
29c6d1bb | 6674 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6675 | if (!err) { |
6676 | conf->scribble_disks = max(conf->raid_disks, | |
6677 | conf->previous_raid_disks); | |
6678 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6679 | conf->prev_chunk_sectors); | |
6680 | } | |
36d1c647 DW |
6681 | return err; |
6682 | } | |
6683 | ||
edbe83ab N |
6684 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6685 | struct shrink_control *sc) | |
6686 | { | |
6687 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6688 | unsigned long ret = SHRINK_STOP; |
6689 | ||
6690 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6691 | ret= 0; | |
49895bcc N |
6692 | while (ret < sc->nr_to_scan && |
6693 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6694 | if (drop_one_stripe(conf) == 0) { |
6695 | ret = SHRINK_STOP; | |
6696 | break; | |
6697 | } | |
6698 | ret++; | |
6699 | } | |
6700 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6701 | } |
6702 | return ret; | |
6703 | } | |
6704 | ||
6705 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6706 | struct shrink_control *sc) | |
6707 | { | |
6708 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6709 | ||
6710 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6711 | /* unlikely, but not impossible */ | |
6712 | return 0; | |
6713 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6714 | } | |
6715 | ||
d1688a6d | 6716 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6717 | { |
d1688a6d | 6718 | struct r5conf *conf; |
5e5e3e78 | 6719 | int raid_disk, memory, max_disks; |
3cb03002 | 6720 | struct md_rdev *rdev; |
1da177e4 | 6721 | struct disk_info *disk; |
0232605d | 6722 | char pers_name[6]; |
566c09c5 | 6723 | int i; |
60aaf933 | 6724 | int group_cnt, worker_cnt_per_group; |
6725 | struct r5worker_group *new_group; | |
1da177e4 | 6726 | |
91adb564 N |
6727 | if (mddev->new_level != 5 |
6728 | && mddev->new_level != 4 | |
6729 | && mddev->new_level != 6) { | |
cc6167b4 N |
6730 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6731 | mdname(mddev), mddev->new_level); | |
91adb564 | 6732 | return ERR_PTR(-EIO); |
1da177e4 | 6733 | } |
91adb564 N |
6734 | if ((mddev->new_level == 5 |
6735 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6736 | (mddev->new_level == 6 | |
6737 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6738 | pr_warn("md/raid:%s: layout %d not supported\n", |
6739 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6740 | return ERR_PTR(-EIO); |
99c0fb5f | 6741 | } |
91adb564 | 6742 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6743 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6744 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6745 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6746 | } |
6747 | ||
664e7c41 AN |
6748 | if (!mddev->new_chunk_sectors || |
6749 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6750 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6751 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6752 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6753 | return ERR_PTR(-EINVAL); |
f6705578 N |
6754 | } |
6755 | ||
d1688a6d | 6756 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6757 | if (conf == NULL) |
1da177e4 | 6758 | goto abort; |
851c30c9 | 6759 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6760 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6761 | &new_group)) { | |
6762 | conf->group_cnt = group_cnt; | |
6763 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6764 | conf->worker_groups = new_group; | |
6765 | } else | |
851c30c9 | 6766 | goto abort; |
f5efd45a | 6767 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6768 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6769 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6770 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6771 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6772 | init_waitqueue_head(&conf->wait_for_overlap); |
6773 | INIT_LIST_HEAD(&conf->handle_list); | |
6774 | INIT_LIST_HEAD(&conf->hold_list); | |
6775 | INIT_LIST_HEAD(&conf->delayed_list); | |
6776 | INIT_LIST_HEAD(&conf->bitmap_list); | |
c3cce6cd | 6777 | bio_list_init(&conf->return_bi); |
773ca82f | 6778 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6779 | atomic_set(&conf->active_stripes, 0); |
6780 | atomic_set(&conf->preread_active_stripes, 0); | |
6781 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
6782 | bio_list_init(&conf->pending_bios); |
6783 | spin_lock_init(&conf->pending_bios_lock); | |
6784 | conf->batch_bio_dispatch = true; | |
6785 | rdev_for_each(rdev, mddev) { | |
6786 | if (test_bit(Journal, &rdev->flags)) | |
6787 | continue; | |
6788 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
6789 | conf->batch_bio_dispatch = false; | |
6790 | break; | |
6791 | } | |
6792 | } | |
6793 | ||
f5efd45a | 6794 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 6795 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6796 | |
6797 | conf->raid_disks = mddev->raid_disks; | |
6798 | if (mddev->reshape_position == MaxSector) | |
6799 | conf->previous_raid_disks = mddev->raid_disks; | |
6800 | else | |
f6705578 | 6801 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6802 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6803 | |
5e5e3e78 | 6804 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc | 6805 | GFP_KERNEL); |
d7bd398e | 6806 | |
b55e6bfc N |
6807 | if (!conf->disks) |
6808 | goto abort; | |
9ffae0cf | 6809 | |
d7bd398e SL |
6810 | for (i = 0; i < max_disks; i++) { |
6811 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
6812 | if (!conf->disks[i].extra_page) | |
6813 | goto abort; | |
6814 | } | |
6815 | ||
1da177e4 LT |
6816 | conf->mddev = mddev; |
6817 | ||
fccddba0 | 6818 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6819 | goto abort; |
1da177e4 | 6820 | |
566c09c5 SL |
6821 | /* We init hash_locks[0] separately to that it can be used |
6822 | * as the reference lock in the spin_lock_nest_lock() call | |
6823 | * in lock_all_device_hash_locks_irq in order to convince | |
6824 | * lockdep that we know what we are doing. | |
6825 | */ | |
6826 | spin_lock_init(conf->hash_locks); | |
6827 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6828 | spin_lock_init(conf->hash_locks + i); | |
6829 | ||
6830 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6831 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6832 | ||
6833 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6834 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6835 | ||
1e6d690b SL |
6836 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
6837 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
6838 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
6839 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
6840 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
6841 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 6842 | |
36d1c647 | 6843 | conf->level = mddev->new_level; |
46d5b785 | 6844 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6845 | if (raid5_alloc_percpu(conf) != 0) |
6846 | goto abort; | |
6847 | ||
0c55e022 | 6848 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6849 | |
dafb20fa | 6850 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6851 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6852 | if (raid_disk >= max_disks |
f2076e7d | 6853 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
6854 | continue; |
6855 | disk = conf->disks + raid_disk; | |
6856 | ||
17045f52 N |
6857 | if (test_bit(Replacement, &rdev->flags)) { |
6858 | if (disk->replacement) | |
6859 | goto abort; | |
6860 | disk->replacement = rdev; | |
6861 | } else { | |
6862 | if (disk->rdev) | |
6863 | goto abort; | |
6864 | disk->rdev = rdev; | |
6865 | } | |
1da177e4 | 6866 | |
b2d444d7 | 6867 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6868 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
6869 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
6870 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6871 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6872 | /* Cannot rely on bitmap to complete recovery */ |
6873 | conf->fullsync = 1; | |
1da177e4 LT |
6874 | } |
6875 | ||
91adb564 | 6876 | conf->level = mddev->new_level; |
584acdd4 | 6877 | if (conf->level == 6) { |
16a53ecc | 6878 | conf->max_degraded = 2; |
584acdd4 MS |
6879 | if (raid6_call.xor_syndrome) |
6880 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6881 | else | |
6882 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6883 | } else { | |
16a53ecc | 6884 | conf->max_degraded = 1; |
584acdd4 MS |
6885 | conf->rmw_level = PARITY_ENABLE_RMW; |
6886 | } | |
91adb564 | 6887 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6888 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6889 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6890 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 6891 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
6892 | } else { |
6893 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
6894 | conf->prev_algo = conf->algorithm; | |
e183eaed | 6895 | } |
1da177e4 | 6896 | |
edbe83ab | 6897 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
6898 | if (mddev->reshape_position != MaxSector) { |
6899 | int stripes = max_t(int, | |
6900 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
6901 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
6902 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
6903 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 6904 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
6905 | mdname(mddev), conf->min_nr_stripes); |
6906 | } | |
edbe83ab | 6907 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 6908 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6909 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6910 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
6911 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
6912 | mdname(mddev), memory); | |
91adb564 N |
6913 | goto abort; |
6914 | } else | |
cc6167b4 | 6915 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
6916 | /* |
6917 | * Losing a stripe head costs more than the time to refill it, | |
6918 | * it reduces the queue depth and so can hurt throughput. | |
6919 | * So set it rather large, scaled by number of devices. | |
6920 | */ | |
6921 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6922 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6923 | conf->shrinker.count_objects = raid5_cache_count; | |
6924 | conf->shrinker.batch = 128; | |
6925 | conf->shrinker.flags = 0; | |
6a0f53ff | 6926 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
6927 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
6928 | mdname(mddev)); | |
6a0f53ff CY |
6929 | goto abort; |
6930 | } | |
1da177e4 | 6931 | |
0232605d N |
6932 | sprintf(pers_name, "raid%d", mddev->new_level); |
6933 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 6934 | if (!conf->thread) { |
cc6167b4 N |
6935 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
6936 | mdname(mddev)); | |
16a53ecc N |
6937 | goto abort; |
6938 | } | |
91adb564 N |
6939 | |
6940 | return conf; | |
6941 | ||
6942 | abort: | |
6943 | if (conf) { | |
95fc17aa | 6944 | free_conf(conf); |
91adb564 N |
6945 | return ERR_PTR(-EIO); |
6946 | } else | |
6947 | return ERR_PTR(-ENOMEM); | |
6948 | } | |
6949 | ||
c148ffdc N |
6950 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6951 | { | |
6952 | switch (algo) { | |
6953 | case ALGORITHM_PARITY_0: | |
6954 | if (raid_disk < max_degraded) | |
6955 | return 1; | |
6956 | break; | |
6957 | case ALGORITHM_PARITY_N: | |
6958 | if (raid_disk >= raid_disks - max_degraded) | |
6959 | return 1; | |
6960 | break; | |
6961 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6962 | if (raid_disk == 0 || |
c148ffdc N |
6963 | raid_disk == raid_disks - 1) |
6964 | return 1; | |
6965 | break; | |
6966 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6967 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6968 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6969 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6970 | if (raid_disk == raid_disks - 1) | |
6971 | return 1; | |
6972 | } | |
6973 | return 0; | |
6974 | } | |
6975 | ||
849674e4 | 6976 | static int raid5_run(struct mddev *mddev) |
91adb564 | 6977 | { |
d1688a6d | 6978 | struct r5conf *conf; |
9f7c2220 | 6979 | int working_disks = 0; |
c148ffdc | 6980 | int dirty_parity_disks = 0; |
3cb03002 | 6981 | struct md_rdev *rdev; |
713cf5a6 | 6982 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 6983 | sector_t reshape_offset = 0; |
17045f52 | 6984 | int i; |
b5254dd5 N |
6985 | long long min_offset_diff = 0; |
6986 | int first = 1; | |
91adb564 | 6987 | |
8c6ac868 | 6988 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
6989 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
6990 | mdname(mddev)); | |
b5254dd5 N |
6991 | |
6992 | rdev_for_each(rdev, mddev) { | |
6993 | long long diff; | |
713cf5a6 | 6994 | |
f2076e7d | 6995 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 6996 | journal_dev = rdev; |
f2076e7d SL |
6997 | continue; |
6998 | } | |
b5254dd5 N |
6999 | if (rdev->raid_disk < 0) |
7000 | continue; | |
7001 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7002 | if (first) { | |
7003 | min_offset_diff = diff; | |
7004 | first = 0; | |
7005 | } else if (mddev->reshape_backwards && | |
7006 | diff < min_offset_diff) | |
7007 | min_offset_diff = diff; | |
7008 | else if (!mddev->reshape_backwards && | |
7009 | diff > min_offset_diff) | |
7010 | min_offset_diff = diff; | |
7011 | } | |
7012 | ||
91adb564 N |
7013 | if (mddev->reshape_position != MaxSector) { |
7014 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7015 | * Difficulties arise if the stripe we would write to |
7016 | * next is at or after the stripe we would read from next. | |
7017 | * For a reshape that changes the number of devices, this | |
7018 | * is only possible for a very short time, and mdadm makes | |
7019 | * sure that time appears to have past before assembling | |
7020 | * the array. So we fail if that time hasn't passed. | |
7021 | * For a reshape that keeps the number of devices the same | |
7022 | * mdadm must be monitoring the reshape can keeping the | |
7023 | * critical areas read-only and backed up. It will start | |
7024 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7025 | */ |
7026 | sector_t here_new, here_old; | |
7027 | int old_disks; | |
18b00334 | 7028 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7029 | int chunk_sectors; |
7030 | int new_data_disks; | |
91adb564 | 7031 | |
713cf5a6 | 7032 | if (journal_dev) { |
cc6167b4 N |
7033 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7034 | mdname(mddev)); | |
713cf5a6 SL |
7035 | return -EINVAL; |
7036 | } | |
7037 | ||
88ce4930 | 7038 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7039 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7040 | mdname(mddev)); | |
91adb564 N |
7041 | return -EINVAL; |
7042 | } | |
91adb564 N |
7043 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7044 | /* reshape_position must be on a new-stripe boundary, and one | |
7045 | * further up in new geometry must map after here in old | |
7046 | * geometry. | |
05256d98 N |
7047 | * If the chunk sizes are different, then as we perform reshape |
7048 | * in units of the largest of the two, reshape_position needs | |
7049 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7050 | */ |
7051 | here_new = mddev->reshape_position; | |
05256d98 N |
7052 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7053 | new_data_disks = mddev->raid_disks - max_degraded; | |
7054 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7055 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7056 | mdname(mddev)); | |
91adb564 N |
7057 | return -EINVAL; |
7058 | } | |
05256d98 | 7059 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7060 | /* here_new is the stripe we will write to */ |
7061 | here_old = mddev->reshape_position; | |
05256d98 | 7062 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7063 | /* here_old is the first stripe that we might need to read |
7064 | * from */ | |
67ac6011 N |
7065 | if (mddev->delta_disks == 0) { |
7066 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7067 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7068 | * and taking constant backups. |
7069 | * mdadm always starts a situation like this in | |
7070 | * readonly mode so it can take control before | |
7071 | * allowing any writes. So just check for that. | |
7072 | */ | |
b5254dd5 N |
7073 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7074 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7075 | /* not really in-place - so OK */; | |
7076 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7077 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7078 | mdname(mddev)); | |
67ac6011 N |
7079 | return -EINVAL; |
7080 | } | |
2c810cdd | 7081 | } else if (mddev->reshape_backwards |
05256d98 N |
7082 | ? (here_new * chunk_sectors + min_offset_diff <= |
7083 | here_old * chunk_sectors) | |
7084 | : (here_new * chunk_sectors >= | |
7085 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7086 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7087 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7088 | mdname(mddev)); | |
91adb564 N |
7089 | return -EINVAL; |
7090 | } | |
cc6167b4 | 7091 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7092 | /* OK, we should be able to continue; */ |
7093 | } else { | |
7094 | BUG_ON(mddev->level != mddev->new_level); | |
7095 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7096 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7097 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7098 | } |
91adb564 | 7099 | |
245f46c2 N |
7100 | if (mddev->private == NULL) |
7101 | conf = setup_conf(mddev); | |
7102 | else | |
7103 | conf = mddev->private; | |
7104 | ||
91adb564 N |
7105 | if (IS_ERR(conf)) |
7106 | return PTR_ERR(conf); | |
7107 | ||
486b0f7b SL |
7108 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7109 | if (!journal_dev) { | |
cc6167b4 N |
7110 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7111 | mdname(mddev)); | |
486b0f7b SL |
7112 | mddev->ro = 1; |
7113 | set_disk_ro(mddev->gendisk, 1); | |
7114 | } else if (mddev->recovery_cp == MaxSector) | |
7115 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7116 | } |
7117 | ||
b5254dd5 | 7118 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7119 | mddev->thread = conf->thread; |
7120 | conf->thread = NULL; | |
7121 | mddev->private = conf; | |
7122 | ||
17045f52 N |
7123 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7124 | i++) { | |
7125 | rdev = conf->disks[i].rdev; | |
7126 | if (!rdev && conf->disks[i].replacement) { | |
7127 | /* The replacement is all we have yet */ | |
7128 | rdev = conf->disks[i].replacement; | |
7129 | conf->disks[i].replacement = NULL; | |
7130 | clear_bit(Replacement, &rdev->flags); | |
7131 | conf->disks[i].rdev = rdev; | |
7132 | } | |
7133 | if (!rdev) | |
c148ffdc | 7134 | continue; |
17045f52 N |
7135 | if (conf->disks[i].replacement && |
7136 | conf->reshape_progress != MaxSector) { | |
7137 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7138 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7139 | goto abort; |
7140 | } | |
2f115882 | 7141 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7142 | working_disks++; |
2f115882 N |
7143 | continue; |
7144 | } | |
c148ffdc N |
7145 | /* This disc is not fully in-sync. However if it |
7146 | * just stored parity (beyond the recovery_offset), | |
7147 | * when we don't need to be concerned about the | |
7148 | * array being dirty. | |
7149 | * When reshape goes 'backwards', we never have | |
7150 | * partially completed devices, so we only need | |
7151 | * to worry about reshape going forwards. | |
7152 | */ | |
7153 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7154 | if (mddev->major_version == 0 && | |
7155 | mddev->minor_version > 90) | |
7156 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7157 | |
c148ffdc N |
7158 | if (rdev->recovery_offset < reshape_offset) { |
7159 | /* We need to check old and new layout */ | |
7160 | if (!only_parity(rdev->raid_disk, | |
7161 | conf->algorithm, | |
7162 | conf->raid_disks, | |
7163 | conf->max_degraded)) | |
7164 | continue; | |
7165 | } | |
7166 | if (!only_parity(rdev->raid_disk, | |
7167 | conf->prev_algo, | |
7168 | conf->previous_raid_disks, | |
7169 | conf->max_degraded)) | |
7170 | continue; | |
7171 | dirty_parity_disks++; | |
7172 | } | |
91adb564 | 7173 | |
17045f52 N |
7174 | /* |
7175 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7176 | */ | |
2e38a37f | 7177 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7178 | |
674806d6 | 7179 | if (has_failed(conf)) { |
cc6167b4 | 7180 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7181 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7182 | goto abort; |
7183 | } | |
7184 | ||
91adb564 | 7185 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7186 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7187 | mddev->resync_max_sectors = mddev->dev_sectors; |
7188 | ||
c148ffdc | 7189 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7190 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec | 7191 | if (mddev->ok_start_degraded) |
cc6167b4 N |
7192 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7193 | mdname(mddev)); | |
6ff8d8ec | 7194 | else { |
cc6167b4 N |
7195 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7196 | mdname(mddev)); | |
6ff8d8ec N |
7197 | goto abort; |
7198 | } | |
1da177e4 LT |
7199 | } |
7200 | ||
cc6167b4 N |
7201 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7202 | mdname(mddev), conf->level, | |
7203 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7204 | mddev->new_layout); | |
1da177e4 LT |
7205 | |
7206 | print_raid5_conf(conf); | |
7207 | ||
fef9c61f | 7208 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7209 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7210 | atomic_set(&conf->reshape_stripes, 0); |
7211 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7212 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7213 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7214 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7215 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7216 | "reshape"); |
f6705578 N |
7217 | } |
7218 | ||
1da177e4 | 7219 | /* Ok, everything is just fine now */ |
a64c876f N |
7220 | if (mddev->to_remove == &raid5_attrs_group) |
7221 | mddev->to_remove = NULL; | |
00bcb4ac N |
7222 | else if (mddev->kobj.sd && |
7223 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7224 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7225 | mdname(mddev)); | |
4a5add49 | 7226 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7227 | |
4a5add49 | 7228 | if (mddev->queue) { |
9f7c2220 | 7229 | int chunk_size; |
620125f2 | 7230 | bool discard_supported = true; |
4a5add49 N |
7231 | /* read-ahead size must cover two whole stripes, which |
7232 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7233 | * number of raid devices | |
7234 | */ | |
7235 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7236 | int stripe = data_disks * | |
7237 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7238 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7239 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7240 | |
9f7c2220 N |
7241 | chunk_size = mddev->chunk_sectors << 9; |
7242 | blk_queue_io_min(mddev->queue, chunk_size); | |
7243 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7244 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7245 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7246 | /* |
7247 | * We can only discard a whole stripe. It doesn't make sense to | |
7248 | * discard data disk but write parity disk | |
7249 | */ | |
7250 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7251 | /* Round up to power of 2, as discard handling |
7252 | * currently assumes that */ | |
7253 | while ((stripe-1) & stripe) | |
7254 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7255 | mddev->queue->limits.discard_alignment = stripe; |
7256 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 KK |
7257 | |
7258 | /* | |
7259 | * We use 16-bit counter of active stripes in bi_phys_segments | |
7260 | * (minus one for over-loaded initialization) | |
7261 | */ | |
7262 | blk_queue_max_hw_sectors(mddev->queue, 0xfffe * STRIPE_SECTORS); | |
7263 | blk_queue_max_discard_sectors(mddev->queue, | |
7264 | 0xfffe * STRIPE_SECTORS); | |
7265 | ||
620125f2 SL |
7266 | /* |
7267 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 7268 | * guarantee discard_zeroes_data |
620125f2 SL |
7269 | */ |
7270 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 7271 | |
5026d7a9 PA |
7272 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
7273 | ||
05616be5 | 7274 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7275 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7276 | rdev->data_offset << 9); | |
05616be5 N |
7277 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7278 | rdev->new_data_offset << 9); | |
620125f2 SL |
7279 | /* |
7280 | * discard_zeroes_data is required, otherwise data | |
7281 | * could be lost. Consider a scenario: discard a stripe | |
7282 | * (the stripe could be inconsistent if | |
7283 | * discard_zeroes_data is 0); write one disk of the | |
7284 | * stripe (the stripe could be inconsistent again | |
7285 | * depending on which disks are used to calculate | |
7286 | * parity); the disk is broken; The stripe data of this | |
7287 | * disk is lost. | |
7288 | */ | |
7289 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
7290 | !bdev_get_queue(rdev->bdev)-> | |
7291 | limits.discard_zeroes_data) | |
7292 | discard_supported = false; | |
8e0e99ba N |
7293 | /* Unfortunately, discard_zeroes_data is not currently |
7294 | * a guarantee - just a hint. So we only allow DISCARD | |
7295 | * if the sysadmin has confirmed that only safe devices | |
7296 | * are in use by setting a module parameter. | |
7297 | */ | |
7298 | if (!devices_handle_discard_safely) { | |
7299 | if (discard_supported) { | |
7300 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
7301 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
7302 | } | |
7303 | discard_supported = false; | |
7304 | } | |
05616be5 | 7305 | } |
620125f2 SL |
7306 | |
7307 | if (discard_supported && | |
e7597e69 JS |
7308 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7309 | mddev->queue->limits.discard_granularity >= stripe) | |
620125f2 SL |
7310 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
7311 | mddev->queue); | |
7312 | else | |
7313 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
7314 | mddev->queue); | |
1dffdddd SL |
7315 | |
7316 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7317 | } |
23032a0e | 7318 | |
5c7e81c3 SL |
7319 | if (journal_dev) { |
7320 | char b[BDEVNAME_SIZE]; | |
7321 | ||
cc6167b4 N |
7322 | pr_debug("md/raid:%s: using device %s as journal\n", |
7323 | mdname(mddev), bdevname(journal_dev->bdev, b)); | |
5aabf7c4 SL |
7324 | if (r5l_init_log(conf, journal_dev)) |
7325 | goto abort; | |
5c7e81c3 SL |
7326 | } |
7327 | ||
1da177e4 LT |
7328 | return 0; |
7329 | abort: | |
01f96c0a | 7330 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7331 | print_raid5_conf(conf); |
7332 | free_conf(conf); | |
1da177e4 | 7333 | mddev->private = NULL; |
cc6167b4 | 7334 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7335 | return -EIO; |
7336 | } | |
7337 | ||
afa0f557 | 7338 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7339 | { |
afa0f557 | 7340 | struct r5conf *conf = priv; |
1da177e4 | 7341 | |
95fc17aa | 7342 | free_conf(conf); |
a64c876f | 7343 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7344 | } |
7345 | ||
849674e4 | 7346 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7347 | { |
d1688a6d | 7348 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7349 | int i; |
7350 | ||
9d8f0363 | 7351 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7352 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7353 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7354 | rcu_read_lock(); |
7355 | for (i = 0; i < conf->raid_disks; i++) { | |
7356 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7357 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7358 | } | |
7359 | rcu_read_unlock(); | |
1da177e4 | 7360 | seq_printf (seq, "]"); |
1da177e4 LT |
7361 | } |
7362 | ||
d1688a6d | 7363 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7364 | { |
7365 | int i; | |
7366 | struct disk_info *tmp; | |
7367 | ||
cc6167b4 | 7368 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7369 | if (!conf) { |
cc6167b4 | 7370 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7371 | return; |
7372 | } | |
cc6167b4 | 7373 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7374 | conf->raid_disks, |
7375 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7376 | |
7377 | for (i = 0; i < conf->raid_disks; i++) { | |
7378 | char b[BDEVNAME_SIZE]; | |
7379 | tmp = conf->disks + i; | |
7380 | if (tmp->rdev) | |
cc6167b4 | 7381 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7382 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7383 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7384 | } |
7385 | } | |
7386 | ||
fd01b88c | 7387 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7388 | { |
7389 | int i; | |
d1688a6d | 7390 | struct r5conf *conf = mddev->private; |
1da177e4 | 7391 | struct disk_info *tmp; |
6b965620 N |
7392 | int count = 0; |
7393 | unsigned long flags; | |
1da177e4 LT |
7394 | |
7395 | for (i = 0; i < conf->raid_disks; i++) { | |
7396 | tmp = conf->disks + i; | |
dd054fce N |
7397 | if (tmp->replacement |
7398 | && tmp->replacement->recovery_offset == MaxSector | |
7399 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7400 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7401 | /* Replacement has just become active. */ | |
7402 | if (!tmp->rdev | |
7403 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7404 | count++; | |
7405 | if (tmp->rdev) { | |
7406 | /* Replaced device not technically faulty, | |
7407 | * but we need to be sure it gets removed | |
7408 | * and never re-added. | |
7409 | */ | |
7410 | set_bit(Faulty, &tmp->rdev->flags); | |
7411 | sysfs_notify_dirent_safe( | |
7412 | tmp->rdev->sysfs_state); | |
7413 | } | |
7414 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7415 | } else if (tmp->rdev | |
70fffd0b | 7416 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7417 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7418 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7419 | count++; |
43c73ca4 | 7420 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7421 | } |
7422 | } | |
6b965620 | 7423 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7424 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7425 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7426 | print_raid5_conf(conf); |
6b965620 | 7427 | return count; |
1da177e4 LT |
7428 | } |
7429 | ||
b8321b68 | 7430 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7431 | { |
d1688a6d | 7432 | struct r5conf *conf = mddev->private; |
1da177e4 | 7433 | int err = 0; |
b8321b68 | 7434 | int number = rdev->raid_disk; |
657e3e4d | 7435 | struct md_rdev **rdevp; |
1da177e4 LT |
7436 | struct disk_info *p = conf->disks + number; |
7437 | ||
7438 | print_raid5_conf(conf); | |
f6b6ec5c SL |
7439 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
7440 | struct r5l_log *log; | |
c2bb6242 | 7441 | /* |
f6b6ec5c SL |
7442 | * we can't wait pending write here, as this is called in |
7443 | * raid5d, wait will deadlock. | |
c2bb6242 | 7444 | */ |
f6b6ec5c SL |
7445 | if (atomic_read(&mddev->writes_pending)) |
7446 | return -EBUSY; | |
7447 | log = conf->log; | |
7448 | conf->log = NULL; | |
7449 | synchronize_rcu(); | |
7450 | r5l_exit_log(log); | |
7451 | return 0; | |
c2bb6242 | 7452 | } |
657e3e4d N |
7453 | if (rdev == p->rdev) |
7454 | rdevp = &p->rdev; | |
7455 | else if (rdev == p->replacement) | |
7456 | rdevp = &p->replacement; | |
7457 | else | |
7458 | return 0; | |
7459 | ||
7460 | if (number >= conf->raid_disks && | |
7461 | conf->reshape_progress == MaxSector) | |
7462 | clear_bit(In_sync, &rdev->flags); | |
7463 | ||
7464 | if (test_bit(In_sync, &rdev->flags) || | |
7465 | atomic_read(&rdev->nr_pending)) { | |
7466 | err = -EBUSY; | |
7467 | goto abort; | |
7468 | } | |
7469 | /* Only remove non-faulty devices if recovery | |
7470 | * isn't possible. | |
7471 | */ | |
7472 | if (!test_bit(Faulty, &rdev->flags) && | |
7473 | mddev->recovery_disabled != conf->recovery_disabled && | |
7474 | !has_failed(conf) && | |
dd054fce | 7475 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7476 | number < conf->raid_disks) { |
7477 | err = -EBUSY; | |
7478 | goto abort; | |
7479 | } | |
7480 | *rdevp = NULL; | |
d787be40 N |
7481 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7482 | synchronize_rcu(); | |
7483 | if (atomic_read(&rdev->nr_pending)) { | |
7484 | /* lost the race, try later */ | |
7485 | err = -EBUSY; | |
7486 | *rdevp = rdev; | |
7487 | } | |
7488 | } | |
7489 | if (p->replacement) { | |
dd054fce N |
7490 | /* We must have just cleared 'rdev' */ |
7491 | p->rdev = p->replacement; | |
7492 | clear_bit(Replacement, &p->replacement->flags); | |
7493 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7494 | * but will never see neither - if they are careful | |
7495 | */ | |
7496 | p->replacement = NULL; | |
7497 | clear_bit(WantReplacement, &rdev->flags); | |
7498 | } else | |
7499 | /* We might have just removed the Replacement as faulty- | |
7500 | * clear the bit just in case | |
7501 | */ | |
7502 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7503 | abort: |
7504 | ||
7505 | print_raid5_conf(conf); | |
7506 | return err; | |
7507 | } | |
7508 | ||
fd01b88c | 7509 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7510 | { |
d1688a6d | 7511 | struct r5conf *conf = mddev->private; |
199050ea | 7512 | int err = -EEXIST; |
1da177e4 LT |
7513 | int disk; |
7514 | struct disk_info *p; | |
6c2fce2e NB |
7515 | int first = 0; |
7516 | int last = conf->raid_disks - 1; | |
1da177e4 | 7517 | |
f6b6ec5c SL |
7518 | if (test_bit(Journal, &rdev->flags)) { |
7519 | char b[BDEVNAME_SIZE]; | |
7520 | if (conf->log) | |
7521 | return -EBUSY; | |
7522 | ||
7523 | rdev->raid_disk = 0; | |
7524 | /* | |
7525 | * The array is in readonly mode if journal is missing, so no | |
7526 | * write requests running. We should be safe | |
7527 | */ | |
7528 | r5l_init_log(conf, rdev); | |
cc6167b4 N |
7529 | pr_debug("md/raid:%s: using device %s as journal\n", |
7530 | mdname(mddev), bdevname(rdev->bdev, b)); | |
f6b6ec5c SL |
7531 | return 0; |
7532 | } | |
7f0da59b N |
7533 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7534 | return -EBUSY; | |
7535 | ||
dc10c643 | 7536 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7537 | /* no point adding a device */ |
199050ea | 7538 | return -EINVAL; |
1da177e4 | 7539 | |
6c2fce2e NB |
7540 | if (rdev->raid_disk >= 0) |
7541 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7542 | |
7543 | /* | |
16a53ecc N |
7544 | * find the disk ... but prefer rdev->saved_raid_disk |
7545 | * if possible. | |
1da177e4 | 7546 | */ |
16a53ecc | 7547 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7548 | rdev->saved_raid_disk >= first && |
16a53ecc | 7549 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7550 | first = rdev->saved_raid_disk; |
7551 | ||
7552 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7553 | p = conf->disks + disk; |
7554 | if (p->rdev == NULL) { | |
b2d444d7 | 7555 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7556 | rdev->raid_disk = disk; |
199050ea | 7557 | err = 0; |
72626685 N |
7558 | if (rdev->saved_raid_disk != disk) |
7559 | conf->fullsync = 1; | |
d6065f7b | 7560 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7561 | goto out; |
1da177e4 | 7562 | } |
5cfb22a1 N |
7563 | } |
7564 | for (disk = first; disk <= last; disk++) { | |
7565 | p = conf->disks + disk; | |
7bfec5f3 N |
7566 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7567 | p->replacement == NULL) { | |
7568 | clear_bit(In_sync, &rdev->flags); | |
7569 | set_bit(Replacement, &rdev->flags); | |
7570 | rdev->raid_disk = disk; | |
7571 | err = 0; | |
7572 | conf->fullsync = 1; | |
7573 | rcu_assign_pointer(p->replacement, rdev); | |
7574 | break; | |
7575 | } | |
7576 | } | |
5cfb22a1 | 7577 | out: |
1da177e4 | 7578 | print_raid5_conf(conf); |
199050ea | 7579 | return err; |
1da177e4 LT |
7580 | } |
7581 | ||
fd01b88c | 7582 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7583 | { |
7584 | /* no resync is happening, and there is enough space | |
7585 | * on all devices, so we can resize. | |
7586 | * We need to make sure resync covers any new space. | |
7587 | * If the array is shrinking we should possibly wait until | |
7588 | * any io in the removed space completes, but it hardly seems | |
7589 | * worth it. | |
7590 | */ | |
a4a6125a | 7591 | sector_t newsize; |
3cb5edf4 N |
7592 | struct r5conf *conf = mddev->private; |
7593 | ||
713cf5a6 SL |
7594 | if (conf->log) |
7595 | return -EINVAL; | |
3cb5edf4 | 7596 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7597 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7598 | if (mddev->external_size && | |
7599 | mddev->array_sectors > newsize) | |
b522adcd | 7600 | return -EINVAL; |
a4a6125a N |
7601 | if (mddev->bitmap) { |
7602 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7603 | if (ret) | |
7604 | return ret; | |
7605 | } | |
7606 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7607 | if (sectors > mddev->dev_sectors && |
7608 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7609 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7610 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7611 | } | |
58c0fed4 | 7612 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7613 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7614 | return 0; |
7615 | } | |
7616 | ||
fd01b88c | 7617 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7618 | { |
7619 | /* Can only proceed if there are plenty of stripe_heads. | |
7620 | * We need a minimum of one full stripe,, and for sensible progress | |
7621 | * it is best to have about 4 times that. | |
7622 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7623 | * allow for chunk sizes up to 256K, which is probably OK. | |
7624 | * If the chunk size is greater, user-space should request more | |
7625 | * stripe_heads first. | |
7626 | */ | |
d1688a6d | 7627 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7628 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7629 | > conf->min_nr_stripes || |
01ee22b4 | 7630 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7631 | > conf->min_nr_stripes) { |
cc6167b4 N |
7632 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7633 | mdname(mddev), | |
7634 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
7635 | / STRIPE_SIZE)*4); | |
01ee22b4 N |
7636 | return 0; |
7637 | } | |
7638 | return 1; | |
7639 | } | |
7640 | ||
fd01b88c | 7641 | static int check_reshape(struct mddev *mddev) |
29269553 | 7642 | { |
d1688a6d | 7643 | struct r5conf *conf = mddev->private; |
29269553 | 7644 | |
713cf5a6 SL |
7645 | if (conf->log) |
7646 | return -EINVAL; | |
88ce4930 N |
7647 | if (mddev->delta_disks == 0 && |
7648 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7649 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7650 | return 0; /* nothing to do */ |
674806d6 | 7651 | if (has_failed(conf)) |
ec32a2bd | 7652 | return -EINVAL; |
fdcfbbb6 | 7653 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7654 | /* We might be able to shrink, but the devices must |
7655 | * be made bigger first. | |
7656 | * For raid6, 4 is the minimum size. | |
7657 | * Otherwise 2 is the minimum | |
7658 | */ | |
7659 | int min = 2; | |
7660 | if (mddev->level == 6) | |
7661 | min = 4; | |
7662 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7663 | return -EINVAL; | |
7664 | } | |
29269553 | 7665 | |
01ee22b4 | 7666 | if (!check_stripe_cache(mddev)) |
29269553 | 7667 | return -ENOSPC; |
29269553 | 7668 | |
738a2738 N |
7669 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7670 | mddev->delta_disks > 0) | |
7671 | if (resize_chunks(conf, | |
7672 | conf->previous_raid_disks | |
7673 | + max(0, mddev->delta_disks), | |
7674 | max(mddev->new_chunk_sectors, | |
7675 | mddev->chunk_sectors) | |
7676 | ) < 0) | |
7677 | return -ENOMEM; | |
e56108d6 N |
7678 | return resize_stripes(conf, (conf->previous_raid_disks |
7679 | + mddev->delta_disks)); | |
63c70c4f N |
7680 | } |
7681 | ||
fd01b88c | 7682 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7683 | { |
d1688a6d | 7684 | struct r5conf *conf = mddev->private; |
3cb03002 | 7685 | struct md_rdev *rdev; |
63c70c4f | 7686 | int spares = 0; |
c04be0aa | 7687 | unsigned long flags; |
63c70c4f | 7688 | |
f416885e | 7689 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7690 | return -EBUSY; |
7691 | ||
01ee22b4 N |
7692 | if (!check_stripe_cache(mddev)) |
7693 | return -ENOSPC; | |
7694 | ||
30b67645 N |
7695 | if (has_failed(conf)) |
7696 | return -EINVAL; | |
7697 | ||
c6563a8c | 7698 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7699 | if (!test_bit(In_sync, &rdev->flags) |
7700 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7701 | spares++; |
c6563a8c | 7702 | } |
63c70c4f | 7703 | |
f416885e | 7704 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7705 | /* Not enough devices even to make a degraded array |
7706 | * of that size | |
7707 | */ | |
7708 | return -EINVAL; | |
7709 | ||
ec32a2bd N |
7710 | /* Refuse to reduce size of the array. Any reductions in |
7711 | * array size must be through explicit setting of array_size | |
7712 | * attribute. | |
7713 | */ | |
7714 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7715 | < mddev->array_sectors) { | |
cc6167b4 N |
7716 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7717 | mdname(mddev)); | |
ec32a2bd N |
7718 | return -EINVAL; |
7719 | } | |
7720 | ||
f6705578 | 7721 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7722 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7723 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7724 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7725 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7726 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7727 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7728 | conf->prev_algo = conf->algorithm; |
7729 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7730 | conf->generation++; |
7731 | /* Code that selects data_offset needs to see the generation update | |
7732 | * if reshape_progress has been set - so a memory barrier needed. | |
7733 | */ | |
7734 | smp_mb(); | |
2c810cdd | 7735 | if (mddev->reshape_backwards) |
fef9c61f N |
7736 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7737 | else | |
7738 | conf->reshape_progress = 0; | |
7739 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7740 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7741 | spin_unlock_irq(&conf->device_lock); |
7742 | ||
4d77e3ba N |
7743 | /* Now make sure any requests that proceeded on the assumption |
7744 | * the reshape wasn't running - like Discard or Read - have | |
7745 | * completed. | |
7746 | */ | |
7747 | mddev_suspend(mddev); | |
7748 | mddev_resume(mddev); | |
7749 | ||
29269553 N |
7750 | /* Add some new drives, as many as will fit. |
7751 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7752 | * Don't add devices if we are reducing the number of |
7753 | * devices in the array. This is because it is not possible | |
7754 | * to correctly record the "partially reconstructed" state of | |
7755 | * such devices during the reshape and confusion could result. | |
29269553 | 7756 | */ |
87a8dec9 | 7757 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7758 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7759 | if (rdev->raid_disk < 0 && |
7760 | !test_bit(Faulty, &rdev->flags)) { | |
7761 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7762 | if (rdev->raid_disk |
9d4c7d87 | 7763 | >= conf->previous_raid_disks) |
87a8dec9 | 7764 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7765 | else |
87a8dec9 | 7766 | rdev->recovery_offset = 0; |
36fad858 NK |
7767 | |
7768 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7769 | /* Failure here is OK */; |
50da0840 | 7770 | } |
87a8dec9 N |
7771 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7772 | && !test_bit(Faulty, &rdev->flags)) { | |
7773 | /* This is a spare that was manually added */ | |
7774 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7775 | } |
29269553 | 7776 | |
87a8dec9 N |
7777 | /* When a reshape changes the number of devices, |
7778 | * ->degraded is measured against the larger of the | |
7779 | * pre and post number of devices. | |
7780 | */ | |
ec32a2bd | 7781 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7782 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
7783 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7784 | } | |
63c70c4f | 7785 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7786 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 7787 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 7788 | |
29269553 N |
7789 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7790 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7791 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7792 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7793 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7794 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7795 | "reshape"); |
29269553 N |
7796 | if (!mddev->sync_thread) { |
7797 | mddev->recovery = 0; | |
7798 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7799 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7800 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7801 | mddev->new_chunk_sectors = |
7802 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7803 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7804 | rdev_for_each(rdev, mddev) |
7805 | rdev->new_data_offset = rdev->data_offset; | |
7806 | smp_wmb(); | |
ba8805b9 | 7807 | conf->generation --; |
fef9c61f | 7808 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7809 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7810 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7811 | spin_unlock_irq(&conf->device_lock); |
7812 | return -EAGAIN; | |
7813 | } | |
c8f517c4 | 7814 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7815 | md_wakeup_thread(mddev->sync_thread); |
7816 | md_new_event(mddev); | |
7817 | return 0; | |
7818 | } | |
29269553 | 7819 | |
ec32a2bd N |
7820 | /* This is called from the reshape thread and should make any |
7821 | * changes needed in 'conf' | |
7822 | */ | |
d1688a6d | 7823 | static void end_reshape(struct r5conf *conf) |
29269553 | 7824 | { |
29269553 | 7825 | |
f6705578 | 7826 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7827 | struct md_rdev *rdev; |
f6705578 | 7828 | |
f6705578 | 7829 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7830 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7831 | rdev_for_each(rdev, conf->mddev) |
7832 | rdev->data_offset = rdev->new_data_offset; | |
7833 | smp_wmb(); | |
fef9c61f | 7834 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7835 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 7836 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7837 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7838 | |
7839 | /* read-ahead size must cover two whole stripes, which is | |
7840 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7841 | */ | |
4a5add49 | 7842 | if (conf->mddev->queue) { |
cea9c228 | 7843 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7844 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7845 | / PAGE_SIZE); |
dc3b17cc JK |
7846 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7847 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 7848 | } |
29269553 | 7849 | } |
29269553 N |
7850 | } |
7851 | ||
ec32a2bd N |
7852 | /* This is called from the raid5d thread with mddev_lock held. |
7853 | * It makes config changes to the device. | |
7854 | */ | |
fd01b88c | 7855 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7856 | { |
d1688a6d | 7857 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7858 | |
7859 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7860 | ||
ec32a2bd N |
7861 | if (mddev->delta_disks > 0) { |
7862 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
fe67d19a HM |
7863 | if (mddev->queue) { |
7864 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
7865 | revalidate_disk(mddev->gendisk); | |
7866 | } | |
ec32a2bd N |
7867 | } else { |
7868 | int d; | |
908f4fbd | 7869 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 7870 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 7871 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
7872 | for (d = conf->raid_disks ; |
7873 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7874 | d++) { |
3cb03002 | 7875 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7876 | if (rdev) |
7877 | clear_bit(In_sync, &rdev->flags); | |
7878 | rdev = conf->disks[d].replacement; | |
7879 | if (rdev) | |
7880 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7881 | } |
cea9c228 | 7882 | } |
88ce4930 | 7883 | mddev->layout = conf->algorithm; |
09c9e5fa | 7884 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7885 | mddev->reshape_position = MaxSector; |
7886 | mddev->delta_disks = 0; | |
2c810cdd | 7887 | mddev->reshape_backwards = 0; |
cea9c228 N |
7888 | } |
7889 | } | |
7890 | ||
fd01b88c | 7891 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7892 | { |
d1688a6d | 7893 | struct r5conf *conf = mddev->private; |
72626685 N |
7894 | |
7895 | switch(state) { | |
e464eafd N |
7896 | case 2: /* resume for a suspend */ |
7897 | wake_up(&conf->wait_for_overlap); | |
7898 | break; | |
7899 | ||
72626685 | 7900 | case 1: /* stop all writes */ |
566c09c5 | 7901 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7902 | /* '2' tells resync/reshape to pause so that all |
7903 | * active stripes can drain | |
7904 | */ | |
a39f7afd | 7905 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 7906 | conf->quiesce = 2; |
b1b46486 | 7907 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7908 | atomic_read(&conf->active_stripes) == 0 && |
7909 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7910 | unlock_all_device_hash_locks_irq(conf), |
7911 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7912 | conf->quiesce = 1; |
566c09c5 | 7913 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7914 | /* allow reshape to continue */ |
7915 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7916 | break; |
7917 | ||
7918 | case 0: /* re-enable writes */ | |
566c09c5 | 7919 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7920 | conf->quiesce = 0; |
b1b46486 | 7921 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7922 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7923 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7924 | break; |
7925 | } | |
e6c033f7 | 7926 | r5l_quiesce(conf->log, state); |
72626685 | 7927 | } |
b15c2e57 | 7928 | |
fd01b88c | 7929 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7930 | { |
e373ab10 | 7931 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7932 | sector_t sectors; |
54071b38 | 7933 | |
f1b29bca | 7934 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7935 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
7936 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7937 | mdname(mddev)); | |
f1b29bca DW |
7938 | return ERR_PTR(-EINVAL); |
7939 | } | |
7940 | ||
e373ab10 N |
7941 | sectors = raid0_conf->strip_zone[0].zone_end; |
7942 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7943 | mddev->dev_sectors = sectors; |
f1b29bca | 7944 | mddev->new_level = level; |
54071b38 TM |
7945 | mddev->new_layout = ALGORITHM_PARITY_N; |
7946 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7947 | mddev->raid_disks += 1; | |
7948 | mddev->delta_disks = 1; | |
7949 | /* make sure it will be not marked as dirty */ | |
7950 | mddev->recovery_cp = MaxSector; | |
7951 | ||
7952 | return setup_conf(mddev); | |
7953 | } | |
7954 | ||
fd01b88c | 7955 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7956 | { |
7957 | int chunksect; | |
6995f0b2 | 7958 | void *ret; |
d562b0c4 N |
7959 | |
7960 | if (mddev->raid_disks != 2 || | |
7961 | mddev->degraded > 1) | |
7962 | return ERR_PTR(-EINVAL); | |
7963 | ||
7964 | /* Should check if there are write-behind devices? */ | |
7965 | ||
7966 | chunksect = 64*2; /* 64K by default */ | |
7967 | ||
7968 | /* The array must be an exact multiple of chunksize */ | |
7969 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
7970 | chunksect >>= 1; | |
7971 | ||
7972 | if ((chunksect<<9) < STRIPE_SIZE) | |
7973 | /* array size does not allow a suitable chunk size */ | |
7974 | return ERR_PTR(-EINVAL); | |
7975 | ||
7976 | mddev->new_level = 5; | |
7977 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 7978 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 7979 | |
6995f0b2 | 7980 | ret = setup_conf(mddev); |
32cd7cbb | 7981 | if (!IS_ERR(ret)) |
394ed8e4 SL |
7982 | mddev_clear_unsupported_flags(mddev, |
7983 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 7984 | return ret; |
d562b0c4 N |
7985 | } |
7986 | ||
fd01b88c | 7987 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
7988 | { |
7989 | int new_layout; | |
7990 | ||
7991 | switch (mddev->layout) { | |
7992 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7993 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
7994 | break; | |
7995 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7996 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
7997 | break; | |
7998 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7999 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8000 | break; | |
8001 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8002 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8003 | break; | |
8004 | case ALGORITHM_PARITY_0_6: | |
8005 | new_layout = ALGORITHM_PARITY_0; | |
8006 | break; | |
8007 | case ALGORITHM_PARITY_N: | |
8008 | new_layout = ALGORITHM_PARITY_N; | |
8009 | break; | |
8010 | default: | |
8011 | return ERR_PTR(-EINVAL); | |
8012 | } | |
8013 | mddev->new_level = 5; | |
8014 | mddev->new_layout = new_layout; | |
8015 | mddev->delta_disks = -1; | |
8016 | mddev->raid_disks -= 1; | |
8017 | return setup_conf(mddev); | |
8018 | } | |
8019 | ||
fd01b88c | 8020 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8021 | { |
88ce4930 N |
8022 | /* For a 2-drive array, the layout and chunk size can be changed |
8023 | * immediately as not restriping is needed. | |
8024 | * For larger arrays we record the new value - after validation | |
8025 | * to be used by a reshape pass. | |
b3546035 | 8026 | */ |
d1688a6d | 8027 | struct r5conf *conf = mddev->private; |
597a711b | 8028 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8029 | |
597a711b | 8030 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8031 | return -EINVAL; |
8032 | if (new_chunk > 0) { | |
0ba459d2 | 8033 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8034 | return -EINVAL; |
597a711b | 8035 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8036 | return -EINVAL; |
597a711b | 8037 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8038 | /* not factor of array size */ |
8039 | return -EINVAL; | |
8040 | } | |
8041 | ||
8042 | /* They look valid */ | |
8043 | ||
88ce4930 | 8044 | if (mddev->raid_disks == 2) { |
597a711b N |
8045 | /* can make the change immediately */ |
8046 | if (mddev->new_layout >= 0) { | |
8047 | conf->algorithm = mddev->new_layout; | |
8048 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8049 | } |
8050 | if (new_chunk > 0) { | |
597a711b N |
8051 | conf->chunk_sectors = new_chunk ; |
8052 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8053 | } |
2953079c | 8054 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8055 | md_wakeup_thread(mddev->thread); |
b3546035 | 8056 | } |
50ac168a | 8057 | return check_reshape(mddev); |
88ce4930 N |
8058 | } |
8059 | ||
fd01b88c | 8060 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8061 | { |
597a711b | 8062 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8063 | |
597a711b | 8064 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8065 | return -EINVAL; |
b3546035 | 8066 | if (new_chunk > 0) { |
0ba459d2 | 8067 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8068 | return -EINVAL; |
597a711b | 8069 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8070 | return -EINVAL; |
597a711b | 8071 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8072 | /* not factor of array size */ |
8073 | return -EINVAL; | |
b3546035 | 8074 | } |
88ce4930 N |
8075 | |
8076 | /* They look valid */ | |
50ac168a | 8077 | return check_reshape(mddev); |
b3546035 N |
8078 | } |
8079 | ||
fd01b88c | 8080 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8081 | { |
8082 | /* raid5 can take over: | |
f1b29bca | 8083 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8084 | * raid1 - if there are two drives. We need to know the chunk size |
8085 | * raid4 - trivial - just use a raid4 layout. | |
8086 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8087 | */ |
f1b29bca DW |
8088 | if (mddev->level == 0) |
8089 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8090 | if (mddev->level == 1) |
8091 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8092 | if (mddev->level == 4) { |
8093 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8094 | mddev->new_level = 5; | |
8095 | return setup_conf(mddev); | |
8096 | } | |
fc9739c6 N |
8097 | if (mddev->level == 6) |
8098 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8099 | |
8100 | return ERR_PTR(-EINVAL); | |
8101 | } | |
8102 | ||
fd01b88c | 8103 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8104 | { |
f1b29bca DW |
8105 | /* raid4 can take over: |
8106 | * raid0 - if there is only one strip zone | |
8107 | * raid5 - if layout is right | |
a78d38a1 | 8108 | */ |
f1b29bca DW |
8109 | if (mddev->level == 0) |
8110 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8111 | if (mddev->level == 5 && |
8112 | mddev->layout == ALGORITHM_PARITY_N) { | |
8113 | mddev->new_layout = 0; | |
8114 | mddev->new_level = 4; | |
8115 | return setup_conf(mddev); | |
8116 | } | |
8117 | return ERR_PTR(-EINVAL); | |
8118 | } | |
d562b0c4 | 8119 | |
84fc4b56 | 8120 | static struct md_personality raid5_personality; |
245f46c2 | 8121 | |
fd01b88c | 8122 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8123 | { |
8124 | /* Currently can only take over a raid5. We map the | |
8125 | * personality to an equivalent raid6 personality | |
8126 | * with the Q block at the end. | |
8127 | */ | |
8128 | int new_layout; | |
8129 | ||
8130 | if (mddev->pers != &raid5_personality) | |
8131 | return ERR_PTR(-EINVAL); | |
8132 | if (mddev->degraded > 1) | |
8133 | return ERR_PTR(-EINVAL); | |
8134 | if (mddev->raid_disks > 253) | |
8135 | return ERR_PTR(-EINVAL); | |
8136 | if (mddev->raid_disks < 3) | |
8137 | return ERR_PTR(-EINVAL); | |
8138 | ||
8139 | switch (mddev->layout) { | |
8140 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8141 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8142 | break; | |
8143 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8144 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8145 | break; | |
8146 | case ALGORITHM_LEFT_SYMMETRIC: | |
8147 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8148 | break; | |
8149 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8150 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8151 | break; | |
8152 | case ALGORITHM_PARITY_0: | |
8153 | new_layout = ALGORITHM_PARITY_0_6; | |
8154 | break; | |
8155 | case ALGORITHM_PARITY_N: | |
8156 | new_layout = ALGORITHM_PARITY_N; | |
8157 | break; | |
8158 | default: | |
8159 | return ERR_PTR(-EINVAL); | |
8160 | } | |
8161 | mddev->new_level = 6; | |
8162 | mddev->new_layout = new_layout; | |
8163 | mddev->delta_disks = 1; | |
8164 | mddev->raid_disks += 1; | |
8165 | return setup_conf(mddev); | |
8166 | } | |
8167 | ||
84fc4b56 | 8168 | static struct md_personality raid6_personality = |
16a53ecc N |
8169 | { |
8170 | .name = "raid6", | |
8171 | .level = 6, | |
8172 | .owner = THIS_MODULE, | |
849674e4 SL |
8173 | .make_request = raid5_make_request, |
8174 | .run = raid5_run, | |
afa0f557 | 8175 | .free = raid5_free, |
849674e4 SL |
8176 | .status = raid5_status, |
8177 | .error_handler = raid5_error, | |
16a53ecc N |
8178 | .hot_add_disk = raid5_add_disk, |
8179 | .hot_remove_disk= raid5_remove_disk, | |
8180 | .spare_active = raid5_spare_active, | |
849674e4 | 8181 | .sync_request = raid5_sync_request, |
16a53ecc | 8182 | .resize = raid5_resize, |
80c3a6ce | 8183 | .size = raid5_size, |
50ac168a | 8184 | .check_reshape = raid6_check_reshape, |
f416885e | 8185 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8186 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8187 | .quiesce = raid5_quiesce, |
245f46c2 | 8188 | .takeover = raid6_takeover, |
5c675f83 | 8189 | .congested = raid5_congested, |
16a53ecc | 8190 | }; |
84fc4b56 | 8191 | static struct md_personality raid5_personality = |
1da177e4 LT |
8192 | { |
8193 | .name = "raid5", | |
2604b703 | 8194 | .level = 5, |
1da177e4 | 8195 | .owner = THIS_MODULE, |
849674e4 SL |
8196 | .make_request = raid5_make_request, |
8197 | .run = raid5_run, | |
afa0f557 | 8198 | .free = raid5_free, |
849674e4 SL |
8199 | .status = raid5_status, |
8200 | .error_handler = raid5_error, | |
1da177e4 LT |
8201 | .hot_add_disk = raid5_add_disk, |
8202 | .hot_remove_disk= raid5_remove_disk, | |
8203 | .spare_active = raid5_spare_active, | |
849674e4 | 8204 | .sync_request = raid5_sync_request, |
1da177e4 | 8205 | .resize = raid5_resize, |
80c3a6ce | 8206 | .size = raid5_size, |
63c70c4f N |
8207 | .check_reshape = raid5_check_reshape, |
8208 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8209 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8210 | .quiesce = raid5_quiesce, |
d562b0c4 | 8211 | .takeover = raid5_takeover, |
5c675f83 | 8212 | .congested = raid5_congested, |
1da177e4 LT |
8213 | }; |
8214 | ||
84fc4b56 | 8215 | static struct md_personality raid4_personality = |
1da177e4 | 8216 | { |
2604b703 N |
8217 | .name = "raid4", |
8218 | .level = 4, | |
8219 | .owner = THIS_MODULE, | |
849674e4 SL |
8220 | .make_request = raid5_make_request, |
8221 | .run = raid5_run, | |
afa0f557 | 8222 | .free = raid5_free, |
849674e4 SL |
8223 | .status = raid5_status, |
8224 | .error_handler = raid5_error, | |
2604b703 N |
8225 | .hot_add_disk = raid5_add_disk, |
8226 | .hot_remove_disk= raid5_remove_disk, | |
8227 | .spare_active = raid5_spare_active, | |
849674e4 | 8228 | .sync_request = raid5_sync_request, |
2604b703 | 8229 | .resize = raid5_resize, |
80c3a6ce | 8230 | .size = raid5_size, |
3d37890b N |
8231 | .check_reshape = raid5_check_reshape, |
8232 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8233 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8234 | .quiesce = raid5_quiesce, |
a78d38a1 | 8235 | .takeover = raid4_takeover, |
5c675f83 | 8236 | .congested = raid5_congested, |
2604b703 N |
8237 | }; |
8238 | ||
8239 | static int __init raid5_init(void) | |
8240 | { | |
29c6d1bb SAS |
8241 | int ret; |
8242 | ||
851c30c9 SL |
8243 | raid5_wq = alloc_workqueue("raid5wq", |
8244 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8245 | if (!raid5_wq) | |
8246 | return -ENOMEM; | |
29c6d1bb SAS |
8247 | |
8248 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8249 | "md/raid5:prepare", | |
8250 | raid456_cpu_up_prepare, | |
8251 | raid456_cpu_dead); | |
8252 | if (ret) { | |
8253 | destroy_workqueue(raid5_wq); | |
8254 | return ret; | |
8255 | } | |
16a53ecc | 8256 | register_md_personality(&raid6_personality); |
2604b703 N |
8257 | register_md_personality(&raid5_personality); |
8258 | register_md_personality(&raid4_personality); | |
8259 | return 0; | |
1da177e4 LT |
8260 | } |
8261 | ||
2604b703 | 8262 | static void raid5_exit(void) |
1da177e4 | 8263 | { |
16a53ecc | 8264 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8265 | unregister_md_personality(&raid5_personality); |
8266 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8267 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8268 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8269 | } |
8270 | ||
8271 | module_init(raid5_init); | |
8272 | module_exit(raid5_exit); | |
8273 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8274 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8275 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8276 | MODULE_ALIAS("md-raid5"); |
8277 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8278 | MODULE_ALIAS("md-level-5"); |
8279 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8280 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8281 | MODULE_ALIAS("md-raid6"); | |
8282 | MODULE_ALIAS("md-level-6"); | |
8283 | ||
8284 | /* This used to be two separate modules, they were: */ | |
8285 | MODULE_ALIAS("raid5"); | |
8286 | MODULE_ALIAS("raid6"); |