]>
Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * raid5.c : Multiple Devices driver for Linux | |
4 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
5 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 6 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 7 | * |
16a53ecc N |
8 | * RAID-4/5/6 management functions. |
9 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
10 | * by donating a test server! | |
1da177e4 LT |
11 | */ |
12 | ||
ae3c20cc N |
13 | /* |
14 | * BITMAP UNPLUGGING: | |
15 | * | |
16 | * The sequencing for updating the bitmap reliably is a little | |
17 | * subtle (and I got it wrong the first time) so it deserves some | |
18 | * explanation. | |
19 | * | |
20 | * We group bitmap updates into batches. Each batch has a number. | |
21 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
22 | * conf->seq_write is the number of the last batch successfully written. |
23 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
24 | * new additions. |
25 | * When we discover that we will need to write to any block in a stripe | |
26 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 27 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
28 | * When we are ready to do a write, if that batch hasn't been written yet, |
29 | * we plug the array and queue the stripe for later. | |
30 | * When an unplug happens, we increment bm_flush, thus closing the current | |
31 | * batch. | |
32 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
33 | * to the bitmap, and advance bm_write to where bm_flush was. | |
34 | * This may occasionally write a bit out twice, but is sure never to | |
35 | * miss any bits. | |
36 | */ | |
1da177e4 | 37 | |
bff61975 | 38 | #include <linux/blkdev.h> |
f6705578 | 39 | #include <linux/kthread.h> |
f701d589 | 40 | #include <linux/raid/pq.h> |
91c00924 | 41 | #include <linux/async_tx.h> |
056075c7 | 42 | #include <linux/module.h> |
07a3b417 | 43 | #include <linux/async.h> |
bff61975 | 44 | #include <linux/seq_file.h> |
36d1c647 | 45 | #include <linux/cpu.h> |
5a0e3ad6 | 46 | #include <linux/slab.h> |
8bda470e | 47 | #include <linux/ratelimit.h> |
851c30c9 | 48 | #include <linux/nodemask.h> |
3f07c014 | 49 | |
a9add5d9 | 50 | #include <trace/events/block.h> |
aaf9f12e | 51 | #include <linux/list_sort.h> |
a9add5d9 | 52 | |
43b2e5d8 | 53 | #include "md.h" |
bff61975 | 54 | #include "raid5.h" |
54071b38 | 55 | #include "raid0.h" |
935fe098 | 56 | #include "md-bitmap.h" |
ff875738 | 57 | #include "raid5-log.h" |
72626685 | 58 | |
394ed8e4 SL |
59 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
60 | ||
851c30c9 SL |
61 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
62 | #define ANY_GROUP NUMA_NO_NODE | |
63 | ||
8e0e99ba N |
64 | static bool devices_handle_discard_safely = false; |
65 | module_param(devices_handle_discard_safely, bool, 0644); | |
66 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
67 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 68 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 69 | |
d1688a6d | 70 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 | 71 | { |
c911c46c | 72 | int hash = (sect >> RAID5_STRIPE_SHIFT(conf)) & HASH_MASK; |
db298e19 N |
73 | return &conf->stripe_hashtbl[hash]; |
74 | } | |
1da177e4 | 75 | |
c911c46c | 76 | static inline int stripe_hash_locks_hash(struct r5conf *conf, sector_t sect) |
566c09c5 | 77 | { |
c911c46c | 78 | return (sect >> RAID5_STRIPE_SHIFT(conf)) & STRIPE_HASH_LOCKS_MASK; |
566c09c5 SL |
79 | } |
80 | ||
81 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
82 | { | |
83 | spin_lock_irq(conf->hash_locks + hash); | |
84 | spin_lock(&conf->device_lock); | |
85 | } | |
86 | ||
87 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
88 | { | |
89 | spin_unlock(&conf->device_lock); | |
90 | spin_unlock_irq(conf->hash_locks + hash); | |
91 | } | |
92 | ||
93 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
94 | { | |
95 | int i; | |
3d05f3ae | 96 | spin_lock_irq(conf->hash_locks); |
566c09c5 SL |
97 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) |
98 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
99 | spin_lock(&conf->device_lock); | |
100 | } | |
101 | ||
102 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
103 | { | |
104 | int i; | |
105 | spin_unlock(&conf->device_lock); | |
3d05f3ae JC |
106 | for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--) |
107 | spin_unlock(conf->hash_locks + i); | |
108 | spin_unlock_irq(conf->hash_locks); | |
566c09c5 SL |
109 | } |
110 | ||
d0dabf7e N |
111 | /* Find first data disk in a raid6 stripe */ |
112 | static inline int raid6_d0(struct stripe_head *sh) | |
113 | { | |
67cc2b81 N |
114 | if (sh->ddf_layout) |
115 | /* ddf always start from first device */ | |
116 | return 0; | |
117 | /* md starts just after Q block */ | |
d0dabf7e N |
118 | if (sh->qd_idx == sh->disks - 1) |
119 | return 0; | |
120 | else | |
121 | return sh->qd_idx + 1; | |
122 | } | |
16a53ecc N |
123 | static inline int raid6_next_disk(int disk, int raid_disks) |
124 | { | |
125 | disk++; | |
126 | return (disk < raid_disks) ? disk : 0; | |
127 | } | |
a4456856 | 128 | |
d0dabf7e N |
129 | /* When walking through the disks in a raid5, starting at raid6_d0, |
130 | * We need to map each disk to a 'slot', where the data disks are slot | |
131 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
132 | * is raid_disks-1. This help does that mapping. | |
133 | */ | |
67cc2b81 N |
134 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
135 | int *count, int syndrome_disks) | |
d0dabf7e | 136 | { |
6629542e | 137 | int slot = *count; |
67cc2b81 | 138 | |
e4424fee | 139 | if (sh->ddf_layout) |
6629542e | 140 | (*count)++; |
d0dabf7e | 141 | if (idx == sh->pd_idx) |
67cc2b81 | 142 | return syndrome_disks; |
d0dabf7e | 143 | if (idx == sh->qd_idx) |
67cc2b81 | 144 | return syndrome_disks + 1; |
e4424fee | 145 | if (!sh->ddf_layout) |
6629542e | 146 | (*count)++; |
d0dabf7e N |
147 | return slot; |
148 | } | |
149 | ||
d1688a6d | 150 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 151 | |
600aa109 DW |
152 | static int stripe_operations_active(struct stripe_head *sh) |
153 | { | |
154 | return sh->check_state || sh->reconstruct_state || | |
155 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
156 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
157 | } | |
158 | ||
535ae4eb SL |
159 | static bool stripe_is_lowprio(struct stripe_head *sh) |
160 | { | |
161 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
162 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
163 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
164 | } | |
165 | ||
851c30c9 SL |
166 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
167 | { | |
168 | struct r5conf *conf = sh->raid_conf; | |
169 | struct r5worker_group *group; | |
bfc90cb0 | 170 | int thread_cnt; |
851c30c9 SL |
171 | int i, cpu = sh->cpu; |
172 | ||
173 | if (!cpu_online(cpu)) { | |
174 | cpu = cpumask_any(cpu_online_mask); | |
175 | sh->cpu = cpu; | |
176 | } | |
177 | ||
178 | if (list_empty(&sh->lru)) { | |
179 | struct r5worker_group *group; | |
180 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
181 | if (stripe_is_lowprio(sh)) |
182 | list_add_tail(&sh->lru, &group->loprio_list); | |
183 | else | |
184 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
185 | group->stripes_cnt++; |
186 | sh->group = group; | |
851c30c9 SL |
187 | } |
188 | ||
189 | if (conf->worker_cnt_per_group == 0) { | |
190 | md_wakeup_thread(conf->mddev->thread); | |
191 | return; | |
192 | } | |
193 | ||
194 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
195 | ||
bfc90cb0 SL |
196 | group->workers[0].working = true; |
197 | /* at least one worker should run to avoid race */ | |
198 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
199 | ||
200 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
201 | /* wakeup more workers */ | |
202 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
203 | if (group->workers[i].working == false) { | |
204 | group->workers[i].working = true; | |
205 | queue_work_on(sh->cpu, raid5_wq, | |
206 | &group->workers[i].work); | |
207 | thread_cnt--; | |
208 | } | |
209 | } | |
851c30c9 SL |
210 | } |
211 | ||
566c09c5 SL |
212 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
213 | struct list_head *temp_inactive_list) | |
1da177e4 | 214 | { |
1e6d690b SL |
215 | int i; |
216 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
217 | ||
4eb788df SL |
218 | BUG_ON(!list_empty(&sh->lru)); |
219 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
220 | |
221 | if (r5c_is_writeback(conf->log)) | |
222 | for (i = sh->disks; i--; ) | |
223 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
224 | injournal++; | |
a39f7afd | 225 | /* |
5ddf0440 SL |
226 | * In the following cases, the stripe cannot be released to cached |
227 | * lists. Therefore, we make the stripe write out and set | |
228 | * STRIPE_HANDLE: | |
229 | * 1. when quiesce in r5c write back; | |
230 | * 2. when resync is requested fot the stripe. | |
a39f7afd | 231 | */ |
5ddf0440 SL |
232 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) || |
233 | (conf->quiesce && r5c_is_writeback(conf->log) && | |
234 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) { | |
a39f7afd SL |
235 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) |
236 | r5c_make_stripe_write_out(sh); | |
237 | set_bit(STRIPE_HANDLE, &sh->state); | |
238 | } | |
1e6d690b | 239 | |
4eb788df SL |
240 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
241 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 242 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 243 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 244 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
245 | sh->bm_seq - conf->seq_write > 0) |
246 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
247 | else { | |
248 | clear_bit(STRIPE_DELAYED, &sh->state); | |
249 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 250 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
251 | if (stripe_is_lowprio(sh)) |
252 | list_add_tail(&sh->lru, | |
253 | &conf->loprio_list); | |
254 | else | |
255 | list_add_tail(&sh->lru, | |
256 | &conf->handle_list); | |
851c30c9 SL |
257 | } else { |
258 | raid5_wakeup_stripe_thread(sh); | |
259 | return; | |
260 | } | |
4eb788df SL |
261 | } |
262 | md_wakeup_thread(conf->mddev->thread); | |
263 | } else { | |
264 | BUG_ON(stripe_operations_active(sh)); | |
265 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
266 | if (atomic_dec_return(&conf->preread_active_stripes) | |
267 | < IO_THRESHOLD) | |
268 | md_wakeup_thread(conf->mddev->thread); | |
269 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
270 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
271 | if (!r5c_is_writeback(conf->log)) | |
272 | list_add_tail(&sh->lru, temp_inactive_list); | |
273 | else { | |
274 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
275 | if (injournal == 0) | |
276 | list_add_tail(&sh->lru, temp_inactive_list); | |
277 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
278 | /* full stripe */ | |
279 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
280 | atomic_inc(&conf->r5c_cached_full_stripes); | |
281 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
282 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
283 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 284 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
285 | } else |
286 | /* | |
287 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
288 | * r5c_try_caching_write(). No need to | |
289 | * set it again. | |
290 | */ | |
1e6d690b | 291 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
292 | } |
293 | } | |
1da177e4 LT |
294 | } |
295 | } | |
d0dabf7e | 296 | |
566c09c5 SL |
297 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
298 | struct list_head *temp_inactive_list) | |
4eb788df SL |
299 | { |
300 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
301 | do_release_stripe(conf, sh, temp_inactive_list); |
302 | } | |
303 | ||
304 | /* | |
305 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
306 | * | |
307 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
308 | * given time. Adding stripes only takes device lock, while deleting stripes | |
309 | * only takes hash lock. | |
310 | */ | |
311 | static void release_inactive_stripe_list(struct r5conf *conf, | |
312 | struct list_head *temp_inactive_list, | |
313 | int hash) | |
314 | { | |
315 | int size; | |
6ab2a4b8 | 316 | bool do_wakeup = false; |
566c09c5 SL |
317 | unsigned long flags; |
318 | ||
319 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
320 | size = NR_STRIPE_HASH_LOCKS; | |
321 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
322 | } else | |
323 | size = 1; | |
324 | while (size) { | |
325 | struct list_head *list = &temp_inactive_list[size - 1]; | |
326 | ||
327 | /* | |
6d036f7d | 328 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
329 | * remove stripes from the list |
330 | */ | |
331 | if (!list_empty_careful(list)) { | |
332 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
333 | if (list_empty(conf->inactive_list + hash) && |
334 | !list_empty(list)) | |
335 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 336 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 337 | do_wakeup = true; |
566c09c5 SL |
338 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
339 | } | |
340 | size--; | |
341 | hash--; | |
342 | } | |
343 | ||
344 | if (do_wakeup) { | |
6ab2a4b8 | 345 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
346 | if (atomic_read(&conf->active_stripes) == 0) |
347 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
348 | if (conf->retry_read_aligned) |
349 | md_wakeup_thread(conf->mddev->thread); | |
350 | } | |
4eb788df SL |
351 | } |
352 | ||
773ca82f | 353 | /* should hold conf->device_lock already */ |
566c09c5 SL |
354 | static int release_stripe_list(struct r5conf *conf, |
355 | struct list_head *temp_inactive_list) | |
773ca82f | 356 | { |
eae8263f | 357 | struct stripe_head *sh, *t; |
773ca82f SL |
358 | int count = 0; |
359 | struct llist_node *head; | |
360 | ||
361 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 362 | head = llist_reverse_order(head); |
eae8263f | 363 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
364 | int hash; |
365 | ||
773ca82f SL |
366 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
367 | smp_mb(); | |
368 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
369 | /* | |
370 | * Don't worry the bit is set here, because if the bit is set | |
371 | * again, the count is always > 1. This is true for | |
372 | * STRIPE_ON_UNPLUG_LIST bit too. | |
373 | */ | |
566c09c5 SL |
374 | hash = sh->hash_lock_index; |
375 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
376 | count++; |
377 | } | |
378 | ||
379 | return count; | |
380 | } | |
381 | ||
6d036f7d | 382 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 383 | { |
d1688a6d | 384 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 385 | unsigned long flags; |
566c09c5 SL |
386 | struct list_head list; |
387 | int hash; | |
773ca82f | 388 | bool wakeup; |
16a53ecc | 389 | |
cf170f3f ES |
390 | /* Avoid release_list until the last reference. |
391 | */ | |
392 | if (atomic_add_unless(&sh->count, -1, 1)) | |
393 | return; | |
394 | ||
ad4068de | 395 | if (unlikely(!conf->mddev->thread) || |
396 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
397 | goto slow_path; |
398 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
399 | if (wakeup) | |
400 | md_wakeup_thread(conf->mddev->thread); | |
401 | return; | |
402 | slow_path: | |
773ca82f | 403 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
685dbcaa | 404 | if (atomic_dec_and_lock_irqsave(&sh->count, &conf->device_lock, flags)) { |
566c09c5 SL |
405 | INIT_LIST_HEAD(&list); |
406 | hash = sh->hash_lock_index; | |
407 | do_release_stripe(conf, sh, &list); | |
08edaaa6 | 408 | spin_unlock_irqrestore(&conf->device_lock, flags); |
566c09c5 | 409 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df | 410 | } |
1da177e4 LT |
411 | } |
412 | ||
fccddba0 | 413 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 414 | { |
45b4233c DW |
415 | pr_debug("remove_hash(), stripe %llu\n", |
416 | (unsigned long long)sh->sector); | |
1da177e4 | 417 | |
fccddba0 | 418 | hlist_del_init(&sh->hash); |
1da177e4 LT |
419 | } |
420 | ||
d1688a6d | 421 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 422 | { |
fccddba0 | 423 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 424 | |
45b4233c DW |
425 | pr_debug("insert_hash(), stripe %llu\n", |
426 | (unsigned long long)sh->sector); | |
1da177e4 | 427 | |
fccddba0 | 428 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
429 | } |
430 | ||
1da177e4 | 431 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 432 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
433 | { |
434 | struct stripe_head *sh = NULL; | |
435 | struct list_head *first; | |
436 | ||
566c09c5 | 437 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 438 | goto out; |
566c09c5 | 439 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
440 | sh = list_entry(first, struct stripe_head, lru); |
441 | list_del_init(first); | |
442 | remove_hash(sh); | |
443 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 444 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
445 | if (list_empty(conf->inactive_list + hash)) |
446 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
447 | out: |
448 | return sh; | |
449 | } | |
450 | ||
046169f0 YY |
451 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
452 | static void free_stripe_pages(struct stripe_head *sh) | |
1da177e4 | 453 | { |
046169f0 | 454 | int i; |
1da177e4 | 455 | struct page *p; |
046169f0 YY |
456 | |
457 | /* Have not allocate page pool */ | |
458 | if (!sh->pages) | |
459 | return; | |
460 | ||
461 | for (i = 0; i < sh->nr_pages; i++) { | |
462 | p = sh->pages[i]; | |
463 | if (p) | |
464 | put_page(p); | |
465 | sh->pages[i] = NULL; | |
466 | } | |
467 | } | |
468 | ||
469 | static int alloc_stripe_pages(struct stripe_head *sh, gfp_t gfp) | |
470 | { | |
471 | int i; | |
472 | struct page *p; | |
473 | ||
474 | for (i = 0; i < sh->nr_pages; i++) { | |
475 | /* The page have allocated. */ | |
476 | if (sh->pages[i]) | |
477 | continue; | |
478 | ||
479 | p = alloc_page(gfp); | |
480 | if (!p) { | |
481 | free_stripe_pages(sh); | |
482 | return -ENOMEM; | |
483 | } | |
484 | sh->pages[i] = p; | |
485 | } | |
486 | return 0; | |
487 | } | |
488 | ||
489 | static int | |
490 | init_stripe_shared_pages(struct stripe_head *sh, struct r5conf *conf, int disks) | |
491 | { | |
492 | int nr_pages, cnt; | |
493 | ||
494 | if (sh->pages) | |
495 | return 0; | |
496 | ||
497 | /* Each of the sh->dev[i] need one conf->stripe_size */ | |
498 | cnt = PAGE_SIZE / conf->stripe_size; | |
499 | nr_pages = (disks + cnt - 1) / cnt; | |
500 | ||
501 | sh->pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); | |
502 | if (!sh->pages) | |
503 | return -ENOMEM; | |
504 | sh->nr_pages = nr_pages; | |
505 | sh->stripes_per_page = cnt; | |
506 | return 0; | |
507 | } | |
508 | #endif | |
509 | ||
510 | static void shrink_buffers(struct stripe_head *sh) | |
511 | { | |
1da177e4 | 512 | int i; |
e4e11e38 | 513 | int num = sh->raid_conf->pool_size; |
1da177e4 | 514 | |
046169f0 | 515 | #if PAGE_SIZE == DEFAULT_STRIPE_SIZE |
e4e11e38 | 516 | for (i = 0; i < num ; i++) { |
046169f0 YY |
517 | struct page *p; |
518 | ||
d592a996 | 519 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
520 | p = sh->dev[i].page; |
521 | if (!p) | |
522 | continue; | |
523 | sh->dev[i].page = NULL; | |
2d1f3b5d | 524 | put_page(p); |
1da177e4 | 525 | } |
046169f0 YY |
526 | #else |
527 | for (i = 0; i < num; i++) | |
528 | sh->dev[i].page = NULL; | |
529 | free_stripe_pages(sh); /* Free pages */ | |
530 | #endif | |
1da177e4 LT |
531 | } |
532 | ||
a9683a79 | 533 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
534 | { |
535 | int i; | |
e4e11e38 | 536 | int num = sh->raid_conf->pool_size; |
1da177e4 | 537 | |
046169f0 | 538 | #if PAGE_SIZE == DEFAULT_STRIPE_SIZE |
e4e11e38 | 539 | for (i = 0; i < num; i++) { |
1da177e4 LT |
540 | struct page *page; |
541 | ||
a9683a79 | 542 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
543 | return 1; |
544 | } | |
545 | sh->dev[i].page = page; | |
d592a996 | 546 | sh->dev[i].orig_page = page; |
7aba13b7 | 547 | sh->dev[i].offset = 0; |
1da177e4 | 548 | } |
046169f0 YY |
549 | #else |
550 | if (alloc_stripe_pages(sh, gfp)) | |
551 | return -ENOMEM; | |
3418d036 | 552 | |
046169f0 YY |
553 | for (i = 0; i < num; i++) { |
554 | sh->dev[i].page = raid5_get_dev_page(sh, i); | |
555 | sh->dev[i].orig_page = sh->dev[i].page; | |
556 | sh->dev[i].offset = raid5_get_page_offset(sh, i); | |
557 | } | |
558 | #endif | |
1da177e4 LT |
559 | return 0; |
560 | } | |
561 | ||
d1688a6d | 562 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 563 | struct stripe_head *sh); |
1da177e4 | 564 | |
b5663ba4 | 565 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 566 | { |
d1688a6d | 567 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 568 | int i, seq; |
1da177e4 | 569 | |
78bafebd ES |
570 | BUG_ON(atomic_read(&sh->count) != 0); |
571 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 572 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 573 | BUG_ON(sh->batch_head); |
d84e0f10 | 574 | |
45b4233c | 575 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 576 | (unsigned long long)sector); |
566c09c5 SL |
577 | retry: |
578 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 579 | sh->generation = conf->generation - previous; |
b5663ba4 | 580 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 581 | sh->sector = sector; |
911d4ee8 | 582 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
583 | sh->state = 0; |
584 | ||
7ecaa1e6 | 585 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
586 | struct r5dev *dev = &sh->dev[i]; |
587 | ||
d84e0f10 | 588 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 589 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 590 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 591 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 592 | dev->read, dev->towrite, dev->written, |
1da177e4 | 593 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 594 | WARN_ON(1); |
1da177e4 LT |
595 | } |
596 | dev->flags = 0; | |
27a4ff8f | 597 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 | 598 | } |
566c09c5 SL |
599 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
600 | goto retry; | |
7a87f434 | 601 | sh->overwrite_disks = 0; |
1da177e4 | 602 | insert_hash(conf, sh); |
851c30c9 | 603 | sh->cpu = smp_processor_id(); |
da41ba65 | 604 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
605 | } |
606 | ||
d1688a6d | 607 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 608 | short generation) |
1da177e4 LT |
609 | { |
610 | struct stripe_head *sh; | |
611 | ||
45b4233c | 612 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 613 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 614 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 615 | return sh; |
45b4233c | 616 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
617 | return NULL; |
618 | } | |
619 | ||
674806d6 N |
620 | /* |
621 | * Need to check if array has failed when deciding whether to: | |
622 | * - start an array | |
623 | * - remove non-faulty devices | |
624 | * - add a spare | |
625 | * - allow a reshape | |
626 | * This determination is simple when no reshape is happening. | |
627 | * However if there is a reshape, we need to carefully check | |
628 | * both the before and after sections. | |
629 | * This is because some failed devices may only affect one | |
630 | * of the two sections, and some non-in_sync devices may | |
631 | * be insync in the section most affected by failed devices. | |
632 | */ | |
2e38a37f | 633 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 634 | { |
908f4fbd | 635 | int degraded, degraded2; |
674806d6 | 636 | int i; |
674806d6 N |
637 | |
638 | rcu_read_lock(); | |
639 | degraded = 0; | |
640 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 641 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
642 | if (rdev && test_bit(Faulty, &rdev->flags)) |
643 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
644 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
645 | degraded++; | |
646 | else if (test_bit(In_sync, &rdev->flags)) | |
647 | ; | |
648 | else | |
649 | /* not in-sync or faulty. | |
650 | * If the reshape increases the number of devices, | |
651 | * this is being recovered by the reshape, so | |
652 | * this 'previous' section is not in_sync. | |
653 | * If the number of devices is being reduced however, | |
654 | * the device can only be part of the array if | |
655 | * we are reverting a reshape, so this section will | |
656 | * be in-sync. | |
657 | */ | |
658 | if (conf->raid_disks >= conf->previous_raid_disks) | |
659 | degraded++; | |
660 | } | |
661 | rcu_read_unlock(); | |
908f4fbd N |
662 | if (conf->raid_disks == conf->previous_raid_disks) |
663 | return degraded; | |
674806d6 | 664 | rcu_read_lock(); |
908f4fbd | 665 | degraded2 = 0; |
674806d6 | 666 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 667 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
668 | if (rdev && test_bit(Faulty, &rdev->flags)) |
669 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 670 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 671 | degraded2++; |
674806d6 N |
672 | else if (test_bit(In_sync, &rdev->flags)) |
673 | ; | |
674 | else | |
675 | /* not in-sync or faulty. | |
676 | * If reshape increases the number of devices, this | |
677 | * section has already been recovered, else it | |
678 | * almost certainly hasn't. | |
679 | */ | |
680 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 681 | degraded2++; |
674806d6 N |
682 | } |
683 | rcu_read_unlock(); | |
908f4fbd N |
684 | if (degraded2 > degraded) |
685 | return degraded2; | |
686 | return degraded; | |
687 | } | |
688 | ||
ba97a054 | 689 | static bool has_failed(struct r5conf *conf) |
908f4fbd | 690 | { |
ba97a054 | 691 | int degraded = conf->mddev->degraded; |
908f4fbd | 692 | |
ba97a054 MT |
693 | if (test_bit(MD_BROKEN, &conf->mddev->flags)) |
694 | return true; | |
908f4fbd | 695 | |
ba97a054 MT |
696 | if (conf->mddev->reshape_position != MaxSector) |
697 | degraded = raid5_calc_degraded(conf); | |
698 | ||
699 | return degraded > conf->max_degraded; | |
674806d6 N |
700 | } |
701 | ||
6d036f7d SL |
702 | struct stripe_head * |
703 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
704 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
705 | { |
706 | struct stripe_head *sh; | |
c911c46c | 707 | int hash = stripe_hash_locks_hash(conf, sector); |
ff00d3b4 | 708 | int inc_empty_inactive_list_flag; |
1da177e4 | 709 | |
45b4233c | 710 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 711 | |
566c09c5 | 712 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
713 | |
714 | do { | |
b1b46486 | 715 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 716 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 717 | *(conf->hash_locks + hash)); |
86b42c71 | 718 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 719 | if (!sh) { |
edbe83ab | 720 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 721 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
722 | if (!sh && !test_bit(R5_DID_ALLOC, |
723 | &conf->cache_state)) | |
edbe83ab N |
724 | set_bit(R5_ALLOC_MORE, |
725 | &conf->cache_state); | |
726 | } | |
1da177e4 LT |
727 | if (noblock && sh == NULL) |
728 | break; | |
a39f7afd SL |
729 | |
730 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 731 | if (!sh) { |
5423399a N |
732 | set_bit(R5_INACTIVE_BLOCKED, |
733 | &conf->cache_state); | |
a39f7afd | 734 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
735 | wait_event_lock_irq( |
736 | conf->wait_for_stripe, | |
566c09c5 SL |
737 | !list_empty(conf->inactive_list + hash) && |
738 | (atomic_read(&conf->active_stripes) | |
739 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
740 | || !test_bit(R5_INACTIVE_BLOCKED, |
741 | &conf->cache_state)), | |
6ab2a4b8 | 742 | *(conf->hash_locks + hash)); |
5423399a N |
743 | clear_bit(R5_INACTIVE_BLOCKED, |
744 | &conf->cache_state); | |
7da9d450 | 745 | } else { |
b5663ba4 | 746 | init_stripe(sh, sector, previous); |
7da9d450 N |
747 | atomic_inc(&sh->count); |
748 | } | |
e240c183 | 749 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 750 | spin_lock(&conf->device_lock); |
e240c183 | 751 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
752 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
753 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
754 | BUG_ON(list_empty(&sh->lru) && |
755 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
756 | inc_empty_inactive_list_flag = 0; |
757 | if (!list_empty(conf->inactive_list + hash)) | |
758 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 759 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
760 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
761 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
762 | if (sh->group) { |
763 | sh->group->stripes_cnt--; | |
764 | sh->group = NULL; | |
765 | } | |
1da177e4 | 766 | } |
7da9d450 | 767 | atomic_inc(&sh->count); |
6d183de4 | 768 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
769 | } |
770 | } while (sh == NULL); | |
771 | ||
566c09c5 | 772 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
773 | return sh; |
774 | } | |
775 | ||
7a87f434 | 776 | static bool is_full_stripe_write(struct stripe_head *sh) |
777 | { | |
778 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
779 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
780 | } | |
781 | ||
59fc630b | 782 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
368ecade CH |
783 | __acquires(&sh1->stripe_lock) |
784 | __acquires(&sh2->stripe_lock) | |
59fc630b | 785 | { |
59fc630b | 786 | if (sh1 > sh2) { |
3d05f3ae | 787 | spin_lock_irq(&sh2->stripe_lock); |
59fc630b | 788 | spin_lock_nested(&sh1->stripe_lock, 1); |
789 | } else { | |
3d05f3ae | 790 | spin_lock_irq(&sh1->stripe_lock); |
59fc630b | 791 | spin_lock_nested(&sh2->stripe_lock, 1); |
792 | } | |
793 | } | |
794 | ||
795 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
368ecade CH |
796 | __releases(&sh1->stripe_lock) |
797 | __releases(&sh2->stripe_lock) | |
59fc630b | 798 | { |
799 | spin_unlock(&sh1->stripe_lock); | |
3d05f3ae | 800 | spin_unlock_irq(&sh2->stripe_lock); |
59fc630b | 801 | } |
802 | ||
803 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
804 | static bool stripe_can_batch(struct stripe_head *sh) | |
805 | { | |
9c3e333d SL |
806 | struct r5conf *conf = sh->raid_conf; |
807 | ||
e254de6b | 808 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
9c3e333d | 809 | return false; |
59fc630b | 810 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 811 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 812 | is_full_stripe_write(sh); |
813 | } | |
814 | ||
815 | /* we only do back search */ | |
816 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
817 | { | |
818 | struct stripe_head *head; | |
819 | sector_t head_sector, tmp_sec; | |
820 | int hash; | |
821 | int dd_idx; | |
ff00d3b4 | 822 | int inc_empty_inactive_list_flag; |
59fc630b | 823 | |
59fc630b | 824 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
825 | tmp_sec = sh->sector; | |
826 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
827 | return; | |
c911c46c | 828 | head_sector = sh->sector - RAID5_STRIPE_SECTORS(conf); |
59fc630b | 829 | |
c911c46c | 830 | hash = stripe_hash_locks_hash(conf, head_sector); |
59fc630b | 831 | spin_lock_irq(conf->hash_locks + hash); |
832 | head = __find_stripe(conf, head_sector, conf->generation); | |
833 | if (head && !atomic_inc_not_zero(&head->count)) { | |
834 | spin_lock(&conf->device_lock); | |
835 | if (!atomic_read(&head->count)) { | |
836 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
837 | atomic_inc(&conf->active_stripes); | |
838 | BUG_ON(list_empty(&head->lru) && | |
839 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
840 | inc_empty_inactive_list_flag = 0; |
841 | if (!list_empty(conf->inactive_list + hash)) | |
842 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 843 | list_del_init(&head->lru); |
ff00d3b4 ZL |
844 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
845 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 846 | if (head->group) { |
847 | head->group->stripes_cnt--; | |
848 | head->group = NULL; | |
849 | } | |
850 | } | |
851 | atomic_inc(&head->count); | |
852 | spin_unlock(&conf->device_lock); | |
853 | } | |
854 | spin_unlock_irq(conf->hash_locks + hash); | |
855 | ||
856 | if (!head) | |
857 | return; | |
858 | if (!stripe_can_batch(head)) | |
859 | goto out; | |
860 | ||
861 | lock_two_stripes(head, sh); | |
862 | /* clear_batch_ready clear the flag */ | |
863 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
864 | goto unlock_out; | |
865 | ||
866 | if (sh->batch_head) | |
867 | goto unlock_out; | |
868 | ||
869 | dd_idx = 0; | |
870 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
871 | dd_idx++; | |
1eff9d32 | 872 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 873 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 874 | goto unlock_out; |
875 | ||
876 | if (head->batch_head) { | |
877 | spin_lock(&head->batch_head->batch_lock); | |
878 | /* This batch list is already running */ | |
879 | if (!stripe_can_batch(head)) { | |
880 | spin_unlock(&head->batch_head->batch_lock); | |
881 | goto unlock_out; | |
882 | } | |
3664847d SL |
883 | /* |
884 | * We must assign batch_head of this stripe within the | |
885 | * batch_lock, otherwise clear_batch_ready of batch head | |
886 | * stripe could clear BATCH_READY bit of this stripe and | |
887 | * this stripe->batch_head doesn't get assigned, which | |
888 | * could confuse clear_batch_ready for this stripe | |
889 | */ | |
890 | sh->batch_head = head->batch_head; | |
59fc630b | 891 | |
892 | /* | |
893 | * at this point, head's BATCH_READY could be cleared, but we | |
894 | * can still add the stripe to batch list | |
895 | */ | |
896 | list_add(&sh->batch_list, &head->batch_list); | |
897 | spin_unlock(&head->batch_head->batch_lock); | |
59fc630b | 898 | } else { |
899 | head->batch_head = head; | |
900 | sh->batch_head = head->batch_head; | |
901 | spin_lock(&head->batch_lock); | |
902 | list_add_tail(&sh->batch_list, &head->batch_list); | |
903 | spin_unlock(&head->batch_lock); | |
904 | } | |
905 | ||
906 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
907 | if (atomic_dec_return(&conf->preread_active_stripes) | |
908 | < IO_THRESHOLD) | |
909 | md_wakeup_thread(conf->mddev->thread); | |
910 | ||
2b6b2457 N |
911 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
912 | int seq = sh->bm_seq; | |
913 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
914 | sh->batch_head->bm_seq > seq) | |
915 | seq = sh->batch_head->bm_seq; | |
916 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
917 | sh->batch_head->bm_seq = seq; | |
918 | } | |
919 | ||
59fc630b | 920 | atomic_inc(&sh->count); |
921 | unlock_out: | |
922 | unlock_two_stripes(head, sh); | |
923 | out: | |
6d036f7d | 924 | raid5_release_stripe(head); |
59fc630b | 925 | } |
926 | ||
05616be5 N |
927 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
928 | * in this stripe_head. | |
929 | */ | |
930 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
931 | { | |
932 | sector_t progress = conf->reshape_progress; | |
933 | /* Need a memory barrier to make sure we see the value | |
934 | * of conf->generation, or ->data_offset that was set before | |
935 | * reshape_progress was updated. | |
936 | */ | |
937 | smp_rmb(); | |
938 | if (progress == MaxSector) | |
939 | return 0; | |
940 | if (sh->generation == conf->generation - 1) | |
941 | return 0; | |
942 | /* We are in a reshape, and this is a new-generation stripe, | |
943 | * so use new_data_offset. | |
944 | */ | |
945 | return 1; | |
946 | } | |
947 | ||
aaf9f12e | 948 | static void dispatch_bio_list(struct bio_list *tmp) |
765d704d | 949 | { |
765d704d SL |
950 | struct bio *bio; |
951 | ||
aaf9f12e | 952 | while ((bio = bio_list_pop(tmp))) |
ed00aabd | 953 | submit_bio_noacct(bio); |
aaf9f12e SL |
954 | } |
955 | ||
4f0f586b ST |
956 | static int cmp_stripe(void *priv, const struct list_head *a, |
957 | const struct list_head *b) | |
aaf9f12e SL |
958 | { |
959 | const struct r5pending_data *da = list_entry(a, | |
960 | struct r5pending_data, sibling); | |
961 | const struct r5pending_data *db = list_entry(b, | |
962 | struct r5pending_data, sibling); | |
963 | if (da->sector > db->sector) | |
964 | return 1; | |
965 | if (da->sector < db->sector) | |
966 | return -1; | |
967 | return 0; | |
968 | } | |
969 | ||
970 | static void dispatch_defer_bios(struct r5conf *conf, int target, | |
971 | struct bio_list *list) | |
972 | { | |
973 | struct r5pending_data *data; | |
974 | struct list_head *first, *next = NULL; | |
975 | int cnt = 0; | |
976 | ||
977 | if (conf->pending_data_cnt == 0) | |
978 | return; | |
979 | ||
980 | list_sort(NULL, &conf->pending_list, cmp_stripe); | |
981 | ||
982 | first = conf->pending_list.next; | |
983 | ||
984 | /* temporarily move the head */ | |
985 | if (conf->next_pending_data) | |
986 | list_move_tail(&conf->pending_list, | |
987 | &conf->next_pending_data->sibling); | |
988 | ||
989 | while (!list_empty(&conf->pending_list)) { | |
990 | data = list_first_entry(&conf->pending_list, | |
991 | struct r5pending_data, sibling); | |
992 | if (&data->sibling == first) | |
993 | first = data->sibling.next; | |
994 | next = data->sibling.next; | |
995 | ||
996 | bio_list_merge(list, &data->bios); | |
997 | list_move(&data->sibling, &conf->free_list); | |
998 | cnt++; | |
999 | if (cnt >= target) | |
1000 | break; | |
1001 | } | |
1002 | conf->pending_data_cnt -= cnt; | |
1003 | BUG_ON(conf->pending_data_cnt < 0 || cnt < target); | |
1004 | ||
1005 | if (next != &conf->pending_list) | |
1006 | conf->next_pending_data = list_entry(next, | |
1007 | struct r5pending_data, sibling); | |
1008 | else | |
1009 | conf->next_pending_data = NULL; | |
1010 | /* list isn't empty */ | |
1011 | if (first != &conf->pending_list) | |
1012 | list_move_tail(&conf->pending_list, first); | |
1013 | } | |
1014 | ||
1015 | static void flush_deferred_bios(struct r5conf *conf) | |
1016 | { | |
1017 | struct bio_list tmp = BIO_EMPTY_LIST; | |
1018 | ||
1019 | if (conf->pending_data_cnt == 0) | |
765d704d SL |
1020 | return; |
1021 | ||
765d704d | 1022 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
1023 | dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp); |
1024 | BUG_ON(conf->pending_data_cnt != 0); | |
765d704d SL |
1025 | spin_unlock(&conf->pending_bios_lock); |
1026 | ||
aaf9f12e | 1027 | dispatch_bio_list(&tmp); |
765d704d SL |
1028 | } |
1029 | ||
aaf9f12e SL |
1030 | static void defer_issue_bios(struct r5conf *conf, sector_t sector, |
1031 | struct bio_list *bios) | |
765d704d | 1032 | { |
aaf9f12e SL |
1033 | struct bio_list tmp = BIO_EMPTY_LIST; |
1034 | struct r5pending_data *ent; | |
1035 | ||
765d704d | 1036 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
1037 | ent = list_first_entry(&conf->free_list, struct r5pending_data, |
1038 | sibling); | |
1039 | list_move_tail(&ent->sibling, &conf->pending_list); | |
1040 | ent->sector = sector; | |
1041 | bio_list_init(&ent->bios); | |
1042 | bio_list_merge(&ent->bios, bios); | |
1043 | conf->pending_data_cnt++; | |
1044 | if (conf->pending_data_cnt >= PENDING_IO_MAX) | |
1045 | dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp); | |
1046 | ||
765d704d | 1047 | spin_unlock(&conf->pending_bios_lock); |
aaf9f12e SL |
1048 | |
1049 | dispatch_bio_list(&tmp); | |
765d704d SL |
1050 | } |
1051 | ||
6712ecf8 | 1052 | static void |
4246a0b6 | 1053 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 1054 | static void |
4246a0b6 | 1055 | raid5_end_write_request(struct bio *bi); |
91c00924 | 1056 | |
c4e5ac0a | 1057 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 1058 | { |
d1688a6d | 1059 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1060 | int i, disks = sh->disks; |
59fc630b | 1061 | struct stripe_head *head_sh = sh; |
aaf9f12e SL |
1062 | struct bio_list pending_bios = BIO_EMPTY_LIST; |
1063 | bool should_defer; | |
91c00924 DW |
1064 | |
1065 | might_sleep(); | |
1066 | ||
ff875738 AP |
1067 | if (log_stripe(sh, s) == 0) |
1068 | return; | |
1e6d690b | 1069 | |
aaf9f12e | 1070 | should_defer = conf->batch_bio_dispatch && conf->group_cnt; |
1e6d690b | 1071 | |
91c00924 | 1072 | for (i = disks; i--; ) { |
796a5cf0 | 1073 | int op, op_flags = 0; |
9a3e1101 | 1074 | int replace_only = 0; |
977df362 N |
1075 | struct bio *bi, *rbi; |
1076 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 1077 | |
1078 | sh = head_sh; | |
e9c7469b | 1079 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 1080 | op = REQ_OP_WRITE; |
e9c7469b | 1081 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 1082 | op_flags = REQ_FUA; |
9e444768 | 1083 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 1084 | op = REQ_OP_DISCARD; |
e9c7469b | 1085 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 1086 | op = REQ_OP_READ; |
9a3e1101 N |
1087 | else if (test_and_clear_bit(R5_WantReplace, |
1088 | &sh->dev[i].flags)) { | |
796a5cf0 | 1089 | op = REQ_OP_WRITE; |
9a3e1101 N |
1090 | replace_only = 1; |
1091 | } else | |
91c00924 | 1092 | continue; |
bc0934f0 | 1093 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 1094 | op_flags |= REQ_SYNC; |
91c00924 | 1095 | |
59fc630b | 1096 | again: |
91c00924 | 1097 | bi = &sh->dev[i].req; |
977df362 | 1098 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 1099 | |
91c00924 | 1100 | rcu_read_lock(); |
9a3e1101 | 1101 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
1102 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
1103 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1104 | if (!rdev) { | |
1105 | rdev = rrdev; | |
1106 | rrdev = NULL; | |
1107 | } | |
796a5cf0 | 1108 | if (op_is_write(op)) { |
9a3e1101 N |
1109 | if (replace_only) |
1110 | rdev = NULL; | |
dd054fce N |
1111 | if (rdev == rrdev) |
1112 | /* We raced and saw duplicates */ | |
1113 | rrdev = NULL; | |
9a3e1101 | 1114 | } else { |
59fc630b | 1115 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
1116 | rdev = rrdev; |
1117 | rrdev = NULL; | |
1118 | } | |
977df362 | 1119 | |
91c00924 DW |
1120 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1121 | rdev = NULL; | |
1122 | if (rdev) | |
1123 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
1124 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
1125 | rrdev = NULL; | |
1126 | if (rrdev) | |
1127 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1128 | rcu_read_unlock(); |
1129 | ||
73e92e51 | 1130 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1131 | * need to check for writes. We never accept write errors |
1132 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1133 | */ |
796a5cf0 | 1134 | while (op_is_write(op) && rdev && |
73e92e51 N |
1135 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1136 | sector_t first_bad; | |
1137 | int bad_sectors; | |
c911c46c | 1138 | int bad = is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), |
73e92e51 N |
1139 | &first_bad, &bad_sectors); |
1140 | if (!bad) | |
1141 | break; | |
1142 | ||
1143 | if (bad < 0) { | |
1144 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1145 | if (!conf->mddev->external && | |
2953079c | 1146 | conf->mddev->sb_flags) { |
73e92e51 N |
1147 | /* It is very unlikely, but we might |
1148 | * still need to write out the | |
1149 | * bad block log - better give it | |
1150 | * a chance*/ | |
1151 | md_check_recovery(conf->mddev); | |
1152 | } | |
1850753d | 1153 | /* |
1154 | * Because md_wait_for_blocked_rdev | |
1155 | * will dec nr_pending, we must | |
1156 | * increment it first. | |
1157 | */ | |
1158 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1159 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1160 | } else { | |
1161 | /* Acknowledged bad block - skip the write */ | |
1162 | rdev_dec_pending(rdev, conf->mddev); | |
1163 | rdev = NULL; | |
1164 | } | |
1165 | } | |
1166 | ||
91c00924 | 1167 | if (rdev) { |
9a3e1101 N |
1168 | if (s->syncing || s->expanding || s->expanded |
1169 | || s->replacing) | |
c911c46c | 1170 | md_sync_acct(rdev->bdev, RAID5_STRIPE_SECTORS(conf)); |
91c00924 | 1171 | |
2b7497f0 DW |
1172 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1173 | ||
74d46992 | 1174 | bio_set_dev(bi, rdev->bdev); |
796a5cf0 MC |
1175 | bio_set_op_attrs(bi, op, op_flags); |
1176 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1177 | ? raid5_end_write_request |
1178 | : raid5_end_read_request; | |
1179 | bi->bi_private = sh; | |
1180 | ||
6296b960 | 1181 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1182 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1183 | bi->bi_opf, i); |
91c00924 | 1184 | atomic_inc(&sh->count); |
59fc630b | 1185 | if (sh != head_sh) |
1186 | atomic_inc(&head_sh->count); | |
05616be5 | 1187 | if (use_new_offset(conf, sh)) |
4f024f37 | 1188 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1189 | + rdev->new_data_offset); |
1190 | else | |
4f024f37 | 1191 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1192 | + rdev->data_offset); |
59fc630b | 1193 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1194 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1195 | |
d592a996 SL |
1196 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1197 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1198 | |
1199 | if (!op_is_write(op) && | |
1200 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1201 | /* | |
1202 | * issuing read for a page in journal, this | |
1203 | * must be preparing for prexor in rmw; read | |
1204 | * the data into orig_page | |
1205 | */ | |
1206 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1207 | else | |
1208 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1209 | bi->bi_vcnt = 1; |
c911c46c | 1210 | bi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf); |
7aba13b7 | 1211 | bi->bi_io_vec[0].bv_offset = sh->dev[i].offset; |
c911c46c | 1212 | bi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf); |
2cd259a7 MD |
1213 | bi->bi_write_hint = sh->dev[i].write_hint; |
1214 | if (!rrdev) | |
f1934892 | 1215 | sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET; |
37c61ff3 SL |
1216 | /* |
1217 | * If this is discard request, set bi_vcnt 0. We don't | |
1218 | * want to confuse SCSI because SCSI will replace payload | |
1219 | */ | |
796a5cf0 | 1220 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1221 | bi->bi_vcnt = 0; |
977df362 N |
1222 | if (rrdev) |
1223 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1224 | |
1225 | if (conf->mddev->gendisk) | |
1c02fca6 CH |
1226 | trace_block_bio_remap(bi, |
1227 | disk_devt(conf->mddev->gendisk), | |
1228 | sh->dev[i].sector); | |
aaf9f12e SL |
1229 | if (should_defer && op_is_write(op)) |
1230 | bio_list_add(&pending_bios, bi); | |
1231 | else | |
ed00aabd | 1232 | submit_bio_noacct(bi); |
977df362 N |
1233 | } |
1234 | if (rrdev) { | |
9a3e1101 N |
1235 | if (s->syncing || s->expanding || s->expanded |
1236 | || s->replacing) | |
c911c46c | 1237 | md_sync_acct(rrdev->bdev, RAID5_STRIPE_SECTORS(conf)); |
977df362 N |
1238 | |
1239 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1240 | ||
74d46992 | 1241 | bio_set_dev(rbi, rrdev->bdev); |
796a5cf0 MC |
1242 | bio_set_op_attrs(rbi, op, op_flags); |
1243 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1244 | rbi->bi_end_io = raid5_end_write_request; |
1245 | rbi->bi_private = sh; | |
1246 | ||
6296b960 | 1247 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1248 | "replacement disc %d\n", |
1249 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1250 | rbi->bi_opf, i); |
977df362 | 1251 | atomic_inc(&sh->count); |
59fc630b | 1252 | if (sh != head_sh) |
1253 | atomic_inc(&head_sh->count); | |
05616be5 | 1254 | if (use_new_offset(conf, sh)) |
4f024f37 | 1255 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1256 | + rrdev->new_data_offset); |
1257 | else | |
4f024f37 | 1258 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1259 | + rrdev->data_offset); |
d592a996 SL |
1260 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1261 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1262 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1263 | rbi->bi_vcnt = 1; |
c911c46c | 1264 | rbi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf); |
7aba13b7 | 1265 | rbi->bi_io_vec[0].bv_offset = sh->dev[i].offset; |
c911c46c | 1266 | rbi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf); |
2cd259a7 | 1267 | rbi->bi_write_hint = sh->dev[i].write_hint; |
f1934892 | 1268 | sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET; |
37c61ff3 SL |
1269 | /* |
1270 | * If this is discard request, set bi_vcnt 0. We don't | |
1271 | * want to confuse SCSI because SCSI will replace payload | |
1272 | */ | |
796a5cf0 | 1273 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1274 | rbi->bi_vcnt = 0; |
e3620a3a | 1275 | if (conf->mddev->gendisk) |
1c02fca6 CH |
1276 | trace_block_bio_remap(rbi, |
1277 | disk_devt(conf->mddev->gendisk), | |
1278 | sh->dev[i].sector); | |
aaf9f12e SL |
1279 | if (should_defer && op_is_write(op)) |
1280 | bio_list_add(&pending_bios, rbi); | |
1281 | else | |
ed00aabd | 1282 | submit_bio_noacct(rbi); |
977df362 N |
1283 | } |
1284 | if (!rdev && !rrdev) { | |
796a5cf0 | 1285 | if (op_is_write(op)) |
91c00924 | 1286 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1287 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1288 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1289 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1290 | set_bit(STRIPE_HANDLE, &sh->state); | |
1291 | } | |
59fc630b | 1292 | |
1293 | if (!head_sh->batch_head) | |
1294 | continue; | |
1295 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1296 | batch_list); | |
1297 | if (sh != head_sh) | |
1298 | goto again; | |
91c00924 | 1299 | } |
aaf9f12e SL |
1300 | |
1301 | if (should_defer && !bio_list_empty(&pending_bios)) | |
1302 | defer_issue_bios(conf, head_sh->sector, &pending_bios); | |
91c00924 DW |
1303 | } |
1304 | ||
1305 | static struct dma_async_tx_descriptor * | |
d592a996 | 1306 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
248728dd | 1307 | unsigned int poff, sector_t sector, struct dma_async_tx_descriptor *tx, |
1e6d690b | 1308 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1309 | { |
7988613b KO |
1310 | struct bio_vec bvl; |
1311 | struct bvec_iter iter; | |
91c00924 | 1312 | struct page *bio_page; |
91c00924 | 1313 | int page_offset; |
a08abd8c | 1314 | struct async_submit_ctl submit; |
0403e382 | 1315 | enum async_tx_flags flags = 0; |
c911c46c | 1316 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1317 | |
4f024f37 KO |
1318 | if (bio->bi_iter.bi_sector >= sector) |
1319 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1320 | else |
4f024f37 | 1321 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1322 | |
0403e382 DW |
1323 | if (frombio) |
1324 | flags |= ASYNC_TX_FENCE; | |
1325 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1326 | ||
7988613b KO |
1327 | bio_for_each_segment(bvl, bio, iter) { |
1328 | int len = bvl.bv_len; | |
91c00924 DW |
1329 | int clen; |
1330 | int b_offset = 0; | |
1331 | ||
1332 | if (page_offset < 0) { | |
1333 | b_offset = -page_offset; | |
1334 | page_offset += b_offset; | |
1335 | len -= b_offset; | |
1336 | } | |
1337 | ||
c911c46c YY |
1338 | if (len > 0 && page_offset + len > RAID5_STRIPE_SIZE(conf)) |
1339 | clen = RAID5_STRIPE_SIZE(conf) - page_offset; | |
91c00924 DW |
1340 | else |
1341 | clen = len; | |
1342 | ||
1343 | if (clen > 0) { | |
7988613b KO |
1344 | b_offset += bvl.bv_offset; |
1345 | bio_page = bvl.bv_page; | |
d592a996 | 1346 | if (frombio) { |
c911c46c | 1347 | if (conf->skip_copy && |
d592a996 | 1348 | b_offset == 0 && page_offset == 0 && |
c911c46c | 1349 | clen == RAID5_STRIPE_SIZE(conf) && |
1e6d690b | 1350 | !no_skipcopy) |
d592a996 SL |
1351 | *page = bio_page; |
1352 | else | |
248728dd | 1353 | tx = async_memcpy(*page, bio_page, page_offset + poff, |
a08abd8c | 1354 | b_offset, clen, &submit); |
d592a996 SL |
1355 | } else |
1356 | tx = async_memcpy(bio_page, *page, b_offset, | |
248728dd | 1357 | page_offset + poff, clen, &submit); |
91c00924 | 1358 | } |
a08abd8c DW |
1359 | /* chain the operations */ |
1360 | submit.depend_tx = tx; | |
1361 | ||
91c00924 DW |
1362 | if (clen < len) /* hit end of page */ |
1363 | break; | |
1364 | page_offset += len; | |
1365 | } | |
1366 | ||
1367 | return tx; | |
1368 | } | |
1369 | ||
1370 | static void ops_complete_biofill(void *stripe_head_ref) | |
1371 | { | |
1372 | struct stripe_head *sh = stripe_head_ref; | |
e4d84909 | 1373 | int i; |
c911c46c | 1374 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1375 | |
e46b272b | 1376 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1377 | (unsigned long long)sh->sector); |
1378 | ||
1379 | /* clear completed biofills */ | |
1380 | for (i = sh->disks; i--; ) { | |
1381 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1382 | |
1383 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1384 | /* and check if we need to reply to a read request, |
1385 | * new R5_Wantfill requests are held off until | |
83de75cc | 1386 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1387 | */ |
1388 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1389 | struct bio *rbi, *rbi2; |
91c00924 | 1390 | |
91c00924 DW |
1391 | BUG_ON(!dev->read); |
1392 | rbi = dev->read; | |
1393 | dev->read = NULL; | |
4f024f37 | 1394 | while (rbi && rbi->bi_iter.bi_sector < |
c911c46c YY |
1395 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
1396 | rbi2 = r5_next_bio(conf, rbi, dev->sector); | |
016c76ac | 1397 | bio_endio(rbi); |
91c00924 DW |
1398 | rbi = rbi2; |
1399 | } | |
1400 | } | |
1401 | } | |
83de75cc | 1402 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1403 | |
e4d84909 | 1404 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1405 | raid5_release_stripe(sh); |
91c00924 DW |
1406 | } |
1407 | ||
1408 | static void ops_run_biofill(struct stripe_head *sh) | |
1409 | { | |
1410 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1411 | struct async_submit_ctl submit; |
91c00924 | 1412 | int i; |
c911c46c | 1413 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1414 | |
59fc630b | 1415 | BUG_ON(sh->batch_head); |
e46b272b | 1416 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1417 | (unsigned long long)sh->sector); |
1418 | ||
1419 | for (i = sh->disks; i--; ) { | |
1420 | struct r5dev *dev = &sh->dev[i]; | |
1421 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1422 | struct bio *rbi; | |
b17459c0 | 1423 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1424 | dev->read = rbi = dev->toread; |
1425 | dev->toread = NULL; | |
b17459c0 | 1426 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1427 | while (rbi && rbi->bi_iter.bi_sector < |
c911c46c | 1428 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
d592a996 | 1429 | tx = async_copy_data(0, rbi, &dev->page, |
248728dd | 1430 | dev->offset, |
1e6d690b | 1431 | dev->sector, tx, sh, 0); |
c911c46c | 1432 | rbi = r5_next_bio(conf, rbi, dev->sector); |
91c00924 DW |
1433 | } |
1434 | } | |
1435 | } | |
1436 | ||
1437 | atomic_inc(&sh->count); | |
a08abd8c DW |
1438 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1439 | async_trigger_callback(&submit); | |
91c00924 DW |
1440 | } |
1441 | ||
4e7d2c0a | 1442 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1443 | { |
4e7d2c0a | 1444 | struct r5dev *tgt; |
91c00924 | 1445 | |
4e7d2c0a DW |
1446 | if (target < 0) |
1447 | return; | |
91c00924 | 1448 | |
4e7d2c0a | 1449 | tgt = &sh->dev[target]; |
91c00924 DW |
1450 | set_bit(R5_UPTODATE, &tgt->flags); |
1451 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1452 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1453 | } |
1454 | ||
ac6b53b6 | 1455 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1456 | { |
1457 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1458 | |
e46b272b | 1459 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1460 | (unsigned long long)sh->sector); |
1461 | ||
ac6b53b6 | 1462 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1463 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1464 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1465 | |
ecc65c9b DW |
1466 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1467 | if (sh->check_state == check_state_compute_run) | |
1468 | sh->check_state = check_state_compute_result; | |
91c00924 | 1469 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1470 | raid5_release_stripe(sh); |
91c00924 DW |
1471 | } |
1472 | ||
d6f38f31 | 1473 | /* return a pointer to the address conversion region of the scribble buffer */ |
b330e6a4 | 1474 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) |
d6f38f31 | 1475 | { |
b330e6a4 | 1476 | return percpu->scribble + i * percpu->scribble_obj_size; |
46d5b785 | 1477 | } |
1478 | ||
1479 | /* return a pointer to the address conversion region of the scribble buffer */ | |
b330e6a4 KO |
1480 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, |
1481 | struct raid5_percpu *percpu, int i) | |
46d5b785 | 1482 | { |
b330e6a4 | 1483 | return (void *) (to_addr_page(percpu, i) + sh->disks + 2); |
d6f38f31 DW |
1484 | } |
1485 | ||
7aba13b7 YY |
1486 | /* |
1487 | * Return a pointer to record offset address. | |
1488 | */ | |
1489 | static unsigned int * | |
1490 | to_addr_offs(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1491 | { | |
1492 | return (unsigned int *) (to_addr_conv(sh, percpu, 0) + sh->disks + 2); | |
1493 | } | |
1494 | ||
d6f38f31 DW |
1495 | static struct dma_async_tx_descriptor * |
1496 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1497 | { |
91c00924 | 1498 | int disks = sh->disks; |
46d5b785 | 1499 | struct page **xor_srcs = to_addr_page(percpu, 0); |
7aba13b7 | 1500 | unsigned int *off_srcs = to_addr_offs(sh, percpu); |
91c00924 DW |
1501 | int target = sh->ops.target; |
1502 | struct r5dev *tgt = &sh->dev[target]; | |
1503 | struct page *xor_dest = tgt->page; | |
7aba13b7 | 1504 | unsigned int off_dest = tgt->offset; |
91c00924 DW |
1505 | int count = 0; |
1506 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1507 | struct async_submit_ctl submit; |
91c00924 DW |
1508 | int i; |
1509 | ||
59fc630b | 1510 | BUG_ON(sh->batch_head); |
1511 | ||
91c00924 | 1512 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1513 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1514 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1515 | ||
7aba13b7 YY |
1516 | for (i = disks; i--; ) { |
1517 | if (i != target) { | |
1518 | off_srcs[count] = sh->dev[i].offset; | |
91c00924 | 1519 | xor_srcs[count++] = sh->dev[i].page; |
7aba13b7 YY |
1520 | } |
1521 | } | |
91c00924 DW |
1522 | |
1523 | atomic_inc(&sh->count); | |
1524 | ||
0403e382 | 1525 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1526 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1527 | if (unlikely(count == 1)) |
7aba13b7 | 1528 | tx = async_memcpy(xor_dest, xor_srcs[0], off_dest, off_srcs[0], |
c911c46c | 1529 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
91c00924 | 1530 | else |
a7c224a8 | 1531 | tx = async_xor_offs(xor_dest, off_dest, xor_srcs, off_srcs, count, |
c911c46c | 1532 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
91c00924 | 1533 | |
91c00924 DW |
1534 | return tx; |
1535 | } | |
1536 | ||
ac6b53b6 DW |
1537 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1538 | * @srcs - (struct page *) array of size sh->disks | |
d69454bc | 1539 | * @offs - (unsigned int) array of offset for each page |
ac6b53b6 DW |
1540 | * @sh - stripe_head to parse |
1541 | * | |
1542 | * Populates srcs in proper layout order for the stripe and returns the | |
1543 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1544 | * destination buffer is recorded in srcs[count] and the Q destination | |
1545 | * is recorded in srcs[count+1]]. | |
1546 | */ | |
584acdd4 | 1547 | static int set_syndrome_sources(struct page **srcs, |
d69454bc | 1548 | unsigned int *offs, |
584acdd4 MS |
1549 | struct stripe_head *sh, |
1550 | int srctype) | |
ac6b53b6 DW |
1551 | { |
1552 | int disks = sh->disks; | |
1553 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1554 | int d0_idx = raid6_d0(sh); | |
1555 | int count; | |
1556 | int i; | |
1557 | ||
1558 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1559 | srcs[i] = NULL; |
ac6b53b6 DW |
1560 | |
1561 | count = 0; | |
1562 | i = d0_idx; | |
1563 | do { | |
1564 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1565 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1566 | |
584acdd4 MS |
1567 | if (i == sh->qd_idx || i == sh->pd_idx || |
1568 | (srctype == SYNDROME_SRC_ALL) || | |
1569 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1570 | (test_bit(R5_Wantdrain, &dev->flags) || |
1571 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1572 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1573 | (dev->written || |
1574 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1575 | if (test_bit(R5_InJournal, &dev->flags)) |
1576 | srcs[slot] = sh->dev[i].orig_page; | |
1577 | else | |
1578 | srcs[slot] = sh->dev[i].page; | |
d69454bc YY |
1579 | /* |
1580 | * For R5_InJournal, PAGE_SIZE must be 4KB and will | |
1581 | * not shared page. In that case, dev[i].offset | |
1582 | * is 0. | |
1583 | */ | |
1584 | offs[slot] = sh->dev[i].offset; | |
1e6d690b | 1585 | } |
ac6b53b6 DW |
1586 | i = raid6_next_disk(i, disks); |
1587 | } while (i != d0_idx); | |
ac6b53b6 | 1588 | |
e4424fee | 1589 | return syndrome_disks; |
ac6b53b6 DW |
1590 | } |
1591 | ||
1592 | static struct dma_async_tx_descriptor * | |
1593 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1594 | { | |
1595 | int disks = sh->disks; | |
46d5b785 | 1596 | struct page **blocks = to_addr_page(percpu, 0); |
a7c224a8 | 1597 | unsigned int *offs = to_addr_offs(sh, percpu); |
ac6b53b6 DW |
1598 | int target; |
1599 | int qd_idx = sh->qd_idx; | |
1600 | struct dma_async_tx_descriptor *tx; | |
1601 | struct async_submit_ctl submit; | |
1602 | struct r5dev *tgt; | |
1603 | struct page *dest; | |
a7c224a8 | 1604 | unsigned int dest_off; |
ac6b53b6 DW |
1605 | int i; |
1606 | int count; | |
1607 | ||
59fc630b | 1608 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1609 | if (sh->ops.target < 0) |
1610 | target = sh->ops.target2; | |
1611 | else if (sh->ops.target2 < 0) | |
1612 | target = sh->ops.target; | |
91c00924 | 1613 | else |
ac6b53b6 DW |
1614 | /* we should only have one valid target */ |
1615 | BUG(); | |
1616 | BUG_ON(target < 0); | |
1617 | pr_debug("%s: stripe %llu block: %d\n", | |
1618 | __func__, (unsigned long long)sh->sector, target); | |
1619 | ||
1620 | tgt = &sh->dev[target]; | |
1621 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1622 | dest = tgt->page; | |
a7c224a8 | 1623 | dest_off = tgt->offset; |
ac6b53b6 DW |
1624 | |
1625 | atomic_inc(&sh->count); | |
1626 | ||
1627 | if (target == qd_idx) { | |
d69454bc | 1628 | count = set_syndrome_sources(blocks, offs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1629 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1630 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1631 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1632 | ops_complete_compute, sh, | |
46d5b785 | 1633 | to_addr_conv(sh, percpu, 0)); |
d69454bc | 1634 | tx = async_gen_syndrome(blocks, offs, count+2, |
c911c46c | 1635 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
ac6b53b6 DW |
1636 | } else { |
1637 | /* Compute any data- or p-drive using XOR */ | |
1638 | count = 0; | |
1639 | for (i = disks; i-- ; ) { | |
1640 | if (i == target || i == qd_idx) | |
1641 | continue; | |
a7c224a8 | 1642 | offs[count] = sh->dev[i].offset; |
ac6b53b6 DW |
1643 | blocks[count++] = sh->dev[i].page; |
1644 | } | |
1645 | ||
0403e382 DW |
1646 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1647 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1648 | to_addr_conv(sh, percpu, 0)); |
a7c224a8 | 1649 | tx = async_xor_offs(dest, dest_off, blocks, offs, count, |
c911c46c | 1650 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
ac6b53b6 | 1651 | } |
91c00924 | 1652 | |
91c00924 DW |
1653 | return tx; |
1654 | } | |
1655 | ||
ac6b53b6 DW |
1656 | static struct dma_async_tx_descriptor * |
1657 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1658 | { | |
1659 | int i, count, disks = sh->disks; | |
1660 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1661 | int d0_idx = raid6_d0(sh); | |
1662 | int faila = -1, failb = -1; | |
1663 | int target = sh->ops.target; | |
1664 | int target2 = sh->ops.target2; | |
1665 | struct r5dev *tgt = &sh->dev[target]; | |
1666 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1667 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1668 | struct page **blocks = to_addr_page(percpu, 0); |
a7c224a8 | 1669 | unsigned int *offs = to_addr_offs(sh, percpu); |
ac6b53b6 DW |
1670 | struct async_submit_ctl submit; |
1671 | ||
59fc630b | 1672 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1673 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1674 | __func__, (unsigned long long)sh->sector, target, target2); | |
1675 | BUG_ON(target < 0 || target2 < 0); | |
1676 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1677 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1678 | ||
6c910a78 | 1679 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1680 | * slot number conversion for 'faila' and 'failb' |
1681 | */ | |
a7c224a8 YY |
1682 | for (i = 0; i < disks ; i++) { |
1683 | offs[i] = 0; | |
5dd33c9a | 1684 | blocks[i] = NULL; |
a7c224a8 | 1685 | } |
ac6b53b6 DW |
1686 | count = 0; |
1687 | i = d0_idx; | |
1688 | do { | |
1689 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1690 | ||
a7c224a8 | 1691 | offs[slot] = sh->dev[i].offset; |
ac6b53b6 DW |
1692 | blocks[slot] = sh->dev[i].page; |
1693 | ||
1694 | if (i == target) | |
1695 | faila = slot; | |
1696 | if (i == target2) | |
1697 | failb = slot; | |
1698 | i = raid6_next_disk(i, disks); | |
1699 | } while (i != d0_idx); | |
ac6b53b6 DW |
1700 | |
1701 | BUG_ON(faila == failb); | |
1702 | if (failb < faila) | |
1703 | swap(faila, failb); | |
1704 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1705 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1706 | ||
1707 | atomic_inc(&sh->count); | |
1708 | ||
1709 | if (failb == syndrome_disks+1) { | |
1710 | /* Q disk is one of the missing disks */ | |
1711 | if (faila == syndrome_disks) { | |
1712 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1713 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1714 | ops_complete_compute, sh, | |
46d5b785 | 1715 | to_addr_conv(sh, percpu, 0)); |
d69454bc | 1716 | return async_gen_syndrome(blocks, offs, syndrome_disks+2, |
c911c46c YY |
1717 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1718 | &submit); | |
ac6b53b6 DW |
1719 | } else { |
1720 | struct page *dest; | |
a7c224a8 | 1721 | unsigned int dest_off; |
ac6b53b6 DW |
1722 | int data_target; |
1723 | int qd_idx = sh->qd_idx; | |
1724 | ||
1725 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1726 | if (target == qd_idx) | |
1727 | data_target = target2; | |
1728 | else | |
1729 | data_target = target; | |
1730 | ||
1731 | count = 0; | |
1732 | for (i = disks; i-- ; ) { | |
1733 | if (i == data_target || i == qd_idx) | |
1734 | continue; | |
a7c224a8 | 1735 | offs[count] = sh->dev[i].offset; |
ac6b53b6 DW |
1736 | blocks[count++] = sh->dev[i].page; |
1737 | } | |
1738 | dest = sh->dev[data_target].page; | |
a7c224a8 | 1739 | dest_off = sh->dev[data_target].offset; |
0403e382 DW |
1740 | init_async_submit(&submit, |
1741 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1742 | NULL, NULL, NULL, | |
46d5b785 | 1743 | to_addr_conv(sh, percpu, 0)); |
a7c224a8 | 1744 | tx = async_xor_offs(dest, dest_off, blocks, offs, count, |
c911c46c | 1745 | RAID5_STRIPE_SIZE(sh->raid_conf), |
ac6b53b6 DW |
1746 | &submit); |
1747 | ||
d69454bc | 1748 | count = set_syndrome_sources(blocks, offs, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1749 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1750 | ops_complete_compute, sh, | |
46d5b785 | 1751 | to_addr_conv(sh, percpu, 0)); |
d69454bc | 1752 | return async_gen_syndrome(blocks, offs, count+2, |
c911c46c YY |
1753 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1754 | &submit); | |
ac6b53b6 | 1755 | } |
ac6b53b6 | 1756 | } else { |
6c910a78 DW |
1757 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1758 | ops_complete_compute, sh, | |
46d5b785 | 1759 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1760 | if (failb == syndrome_disks) { |
1761 | /* We're missing D+P. */ | |
1762 | return async_raid6_datap_recov(syndrome_disks+2, | |
c911c46c YY |
1763 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1764 | faila, | |
4f86ff55 | 1765 | blocks, offs, &submit); |
6c910a78 DW |
1766 | } else { |
1767 | /* We're missing D+D. */ | |
1768 | return async_raid6_2data_recov(syndrome_disks+2, | |
c911c46c YY |
1769 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1770 | faila, failb, | |
4f86ff55 | 1771 | blocks, offs, &submit); |
6c910a78 | 1772 | } |
ac6b53b6 DW |
1773 | } |
1774 | } | |
1775 | ||
91c00924 DW |
1776 | static void ops_complete_prexor(void *stripe_head_ref) |
1777 | { | |
1778 | struct stripe_head *sh = stripe_head_ref; | |
1779 | ||
e46b272b | 1780 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1781 | (unsigned long long)sh->sector); |
1e6d690b SL |
1782 | |
1783 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1784 | /* | |
1785 | * raid5-cache write back uses orig_page during prexor. | |
1786 | * After prexor, it is time to free orig_page | |
1787 | */ | |
1788 | r5c_release_extra_page(sh); | |
91c00924 DW |
1789 | } |
1790 | ||
1791 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1792 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1793 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1794 | { |
91c00924 | 1795 | int disks = sh->disks; |
46d5b785 | 1796 | struct page **xor_srcs = to_addr_page(percpu, 0); |
a7c224a8 | 1797 | unsigned int *off_srcs = to_addr_offs(sh, percpu); |
91c00924 | 1798 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1799 | struct async_submit_ctl submit; |
91c00924 DW |
1800 | |
1801 | /* existing parity data subtracted */ | |
a7c224a8 | 1802 | unsigned int off_dest = off_srcs[count] = sh->dev[pd_idx].offset; |
91c00924 DW |
1803 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1804 | ||
59fc630b | 1805 | BUG_ON(sh->batch_head); |
e46b272b | 1806 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1807 | (unsigned long long)sh->sector); |
1808 | ||
1809 | for (i = disks; i--; ) { | |
1810 | struct r5dev *dev = &sh->dev[i]; | |
1811 | /* Only process blocks that are known to be uptodate */ | |
a7c224a8 YY |
1812 | if (test_bit(R5_InJournal, &dev->flags)) { |
1813 | /* | |
1814 | * For this case, PAGE_SIZE must be equal to 4KB and | |
1815 | * page offset is zero. | |
1816 | */ | |
1817 | off_srcs[count] = dev->offset; | |
1e6d690b | 1818 | xor_srcs[count++] = dev->orig_page; |
a7c224a8 YY |
1819 | } else if (test_bit(R5_Wantdrain, &dev->flags)) { |
1820 | off_srcs[count] = dev->offset; | |
91c00924 | 1821 | xor_srcs[count++] = dev->page; |
a7c224a8 | 1822 | } |
91c00924 DW |
1823 | } |
1824 | ||
0403e382 | 1825 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1826 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a7c224a8 | 1827 | tx = async_xor_offs(xor_dest, off_dest, xor_srcs, off_srcs, count, |
c911c46c | 1828 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
91c00924 DW |
1829 | |
1830 | return tx; | |
1831 | } | |
1832 | ||
584acdd4 MS |
1833 | static struct dma_async_tx_descriptor * |
1834 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1835 | struct dma_async_tx_descriptor *tx) | |
1836 | { | |
1837 | struct page **blocks = to_addr_page(percpu, 0); | |
d69454bc | 1838 | unsigned int *offs = to_addr_offs(sh, percpu); |
584acdd4 MS |
1839 | int count; |
1840 | struct async_submit_ctl submit; | |
1841 | ||
1842 | pr_debug("%s: stripe %llu\n", __func__, | |
1843 | (unsigned long long)sh->sector); | |
1844 | ||
d69454bc | 1845 | count = set_syndrome_sources(blocks, offs, sh, SYNDROME_SRC_WANT_DRAIN); |
584acdd4 MS |
1846 | |
1847 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1848 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
d69454bc | 1849 | tx = async_gen_syndrome(blocks, offs, count+2, |
c911c46c | 1850 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
584acdd4 MS |
1851 | |
1852 | return tx; | |
1853 | } | |
1854 | ||
91c00924 | 1855 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1856 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1857 | { |
1e6d690b | 1858 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1859 | int disks = sh->disks; |
d8ee0728 | 1860 | int i; |
59fc630b | 1861 | struct stripe_head *head_sh = sh; |
91c00924 | 1862 | |
e46b272b | 1863 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1864 | (unsigned long long)sh->sector); |
1865 | ||
1866 | for (i = disks; i--; ) { | |
59fc630b | 1867 | struct r5dev *dev; |
91c00924 | 1868 | struct bio *chosen; |
91c00924 | 1869 | |
59fc630b | 1870 | sh = head_sh; |
1871 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1872 | struct bio *wbi; |
1873 | ||
59fc630b | 1874 | again: |
1875 | dev = &sh->dev[i]; | |
1e6d690b SL |
1876 | /* |
1877 | * clear R5_InJournal, so when rewriting a page in | |
1878 | * journal, it is not skipped by r5l_log_stripe() | |
1879 | */ | |
1880 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1881 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1882 | chosen = dev->towrite; |
1883 | dev->towrite = NULL; | |
7a87f434 | 1884 | sh->overwrite_disks = 0; |
91c00924 DW |
1885 | BUG_ON(dev->written); |
1886 | wbi = dev->written = chosen; | |
b17459c0 | 1887 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1888 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1889 | |
4f024f37 | 1890 | while (wbi && wbi->bi_iter.bi_sector < |
c911c46c | 1891 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
1eff9d32 | 1892 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1893 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1894 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1895 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1896 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1897 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1898 | else { |
1899 | tx = async_copy_data(1, wbi, &dev->page, | |
248728dd | 1900 | dev->offset, |
1e6d690b SL |
1901 | dev->sector, tx, sh, |
1902 | r5c_is_writeback(conf->log)); | |
1903 | if (dev->page != dev->orig_page && | |
1904 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1905 | set_bit(R5_SkipCopy, &dev->flags); |
1906 | clear_bit(R5_UPTODATE, &dev->flags); | |
1907 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1908 | } | |
1909 | } | |
c911c46c | 1910 | wbi = r5_next_bio(conf, wbi, dev->sector); |
91c00924 | 1911 | } |
59fc630b | 1912 | |
1913 | if (head_sh->batch_head) { | |
1914 | sh = list_first_entry(&sh->batch_list, | |
1915 | struct stripe_head, | |
1916 | batch_list); | |
1917 | if (sh == head_sh) | |
1918 | continue; | |
1919 | goto again; | |
1920 | } | |
91c00924 DW |
1921 | } |
1922 | } | |
1923 | ||
1924 | return tx; | |
1925 | } | |
1926 | ||
ac6b53b6 | 1927 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1928 | { |
1929 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1930 | int disks = sh->disks; |
1931 | int pd_idx = sh->pd_idx; | |
1932 | int qd_idx = sh->qd_idx; | |
1933 | int i; | |
9e444768 | 1934 | bool fua = false, sync = false, discard = false; |
91c00924 | 1935 | |
e46b272b | 1936 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1937 | (unsigned long long)sh->sector); |
1938 | ||
bc0934f0 | 1939 | for (i = disks; i--; ) { |
e9c7469b | 1940 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1941 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1942 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1943 | } |
e9c7469b | 1944 | |
91c00924 DW |
1945 | for (i = disks; i--; ) { |
1946 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1947 | |
e9c7469b | 1948 | if (dev->written || i == pd_idx || i == qd_idx) { |
235b6003 | 1949 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) { |
9e444768 | 1950 | set_bit(R5_UPTODATE, &dev->flags); |
235b6003 N |
1951 | if (test_bit(STRIPE_EXPAND_READY, &sh->state)) |
1952 | set_bit(R5_Expanded, &dev->flags); | |
1953 | } | |
e9c7469b TH |
1954 | if (fua) |
1955 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1956 | if (sync) |
1957 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1958 | } |
91c00924 DW |
1959 | } |
1960 | ||
d8ee0728 DW |
1961 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1962 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1963 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1964 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1965 | else { | |
1966 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1967 | sh->reconstruct_state = reconstruct_state_result; | |
1968 | } | |
91c00924 DW |
1969 | |
1970 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1971 | raid5_release_stripe(sh); |
91c00924 DW |
1972 | } |
1973 | ||
1974 | static void | |
ac6b53b6 DW |
1975 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1976 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1977 | { |
91c00924 | 1978 | int disks = sh->disks; |
59fc630b | 1979 | struct page **xor_srcs; |
7aba13b7 | 1980 | unsigned int *off_srcs; |
a08abd8c | 1981 | struct async_submit_ctl submit; |
59fc630b | 1982 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1983 | struct page *xor_dest; |
7aba13b7 | 1984 | unsigned int off_dest; |
d8ee0728 | 1985 | int prexor = 0; |
91c00924 | 1986 | unsigned long flags; |
59fc630b | 1987 | int j = 0; |
1988 | struct stripe_head *head_sh = sh; | |
1989 | int last_stripe; | |
91c00924 | 1990 | |
e46b272b | 1991 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1992 | (unsigned long long)sh->sector); |
1993 | ||
620125f2 SL |
1994 | for (i = 0; i < sh->disks; i++) { |
1995 | if (pd_idx == i) | |
1996 | continue; | |
1997 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1998 | break; | |
1999 | } | |
2000 | if (i >= sh->disks) { | |
2001 | atomic_inc(&sh->count); | |
620125f2 SL |
2002 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
2003 | ops_complete_reconstruct(sh); | |
2004 | return; | |
2005 | } | |
59fc630b | 2006 | again: |
2007 | count = 0; | |
2008 | xor_srcs = to_addr_page(percpu, j); | |
7aba13b7 | 2009 | off_srcs = to_addr_offs(sh, percpu); |
91c00924 DW |
2010 | /* check if prexor is active which means only process blocks |
2011 | * that are part of a read-modify-write (written) | |
2012 | */ | |
59fc630b | 2013 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 2014 | prexor = 1; |
7aba13b7 | 2015 | off_dest = off_srcs[count] = sh->dev[pd_idx].offset; |
91c00924 DW |
2016 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
2017 | for (i = disks; i--; ) { | |
2018 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b | 2019 | if (head_sh->dev[i].written || |
7aba13b7 YY |
2020 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) { |
2021 | off_srcs[count] = dev->offset; | |
91c00924 | 2022 | xor_srcs[count++] = dev->page; |
7aba13b7 | 2023 | } |
91c00924 DW |
2024 | } |
2025 | } else { | |
2026 | xor_dest = sh->dev[pd_idx].page; | |
7aba13b7 | 2027 | off_dest = sh->dev[pd_idx].offset; |
91c00924 DW |
2028 | for (i = disks; i--; ) { |
2029 | struct r5dev *dev = &sh->dev[i]; | |
7aba13b7 YY |
2030 | if (i != pd_idx) { |
2031 | off_srcs[count] = dev->offset; | |
91c00924 | 2032 | xor_srcs[count++] = dev->page; |
7aba13b7 | 2033 | } |
91c00924 DW |
2034 | } |
2035 | } | |
2036 | ||
91c00924 DW |
2037 | /* 1/ if we prexor'd then the dest is reused as a source |
2038 | * 2/ if we did not prexor then we are redoing the parity | |
2039 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
2040 | * for the synchronous xor case | |
2041 | */ | |
59fc630b | 2042 | last_stripe = !head_sh->batch_head || |
2043 | list_first_entry(&sh->batch_list, | |
2044 | struct stripe_head, batch_list) == head_sh; | |
2045 | if (last_stripe) { | |
2046 | flags = ASYNC_TX_ACK | | |
2047 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
2048 | ||
2049 | atomic_inc(&head_sh->count); | |
2050 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
2051 | to_addr_conv(sh, percpu, j)); | |
2052 | } else { | |
2053 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
2054 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
2055 | to_addr_conv(sh, percpu, j)); | |
2056 | } | |
91c00924 | 2057 | |
a08abd8c | 2058 | if (unlikely(count == 1)) |
7aba13b7 | 2059 | tx = async_memcpy(xor_dest, xor_srcs[0], off_dest, off_srcs[0], |
c911c46c | 2060 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
a08abd8c | 2061 | else |
a7c224a8 | 2062 | tx = async_xor_offs(xor_dest, off_dest, xor_srcs, off_srcs, count, |
c911c46c | 2063 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
59fc630b | 2064 | if (!last_stripe) { |
2065 | j++; | |
2066 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
2067 | batch_list); | |
2068 | goto again; | |
2069 | } | |
91c00924 DW |
2070 | } |
2071 | ||
ac6b53b6 DW |
2072 | static void |
2073 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
2074 | struct dma_async_tx_descriptor *tx) | |
2075 | { | |
2076 | struct async_submit_ctl submit; | |
59fc630b | 2077 | struct page **blocks; |
d69454bc | 2078 | unsigned int *offs; |
59fc630b | 2079 | int count, i, j = 0; |
2080 | struct stripe_head *head_sh = sh; | |
2081 | int last_stripe; | |
584acdd4 MS |
2082 | int synflags; |
2083 | unsigned long txflags; | |
ac6b53b6 DW |
2084 | |
2085 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
2086 | ||
620125f2 SL |
2087 | for (i = 0; i < sh->disks; i++) { |
2088 | if (sh->pd_idx == i || sh->qd_idx == i) | |
2089 | continue; | |
2090 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
2091 | break; | |
2092 | } | |
2093 | if (i >= sh->disks) { | |
2094 | atomic_inc(&sh->count); | |
620125f2 SL |
2095 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
2096 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
2097 | ops_complete_reconstruct(sh); | |
2098 | return; | |
2099 | } | |
2100 | ||
59fc630b | 2101 | again: |
2102 | blocks = to_addr_page(percpu, j); | |
d69454bc | 2103 | offs = to_addr_offs(sh, percpu); |
584acdd4 MS |
2104 | |
2105 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
2106 | synflags = SYNDROME_SRC_WRITTEN; | |
2107 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
2108 | } else { | |
2109 | synflags = SYNDROME_SRC_ALL; | |
2110 | txflags = ASYNC_TX_ACK; | |
2111 | } | |
2112 | ||
d69454bc | 2113 | count = set_syndrome_sources(blocks, offs, sh, synflags); |
59fc630b | 2114 | last_stripe = !head_sh->batch_head || |
2115 | list_first_entry(&sh->batch_list, | |
2116 | struct stripe_head, batch_list) == head_sh; | |
2117 | ||
2118 | if (last_stripe) { | |
2119 | atomic_inc(&head_sh->count); | |
584acdd4 | 2120 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 2121 | head_sh, to_addr_conv(sh, percpu, j)); |
2122 | } else | |
2123 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
2124 | to_addr_conv(sh, percpu, j)); | |
d69454bc | 2125 | tx = async_gen_syndrome(blocks, offs, count+2, |
c911c46c | 2126 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); |
59fc630b | 2127 | if (!last_stripe) { |
2128 | j++; | |
2129 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
2130 | batch_list); | |
2131 | goto again; | |
2132 | } | |
91c00924 DW |
2133 | } |
2134 | ||
2135 | static void ops_complete_check(void *stripe_head_ref) | |
2136 | { | |
2137 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 2138 | |
e46b272b | 2139 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2140 | (unsigned long long)sh->sector); |
2141 | ||
ecc65c9b | 2142 | sh->check_state = check_state_check_result; |
91c00924 | 2143 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2144 | raid5_release_stripe(sh); |
91c00924 DW |
2145 | } |
2146 | ||
ac6b53b6 | 2147 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 2148 | { |
91c00924 | 2149 | int disks = sh->disks; |
ac6b53b6 DW |
2150 | int pd_idx = sh->pd_idx; |
2151 | int qd_idx = sh->qd_idx; | |
2152 | struct page *xor_dest; | |
a7c224a8 | 2153 | unsigned int off_dest; |
46d5b785 | 2154 | struct page **xor_srcs = to_addr_page(percpu, 0); |
a7c224a8 | 2155 | unsigned int *off_srcs = to_addr_offs(sh, percpu); |
91c00924 | 2156 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 2157 | struct async_submit_ctl submit; |
ac6b53b6 DW |
2158 | int count; |
2159 | int i; | |
91c00924 | 2160 | |
e46b272b | 2161 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2162 | (unsigned long long)sh->sector); |
2163 | ||
59fc630b | 2164 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
2165 | count = 0; |
2166 | xor_dest = sh->dev[pd_idx].page; | |
a7c224a8 YY |
2167 | off_dest = sh->dev[pd_idx].offset; |
2168 | off_srcs[count] = off_dest; | |
ac6b53b6 | 2169 | xor_srcs[count++] = xor_dest; |
91c00924 | 2170 | for (i = disks; i--; ) { |
ac6b53b6 DW |
2171 | if (i == pd_idx || i == qd_idx) |
2172 | continue; | |
a7c224a8 | 2173 | off_srcs[count] = sh->dev[i].offset; |
ac6b53b6 | 2174 | xor_srcs[count++] = sh->dev[i].page; |
91c00924 DW |
2175 | } |
2176 | ||
d6f38f31 | 2177 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 2178 | to_addr_conv(sh, percpu, 0)); |
a7c224a8 | 2179 | tx = async_xor_val_offs(xor_dest, off_dest, xor_srcs, off_srcs, count, |
c911c46c | 2180 | RAID5_STRIPE_SIZE(sh->raid_conf), |
a08abd8c | 2181 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 2182 | |
91c00924 | 2183 | atomic_inc(&sh->count); |
a08abd8c DW |
2184 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
2185 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
2186 | } |
2187 | ||
ac6b53b6 DW |
2188 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
2189 | { | |
46d5b785 | 2190 | struct page **srcs = to_addr_page(percpu, 0); |
d69454bc | 2191 | unsigned int *offs = to_addr_offs(sh, percpu); |
ac6b53b6 DW |
2192 | struct async_submit_ctl submit; |
2193 | int count; | |
2194 | ||
2195 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
2196 | (unsigned long long)sh->sector, checkp); | |
2197 | ||
59fc630b | 2198 | BUG_ON(sh->batch_head); |
d69454bc | 2199 | count = set_syndrome_sources(srcs, offs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
2200 | if (!checkp) |
2201 | srcs[count] = NULL; | |
91c00924 | 2202 | |
91c00924 | 2203 | atomic_inc(&sh->count); |
ac6b53b6 | 2204 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 2205 | sh, to_addr_conv(sh, percpu, 0)); |
d69454bc | 2206 | async_syndrome_val(srcs, offs, count+2, |
c911c46c | 2207 | RAID5_STRIPE_SIZE(sh->raid_conf), |
d69454bc | 2208 | &sh->ops.zero_sum_result, percpu->spare_page, 0, &submit); |
91c00924 DW |
2209 | } |
2210 | ||
51acbcec | 2211 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
2212 | { |
2213 | int overlap_clear = 0, i, disks = sh->disks; | |
2214 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 2215 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2216 | int level = conf->level; |
d6f38f31 DW |
2217 | struct raid5_percpu *percpu; |
2218 | unsigned long cpu; | |
91c00924 | 2219 | |
d6f38f31 DW |
2220 | cpu = get_cpu(); |
2221 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2222 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2223 | ops_run_biofill(sh); |
2224 | overlap_clear++; | |
2225 | } | |
2226 | ||
7b3a871e | 2227 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2228 | if (level < 6) |
2229 | tx = ops_run_compute5(sh, percpu); | |
2230 | else { | |
2231 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2232 | tx = ops_run_compute6_1(sh, percpu); | |
2233 | else | |
2234 | tx = ops_run_compute6_2(sh, percpu); | |
2235 | } | |
2236 | /* terminate the chain if reconstruct is not set to be run */ | |
2237 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2238 | async_tx_ack(tx); |
2239 | } | |
91c00924 | 2240 | |
584acdd4 MS |
2241 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2242 | if (level < 6) | |
2243 | tx = ops_run_prexor5(sh, percpu, tx); | |
2244 | else | |
2245 | tx = ops_run_prexor6(sh, percpu, tx); | |
2246 | } | |
91c00924 | 2247 | |
ae1713e2 AP |
2248 | if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request)) |
2249 | tx = ops_run_partial_parity(sh, percpu, tx); | |
2250 | ||
600aa109 | 2251 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2252 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2253 | overlap_clear++; |
2254 | } | |
2255 | ||
ac6b53b6 DW |
2256 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2257 | if (level < 6) | |
2258 | ops_run_reconstruct5(sh, percpu, tx); | |
2259 | else | |
2260 | ops_run_reconstruct6(sh, percpu, tx); | |
2261 | } | |
91c00924 | 2262 | |
ac6b53b6 DW |
2263 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2264 | if (sh->check_state == check_state_run) | |
2265 | ops_run_check_p(sh, percpu); | |
2266 | else if (sh->check_state == check_state_run_q) | |
2267 | ops_run_check_pq(sh, percpu, 0); | |
2268 | else if (sh->check_state == check_state_run_pq) | |
2269 | ops_run_check_pq(sh, percpu, 1); | |
2270 | else | |
2271 | BUG(); | |
2272 | } | |
91c00924 | 2273 | |
59fc630b | 2274 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2275 | for (i = disks; i--; ) { |
2276 | struct r5dev *dev = &sh->dev[i]; | |
2277 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2278 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2279 | } | |
d6f38f31 | 2280 | put_cpu(); |
91c00924 DW |
2281 | } |
2282 | ||
845b9e22 AP |
2283 | static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh) |
2284 | { | |
046169f0 YY |
2285 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
2286 | kfree(sh->pages); | |
2287 | #endif | |
845b9e22 AP |
2288 | if (sh->ppl_page) |
2289 | __free_page(sh->ppl_page); | |
2290 | kmem_cache_free(sc, sh); | |
2291 | } | |
2292 | ||
5f9d1fde | 2293 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
845b9e22 | 2294 | int disks, struct r5conf *conf) |
f18c1a35 N |
2295 | { |
2296 | struct stripe_head *sh; | |
5f9d1fde | 2297 | int i; |
f18c1a35 N |
2298 | |
2299 | sh = kmem_cache_zalloc(sc, gfp); | |
2300 | if (sh) { | |
2301 | spin_lock_init(&sh->stripe_lock); | |
2302 | spin_lock_init(&sh->batch_lock); | |
2303 | INIT_LIST_HEAD(&sh->batch_list); | |
2304 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2305 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2306 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2307 | atomic_set(&sh->count, 1); |
845b9e22 | 2308 | sh->raid_conf = conf; |
a39f7afd | 2309 | sh->log_start = MaxSector; |
5f9d1fde SL |
2310 | for (i = 0; i < disks; i++) { |
2311 | struct r5dev *dev = &sh->dev[i]; | |
2312 | ||
3a83f467 ML |
2313 | bio_init(&dev->req, &dev->vec, 1); |
2314 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2315 | } |
845b9e22 AP |
2316 | |
2317 | if (raid5_has_ppl(conf)) { | |
2318 | sh->ppl_page = alloc_page(gfp); | |
2319 | if (!sh->ppl_page) { | |
2320 | free_stripe(sc, sh); | |
046169f0 | 2321 | return NULL; |
845b9e22 AP |
2322 | } |
2323 | } | |
046169f0 YY |
2324 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
2325 | if (init_stripe_shared_pages(sh, conf, disks)) { | |
2326 | free_stripe(sc, sh); | |
2327 | return NULL; | |
2328 | } | |
2329 | #endif | |
f18c1a35 N |
2330 | } |
2331 | return sh; | |
2332 | } | |
486f0644 | 2333 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2334 | { |
2335 | struct stripe_head *sh; | |
f18c1a35 | 2336 | |
845b9e22 | 2337 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf); |
3f294f4f N |
2338 | if (!sh) |
2339 | return 0; | |
6ce32846 | 2340 | |
a9683a79 | 2341 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2342 | shrink_buffers(sh); |
845b9e22 | 2343 | free_stripe(conf->slab_cache, sh); |
3f294f4f N |
2344 | return 0; |
2345 | } | |
486f0644 N |
2346 | sh->hash_lock_index = |
2347 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2348 | /* we just created an active stripe so... */ |
3f294f4f | 2349 | atomic_inc(&conf->active_stripes); |
59fc630b | 2350 | |
6d036f7d | 2351 | raid5_release_stripe(sh); |
486f0644 | 2352 | conf->max_nr_stripes++; |
3f294f4f N |
2353 | return 1; |
2354 | } | |
2355 | ||
d1688a6d | 2356 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2357 | { |
e18b890b | 2358 | struct kmem_cache *sc; |
53b8d89d | 2359 | size_t namelen = sizeof(conf->cache_name[0]); |
5e5e3e78 | 2360 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2361 | |
f4be6b43 | 2362 | if (conf->mddev->gendisk) |
53b8d89d | 2363 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 N |
2364 | "raid%d-%s", conf->level, mdname(conf->mddev)); |
2365 | else | |
53b8d89d | 2366 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 | 2367 | "raid%d-%p", conf->level, conf->mddev); |
53b8d89d | 2368 | snprintf(conf->cache_name[1], namelen, "%.27s-alt", conf->cache_name[0]); |
f4be6b43 | 2369 | |
ad01c9e3 N |
2370 | conf->active_name = 0; |
2371 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2372 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2373 | 0, 0, NULL); |
1da177e4 LT |
2374 | if (!sc) |
2375 | return 1; | |
2376 | conf->slab_cache = sc; | |
ad01c9e3 | 2377 | conf->pool_size = devs; |
486f0644 N |
2378 | while (num--) |
2379 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2380 | return 1; |
486f0644 | 2381 | |
1da177e4 LT |
2382 | return 0; |
2383 | } | |
29269553 | 2384 | |
d6f38f31 | 2385 | /** |
7f8a30e5 CL |
2386 | * scribble_alloc - allocate percpu scribble buffer for required size |
2387 | * of the scribble region | |
2aada5b1 DLM |
2388 | * @percpu: from for_each_present_cpu() of the caller |
2389 | * @num: total number of disks in the array | |
2390 | * @cnt: scribble objs count for required size of the scribble region | |
d6f38f31 | 2391 | * |
7f8a30e5 | 2392 | * The scribble buffer size must be enough to contain: |
d6f38f31 DW |
2393 | * 1/ a struct page pointer for each device in the array +2 |
2394 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2395 | * (dma_map_page()) or page (page_address()) address. | |
2396 | * | |
2397 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2398 | * calculate over all devices (not just the data blocks), using zeros in place | |
2399 | * of the P and Q blocks. | |
2400 | */ | |
b330e6a4 | 2401 | static int scribble_alloc(struct raid5_percpu *percpu, |
ba54d4d4 | 2402 | int num, int cnt) |
d6f38f31 | 2403 | { |
b330e6a4 | 2404 | size_t obj_size = |
7aba13b7 YY |
2405 | sizeof(struct page *) * (num + 2) + |
2406 | sizeof(addr_conv_t) * (num + 2) + | |
2407 | sizeof(unsigned int) * (num + 2); | |
b330e6a4 | 2408 | void *scribble; |
d6f38f31 | 2409 | |
ba54d4d4 CL |
2410 | /* |
2411 | * If here is in raid array suspend context, it is in memalloc noio | |
2412 | * context as well, there is no potential recursive memory reclaim | |
2413 | * I/Os with the GFP_KERNEL flag. | |
2414 | */ | |
2415 | scribble = kvmalloc_array(cnt, obj_size, GFP_KERNEL); | |
b330e6a4 KO |
2416 | if (!scribble) |
2417 | return -ENOMEM; | |
2418 | ||
2419 | kvfree(percpu->scribble); | |
2420 | ||
2421 | percpu->scribble = scribble; | |
2422 | percpu->scribble_obj_size = obj_size; | |
2423 | return 0; | |
d6f38f31 DW |
2424 | } |
2425 | ||
738a2738 N |
2426 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2427 | { | |
2428 | unsigned long cpu; | |
2429 | int err = 0; | |
2430 | ||
27a353c0 SL |
2431 | /* |
2432 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2433 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2434 | * should equal to new_disks and new_sectors | |
2435 | */ | |
2436 | if (conf->scribble_disks >= new_disks && | |
2437 | conf->scribble_sectors >= new_sectors) | |
2438 | return 0; | |
738a2738 | 2439 | mddev_suspend(conf->mddev); |
252034e0 | 2440 | cpus_read_lock(); |
b330e6a4 | 2441 | |
738a2738 N |
2442 | for_each_present_cpu(cpu) { |
2443 | struct raid5_percpu *percpu; | |
738a2738 N |
2444 | |
2445 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
b330e6a4 | 2446 | err = scribble_alloc(percpu, new_disks, |
c911c46c | 2447 | new_sectors / RAID5_STRIPE_SECTORS(conf)); |
b330e6a4 | 2448 | if (err) |
738a2738 | 2449 | break; |
738a2738 | 2450 | } |
b330e6a4 | 2451 | |
252034e0 | 2452 | cpus_read_unlock(); |
738a2738 | 2453 | mddev_resume(conf->mddev); |
27a353c0 SL |
2454 | if (!err) { |
2455 | conf->scribble_disks = new_disks; | |
2456 | conf->scribble_sectors = new_sectors; | |
2457 | } | |
738a2738 N |
2458 | return err; |
2459 | } | |
2460 | ||
d1688a6d | 2461 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2462 | { |
2463 | /* Make all the stripes able to hold 'newsize' devices. | |
2464 | * New slots in each stripe get 'page' set to a new page. | |
2465 | * | |
2466 | * This happens in stages: | |
2467 | * 1/ create a new kmem_cache and allocate the required number of | |
2468 | * stripe_heads. | |
83f0d77a | 2469 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2470 | * to the new stripe_heads. This will have the side effect of |
2471 | * freezing the array as once all stripe_heads have been collected, | |
2472 | * no IO will be possible. Old stripe heads are freed once their | |
2473 | * pages have been transferred over, and the old kmem_cache is | |
2474 | * freed when all stripes are done. | |
2475 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
3560741e | 2476 | * we simple return a failure status - no need to clean anything up. |
ad01c9e3 N |
2477 | * 4/ allocate new pages for the new slots in the new stripe_heads. |
2478 | * If this fails, we don't bother trying the shrink the | |
2479 | * stripe_heads down again, we just leave them as they are. | |
2480 | * As each stripe_head is processed the new one is released into | |
2481 | * active service. | |
2482 | * | |
2483 | * Once step2 is started, we cannot afford to wait for a write, | |
2484 | * so we use GFP_NOIO allocations. | |
2485 | */ | |
2486 | struct stripe_head *osh, *nsh; | |
2487 | LIST_HEAD(newstripes); | |
2488 | struct disk_info *ndisks; | |
2214c260 | 2489 | int err = 0; |
e18b890b | 2490 | struct kmem_cache *sc; |
ad01c9e3 | 2491 | int i; |
566c09c5 | 2492 | int hash, cnt; |
ad01c9e3 | 2493 | |
2214c260 | 2494 | md_allow_write(conf->mddev); |
2a2275d6 | 2495 | |
ad01c9e3 N |
2496 | /* Step 1 */ |
2497 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2498 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2499 | 0, 0, NULL); |
ad01c9e3 N |
2500 | if (!sc) |
2501 | return -ENOMEM; | |
2502 | ||
2d5b569b N |
2503 | /* Need to ensure auto-resizing doesn't interfere */ |
2504 | mutex_lock(&conf->cache_size_mutex); | |
2505 | ||
ad01c9e3 | 2506 | for (i = conf->max_nr_stripes; i; i--) { |
845b9e22 | 2507 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf); |
ad01c9e3 N |
2508 | if (!nsh) |
2509 | break; | |
2510 | ||
ad01c9e3 N |
2511 | list_add(&nsh->lru, &newstripes); |
2512 | } | |
2513 | if (i) { | |
2514 | /* didn't get enough, give up */ | |
2515 | while (!list_empty(&newstripes)) { | |
2516 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2517 | list_del(&nsh->lru); | |
845b9e22 | 2518 | free_stripe(sc, nsh); |
ad01c9e3 N |
2519 | } |
2520 | kmem_cache_destroy(sc); | |
2d5b569b | 2521 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2522 | return -ENOMEM; |
2523 | } | |
2524 | /* Step 2 - Must use GFP_NOIO now. | |
2525 | * OK, we have enough stripes, start collecting inactive | |
2526 | * stripes and copying them over | |
2527 | */ | |
566c09c5 SL |
2528 | hash = 0; |
2529 | cnt = 0; | |
ad01c9e3 | 2530 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2531 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2532 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2533 | !list_empty(conf->inactive_list + hash), |
2534 | unlock_device_hash_lock(conf, hash), | |
2535 | lock_device_hash_lock(conf, hash)); | |
2536 | osh = get_free_stripe(conf, hash); | |
2537 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2538 | |
f16acaf3 YY |
2539 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
2540 | for (i = 0; i < osh->nr_pages; i++) { | |
2541 | nsh->pages[i] = osh->pages[i]; | |
2542 | osh->pages[i] = NULL; | |
2543 | } | |
2544 | #endif | |
d592a996 | 2545 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2546 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 | 2547 | nsh->dev[i].orig_page = osh->dev[i].page; |
7aba13b7 | 2548 | nsh->dev[i].offset = osh->dev[i].offset; |
d592a996 | 2549 | } |
566c09c5 | 2550 | nsh->hash_lock_index = hash; |
845b9e22 | 2551 | free_stripe(conf->slab_cache, osh); |
566c09c5 SL |
2552 | cnt++; |
2553 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2554 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2555 | hash++; | |
2556 | cnt = 0; | |
2557 | } | |
ad01c9e3 N |
2558 | } |
2559 | kmem_cache_destroy(conf->slab_cache); | |
2560 | ||
2561 | /* Step 3. | |
2562 | * At this point, we are holding all the stripes so the array | |
2563 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2564 | * conf->disks and the scribble region |
ad01c9e3 | 2565 | */ |
6396bb22 | 2566 | ndisks = kcalloc(newsize, sizeof(struct disk_info), GFP_NOIO); |
ad01c9e3 | 2567 | if (ndisks) { |
d7bd398e | 2568 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2569 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2570 | |
2571 | for (i = conf->pool_size; i < newsize; i++) { | |
2572 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2573 | if (!ndisks[i].extra_page) | |
2574 | err = -ENOMEM; | |
2575 | } | |
2576 | ||
2577 | if (err) { | |
2578 | for (i = conf->pool_size; i < newsize; i++) | |
2579 | if (ndisks[i].extra_page) | |
2580 | put_page(ndisks[i].extra_page); | |
2581 | kfree(ndisks); | |
2582 | } else { | |
2583 | kfree(conf->disks); | |
2584 | conf->disks = ndisks; | |
2585 | } | |
ad01c9e3 N |
2586 | } else |
2587 | err = -ENOMEM; | |
2588 | ||
583da48e DY |
2589 | conf->slab_cache = sc; |
2590 | conf->active_name = 1-conf->active_name; | |
2591 | ||
ad01c9e3 N |
2592 | /* Step 4, return new stripes to service */ |
2593 | while(!list_empty(&newstripes)) { | |
2594 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2595 | list_del_init(&nsh->lru); | |
d6f38f31 | 2596 | |
f16acaf3 YY |
2597 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
2598 | for (i = 0; i < nsh->nr_pages; i++) { | |
2599 | if (nsh->pages[i]) | |
2600 | continue; | |
2601 | nsh->pages[i] = alloc_page(GFP_NOIO); | |
2602 | if (!nsh->pages[i]) | |
2603 | err = -ENOMEM; | |
2604 | } | |
2605 | ||
2606 | for (i = conf->raid_disks; i < newsize; i++) { | |
2607 | if (nsh->dev[i].page) | |
2608 | continue; | |
2609 | nsh->dev[i].page = raid5_get_dev_page(nsh, i); | |
2610 | nsh->dev[i].orig_page = nsh->dev[i].page; | |
2611 | nsh->dev[i].offset = raid5_get_page_offset(nsh, i); | |
2612 | } | |
2613 | #else | |
ad01c9e3 N |
2614 | for (i=conf->raid_disks; i < newsize; i++) |
2615 | if (nsh->dev[i].page == NULL) { | |
2616 | struct page *p = alloc_page(GFP_NOIO); | |
2617 | nsh->dev[i].page = p; | |
d592a996 | 2618 | nsh->dev[i].orig_page = p; |
7aba13b7 | 2619 | nsh->dev[i].offset = 0; |
ad01c9e3 N |
2620 | if (!p) |
2621 | err = -ENOMEM; | |
2622 | } | |
f16acaf3 | 2623 | #endif |
6d036f7d | 2624 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2625 | } |
2626 | /* critical section pass, GFP_NOIO no longer needed */ | |
2627 | ||
6e9eac2d N |
2628 | if (!err) |
2629 | conf->pool_size = newsize; | |
b44c018c SL |
2630 | mutex_unlock(&conf->cache_size_mutex); |
2631 | ||
ad01c9e3 N |
2632 | return err; |
2633 | } | |
1da177e4 | 2634 | |
486f0644 | 2635 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2636 | { |
2637 | struct stripe_head *sh; | |
49895bcc | 2638 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2639 | |
566c09c5 SL |
2640 | spin_lock_irq(conf->hash_locks + hash); |
2641 | sh = get_free_stripe(conf, hash); | |
2642 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2643 | if (!sh) |
2644 | return 0; | |
78bafebd | 2645 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2646 | shrink_buffers(sh); |
845b9e22 | 2647 | free_stripe(conf->slab_cache, sh); |
3f294f4f | 2648 | atomic_dec(&conf->active_stripes); |
486f0644 | 2649 | conf->max_nr_stripes--; |
3f294f4f N |
2650 | return 1; |
2651 | } | |
2652 | ||
d1688a6d | 2653 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2654 | { |
486f0644 N |
2655 | while (conf->max_nr_stripes && |
2656 | drop_one_stripe(conf)) | |
2657 | ; | |
3f294f4f | 2658 | |
644df1a8 | 2659 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2660 | conf->slab_cache = NULL; |
2661 | } | |
2662 | ||
4246a0b6 | 2663 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2664 | { |
99c0fb5f | 2665 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2666 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2667 | int disks = sh->disks, i; |
d6950432 | 2668 | char b[BDEVNAME_SIZE]; |
dd054fce | 2669 | struct md_rdev *rdev = NULL; |
05616be5 | 2670 | sector_t s; |
1da177e4 LT |
2671 | |
2672 | for (i=0 ; i<disks; i++) | |
2673 | if (bi == &sh->dev[i].req) | |
2674 | break; | |
2675 | ||
4246a0b6 | 2676 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2677 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2678 | bi->bi_status); |
1da177e4 | 2679 | if (i == disks) { |
5f9d1fde | 2680 | bio_reset(bi); |
1da177e4 | 2681 | BUG(); |
6712ecf8 | 2682 | return; |
1da177e4 | 2683 | } |
14a75d3e | 2684 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2685 | /* If replacement finished while this request was outstanding, |
2686 | * 'replacement' might be NULL already. | |
2687 | * In that case it moved down to 'rdev'. | |
2688 | * rdev is not removed until all requests are finished. | |
2689 | */ | |
14a75d3e | 2690 | rdev = conf->disks[i].replacement; |
dd054fce | 2691 | if (!rdev) |
14a75d3e | 2692 | rdev = conf->disks[i].rdev; |
1da177e4 | 2693 | |
05616be5 N |
2694 | if (use_new_offset(conf, sh)) |
2695 | s = sh->sector + rdev->new_data_offset; | |
2696 | else | |
2697 | s = sh->sector + rdev->data_offset; | |
4e4cbee9 | 2698 | if (!bi->bi_status) { |
1da177e4 | 2699 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2700 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2701 | /* Note that this cannot happen on a |
2702 | * replacement device. We just fail those on | |
2703 | * any error | |
2704 | */ | |
cc6167b4 N |
2705 | pr_info_ratelimited( |
2706 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
c911c46c | 2707 | mdname(conf->mddev), RAID5_STRIPE_SECTORS(conf), |
05616be5 | 2708 | (unsigned long long)s, |
8bda470e | 2709 | bdevname(rdev->bdev, b)); |
c911c46c | 2710 | atomic_add(RAID5_STRIPE_SECTORS(conf), &rdev->corrected_errors); |
4e5314b5 N |
2711 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2712 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2713 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2714 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2715 | ||
86aa1397 SL |
2716 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2717 | /* | |
2718 | * end read for a page in journal, this | |
2719 | * must be preparing for prexor in rmw | |
2720 | */ | |
2721 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2722 | ||
14a75d3e N |
2723 | if (atomic_read(&rdev->read_errors)) |
2724 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2725 | } else { |
14a75d3e | 2726 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2727 | int retry = 0; |
2e8ac303 | 2728 | int set_bad = 0; |
d6950432 | 2729 | |
1da177e4 | 2730 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
b76b4715 NC |
2731 | if (!(bi->bi_status == BLK_STS_PROTECTION)) |
2732 | atomic_inc(&rdev->read_errors); | |
14a75d3e | 2733 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2734 | pr_warn_ratelimited( |
2735 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2736 | mdname(conf->mddev), |
05616be5 | 2737 | (unsigned long long)s, |
14a75d3e | 2738 | bdn); |
2e8ac303 | 2739 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2740 | set_bad = 1; | |
cc6167b4 N |
2741 | pr_warn_ratelimited( |
2742 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2743 | mdname(conf->mddev), |
05616be5 | 2744 | (unsigned long long)s, |
8bda470e | 2745 | bdn); |
2e8ac303 | 2746 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2747 | /* Oh, no!!! */ |
2e8ac303 | 2748 | set_bad = 1; |
cc6167b4 N |
2749 | pr_warn_ratelimited( |
2750 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2751 | mdname(conf->mddev), |
05616be5 | 2752 | (unsigned long long)s, |
8bda470e | 2753 | bdn); |
2e8ac303 | 2754 | } else if (atomic_read(&rdev->read_errors) |
0009fad0 NC |
2755 | > conf->max_nr_stripes) { |
2756 | if (!test_bit(Faulty, &rdev->flags)) { | |
2757 | pr_warn("md/raid:%s: %d read_errors > %d stripes\n", | |
2758 | mdname(conf->mddev), | |
2759 | atomic_read(&rdev->read_errors), | |
2760 | conf->max_nr_stripes); | |
2761 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", | |
2762 | mdname(conf->mddev), bdn); | |
2763 | } | |
2764 | } else | |
ba22dcbf | 2765 | retry = 1; |
edfa1f65 BY |
2766 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2767 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2768 | retry = 1; | |
ba22dcbf | 2769 | if (retry) |
143f6e73 XN |
2770 | if (sh->qd_idx >= 0 && sh->pd_idx == i) |
2771 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2772 | else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { | |
3f9e7c14 | 2773 | set_bit(R5_ReadError, &sh->dev[i].flags); |
2774 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2775 | } else | |
2776 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2777 | else { |
4e5314b5 N |
2778 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2779 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2780 | if (!(set_bad |
2781 | && test_bit(In_sync, &rdev->flags) | |
2782 | && rdev_set_badblocks( | |
c911c46c | 2783 | rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), 0))) |
2e8ac303 | 2784 | md_error(conf->mddev, rdev); |
ba22dcbf | 2785 | } |
1da177e4 | 2786 | } |
14a75d3e | 2787 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2788 | bio_reset(bi); |
1da177e4 LT |
2789 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2790 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2791 | raid5_release_stripe(sh); |
1da177e4 LT |
2792 | } |
2793 | ||
4246a0b6 | 2794 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2795 | { |
99c0fb5f | 2796 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2797 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2798 | int disks = sh->disks, i; |
3f649ab7 | 2799 | struct md_rdev *rdev; |
b84db560 N |
2800 | sector_t first_bad; |
2801 | int bad_sectors; | |
977df362 | 2802 | int replacement = 0; |
1da177e4 | 2803 | |
977df362 N |
2804 | for (i = 0 ; i < disks; i++) { |
2805 | if (bi == &sh->dev[i].req) { | |
2806 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2807 | break; |
977df362 N |
2808 | } |
2809 | if (bi == &sh->dev[i].rreq) { | |
2810 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2811 | if (rdev) |
2812 | replacement = 1; | |
2813 | else | |
2814 | /* rdev was removed and 'replacement' | |
2815 | * replaced it. rdev is not removed | |
2816 | * until all requests are finished. | |
2817 | */ | |
2818 | rdev = conf->disks[i].rdev; | |
977df362 N |
2819 | break; |
2820 | } | |
2821 | } | |
4246a0b6 | 2822 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2823 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2824 | bi->bi_status); |
1da177e4 | 2825 | if (i == disks) { |
5f9d1fde | 2826 | bio_reset(bi); |
1da177e4 | 2827 | BUG(); |
6712ecf8 | 2828 | return; |
1da177e4 LT |
2829 | } |
2830 | ||
977df362 | 2831 | if (replacement) { |
4e4cbee9 | 2832 | if (bi->bi_status) |
977df362 N |
2833 | md_error(conf->mddev, rdev); |
2834 | else if (is_badblock(rdev, sh->sector, | |
c911c46c | 2835 | RAID5_STRIPE_SECTORS(conf), |
977df362 N |
2836 | &first_bad, &bad_sectors)) |
2837 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2838 | } else { | |
4e4cbee9 | 2839 | if (bi->bi_status) { |
9f97e4b1 | 2840 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2841 | set_bit(WriteErrorSeen, &rdev->flags); |
2842 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2843 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2844 | set_bit(MD_RECOVERY_NEEDED, | |
2845 | &rdev->mddev->recovery); | |
977df362 | 2846 | } else if (is_badblock(rdev, sh->sector, |
c911c46c | 2847 | RAID5_STRIPE_SECTORS(conf), |
c0b32972 | 2848 | &first_bad, &bad_sectors)) { |
977df362 | 2849 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2850 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2851 | /* That was a successful write so make | |
2852 | * sure it looks like we already did | |
2853 | * a re-write. | |
2854 | */ | |
2855 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2856 | } | |
977df362 N |
2857 | } |
2858 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2859 | |
4e4cbee9 | 2860 | if (sh->batch_head && bi->bi_status && !replacement) |
72ac7330 | 2861 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2862 | ||
c9445555 | 2863 | bio_reset(bi); |
977df362 N |
2864 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2865 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2866 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2867 | raid5_release_stripe(sh); |
59fc630b | 2868 | |
2869 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2870 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2871 | } |
2872 | ||
849674e4 | 2873 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2874 | { |
2875 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2876 | struct r5conf *conf = mddev->private; |
908f4fbd | 2877 | unsigned long flags; |
0c55e022 | 2878 | pr_debug("raid456: error called\n"); |
1da177e4 | 2879 | |
ba97a054 MT |
2880 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n", |
2881 | mdname(mddev), bdevname(rdev->bdev, b)); | |
2882 | ||
908f4fbd | 2883 | spin_lock_irqsave(&conf->device_lock, flags); |
ba97a054 MT |
2884 | set_bit(Faulty, &rdev->flags); |
2885 | clear_bit(In_sync, &rdev->flags); | |
2886 | mddev->degraded = raid5_calc_degraded(conf); | |
fb73b357 | 2887 | |
ba97a054 MT |
2888 | if (has_failed(conf)) { |
2889 | set_bit(MD_BROKEN, &conf->mddev->flags); | |
fb73b357 | 2890 | conf->recovery_disabled = mddev->recovery_disabled; |
ba97a054 MT |
2891 | |
2892 | pr_crit("md/raid:%s: Cannot continue operation (%d/%d failed).\n", | |
2893 | mdname(mddev), mddev->degraded, conf->raid_disks); | |
2894 | } else { | |
2895 | pr_crit("md/raid:%s: Operation continuing on %d devices.\n", | |
2896 | mdname(mddev), conf->raid_disks - mddev->degraded); | |
fb73b357 MT |
2897 | } |
2898 | ||
908f4fbd N |
2899 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2900 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2901 | ||
de393cde | 2902 | set_bit(Blocked, &rdev->flags); |
2953079c SL |
2903 | set_mask_bits(&mddev->sb_flags, 0, |
2904 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
70d466f7 | 2905 | r5c_update_on_rdev_error(mddev, rdev); |
16a53ecc | 2906 | } |
1da177e4 LT |
2907 | |
2908 | /* | |
2909 | * Input: a 'big' sector number, | |
2910 | * Output: index of the data and parity disk, and the sector # in them. | |
2911 | */ | |
6d036f7d SL |
2912 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2913 | int previous, int *dd_idx, | |
2914 | struct stripe_head *sh) | |
1da177e4 | 2915 | { |
6e3b96ed | 2916 | sector_t stripe, stripe2; |
35f2a591 | 2917 | sector_t chunk_number; |
1da177e4 | 2918 | unsigned int chunk_offset; |
911d4ee8 | 2919 | int pd_idx, qd_idx; |
67cc2b81 | 2920 | int ddf_layout = 0; |
1da177e4 | 2921 | sector_t new_sector; |
e183eaed N |
2922 | int algorithm = previous ? conf->prev_algo |
2923 | : conf->algorithm; | |
09c9e5fa AN |
2924 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2925 | : conf->chunk_sectors; | |
112bf897 N |
2926 | int raid_disks = previous ? conf->previous_raid_disks |
2927 | : conf->raid_disks; | |
2928 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2929 | |
2930 | /* First compute the information on this sector */ | |
2931 | ||
2932 | /* | |
2933 | * Compute the chunk number and the sector offset inside the chunk | |
2934 | */ | |
2935 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2936 | chunk_number = r_sector; | |
1da177e4 LT |
2937 | |
2938 | /* | |
2939 | * Compute the stripe number | |
2940 | */ | |
35f2a591 N |
2941 | stripe = chunk_number; |
2942 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2943 | stripe2 = stripe; |
1da177e4 LT |
2944 | /* |
2945 | * Select the parity disk based on the user selected algorithm. | |
2946 | */ | |
84789554 | 2947 | pd_idx = qd_idx = -1; |
16a53ecc N |
2948 | switch(conf->level) { |
2949 | case 4: | |
911d4ee8 | 2950 | pd_idx = data_disks; |
16a53ecc N |
2951 | break; |
2952 | case 5: | |
e183eaed | 2953 | switch (algorithm) { |
1da177e4 | 2954 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2955 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2956 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2957 | (*dd_idx)++; |
2958 | break; | |
2959 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2960 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2961 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2962 | (*dd_idx)++; |
2963 | break; | |
2964 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2965 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2966 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2967 | break; |
2968 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2969 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2970 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2971 | break; |
99c0fb5f N |
2972 | case ALGORITHM_PARITY_0: |
2973 | pd_idx = 0; | |
2974 | (*dd_idx)++; | |
2975 | break; | |
2976 | case ALGORITHM_PARITY_N: | |
2977 | pd_idx = data_disks; | |
2978 | break; | |
1da177e4 | 2979 | default: |
99c0fb5f | 2980 | BUG(); |
16a53ecc N |
2981 | } |
2982 | break; | |
2983 | case 6: | |
2984 | ||
e183eaed | 2985 | switch (algorithm) { |
16a53ecc | 2986 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2987 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2988 | qd_idx = pd_idx + 1; |
2989 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2990 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2991 | qd_idx = 0; |
2992 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2993 | (*dd_idx) += 2; /* D D P Q D */ |
2994 | break; | |
2995 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2996 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2997 | qd_idx = pd_idx + 1; |
2998 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2999 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
3000 | qd_idx = 0; |
3001 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
3002 | (*dd_idx) += 2; /* D D P Q D */ |
3003 | break; | |
3004 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 3005 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
3006 | qd_idx = (pd_idx + 1) % raid_disks; |
3007 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
3008 | break; |
3009 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 3010 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
3011 | qd_idx = (pd_idx + 1) % raid_disks; |
3012 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 3013 | break; |
99c0fb5f N |
3014 | |
3015 | case ALGORITHM_PARITY_0: | |
3016 | pd_idx = 0; | |
3017 | qd_idx = 1; | |
3018 | (*dd_idx) += 2; | |
3019 | break; | |
3020 | case ALGORITHM_PARITY_N: | |
3021 | pd_idx = data_disks; | |
3022 | qd_idx = data_disks + 1; | |
3023 | break; | |
3024 | ||
3025 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
3026 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
3027 | * of blocks for computing Q is different. | |
3028 | */ | |
6e3b96ed | 3029 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
3030 | qd_idx = pd_idx + 1; |
3031 | if (pd_idx == raid_disks-1) { | |
3032 | (*dd_idx)++; /* Q D D D P */ | |
3033 | qd_idx = 0; | |
3034 | } else if (*dd_idx >= pd_idx) | |
3035 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 3036 | ddf_layout = 1; |
99c0fb5f N |
3037 | break; |
3038 | ||
3039 | case ALGORITHM_ROTATING_N_RESTART: | |
3040 | /* Same a left_asymmetric, by first stripe is | |
3041 | * D D D P Q rather than | |
3042 | * Q D D D P | |
3043 | */ | |
6e3b96ed N |
3044 | stripe2 += 1; |
3045 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
3046 | qd_idx = pd_idx + 1; |
3047 | if (pd_idx == raid_disks-1) { | |
3048 | (*dd_idx)++; /* Q D D D P */ | |
3049 | qd_idx = 0; | |
3050 | } else if (*dd_idx >= pd_idx) | |
3051 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 3052 | ddf_layout = 1; |
99c0fb5f N |
3053 | break; |
3054 | ||
3055 | case ALGORITHM_ROTATING_N_CONTINUE: | |
3056 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 3057 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
3058 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
3059 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 3060 | ddf_layout = 1; |
99c0fb5f N |
3061 | break; |
3062 | ||
3063 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
3064 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 3065 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
3066 | if (*dd_idx >= pd_idx) |
3067 | (*dd_idx)++; | |
3068 | qd_idx = raid_disks - 1; | |
3069 | break; | |
3070 | ||
3071 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 3072 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
3073 | if (*dd_idx >= pd_idx) |
3074 | (*dd_idx)++; | |
3075 | qd_idx = raid_disks - 1; | |
3076 | break; | |
3077 | ||
3078 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 3079 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
3080 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
3081 | qd_idx = raid_disks - 1; | |
3082 | break; | |
3083 | ||
3084 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 3085 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
3086 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
3087 | qd_idx = raid_disks - 1; | |
3088 | break; | |
3089 | ||
3090 | case ALGORITHM_PARITY_0_6: | |
3091 | pd_idx = 0; | |
3092 | (*dd_idx)++; | |
3093 | qd_idx = raid_disks - 1; | |
3094 | break; | |
3095 | ||
16a53ecc | 3096 | default: |
99c0fb5f | 3097 | BUG(); |
16a53ecc N |
3098 | } |
3099 | break; | |
1da177e4 LT |
3100 | } |
3101 | ||
911d4ee8 N |
3102 | if (sh) { |
3103 | sh->pd_idx = pd_idx; | |
3104 | sh->qd_idx = qd_idx; | |
67cc2b81 | 3105 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 3106 | } |
1da177e4 LT |
3107 | /* |
3108 | * Finally, compute the new sector number | |
3109 | */ | |
3110 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
3111 | return new_sector; | |
3112 | } | |
3113 | ||
6d036f7d | 3114 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 3115 | { |
d1688a6d | 3116 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
3117 | int raid_disks = sh->disks; |
3118 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 3119 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
3120 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
3121 | : conf->chunk_sectors; | |
e183eaed N |
3122 | int algorithm = previous ? conf->prev_algo |
3123 | : conf->algorithm; | |
1da177e4 LT |
3124 | sector_t stripe; |
3125 | int chunk_offset; | |
35f2a591 N |
3126 | sector_t chunk_number; |
3127 | int dummy1, dd_idx = i; | |
1da177e4 | 3128 | sector_t r_sector; |
911d4ee8 | 3129 | struct stripe_head sh2; |
1da177e4 LT |
3130 | |
3131 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
3132 | stripe = new_sector; | |
1da177e4 | 3133 | |
16a53ecc N |
3134 | if (i == sh->pd_idx) |
3135 | return 0; | |
3136 | switch(conf->level) { | |
3137 | case 4: break; | |
3138 | case 5: | |
e183eaed | 3139 | switch (algorithm) { |
1da177e4 LT |
3140 | case ALGORITHM_LEFT_ASYMMETRIC: |
3141 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
3142 | if (i > sh->pd_idx) | |
3143 | i--; | |
3144 | break; | |
3145 | case ALGORITHM_LEFT_SYMMETRIC: | |
3146 | case ALGORITHM_RIGHT_SYMMETRIC: | |
3147 | if (i < sh->pd_idx) | |
3148 | i += raid_disks; | |
3149 | i -= (sh->pd_idx + 1); | |
3150 | break; | |
99c0fb5f N |
3151 | case ALGORITHM_PARITY_0: |
3152 | i -= 1; | |
3153 | break; | |
3154 | case ALGORITHM_PARITY_N: | |
3155 | break; | |
1da177e4 | 3156 | default: |
99c0fb5f | 3157 | BUG(); |
16a53ecc N |
3158 | } |
3159 | break; | |
3160 | case 6: | |
d0dabf7e | 3161 | if (i == sh->qd_idx) |
16a53ecc | 3162 | return 0; /* It is the Q disk */ |
e183eaed | 3163 | switch (algorithm) { |
16a53ecc N |
3164 | case ALGORITHM_LEFT_ASYMMETRIC: |
3165 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
3166 | case ALGORITHM_ROTATING_ZERO_RESTART: |
3167 | case ALGORITHM_ROTATING_N_RESTART: | |
3168 | if (sh->pd_idx == raid_disks-1) | |
3169 | i--; /* Q D D D P */ | |
16a53ecc N |
3170 | else if (i > sh->pd_idx) |
3171 | i -= 2; /* D D P Q D */ | |
3172 | break; | |
3173 | case ALGORITHM_LEFT_SYMMETRIC: | |
3174 | case ALGORITHM_RIGHT_SYMMETRIC: | |
3175 | if (sh->pd_idx == raid_disks-1) | |
3176 | i--; /* Q D D D P */ | |
3177 | else { | |
3178 | /* D D P Q D */ | |
3179 | if (i < sh->pd_idx) | |
3180 | i += raid_disks; | |
3181 | i -= (sh->pd_idx + 2); | |
3182 | } | |
3183 | break; | |
99c0fb5f N |
3184 | case ALGORITHM_PARITY_0: |
3185 | i -= 2; | |
3186 | break; | |
3187 | case ALGORITHM_PARITY_N: | |
3188 | break; | |
3189 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 3190 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
3191 | if (sh->pd_idx == 0) |
3192 | i--; /* P D D D Q */ | |
e4424fee N |
3193 | else { |
3194 | /* D D Q P D */ | |
3195 | if (i < sh->pd_idx) | |
3196 | i += raid_disks; | |
3197 | i -= (sh->pd_idx + 1); | |
3198 | } | |
99c0fb5f N |
3199 | break; |
3200 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
3201 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
3202 | if (i > sh->pd_idx) | |
3203 | i--; | |
3204 | break; | |
3205 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
3206 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
3207 | if (i < sh->pd_idx) | |
3208 | i += data_disks + 1; | |
3209 | i -= (sh->pd_idx + 1); | |
3210 | break; | |
3211 | case ALGORITHM_PARITY_0_6: | |
3212 | i -= 1; | |
3213 | break; | |
16a53ecc | 3214 | default: |
99c0fb5f | 3215 | BUG(); |
16a53ecc N |
3216 | } |
3217 | break; | |
1da177e4 LT |
3218 | } |
3219 | ||
3220 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 3221 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 3222 | |
112bf897 | 3223 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 3224 | previous, &dummy1, &sh2); |
911d4ee8 N |
3225 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
3226 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
3227 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
3228 | mdname(conf->mddev)); | |
1da177e4 LT |
3229 | return 0; |
3230 | } | |
3231 | return r_sector; | |
3232 | } | |
3233 | ||
07e83364 SL |
3234 | /* |
3235 | * There are cases where we want handle_stripe_dirtying() and | |
3236 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
3237 | * | |
3238 | * This function checks whether we want to delay the towrite. Specifically, | |
3239 | * we delay the towrite when: | |
3240 | * | |
3241 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
3242 | * stripe has data in journal (for other devices). | |
3243 | * | |
3244 | * In this case, when reading data for the non-overwrite dev, it is | |
3245 | * necessary to handle complex rmw of write back cache (prexor with | |
3246 | * orig_page, and xor with page). To keep read path simple, we would | |
3247 | * like to flush data in journal to RAID disks first, so complex rmw | |
3248 | * is handled in the write patch (handle_stripe_dirtying). | |
3249 | * | |
39b99586 SL |
3250 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
3251 | * | |
3252 | * It is important to be able to flush all stripes in raid5-cache. | |
3253 | * Therefore, we need reserve some space on the journal device for | |
3254 | * these flushes. If flush operation includes pending writes to the | |
3255 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
3256 | * for the flush out. If we exclude these pending writes from flush | |
3257 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
3258 | * Therefore, excluding pending writes in these cases enables more | |
3259 | * efficient use of the journal device. | |
3260 | * | |
3261 | * Note: To make sure the stripe makes progress, we only delay | |
3262 | * towrite for stripes with data already in journal (injournal > 0). | |
3263 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
3264 | * no_space_stripes list. | |
3265 | * | |
70d466f7 SL |
3266 | * 3. during journal failure |
3267 | * In journal failure, we try to flush all cached data to raid disks | |
3268 | * based on data in stripe cache. The array is read-only to upper | |
3269 | * layers, so we would skip all pending writes. | |
3270 | * | |
07e83364 | 3271 | */ |
39b99586 SL |
3272 | static inline bool delay_towrite(struct r5conf *conf, |
3273 | struct r5dev *dev, | |
3274 | struct stripe_head_state *s) | |
07e83364 | 3275 | { |
39b99586 SL |
3276 | /* case 1 above */ |
3277 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
3278 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
3279 | return true; | |
3280 | /* case 2 above */ | |
3281 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
3282 | s->injournal > 0) | |
3283 | return true; | |
70d466f7 SL |
3284 | /* case 3 above */ |
3285 | if (s->log_failed && s->injournal) | |
3286 | return true; | |
39b99586 | 3287 | return false; |
07e83364 SL |
3288 | } |
3289 | ||
600aa109 | 3290 | static void |
c0f7bddb | 3291 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3292 | int rcw, int expand) |
e33129d8 | 3293 | { |
584acdd4 | 3294 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3295 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3296 | int level = conf->level; |
e33129d8 DW |
3297 | |
3298 | if (rcw) { | |
1e6d690b SL |
3299 | /* |
3300 | * In some cases, handle_stripe_dirtying initially decided to | |
3301 | * run rmw and allocates extra page for prexor. However, rcw is | |
3302 | * cheaper later on. We need to free the extra page now, | |
3303 | * because we won't be able to do that in ops_complete_prexor(). | |
3304 | */ | |
3305 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3306 | |
3307 | for (i = disks; i--; ) { | |
3308 | struct r5dev *dev = &sh->dev[i]; | |
3309 | ||
39b99586 | 3310 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3311 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3312 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3313 | if (!expand) |
3314 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3315 | s->locked++; |
1e6d690b SL |
3316 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3317 | set_bit(R5_LOCKED, &dev->flags); | |
3318 | s->locked++; | |
e33129d8 DW |
3319 | } |
3320 | } | |
ce7d363a N |
3321 | /* if we are not expanding this is a proper write request, and |
3322 | * there will be bios with new data to be drained into the | |
3323 | * stripe cache | |
3324 | */ | |
3325 | if (!expand) { | |
3326 | if (!s->locked) | |
3327 | /* False alarm, nothing to do */ | |
3328 | return; | |
3329 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3330 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3331 | } else | |
3332 | sh->reconstruct_state = reconstruct_state_run; | |
3333 | ||
3334 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3335 | ||
c0f7bddb | 3336 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3337 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3338 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3339 | } else { |
3340 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3341 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3342 | BUG_ON(level == 6 && |
3343 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3344 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3345 | |
e33129d8 DW |
3346 | for (i = disks; i--; ) { |
3347 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3348 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3349 | continue; |
3350 | ||
e33129d8 DW |
3351 | if (dev->towrite && |
3352 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3353 | test_bit(R5_Wantcompute, &dev->flags))) { |
3354 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3355 | set_bit(R5_LOCKED, &dev->flags); |
3356 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3357 | s->locked++; |
1e6d690b SL |
3358 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3359 | set_bit(R5_LOCKED, &dev->flags); | |
3360 | s->locked++; | |
e33129d8 DW |
3361 | } |
3362 | } | |
ce7d363a N |
3363 | if (!s->locked) |
3364 | /* False alarm - nothing to do */ | |
3365 | return; | |
3366 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3367 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3368 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3369 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3370 | } |
3371 | ||
c0f7bddb | 3372 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3373 | * are in flight |
3374 | */ | |
3375 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3376 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3377 | s->locked++; |
e33129d8 | 3378 | |
c0f7bddb YT |
3379 | if (level == 6) { |
3380 | int qd_idx = sh->qd_idx; | |
3381 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3382 | ||
3383 | set_bit(R5_LOCKED, &dev->flags); | |
3384 | clear_bit(R5_UPTODATE, &dev->flags); | |
3385 | s->locked++; | |
3386 | } | |
3387 | ||
845b9e22 | 3388 | if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page && |
3418d036 AP |
3389 | test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) && |
3390 | !test_bit(STRIPE_FULL_WRITE, &sh->state) && | |
3391 | test_bit(R5_Insync, &sh->dev[pd_idx].flags)) | |
3392 | set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request); | |
3393 | ||
600aa109 | 3394 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3395 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3396 | s->locked, s->ops_request); |
e33129d8 | 3397 | } |
16a53ecc | 3398 | |
1da177e4 LT |
3399 | /* |
3400 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3401 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3402 | * The bi_next chain must be in order. |
3403 | */ | |
da41ba65 | 3404 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3405 | int forwrite, int previous) | |
1da177e4 LT |
3406 | { |
3407 | struct bio **bip; | |
d1688a6d | 3408 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3409 | int firstwrite=0; |
1da177e4 | 3410 | |
cbe47ec5 | 3411 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3412 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3413 | (unsigned long long)sh->sector); |
3414 | ||
b17459c0 | 3415 | spin_lock_irq(&sh->stripe_lock); |
2cd259a7 | 3416 | sh->dev[dd_idx].write_hint = bi->bi_write_hint; |
59fc630b | 3417 | /* Don't allow new IO added to stripes in batch list */ |
3418 | if (sh->batch_head) | |
3419 | goto overlap; | |
72626685 | 3420 | if (forwrite) { |
1da177e4 | 3421 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3422 | if (*bip == NULL) |
72626685 N |
3423 | firstwrite = 1; |
3424 | } else | |
1da177e4 | 3425 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3426 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3427 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3428 | goto overlap; |
3429 | bip = & (*bip)->bi_next; | |
3430 | } | |
4f024f37 | 3431 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3432 | goto overlap; |
3433 | ||
3418d036 AP |
3434 | if (forwrite && raid5_has_ppl(conf)) { |
3435 | /* | |
3436 | * With PPL only writes to consecutive data chunks within a | |
3437 | * stripe are allowed because for a single stripe_head we can | |
3438 | * only have one PPL entry at a time, which describes one data | |
3439 | * range. Not really an overlap, but wait_for_overlap can be | |
3440 | * used to handle this. | |
3441 | */ | |
3442 | sector_t sector; | |
3443 | sector_t first = 0; | |
3444 | sector_t last = 0; | |
3445 | int count = 0; | |
3446 | int i; | |
3447 | ||
3448 | for (i = 0; i < sh->disks; i++) { | |
3449 | if (i != sh->pd_idx && | |
3450 | (i == dd_idx || sh->dev[i].towrite)) { | |
3451 | sector = sh->dev[i].sector; | |
3452 | if (count == 0 || sector < first) | |
3453 | first = sector; | |
3454 | if (sector > last) | |
3455 | last = sector; | |
3456 | count++; | |
3457 | } | |
3458 | } | |
3459 | ||
3460 | if (first + conf->chunk_sectors * (count - 1) != last) | |
3461 | goto overlap; | |
3462 | } | |
3463 | ||
da41ba65 | 3464 | if (!forwrite || previous) |
3465 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3466 | ||
78bafebd | 3467 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3468 | if (*bip) |
3469 | bi->bi_next = *bip; | |
3470 | *bip = bi; | |
016c76ac | 3471 | bio_inc_remaining(bi); |
49728050 | 3472 | md_write_inc(conf->mddev, bi); |
72626685 | 3473 | |
1da177e4 LT |
3474 | if (forwrite) { |
3475 | /* check if page is covered */ | |
3476 | sector_t sector = sh->dev[dd_idx].sector; | |
3477 | for (bi=sh->dev[dd_idx].towrite; | |
c911c46c | 3478 | sector < sh->dev[dd_idx].sector + RAID5_STRIPE_SECTORS(conf) && |
4f024f37 | 3479 | bi && bi->bi_iter.bi_sector <= sector; |
c911c46c | 3480 | bi = r5_next_bio(conf, bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3481 | if (bio_end_sector(bi) >= sector) |
3482 | sector = bio_end_sector(bi); | |
1da177e4 | 3483 | } |
c911c46c | 3484 | if (sector >= sh->dev[dd_idx].sector + RAID5_STRIPE_SECTORS(conf)) |
7a87f434 | 3485 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3486 | sh->overwrite_disks++; | |
1da177e4 | 3487 | } |
cbe47ec5 N |
3488 | |
3489 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3490 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3491 | (unsigned long long)sh->sector, dd_idx); |
3492 | ||
3493 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3494 | /* Cannot hold spinlock over bitmap_startwrite, |
3495 | * but must ensure this isn't added to a batch until | |
3496 | * we have added to the bitmap and set bm_seq. | |
3497 | * So set STRIPE_BITMAP_PENDING to prevent | |
3498 | * batching. | |
3499 | * If multiple add_stripe_bio() calls race here they | |
3500 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3501 | * to complete "bitmap_startwrite" gets to set | |
3502 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3503 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3504 | * any more. | |
3505 | */ | |
3506 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3507 | spin_unlock_irq(&sh->stripe_lock); | |
e64e4018 | 3508 | md_bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3509 | RAID5_STRIPE_SECTORS(conf), 0); |
d0852df5 N |
3510 | spin_lock_irq(&sh->stripe_lock); |
3511 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3512 | if (!sh->batch_head) { | |
3513 | sh->bm_seq = conf->seq_flush+1; | |
3514 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3515 | } | |
cbe47ec5 | 3516 | } |
d0852df5 | 3517 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3518 | |
3519 | if (stripe_can_batch(sh)) | |
3520 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3521 | return 1; |
3522 | ||
3523 | overlap: | |
3524 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3525 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3526 | return 0; |
3527 | } | |
3528 | ||
d1688a6d | 3529 | static void end_reshape(struct r5conf *conf); |
29269553 | 3530 | |
d1688a6d | 3531 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3532 | struct stripe_head *sh) |
ccfcc3c1 | 3533 | { |
784052ec | 3534 | int sectors_per_chunk = |
09c9e5fa | 3535 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3536 | int dd_idx; |
2d2063ce | 3537 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3538 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3539 | |
112bf897 N |
3540 | raid5_compute_sector(conf, |
3541 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3542 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3543 | previous, |
911d4ee8 | 3544 | &dd_idx, sh); |
ccfcc3c1 N |
3545 | } |
3546 | ||
a4456856 | 3547 | static void |
d1688a6d | 3548 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
bd83d0a2 | 3549 | struct stripe_head_state *s, int disks) |
a4456856 DW |
3550 | { |
3551 | int i; | |
59fc630b | 3552 | BUG_ON(sh->batch_head); |
a4456856 DW |
3553 | for (i = disks; i--; ) { |
3554 | struct bio *bi; | |
3555 | int bitmap_end = 0; | |
3556 | ||
3557 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3558 | struct md_rdev *rdev; |
a4456856 DW |
3559 | rcu_read_lock(); |
3560 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3561 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3562 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3563 | atomic_inc(&rdev->nr_pending); |
3564 | else | |
3565 | rdev = NULL; | |
a4456856 | 3566 | rcu_read_unlock(); |
7f0da59b N |
3567 | if (rdev) { |
3568 | if (!rdev_set_badblocks( | |
3569 | rdev, | |
3570 | sh->sector, | |
c911c46c | 3571 | RAID5_STRIPE_SECTORS(conf), 0)) |
7f0da59b N |
3572 | md_error(conf->mddev, rdev); |
3573 | rdev_dec_pending(rdev, conf->mddev); | |
3574 | } | |
a4456856 | 3575 | } |
b17459c0 | 3576 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3577 | /* fail all writes first */ |
3578 | bi = sh->dev[i].towrite; | |
3579 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3580 | sh->overwrite_disks = 0; |
b17459c0 | 3581 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3582 | if (bi) |
a4456856 | 3583 | bitmap_end = 1; |
a4456856 | 3584 | |
ff875738 | 3585 | log_stripe_write_finished(sh); |
0576b1c6 | 3586 | |
a4456856 DW |
3587 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3588 | wake_up(&conf->wait_for_overlap); | |
3589 | ||
4f024f37 | 3590 | while (bi && bi->bi_iter.bi_sector < |
c911c46c YY |
3591 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
3592 | struct bio *nextbi = r5_next_bio(conf, bi, sh->dev[i].sector); | |
4246a0b6 | 3593 | |
49728050 | 3594 | md_write_end(conf->mddev); |
6308d8e3 | 3595 | bio_io_error(bi); |
a4456856 DW |
3596 | bi = nextbi; |
3597 | } | |
7eaf7e8e | 3598 | if (bitmap_end) |
e64e4018 | 3599 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3600 | RAID5_STRIPE_SECTORS(conf), 0, 0); |
7eaf7e8e | 3601 | bitmap_end = 0; |
a4456856 DW |
3602 | /* and fail all 'written' */ |
3603 | bi = sh->dev[i].written; | |
3604 | sh->dev[i].written = NULL; | |
d592a996 SL |
3605 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3606 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3607 | sh->dev[i].page = sh->dev[i].orig_page; | |
3608 | } | |
3609 | ||
a4456856 | 3610 | if (bi) bitmap_end = 1; |
4f024f37 | 3611 | while (bi && bi->bi_iter.bi_sector < |
c911c46c YY |
3612 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
3613 | struct bio *bi2 = r5_next_bio(conf, bi, sh->dev[i].sector); | |
4246a0b6 | 3614 | |
49728050 | 3615 | md_write_end(conf->mddev); |
6308d8e3 | 3616 | bio_io_error(bi); |
a4456856 DW |
3617 | bi = bi2; |
3618 | } | |
3619 | ||
b5e98d65 DW |
3620 | /* fail any reads if this device is non-operational and |
3621 | * the data has not reached the cache yet. | |
3622 | */ | |
3623 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3624 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3625 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3626 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3627 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3628 | bi = sh->dev[i].toread; |
3629 | sh->dev[i].toread = NULL; | |
143c4d05 | 3630 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3631 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3632 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3633 | if (bi) |
3634 | s->to_read--; | |
4f024f37 | 3635 | while (bi && bi->bi_iter.bi_sector < |
c911c46c | 3636 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
a4456856 | 3637 | struct bio *nextbi = |
c911c46c | 3638 | r5_next_bio(conf, bi, sh->dev[i].sector); |
4246a0b6 | 3639 | |
6308d8e3 | 3640 | bio_io_error(bi); |
a4456856 DW |
3641 | bi = nextbi; |
3642 | } | |
3643 | } | |
a4456856 | 3644 | if (bitmap_end) |
e64e4018 | 3645 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3646 | RAID5_STRIPE_SECTORS(conf), 0, 0); |
8cfa7b0f N |
3647 | /* If we were in the middle of a write the parity block might |
3648 | * still be locked - so just clear all R5_LOCKED flags | |
3649 | */ | |
3650 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3651 | } |
ebda780b SL |
3652 | s->to_write = 0; |
3653 | s->written = 0; | |
a4456856 | 3654 | |
8b3e6cdc DW |
3655 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3656 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3657 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3658 | } |
3659 | ||
7f0da59b | 3660 | static void |
d1688a6d | 3661 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3662 | struct stripe_head_state *s) |
3663 | { | |
3664 | int abort = 0; | |
3665 | int i; | |
3666 | ||
59fc630b | 3667 | BUG_ON(sh->batch_head); |
7f0da59b | 3668 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3669 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3670 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3671 | s->syncing = 0; |
9a3e1101 | 3672 | s->replacing = 0; |
7f0da59b | 3673 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3674 | * Don't even need to abort as that is handled elsewhere |
3675 | * if needed, and not always wanted e.g. if there is a known | |
3676 | * bad block here. | |
9a3e1101 | 3677 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3678 | * non-sync devices, or abort the recovery |
3679 | */ | |
18b9837e N |
3680 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3681 | /* During recovery devices cannot be removed, so | |
3682 | * locking and refcounting of rdevs is not needed | |
3683 | */ | |
e50d3992 | 3684 | rcu_read_lock(); |
18b9837e | 3685 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3686 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3687 | if (rdev |
3688 | && !test_bit(Faulty, &rdev->flags) | |
3689 | && !test_bit(In_sync, &rdev->flags) | |
3690 | && !rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 3691 | RAID5_STRIPE_SECTORS(conf), 0)) |
18b9837e | 3692 | abort = 1; |
e50d3992 | 3693 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3694 | if (rdev |
3695 | && !test_bit(Faulty, &rdev->flags) | |
3696 | && !test_bit(In_sync, &rdev->flags) | |
3697 | && !rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 3698 | RAID5_STRIPE_SECTORS(conf), 0)) |
18b9837e N |
3699 | abort = 1; |
3700 | } | |
e50d3992 | 3701 | rcu_read_unlock(); |
18b9837e N |
3702 | if (abort) |
3703 | conf->recovery_disabled = | |
3704 | conf->mddev->recovery_disabled; | |
7f0da59b | 3705 | } |
c911c46c | 3706 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), !abort); |
7f0da59b N |
3707 | } |
3708 | ||
9a3e1101 N |
3709 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3710 | { | |
3711 | struct md_rdev *rdev; | |
3712 | int rv = 0; | |
3f232d6a N |
3713 | |
3714 | rcu_read_lock(); | |
3715 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3716 | if (rdev |
3717 | && !test_bit(Faulty, &rdev->flags) | |
3718 | && !test_bit(In_sync, &rdev->flags) | |
3719 | && (rdev->recovery_offset <= sh->sector | |
3720 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3721 | rv = 1; | |
3f232d6a | 3722 | rcu_read_unlock(); |
9a3e1101 N |
3723 | return rv; |
3724 | } | |
3725 | ||
2c58f06e N |
3726 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3727 | int disk_idx, int disks) | |
a4456856 | 3728 | { |
5599becc | 3729 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3730 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3731 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3732 | int i; |
45a4d8fd | 3733 | bool force_rcw = (sh->raid_conf->rmw_level == PARITY_DISABLE_RMW); |
5599becc | 3734 | |
a79cfe12 N |
3735 | |
3736 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3737 | test_bit(R5_UPTODATE, &dev->flags)) | |
3738 | /* No point reading this as we already have it or have | |
3739 | * decided to get it. | |
3740 | */ | |
3741 | return 0; | |
3742 | ||
3743 | if (dev->toread || | |
3744 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3745 | /* We need this block to directly satisfy a request */ | |
3746 | return 1; | |
3747 | ||
3748 | if (s->syncing || s->expanding || | |
3749 | (s->replacing && want_replace(sh, disk_idx))) | |
3750 | /* When syncing, or expanding we read everything. | |
3751 | * When replacing, we need the replaced block. | |
3752 | */ | |
3753 | return 1; | |
3754 | ||
3755 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3756 | (s->failed >= 2 && fdev[1]->toread)) | |
3757 | /* If we want to read from a failed device, then | |
3758 | * we need to actually read every other device. | |
3759 | */ | |
3760 | return 1; | |
3761 | ||
a9d56950 N |
3762 | /* Sometimes neither read-modify-write nor reconstruct-write |
3763 | * cycles can work. In those cases we read every block we | |
3764 | * can. Then the parity-update is certain to have enough to | |
3765 | * work with. | |
3766 | * This can only be a problem when we need to write something, | |
3767 | * and some device has failed. If either of those tests | |
3768 | * fail we need look no further. | |
3769 | */ | |
3770 | if (!s->failed || !s->to_write) | |
3771 | return 0; | |
3772 | ||
3773 | if (test_bit(R5_Insync, &dev->flags) && | |
3774 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3775 | /* Pre-reads at not permitted until after short delay | |
3776 | * to gather multiple requests. However if this | |
3560741e | 3777 | * device is no Insync, the block could only be computed |
a9d56950 N |
3778 | * and there is no need to delay that. |
3779 | */ | |
3780 | return 0; | |
ea664c82 | 3781 | |
36707bb2 | 3782 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3783 | if (fdev[i]->towrite && |
3784 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3785 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3786 | /* If we have a partial write to a failed | |
3787 | * device, then we will need to reconstruct | |
3788 | * the content of that device, so all other | |
3789 | * devices must be read. | |
3790 | */ | |
3791 | return 1; | |
45a4d8fd CP |
3792 | |
3793 | if (s->failed >= 2 && | |
3794 | (fdev[i]->towrite || | |
3795 | s->failed_num[i] == sh->pd_idx || | |
3796 | s->failed_num[i] == sh->qd_idx) && | |
3797 | !test_bit(R5_UPTODATE, &fdev[i]->flags)) | |
3798 | /* In max degraded raid6, If the failed disk is P, Q, | |
3799 | * or we want to read the failed disk, we need to do | |
3800 | * reconstruct-write. | |
3801 | */ | |
3802 | force_rcw = true; | |
ea664c82 N |
3803 | } |
3804 | ||
45a4d8fd CP |
3805 | /* If we are forced to do a reconstruct-write, because parity |
3806 | * cannot be trusted and we are currently recovering it, there | |
3807 | * is extra need to be careful. | |
ea664c82 N |
3808 | * If one of the devices that we would need to read, because |
3809 | * it is not being overwritten (and maybe not written at all) | |
3810 | * is missing/faulty, then we need to read everything we can. | |
3811 | */ | |
45a4d8fd | 3812 | if (!force_rcw && |
ea664c82 N |
3813 | sh->sector < sh->raid_conf->mddev->recovery_cp) |
3814 | /* reconstruct-write isn't being forced */ | |
3815 | return 0; | |
36707bb2 | 3816 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3817 | if (s->failed_num[i] != sh->pd_idx && |
3818 | s->failed_num[i] != sh->qd_idx && | |
3819 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3820 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3821 | return 1; | |
3822 | } | |
3823 | ||
2c58f06e N |
3824 | return 0; |
3825 | } | |
3826 | ||
ba02684d SL |
3827 | /* fetch_block - checks the given member device to see if its data needs |
3828 | * to be read or computed to satisfy a request. | |
3829 | * | |
3830 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3831 | * 0 to tell the loop in handle_stripe_fill to continue | |
3832 | */ | |
2c58f06e N |
3833 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3834 | int disk_idx, int disks) | |
3835 | { | |
3836 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3837 | ||
3838 | /* is the data in this block needed, and can we get it? */ | |
3839 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3840 | /* we would like to get this block, possibly by computing it, |
3841 | * otherwise read it if the backing disk is insync | |
3842 | */ | |
3843 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3844 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3845 | BUG_ON(sh->batch_head); |
7471fb77 N |
3846 | |
3847 | /* | |
3848 | * In the raid6 case if the only non-uptodate disk is P | |
3849 | * then we already trusted P to compute the other failed | |
3850 | * drives. It is safe to compute rather than re-read P. | |
3851 | * In other cases we only compute blocks from failed | |
3852 | * devices, otherwise check/repair might fail to detect | |
3853 | * a real inconsistency. | |
3854 | */ | |
3855 | ||
5599becc | 3856 | if ((s->uptodate == disks - 1) && |
7471fb77 | 3857 | ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) || |
f2b3b44d | 3858 | (s->failed && (disk_idx == s->failed_num[0] || |
7471fb77 | 3859 | disk_idx == s->failed_num[1])))) { |
5599becc YT |
3860 | /* have disk failed, and we're requested to fetch it; |
3861 | * do compute it | |
a4456856 | 3862 | */ |
5599becc YT |
3863 | pr_debug("Computing stripe %llu block %d\n", |
3864 | (unsigned long long)sh->sector, disk_idx); | |
3865 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3866 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3867 | set_bit(R5_Wantcompute, &dev->flags); | |
3868 | sh->ops.target = disk_idx; | |
3869 | sh->ops.target2 = -1; /* no 2nd target */ | |
3870 | s->req_compute = 1; | |
93b3dbce N |
3871 | /* Careful: from this point on 'uptodate' is in the eye |
3872 | * of raid_run_ops which services 'compute' operations | |
3873 | * before writes. R5_Wantcompute flags a block that will | |
3874 | * be R5_UPTODATE by the time it is needed for a | |
3875 | * subsequent operation. | |
3876 | */ | |
5599becc YT |
3877 | s->uptodate++; |
3878 | return 1; | |
3879 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3880 | /* Computing 2-failure is *very* expensive; only | |
3881 | * do it if failed >= 2 | |
3882 | */ | |
3883 | int other; | |
3884 | for (other = disks; other--; ) { | |
3885 | if (other == disk_idx) | |
3886 | continue; | |
3887 | if (!test_bit(R5_UPTODATE, | |
3888 | &sh->dev[other].flags)) | |
3889 | break; | |
a4456856 | 3890 | } |
5599becc YT |
3891 | BUG_ON(other < 0); |
3892 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3893 | (unsigned long long)sh->sector, | |
3894 | disk_idx, other); | |
3895 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3896 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3897 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3898 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3899 | sh->ops.target = disk_idx; | |
3900 | sh->ops.target2 = other; | |
3901 | s->uptodate += 2; | |
3902 | s->req_compute = 1; | |
3903 | return 1; | |
3904 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3905 | set_bit(R5_LOCKED, &dev->flags); | |
3906 | set_bit(R5_Wantread, &dev->flags); | |
3907 | s->locked++; | |
3908 | pr_debug("Reading block %d (sync=%d)\n", | |
3909 | disk_idx, s->syncing); | |
a4456856 DW |
3910 | } |
3911 | } | |
5599becc YT |
3912 | |
3913 | return 0; | |
3914 | } | |
3915 | ||
2aada5b1 | 3916 | /* |
93b3dbce | 3917 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3918 | */ |
93b3dbce N |
3919 | static void handle_stripe_fill(struct stripe_head *sh, |
3920 | struct stripe_head_state *s, | |
3921 | int disks) | |
5599becc YT |
3922 | { |
3923 | int i; | |
3924 | ||
3925 | /* look for blocks to read/compute, skip this if a compute | |
3926 | * is already in flight, or if the stripe contents are in the | |
3927 | * midst of changing due to a write | |
3928 | */ | |
3929 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3930 | !sh->reconstruct_state) { |
3931 | ||
3932 | /* | |
3933 | * For degraded stripe with data in journal, do not handle | |
3934 | * read requests yet, instead, flush the stripe to raid | |
3935 | * disks first, this avoids handling complex rmw of write | |
3936 | * back cache (prexor with orig_page, and then xor with | |
3937 | * page) in the read path | |
3938 | */ | |
3939 | if (s->injournal && s->failed) { | |
3940 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3941 | r5c_make_stripe_write_out(sh); | |
3942 | goto out; | |
3943 | } | |
3944 | ||
5599becc | 3945 | for (i = disks; i--; ) |
93b3dbce | 3946 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3947 | break; |
07e83364 SL |
3948 | } |
3949 | out: | |
a4456856 DW |
3950 | set_bit(STRIPE_HANDLE, &sh->state); |
3951 | } | |
3952 | ||
787b76fa N |
3953 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3954 | unsigned long handle_flags); | |
1fe797e6 | 3955 | /* handle_stripe_clean_event |
a4456856 DW |
3956 | * any written block on an uptodate or failed drive can be returned. |
3957 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3958 | * never LOCKED, so we don't need to test 'failed' directly. | |
3959 | */ | |
d1688a6d | 3960 | static void handle_stripe_clean_event(struct r5conf *conf, |
bd83d0a2 | 3961 | struct stripe_head *sh, int disks) |
a4456856 DW |
3962 | { |
3963 | int i; | |
3964 | struct r5dev *dev; | |
f8dfcffd | 3965 | int discard_pending = 0; |
59fc630b | 3966 | struct stripe_head *head_sh = sh; |
3967 | bool do_endio = false; | |
a4456856 DW |
3968 | |
3969 | for (i = disks; i--; ) | |
3970 | if (sh->dev[i].written) { | |
3971 | dev = &sh->dev[i]; | |
3972 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3973 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3974 | test_bit(R5_Discard, &dev->flags) || |
3975 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3976 | /* We can return any write requests */ |
3977 | struct bio *wbi, *wbi2; | |
45b4233c | 3978 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3979 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3980 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3981 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3982 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3983 | } |
59fc630b | 3984 | do_endio = true; |
3985 | ||
3986 | returnbi: | |
3987 | dev->page = dev->orig_page; | |
a4456856 DW |
3988 | wbi = dev->written; |
3989 | dev->written = NULL; | |
4f024f37 | 3990 | while (wbi && wbi->bi_iter.bi_sector < |
c911c46c YY |
3991 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
3992 | wbi2 = r5_next_bio(conf, wbi, dev->sector); | |
49728050 | 3993 | md_write_end(conf->mddev); |
016c76ac | 3994 | bio_endio(wbi); |
a4456856 DW |
3995 | wbi = wbi2; |
3996 | } | |
e64e4018 | 3997 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3998 | RAID5_STRIPE_SECTORS(conf), |
e64e4018 AS |
3999 | !test_bit(STRIPE_DEGRADED, &sh->state), |
4000 | 0); | |
59fc630b | 4001 | if (head_sh->batch_head) { |
4002 | sh = list_first_entry(&sh->batch_list, | |
4003 | struct stripe_head, | |
4004 | batch_list); | |
4005 | if (sh != head_sh) { | |
4006 | dev = &sh->dev[i]; | |
4007 | goto returnbi; | |
4008 | } | |
4009 | } | |
4010 | sh = head_sh; | |
4011 | dev = &sh->dev[i]; | |
f8dfcffd N |
4012 | } else if (test_bit(R5_Discard, &dev->flags)) |
4013 | discard_pending = 1; | |
4014 | } | |
f6bed0ef | 4015 | |
ff875738 | 4016 | log_stripe_write_finished(sh); |
0576b1c6 | 4017 | |
f8dfcffd N |
4018 | if (!discard_pending && |
4019 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 4020 | int hash; |
f8dfcffd N |
4021 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
4022 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
4023 | if (sh->qd_idx >= 0) { | |
4024 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
4025 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
4026 | } | |
4027 | /* now that discard is done we can proceed with any sync */ | |
4028 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
4029 | /* |
4030 | * SCSI discard will change some bio fields and the stripe has | |
4031 | * no updated data, so remove it from hash list and the stripe | |
4032 | * will be reinitialized | |
4033 | */ | |
59fc630b | 4034 | unhash: |
b8a9d66d RG |
4035 | hash = sh->hash_lock_index; |
4036 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 4037 | remove_hash(sh); |
b8a9d66d | 4038 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 4039 | if (head_sh->batch_head) { |
4040 | sh = list_first_entry(&sh->batch_list, | |
4041 | struct stripe_head, batch_list); | |
4042 | if (sh != head_sh) | |
4043 | goto unhash; | |
4044 | } | |
59fc630b | 4045 | sh = head_sh; |
4046 | ||
f8dfcffd N |
4047 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
4048 | set_bit(STRIPE_HANDLE, &sh->state); | |
4049 | ||
4050 | } | |
8b3e6cdc DW |
4051 | |
4052 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
4053 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
4054 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 4055 | |
787b76fa N |
4056 | if (head_sh->batch_head && do_endio) |
4057 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
4058 | } |
4059 | ||
86aa1397 SL |
4060 | /* |
4061 | * For RMW in write back cache, we need extra page in prexor to store the | |
4062 | * old data. This page is stored in dev->orig_page. | |
4063 | * | |
4064 | * This function checks whether we have data for prexor. The exact logic | |
4065 | * is: | |
4066 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
4067 | */ | |
4068 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
4069 | { | |
4070 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
4071 | (!test_bit(R5_InJournal, &dev->flags) || | |
4072 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
4073 | } | |
4074 | ||
d7bd398e SL |
4075 | static int handle_stripe_dirtying(struct r5conf *conf, |
4076 | struct stripe_head *sh, | |
4077 | struct stripe_head_state *s, | |
4078 | int disks) | |
a4456856 DW |
4079 | { |
4080 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
4081 | sector_t recovery_cp = conf->mddev->recovery_cp; |
4082 | ||
584acdd4 | 4083 | /* Check whether resync is now happening or should start. |
a7854487 AL |
4084 | * If yes, then the array is dirty (after unclean shutdown or |
4085 | * initial creation), so parity in some stripes might be inconsistent. | |
4086 | * In this case, we need to always do reconstruct-write, to ensure | |
4087 | * that in case of drive failure or read-error correction, we | |
4088 | * generate correct data from the parity. | |
4089 | */ | |
584acdd4 | 4090 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
4091 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
4092 | s->failed == 0)) { | |
a7854487 | 4093 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
4094 | * look like rcw is cheaper |
4095 | */ | |
4096 | rcw = 1; rmw = 2; | |
584acdd4 MS |
4097 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
4098 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 4099 | (unsigned long long)sh->sector); |
c8ac1803 | 4100 | } else for (i = disks; i--; ) { |
a4456856 DW |
4101 | /* would I have to read this buffer for read_modify_write */ |
4102 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 4103 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 4104 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 4105 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 4106 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 4107 | !(uptodate_for_rmw(dev) || |
f38e1219 | 4108 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
4109 | if (test_bit(R5_Insync, &dev->flags)) |
4110 | rmw++; | |
4111 | else | |
4112 | rmw += 2*disks; /* cannot read it */ | |
4113 | } | |
4114 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
4115 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
4116 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 4117 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 4118 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 4119 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
4120 | if (test_bit(R5_Insync, &dev->flags)) |
4121 | rcw++; | |
a4456856 DW |
4122 | else |
4123 | rcw += 2*disks; | |
4124 | } | |
4125 | } | |
1e6d690b | 4126 | |
39b99586 SL |
4127 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
4128 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 4129 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 4130 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 4131 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
4132 | if (conf->mddev->queue) |
4133 | blk_add_trace_msg(conf->mddev->queue, | |
4134 | "raid5 rmw %llu %d", | |
4135 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
4136 | for (i = disks; i--; ) { |
4137 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
4138 | if (test_bit(R5_InJournal, &dev->flags) && |
4139 | dev->page == dev->orig_page && | |
4140 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
4141 | /* alloc page for prexor */ | |
d7bd398e SL |
4142 | struct page *p = alloc_page(GFP_NOIO); |
4143 | ||
4144 | if (p) { | |
4145 | dev->orig_page = p; | |
4146 | continue; | |
4147 | } | |
4148 | ||
4149 | /* | |
4150 | * alloc_page() failed, try use | |
4151 | * disk_info->extra_page | |
4152 | */ | |
4153 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
4154 | &conf->cache_state)) { | |
4155 | r5c_use_extra_page(sh); | |
4156 | break; | |
4157 | } | |
1e6d690b | 4158 | |
d7bd398e SL |
4159 | /* extra_page in use, add to delayed_list */ |
4160 | set_bit(STRIPE_DELAYED, &sh->state); | |
4161 | s->waiting_extra_page = 1; | |
4162 | return -EAGAIN; | |
1e6d690b | 4163 | } |
d7bd398e | 4164 | } |
1e6d690b | 4165 | |
d7bd398e SL |
4166 | for (i = disks; i--; ) { |
4167 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 4168 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
4169 | i == sh->pd_idx || i == sh->qd_idx || |
4170 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 4171 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 4172 | !(uptodate_for_rmw(dev) || |
1e6d690b | 4173 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 4174 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
4175 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
4176 | &sh->state)) { | |
4177 | pr_debug("Read_old block %d for r-m-w\n", | |
4178 | i); | |
a4456856 DW |
4179 | set_bit(R5_LOCKED, &dev->flags); |
4180 | set_bit(R5_Wantread, &dev->flags); | |
4181 | s->locked++; | |
e3914d59 | 4182 | } else |
a4456856 | 4183 | set_bit(STRIPE_DELAYED, &sh->state); |
a4456856 DW |
4184 | } |
4185 | } | |
a9add5d9 | 4186 | } |
41257580 | 4187 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 4188 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 4189 | int qread =0; |
c8ac1803 | 4190 | rcw = 0; |
a4456856 DW |
4191 | for (i = disks; i--; ) { |
4192 | struct r5dev *dev = &sh->dev[i]; | |
4193 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 4194 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 4195 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 4196 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
4197 | test_bit(R5_Wantcompute, &dev->flags))) { |
4198 | rcw++; | |
67f45548 N |
4199 | if (test_bit(R5_Insync, &dev->flags) && |
4200 | test_bit(STRIPE_PREREAD_ACTIVE, | |
4201 | &sh->state)) { | |
45b4233c | 4202 | pr_debug("Read_old block " |
a4456856 DW |
4203 | "%d for Reconstruct\n", i); |
4204 | set_bit(R5_LOCKED, &dev->flags); | |
4205 | set_bit(R5_Wantread, &dev->flags); | |
4206 | s->locked++; | |
a9add5d9 | 4207 | qread++; |
e3914d59 | 4208 | } else |
a4456856 | 4209 | set_bit(STRIPE_DELAYED, &sh->state); |
a4456856 DW |
4210 | } |
4211 | } | |
e3620a3a | 4212 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
4213 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
4214 | (unsigned long long)sh->sector, | |
4215 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 4216 | } |
b1b02fe9 N |
4217 | |
4218 | if (rcw > disks && rmw > disks && | |
4219 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4220 | set_bit(STRIPE_DELAYED, &sh->state); | |
4221 | ||
a4456856 DW |
4222 | /* now if nothing is locked, and if we have enough data, |
4223 | * we can start a write request | |
4224 | */ | |
f38e1219 DW |
4225 | /* since handle_stripe can be called at any time we need to handle the |
4226 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
4227 | * subsequent call wants to start a write request. raid_run_ops only |
4228 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
4229 | * simultaneously. If this is not the case then new writes need to be |
4230 | * held off until the compute completes. | |
4231 | */ | |
976ea8d4 DW |
4232 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
4233 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 4234 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 4235 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 4236 | return 0; |
a4456856 DW |
4237 | } |
4238 | ||
d1688a6d | 4239 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
4240 | struct stripe_head_state *s, int disks) |
4241 | { | |
ecc65c9b | 4242 | struct r5dev *dev = NULL; |
bd2ab670 | 4243 | |
59fc630b | 4244 | BUG_ON(sh->batch_head); |
a4456856 | 4245 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 4246 | |
ecc65c9b DW |
4247 | switch (sh->check_state) { |
4248 | case check_state_idle: | |
4249 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 4250 | if (s->failed == 0) { |
bd2ab670 | 4251 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
4252 | sh->check_state = check_state_run; |
4253 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 4254 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 4255 | s->uptodate--; |
ecc65c9b | 4256 | break; |
bd2ab670 | 4257 | } |
f2b3b44d | 4258 | dev = &sh->dev[s->failed_num[0]]; |
df561f66 | 4259 | fallthrough; |
ecc65c9b DW |
4260 | case check_state_compute_result: |
4261 | sh->check_state = check_state_idle; | |
4262 | if (!dev) | |
4263 | dev = &sh->dev[sh->pd_idx]; | |
4264 | ||
4265 | /* check that a write has not made the stripe insync */ | |
4266 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4267 | break; | |
c8894419 | 4268 | |
a4456856 | 4269 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
4270 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
4271 | BUG_ON(s->uptodate != disks); | |
4272 | ||
4273 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 4274 | s->locked++; |
a4456856 | 4275 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 4276 | |
a4456856 | 4277 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 4278 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
4279 | break; |
4280 | case check_state_run: | |
4281 | break; /* we will be called again upon completion */ | |
4282 | case check_state_check_result: | |
4283 | sh->check_state = check_state_idle; | |
4284 | ||
4285 | /* if a failure occurred during the check operation, leave | |
4286 | * STRIPE_INSYNC not set and let the stripe be handled again | |
4287 | */ | |
4288 | if (s->failed) | |
4289 | break; | |
4290 | ||
4291 | /* handle a successful check operation, if parity is correct | |
4292 | * we are done. Otherwise update the mismatch count and repair | |
4293 | * parity if !MD_RECOVERY_CHECK | |
4294 | */ | |
ad283ea4 | 4295 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
4296 | /* parity is correct (on disc, |
4297 | * not in buffer any more) | |
4298 | */ | |
4299 | set_bit(STRIPE_INSYNC, &sh->state); | |
4300 | else { | |
c911c46c | 4301 | atomic64_add(RAID5_STRIPE_SECTORS(conf), &conf->mddev->resync_mismatches); |
e1539036 | 4302 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
ecc65c9b DW |
4303 | /* don't try to repair!! */ |
4304 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4305 | pr_warn_ratelimited("%s: mismatch sector in range " |
4306 | "%llu-%llu\n", mdname(conf->mddev), | |
4307 | (unsigned long long) sh->sector, | |
4308 | (unsigned long long) sh->sector + | |
c911c46c | 4309 | RAID5_STRIPE_SECTORS(conf)); |
e1539036 | 4310 | } else { |
ecc65c9b | 4311 | sh->check_state = check_state_compute_run; |
976ea8d4 | 4312 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
4313 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
4314 | set_bit(R5_Wantcompute, | |
4315 | &sh->dev[sh->pd_idx].flags); | |
4316 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 4317 | sh->ops.target2 = -1; |
ecc65c9b DW |
4318 | s->uptodate++; |
4319 | } | |
4320 | } | |
4321 | break; | |
4322 | case check_state_compute_run: | |
4323 | break; | |
4324 | default: | |
cc6167b4 | 4325 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
4326 | __func__, sh->check_state, |
4327 | (unsigned long long) sh->sector); | |
4328 | BUG(); | |
a4456856 DW |
4329 | } |
4330 | } | |
4331 | ||
d1688a6d | 4332 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4333 | struct stripe_head_state *s, |
f2b3b44d | 4334 | int disks) |
a4456856 | 4335 | { |
a4456856 | 4336 | int pd_idx = sh->pd_idx; |
34e04e87 | 4337 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4338 | struct r5dev *dev; |
a4456856 | 4339 | |
59fc630b | 4340 | BUG_ON(sh->batch_head); |
a4456856 DW |
4341 | set_bit(STRIPE_HANDLE, &sh->state); |
4342 | ||
4343 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4344 | |
a4456856 DW |
4345 | /* Want to check and possibly repair P and Q. |
4346 | * However there could be one 'failed' device, in which | |
4347 | * case we can only check one of them, possibly using the | |
4348 | * other to generate missing data | |
4349 | */ | |
4350 | ||
d82dfee0 DW |
4351 | switch (sh->check_state) { |
4352 | case check_state_idle: | |
4353 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4354 | if (s->failed == s->q_failed) { |
d82dfee0 | 4355 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4356 | * makes sense to check P (If anything else were failed, |
4357 | * we would have used P to recreate it). | |
4358 | */ | |
d82dfee0 | 4359 | sh->check_state = check_state_run; |
a4456856 | 4360 | } |
f2b3b44d | 4361 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4362 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4363 | * anything, so it makes sense to check it |
4364 | */ | |
d82dfee0 DW |
4365 | if (sh->check_state == check_state_run) |
4366 | sh->check_state = check_state_run_pq; | |
4367 | else | |
4368 | sh->check_state = check_state_run_q; | |
a4456856 | 4369 | } |
a4456856 | 4370 | |
d82dfee0 DW |
4371 | /* discard potentially stale zero_sum_result */ |
4372 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4373 | |
d82dfee0 DW |
4374 | if (sh->check_state == check_state_run) { |
4375 | /* async_xor_zero_sum destroys the contents of P */ | |
4376 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4377 | s->uptodate--; | |
a4456856 | 4378 | } |
d82dfee0 DW |
4379 | if (sh->check_state >= check_state_run && |
4380 | sh->check_state <= check_state_run_pq) { | |
4381 | /* async_syndrome_zero_sum preserves P and Q, so | |
4382 | * no need to mark them !uptodate here | |
4383 | */ | |
4384 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4385 | break; | |
a4456856 DW |
4386 | } |
4387 | ||
d82dfee0 DW |
4388 | /* we have 2-disk failure */ |
4389 | BUG_ON(s->failed != 2); | |
df561f66 | 4390 | fallthrough; |
d82dfee0 DW |
4391 | case check_state_compute_result: |
4392 | sh->check_state = check_state_idle; | |
a4456856 | 4393 | |
d82dfee0 DW |
4394 | /* check that a write has not made the stripe insync */ |
4395 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4396 | break; | |
a4456856 DW |
4397 | |
4398 | /* now write out any block on a failed drive, | |
d82dfee0 | 4399 | * or P or Q if they were recomputed |
a4456856 | 4400 | */ |
b2176a1d | 4401 | dev = NULL; |
a4456856 | 4402 | if (s->failed == 2) { |
f2b3b44d | 4403 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4404 | s->locked++; |
4405 | set_bit(R5_LOCKED, &dev->flags); | |
4406 | set_bit(R5_Wantwrite, &dev->flags); | |
4407 | } | |
4408 | if (s->failed >= 1) { | |
f2b3b44d | 4409 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4410 | s->locked++; |
4411 | set_bit(R5_LOCKED, &dev->flags); | |
4412 | set_bit(R5_Wantwrite, &dev->flags); | |
4413 | } | |
d82dfee0 | 4414 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4415 | dev = &sh->dev[pd_idx]; |
4416 | s->locked++; | |
4417 | set_bit(R5_LOCKED, &dev->flags); | |
4418 | set_bit(R5_Wantwrite, &dev->flags); | |
4419 | } | |
d82dfee0 | 4420 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4421 | dev = &sh->dev[qd_idx]; |
4422 | s->locked++; | |
4423 | set_bit(R5_LOCKED, &dev->flags); | |
4424 | set_bit(R5_Wantwrite, &dev->flags); | |
4425 | } | |
b2176a1d NC |
4426 | if (WARN_ONCE(dev && !test_bit(R5_UPTODATE, &dev->flags), |
4427 | "%s: disk%td not up to date\n", | |
4428 | mdname(conf->mddev), | |
4429 | dev - (struct r5dev *) &sh->dev)) { | |
4430 | clear_bit(R5_LOCKED, &dev->flags); | |
4431 | clear_bit(R5_Wantwrite, &dev->flags); | |
4432 | s->locked--; | |
4433 | } | |
a4456856 DW |
4434 | clear_bit(STRIPE_DEGRADED, &sh->state); |
4435 | ||
4436 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4437 | break; |
4438 | case check_state_run: | |
4439 | case check_state_run_q: | |
4440 | case check_state_run_pq: | |
4441 | break; /* we will be called again upon completion */ | |
4442 | case check_state_check_result: | |
4443 | sh->check_state = check_state_idle; | |
4444 | ||
4445 | /* handle a successful check operation, if parity is correct | |
4446 | * we are done. Otherwise update the mismatch count and repair | |
4447 | * parity if !MD_RECOVERY_CHECK | |
4448 | */ | |
4449 | if (sh->ops.zero_sum_result == 0) { | |
a25d8c32 SL |
4450 | /* both parities are correct */ |
4451 | if (!s->failed) | |
4452 | set_bit(STRIPE_INSYNC, &sh->state); | |
4453 | else { | |
4454 | /* in contrast to the raid5 case we can validate | |
4455 | * parity, but still have a failure to write | |
4456 | * back | |
4457 | */ | |
4458 | sh->check_state = check_state_compute_result; | |
4459 | /* Returning at this point means that we may go | |
4460 | * off and bring p and/or q uptodate again so | |
4461 | * we make sure to check zero_sum_result again | |
4462 | * to verify if p or q need writeback | |
4463 | */ | |
4464 | } | |
d82dfee0 | 4465 | } else { |
c911c46c | 4466 | atomic64_add(RAID5_STRIPE_SECTORS(conf), &conf->mddev->resync_mismatches); |
e1539036 | 4467 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
d82dfee0 DW |
4468 | /* don't try to repair!! */ |
4469 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4470 | pr_warn_ratelimited("%s: mismatch sector in range " |
4471 | "%llu-%llu\n", mdname(conf->mddev), | |
4472 | (unsigned long long) sh->sector, | |
4473 | (unsigned long long) sh->sector + | |
c911c46c | 4474 | RAID5_STRIPE_SECTORS(conf)); |
e1539036 | 4475 | } else { |
d82dfee0 DW |
4476 | int *target = &sh->ops.target; |
4477 | ||
4478 | sh->ops.target = -1; | |
4479 | sh->ops.target2 = -1; | |
4480 | sh->check_state = check_state_compute_run; | |
4481 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4482 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4483 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4484 | set_bit(R5_Wantcompute, | |
4485 | &sh->dev[pd_idx].flags); | |
4486 | *target = pd_idx; | |
4487 | target = &sh->ops.target2; | |
4488 | s->uptodate++; | |
4489 | } | |
4490 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4491 | set_bit(R5_Wantcompute, | |
4492 | &sh->dev[qd_idx].flags); | |
4493 | *target = qd_idx; | |
4494 | s->uptodate++; | |
4495 | } | |
4496 | } | |
4497 | } | |
4498 | break; | |
4499 | case check_state_compute_run: | |
4500 | break; | |
4501 | default: | |
cc6167b4 N |
4502 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4503 | __func__, sh->check_state, | |
4504 | (unsigned long long) sh->sector); | |
d82dfee0 | 4505 | BUG(); |
a4456856 DW |
4506 | } |
4507 | } | |
4508 | ||
d1688a6d | 4509 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4510 | { |
4511 | int i; | |
4512 | ||
4513 | /* We have read all the blocks in this stripe and now we need to | |
4514 | * copy some of them into a target stripe for expand. | |
4515 | */ | |
f0a50d37 | 4516 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4517 | BUG_ON(sh->batch_head); |
a4456856 DW |
4518 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4519 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4520 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4521 | int dd_idx, j; |
a4456856 | 4522 | struct stripe_head *sh2; |
a08abd8c | 4523 | struct async_submit_ctl submit; |
a4456856 | 4524 | |
6d036f7d | 4525 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4526 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4527 | &dd_idx, NULL); | |
6d036f7d | 4528 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4529 | if (sh2 == NULL) |
4530 | /* so far only the early blocks of this stripe | |
4531 | * have been requested. When later blocks | |
4532 | * get requested, we will try again | |
4533 | */ | |
4534 | continue; | |
4535 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4536 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4537 | /* must have already done this block */ | |
6d036f7d | 4538 | raid5_release_stripe(sh2); |
a4456856 DW |
4539 | continue; |
4540 | } | |
f0a50d37 DW |
4541 | |
4542 | /* place all the copies on one channel */ | |
a08abd8c | 4543 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4544 | tx = async_memcpy(sh2->dev[dd_idx].page, |
7aba13b7 YY |
4545 | sh->dev[i].page, sh2->dev[dd_idx].offset, |
4546 | sh->dev[i].offset, RAID5_STRIPE_SIZE(conf), | |
a08abd8c | 4547 | &submit); |
f0a50d37 | 4548 | |
a4456856 DW |
4549 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4550 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4551 | for (j = 0; j < conf->raid_disks; j++) | |
4552 | if (j != sh2->pd_idx && | |
86c374ba | 4553 | j != sh2->qd_idx && |
a4456856 DW |
4554 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4555 | break; | |
4556 | if (j == conf->raid_disks) { | |
4557 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4558 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4559 | } | |
6d036f7d | 4560 | raid5_release_stripe(sh2); |
f0a50d37 | 4561 | |
a4456856 | 4562 | } |
a2e08551 | 4563 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4564 | async_tx_quiesce(&tx); |
a4456856 | 4565 | } |
1da177e4 LT |
4566 | |
4567 | /* | |
4568 | * handle_stripe - do things to a stripe. | |
4569 | * | |
9a3e1101 N |
4570 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4571 | * state of various bits to see what needs to be done. | |
1da177e4 | 4572 | * Possible results: |
9a3e1101 N |
4573 | * return some read requests which now have data |
4574 | * return some write requests which are safely on storage | |
1da177e4 LT |
4575 | * schedule a read on some buffers |
4576 | * schedule a write of some buffers | |
4577 | * return confirmation of parity correctness | |
4578 | * | |
1da177e4 | 4579 | */ |
a4456856 | 4580 | |
acfe726b | 4581 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4582 | { |
d1688a6d | 4583 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4584 | int disks = sh->disks; |
474af965 N |
4585 | struct r5dev *dev; |
4586 | int i; | |
9a3e1101 | 4587 | int do_recovery = 0; |
1da177e4 | 4588 | |
acfe726b N |
4589 | memset(s, 0, sizeof(*s)); |
4590 | ||
dabc4ec6 | 4591 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4592 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4593 | s->failed_num[0] = -1; |
4594 | s->failed_num[1] = -1; | |
6e74a9cf | 4595 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4596 | |
acfe726b | 4597 | /* Now to look around and see what can be done */ |
1da177e4 | 4598 | rcu_read_lock(); |
16a53ecc | 4599 | for (i=disks; i--; ) { |
3cb03002 | 4600 | struct md_rdev *rdev; |
31c176ec N |
4601 | sector_t first_bad; |
4602 | int bad_sectors; | |
4603 | int is_bad = 0; | |
acfe726b | 4604 | |
16a53ecc | 4605 | dev = &sh->dev[i]; |
1da177e4 | 4606 | |
45b4233c | 4607 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4608 | i, dev->flags, |
4609 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4610 | /* maybe we can reply to a read |
4611 | * | |
4612 | * new wantfill requests are only permitted while | |
4613 | * ops_complete_biofill is guaranteed to be inactive | |
4614 | */ | |
4615 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4616 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4617 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4618 | |
16a53ecc | 4619 | /* now count some things */ |
cc94015a N |
4620 | if (test_bit(R5_LOCKED, &dev->flags)) |
4621 | s->locked++; | |
4622 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4623 | s->uptodate++; | |
2d6e4ecc | 4624 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4625 | s->compute++; |
4626 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4627 | } |
1da177e4 | 4628 | |
acfe726b | 4629 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4630 | s->to_fill++; |
acfe726b | 4631 | else if (dev->toread) |
cc94015a | 4632 | s->to_read++; |
16a53ecc | 4633 | if (dev->towrite) { |
cc94015a | 4634 | s->to_write++; |
16a53ecc | 4635 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4636 | s->non_overwrite++; |
16a53ecc | 4637 | } |
a4456856 | 4638 | if (dev->written) |
cc94015a | 4639 | s->written++; |
14a75d3e N |
4640 | /* Prefer to use the replacement for reads, but only |
4641 | * if it is recovered enough and has no bad blocks. | |
4642 | */ | |
4643 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4644 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
c911c46c YY |
4645 | rdev->recovery_offset >= sh->sector + RAID5_STRIPE_SECTORS(conf) && |
4646 | !is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), | |
14a75d3e N |
4647 | &first_bad, &bad_sectors)) |
4648 | set_bit(R5_ReadRepl, &dev->flags); | |
4649 | else { | |
e6030cb0 | 4650 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4651 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4652 | else |
4653 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4654 | rdev = rcu_dereference(conf->disks[i].rdev); |
4655 | clear_bit(R5_ReadRepl, &dev->flags); | |
4656 | } | |
9283d8c5 N |
4657 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4658 | rdev = NULL; | |
31c176ec | 4659 | if (rdev) { |
c911c46c | 4660 | is_bad = is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), |
31c176ec N |
4661 | &first_bad, &bad_sectors); |
4662 | if (s->blocked_rdev == NULL | |
4663 | && (test_bit(Blocked, &rdev->flags) | |
4664 | || is_bad < 0)) { | |
4665 | if (is_bad < 0) | |
4666 | set_bit(BlockedBadBlocks, | |
4667 | &rdev->flags); | |
4668 | s->blocked_rdev = rdev; | |
4669 | atomic_inc(&rdev->nr_pending); | |
4670 | } | |
6bfe0b49 | 4671 | } |
415e72d0 N |
4672 | clear_bit(R5_Insync, &dev->flags); |
4673 | if (!rdev) | |
4674 | /* Not in-sync */; | |
31c176ec N |
4675 | else if (is_bad) { |
4676 | /* also not in-sync */ | |
18b9837e N |
4677 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4678 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4679 | /* treat as in-sync, but with a read error |
4680 | * which we can now try to correct | |
4681 | */ | |
4682 | set_bit(R5_Insync, &dev->flags); | |
4683 | set_bit(R5_ReadError, &dev->flags); | |
4684 | } | |
4685 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4686 | set_bit(R5_Insync, &dev->flags); |
c911c46c | 4687 | else if (sh->sector + RAID5_STRIPE_SECTORS(conf) <= rdev->recovery_offset) |
415e72d0 | 4688 | /* in sync if before recovery_offset */ |
30d7a483 N |
4689 | set_bit(R5_Insync, &dev->flags); |
4690 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4691 | test_bit(R5_Expanded, &dev->flags)) | |
4692 | /* If we've reshaped into here, we assume it is Insync. | |
4693 | * We will shortly update recovery_offset to make | |
4694 | * it official. | |
4695 | */ | |
4696 | set_bit(R5_Insync, &dev->flags); | |
4697 | ||
1cc03eb9 | 4698 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4699 | /* This flag does not apply to '.replacement' |
4700 | * only to .rdev, so make sure to check that*/ | |
4701 | struct md_rdev *rdev2 = rcu_dereference( | |
4702 | conf->disks[i].rdev); | |
4703 | if (rdev2 == rdev) | |
4704 | clear_bit(R5_Insync, &dev->flags); | |
4705 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4706 | s->handle_bad_blocks = 1; |
14a75d3e | 4707 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4708 | } else |
4709 | clear_bit(R5_WriteError, &dev->flags); | |
4710 | } | |
1cc03eb9 | 4711 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4712 | /* This flag does not apply to '.replacement' |
4713 | * only to .rdev, so make sure to check that*/ | |
4714 | struct md_rdev *rdev2 = rcu_dereference( | |
4715 | conf->disks[i].rdev); | |
4716 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4717 | s->handle_bad_blocks = 1; |
14a75d3e | 4718 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4719 | } else |
4720 | clear_bit(R5_MadeGood, &dev->flags); | |
4721 | } | |
977df362 N |
4722 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4723 | struct md_rdev *rdev2 = rcu_dereference( | |
4724 | conf->disks[i].replacement); | |
4725 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4726 | s->handle_bad_blocks = 1; | |
4727 | atomic_inc(&rdev2->nr_pending); | |
4728 | } else | |
4729 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4730 | } | |
415e72d0 | 4731 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4732 | /* The ReadError flag will just be confusing now */ |
4733 | clear_bit(R5_ReadError, &dev->flags); | |
4734 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4735 | } |
415e72d0 N |
4736 | if (test_bit(R5_ReadError, &dev->flags)) |
4737 | clear_bit(R5_Insync, &dev->flags); | |
4738 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4739 | if (s->failed < 2) |
4740 | s->failed_num[s->failed] = i; | |
4741 | s->failed++; | |
9a3e1101 N |
4742 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4743 | do_recovery = 1; | |
d63e2fc8 BC |
4744 | else if (!rdev) { |
4745 | rdev = rcu_dereference( | |
4746 | conf->disks[i].replacement); | |
4747 | if (rdev && !test_bit(Faulty, &rdev->flags)) | |
4748 | do_recovery = 1; | |
4749 | } | |
415e72d0 | 4750 | } |
2ded3703 SL |
4751 | |
4752 | if (test_bit(R5_InJournal, &dev->flags)) | |
4753 | s->injournal++; | |
1e6d690b SL |
4754 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4755 | s->just_cached++; | |
1da177e4 | 4756 | } |
9a3e1101 N |
4757 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4758 | /* If there is a failed device being replaced, | |
4759 | * we must be recovering. | |
4760 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4761 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4762 | * else we can only be replacing |
4763 | * sync and recovery both need to read all devices, and so | |
4764 | * use the same flag. | |
4765 | */ | |
4766 | if (do_recovery || | |
c6d2e084 | 4767 | sh->sector >= conf->mddev->recovery_cp || |
4768 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4769 | s->syncing = 1; |
4770 | else | |
4771 | s->replacing = 1; | |
4772 | } | |
1da177e4 | 4773 | rcu_read_unlock(); |
cc94015a N |
4774 | } |
4775 | ||
cb9902db GJ |
4776 | /* |
4777 | * Return '1' if this is a member of batch, or '0' if it is a lone stripe or | |
4778 | * a head which can now be handled. | |
4779 | */ | |
59fc630b | 4780 | static int clear_batch_ready(struct stripe_head *sh) |
4781 | { | |
4782 | struct stripe_head *tmp; | |
4783 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4784 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4785 | spin_lock(&sh->stripe_lock); |
4786 | if (!sh->batch_head) { | |
4787 | spin_unlock(&sh->stripe_lock); | |
4788 | return 0; | |
4789 | } | |
4790 | ||
4791 | /* | |
4792 | * this stripe could be added to a batch list before we check | |
4793 | * BATCH_READY, skips it | |
4794 | */ | |
4795 | if (sh->batch_head != sh) { | |
4796 | spin_unlock(&sh->stripe_lock); | |
4797 | return 1; | |
4798 | } | |
4799 | spin_lock(&sh->batch_lock); | |
4800 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4801 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4802 | spin_unlock(&sh->batch_lock); | |
4803 | spin_unlock(&sh->stripe_lock); | |
4804 | ||
4805 | /* | |
4806 | * BATCH_READY is cleared, no new stripes can be added. | |
4807 | * batch_list can be accessed without lock | |
4808 | */ | |
4809 | return 0; | |
4810 | } | |
4811 | ||
3960ce79 N |
4812 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4813 | unsigned long handle_flags) | |
72ac7330 | 4814 | { |
4e3d62ff | 4815 | struct stripe_head *sh, *next; |
72ac7330 | 4816 | int i; |
fb642b92 | 4817 | int do_wakeup = 0; |
72ac7330 | 4818 | |
bb27051f N |
4819 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4820 | ||
72ac7330 | 4821 | list_del_init(&sh->batch_list); |
4822 | ||
fb3229d5 | 4823 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4824 | (1 << STRIPE_SYNCING) | |
4825 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4826 | (1 << STRIPE_DELAYED) | |
4827 | (1 << STRIPE_BIT_DELAY) | | |
4828 | (1 << STRIPE_FULL_WRITE) | | |
4829 | (1 << STRIPE_BIOFILL_RUN) | | |
4830 | (1 << STRIPE_COMPUTE_RUN) | | |
1b956f7a N |
4831 | (1 << STRIPE_DISCARD) | |
4832 | (1 << STRIPE_BATCH_READY) | | |
4833 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4834 | (1 << STRIPE_BITMAP_PENDING)), |
4835 | "stripe state: %lx\n", sh->state); | |
4836 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4837 | (1 << STRIPE_REPLACED)), | |
4838 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4839 | |
4840 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4841 | (1 << STRIPE_PREREAD_ACTIVE) | |
184a09eb DY |
4842 | (1 << STRIPE_DEGRADED) | |
4843 | (1 << STRIPE_ON_UNPLUG_LIST)), | |
1b956f7a N |
4844 | head_sh->state & (1 << STRIPE_INSYNC)); |
4845 | ||
72ac7330 | 4846 | sh->check_state = head_sh->check_state; |
4847 | sh->reconstruct_state = head_sh->reconstruct_state; | |
448ec638 AC |
4848 | spin_lock_irq(&sh->stripe_lock); |
4849 | sh->batch_head = NULL; | |
4850 | spin_unlock_irq(&sh->stripe_lock); | |
fb642b92 N |
4851 | for (i = 0; i < sh->disks; i++) { |
4852 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4853 | do_wakeup = 1; | |
72ac7330 | 4854 | sh->dev[i].flags = head_sh->dev[i].flags & |
4855 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4856 | } |
3960ce79 N |
4857 | if (handle_flags == 0 || |
4858 | sh->state & handle_flags) | |
4859 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4860 | raid5_release_stripe(sh); |
72ac7330 | 4861 | } |
fb642b92 N |
4862 | spin_lock_irq(&head_sh->stripe_lock); |
4863 | head_sh->batch_head = NULL; | |
4864 | spin_unlock_irq(&head_sh->stripe_lock); | |
4865 | for (i = 0; i < head_sh->disks; i++) | |
4866 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4867 | do_wakeup = 1; | |
3960ce79 N |
4868 | if (head_sh->state & handle_flags) |
4869 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4870 | |
4871 | if (do_wakeup) | |
4872 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4873 | } |
4874 | ||
cc94015a N |
4875 | static void handle_stripe(struct stripe_head *sh) |
4876 | { | |
4877 | struct stripe_head_state s; | |
d1688a6d | 4878 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4879 | int i; |
84789554 N |
4880 | int prexor; |
4881 | int disks = sh->disks; | |
474af965 | 4882 | struct r5dev *pdev, *qdev; |
cc94015a N |
4883 | |
4884 | clear_bit(STRIPE_HANDLE, &sh->state); | |
a377a472 GJ |
4885 | |
4886 | /* | |
4887 | * handle_stripe should not continue handle the batched stripe, only | |
4888 | * the head of batch list or lone stripe can continue. Otherwise we | |
4889 | * could see break_stripe_batch_list warns about the STRIPE_ACTIVE | |
4890 | * is set for the batched stripe. | |
4891 | */ | |
4892 | if (clear_batch_ready(sh)) | |
4893 | return; | |
4894 | ||
257a4b42 | 4895 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4896 | /* already being handled, ensure it gets handled |
4897 | * again when current action finishes */ | |
4898 | set_bit(STRIPE_HANDLE, &sh->state); | |
4899 | return; | |
4900 | } | |
4901 | ||
4e3d62ff | 4902 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4903 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4904 | |
dabc4ec6 | 4905 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd | 4906 | spin_lock(&sh->stripe_lock); |
5ddf0440 SL |
4907 | /* |
4908 | * Cannot process 'sync' concurrently with 'discard'. | |
4909 | * Flush data in r5cache before 'sync'. | |
4910 | */ | |
4911 | if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) && | |
4912 | !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) && | |
4913 | !test_bit(STRIPE_DISCARD, &sh->state) && | |
f8dfcffd N |
4914 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
4915 | set_bit(STRIPE_SYNCING, &sh->state); | |
4916 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4917 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4918 | } |
4919 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4920 | } |
4921 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4922 | ||
4923 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4924 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4925 | (unsigned long long)sh->sector, sh->state, | |
4926 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4927 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4928 | |
acfe726b | 4929 | analyse_stripe(sh, &s); |
c5a31000 | 4930 | |
b70abcb2 SL |
4931 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4932 | goto finish; | |
4933 | ||
16d997b7 N |
4934 | if (s.handle_bad_blocks || |
4935 | test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { | |
bc2607f3 N |
4936 | set_bit(STRIPE_HANDLE, &sh->state); |
4937 | goto finish; | |
4938 | } | |
4939 | ||
474af965 N |
4940 | if (unlikely(s.blocked_rdev)) { |
4941 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4942 | s.replacing || s.to_write || s.written) { |
474af965 N |
4943 | set_bit(STRIPE_HANDLE, &sh->state); |
4944 | goto finish; | |
4945 | } | |
4946 | /* There is nothing for the blocked_rdev to block */ | |
4947 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4948 | s.blocked_rdev = NULL; | |
4949 | } | |
4950 | ||
4951 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4952 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4953 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4954 | } | |
4955 | ||
4956 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4957 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4958 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4959 | s.failed_num[0], s.failed_num[1]); | |
70d466f7 SL |
4960 | /* |
4961 | * check if the array has lost more than max_degraded devices and, | |
474af965 | 4962 | * if so, some requests might need to be failed. |
70d466f7 SL |
4963 | * |
4964 | * When journal device failed (log_failed), we will only process | |
4965 | * the stripe if there is data need write to raid disks | |
474af965 | 4966 | */ |
70d466f7 SL |
4967 | if (s.failed > conf->max_degraded || |
4968 | (s.log_failed && s.injournal == 0)) { | |
9a3f530f N |
4969 | sh->check_state = 0; |
4970 | sh->reconstruct_state = 0; | |
626f2092 | 4971 | break_stripe_batch_list(sh, 0); |
9a3f530f | 4972 | if (s.to_read+s.to_write+s.written) |
bd83d0a2 | 4973 | handle_failed_stripe(conf, sh, &s, disks); |
9a3e1101 | 4974 | if (s.syncing + s.replacing) |
9a3f530f N |
4975 | handle_failed_sync(conf, sh, &s); |
4976 | } | |
474af965 | 4977 | |
84789554 N |
4978 | /* Now we check to see if any write operations have recently |
4979 | * completed | |
4980 | */ | |
4981 | prexor = 0; | |
4982 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4983 | prexor = 1; | |
4984 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4985 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4986 | sh->reconstruct_state = reconstruct_state_idle; | |
4987 | ||
4988 | /* All the 'written' buffers and the parity block are ready to | |
4989 | * be written back to disk | |
4990 | */ | |
9e444768 SL |
4991 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4992 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4993 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4994 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4995 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4996 | for (i = disks; i--; ) { |
4997 | struct r5dev *dev = &sh->dev[i]; | |
4998 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4999 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
5000 | dev->written || test_bit(R5_InJournal, |
5001 | &dev->flags))) { | |
84789554 N |
5002 | pr_debug("Writing block %d\n", i); |
5003 | set_bit(R5_Wantwrite, &dev->flags); | |
5004 | if (prexor) | |
5005 | continue; | |
9c4bdf69 N |
5006 | if (s.failed > 1) |
5007 | continue; | |
84789554 N |
5008 | if (!test_bit(R5_Insync, &dev->flags) || |
5009 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
5010 | s.failed == 0)) | |
5011 | set_bit(STRIPE_INSYNC, &sh->state); | |
5012 | } | |
5013 | } | |
5014 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5015 | s.dec_preread_active = 1; | |
5016 | } | |
5017 | ||
ef5b7c69 N |
5018 | /* |
5019 | * might be able to return some write requests if the parity blocks | |
5020 | * are safe, or on a failed drive | |
5021 | */ | |
5022 | pdev = &sh->dev[sh->pd_idx]; | |
5023 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
5024 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
5025 | qdev = &sh->dev[sh->qd_idx]; | |
5026 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
5027 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
5028 | || conf->level < 6; | |
5029 | ||
5030 | if (s.written && | |
5031 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
5032 | && !test_bit(R5_LOCKED, &pdev->flags) | |
5033 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
5034 | test_bit(R5_Discard, &pdev->flags))))) && | |
5035 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
5036 | && !test_bit(R5_LOCKED, &qdev->flags) | |
5037 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
5038 | test_bit(R5_Discard, &qdev->flags)))))) | |
bd83d0a2 | 5039 | handle_stripe_clean_event(conf, sh, disks); |
ef5b7c69 | 5040 | |
1e6d690b | 5041 | if (s.just_cached) |
bd83d0a2 | 5042 | r5c_handle_cached_data_endio(conf, sh, disks); |
ff875738 | 5043 | log_stripe_write_finished(sh); |
1e6d690b | 5044 | |
ef5b7c69 N |
5045 | /* Now we might consider reading some blocks, either to check/generate |
5046 | * parity, or to satisfy requests | |
5047 | * or to load a block that is being partially written. | |
5048 | */ | |
5049 | if (s.to_read || s.non_overwrite | |
a1c6ae3d | 5050 | || (s.to_write && s.failed) |
ef5b7c69 N |
5051 | || (s.syncing && (s.uptodate + s.compute < disks)) |
5052 | || s.replacing | |
5053 | || s.expanding) | |
5054 | handle_stripe_fill(sh, &s, disks); | |
5055 | ||
2ded3703 SL |
5056 | /* |
5057 | * When the stripe finishes full journal write cycle (write to journal | |
5058 | * and raid disk), this is the clean up procedure so it is ready for | |
5059 | * next operation. | |
5060 | */ | |
5061 | r5c_finish_stripe_write_out(conf, sh, &s); | |
5062 | ||
5063 | /* | |
5064 | * Now to consider new write requests, cache write back and what else, | |
5065 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
5066 | * 1/ A 'write' operation (copy+xor) is already in flight. |
5067 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
5068 | * block. | |
2ded3703 | 5069 | * 3/ A r5c cache log write is in flight. |
84789554 | 5070 | */ |
2ded3703 SL |
5071 | |
5072 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
5073 | if (!r5c_is_writeback(conf->log)) { | |
5074 | if (s.to_write) | |
5075 | handle_stripe_dirtying(conf, sh, &s, disks); | |
5076 | } else { /* write back cache */ | |
5077 | int ret = 0; | |
5078 | ||
5079 | /* First, try handle writes in caching phase */ | |
5080 | if (s.to_write) | |
5081 | ret = r5c_try_caching_write(conf, sh, &s, | |
5082 | disks); | |
5083 | /* | |
5084 | * If caching phase failed: ret == -EAGAIN | |
5085 | * OR | |
5086 | * stripe under reclaim: !caching && injournal | |
5087 | * | |
5088 | * fall back to handle_stripe_dirtying() | |
5089 | */ | |
5090 | if (ret == -EAGAIN || | |
5091 | /* stripe under reclaim: !caching && injournal */ | |
5092 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
5093 | s.injournal > 0)) { |
5094 | ret = handle_stripe_dirtying(conf, sh, &s, | |
5095 | disks); | |
5096 | if (ret == -EAGAIN) | |
5097 | goto finish; | |
5098 | } | |
2ded3703 SL |
5099 | } |
5100 | } | |
84789554 N |
5101 | |
5102 | /* maybe we need to check and possibly fix the parity for this stripe | |
5103 | * Any reads will already have been scheduled, so we just see if enough | |
5104 | * data is available. The parity check is held off while parity | |
5105 | * dependent operations are in flight. | |
5106 | */ | |
5107 | if (sh->check_state || | |
5108 | (s.syncing && s.locked == 0 && | |
5109 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
5110 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
5111 | if (conf->level == 6) | |
5112 | handle_parity_checks6(conf, sh, &s, disks); | |
5113 | else | |
5114 | handle_parity_checks5(conf, sh, &s, disks); | |
5115 | } | |
c5a31000 | 5116 | |
f94c0b66 N |
5117 | if ((s.replacing || s.syncing) && s.locked == 0 |
5118 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
5119 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
5120 | /* Write out to replacement devices where possible */ |
5121 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
5122 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
5123 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
5124 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
5125 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
5126 | s.locked++; | |
5127 | } | |
f94c0b66 N |
5128 | if (s.replacing) |
5129 | set_bit(STRIPE_INSYNC, &sh->state); | |
5130 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
5131 | } |
5132 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 5133 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 5134 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c911c46c | 5135 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), 1); |
c5a31000 | 5136 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
5137 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
5138 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
5139 | } |
5140 | ||
5141 | /* If the failed drives are just a ReadError, then we might need | |
5142 | * to progress the repair/check process | |
5143 | */ | |
5144 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
5145 | for (i = 0; i < s.failed; i++) { | |
5146 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
5147 | if (test_bit(R5_ReadError, &dev->flags) | |
5148 | && !test_bit(R5_LOCKED, &dev->flags) | |
5149 | && test_bit(R5_UPTODATE, &dev->flags) | |
5150 | ) { | |
5151 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
5152 | set_bit(R5_Wantwrite, &dev->flags); | |
5153 | set_bit(R5_ReWrite, &dev->flags); | |
3a31cf3d | 5154 | } else |
c5a31000 N |
5155 | /* let's read it back */ |
5156 | set_bit(R5_Wantread, &dev->flags); | |
3a31cf3d GJ |
5157 | set_bit(R5_LOCKED, &dev->flags); |
5158 | s.locked++; | |
c5a31000 N |
5159 | } |
5160 | } | |
5161 | ||
3687c061 N |
5162 | /* Finish reconstruct operations initiated by the expansion process */ |
5163 | if (sh->reconstruct_state == reconstruct_state_result) { | |
5164 | struct stripe_head *sh_src | |
6d036f7d | 5165 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
5166 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
5167 | /* sh cannot be written until sh_src has been read. | |
5168 | * so arrange for sh to be delayed a little | |
5169 | */ | |
5170 | set_bit(STRIPE_DELAYED, &sh->state); | |
5171 | set_bit(STRIPE_HANDLE, &sh->state); | |
5172 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
5173 | &sh_src->state)) | |
5174 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 5175 | raid5_release_stripe(sh_src); |
3687c061 N |
5176 | goto finish; |
5177 | } | |
5178 | if (sh_src) | |
6d036f7d | 5179 | raid5_release_stripe(sh_src); |
3687c061 N |
5180 | |
5181 | sh->reconstruct_state = reconstruct_state_idle; | |
5182 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
5183 | for (i = conf->raid_disks; i--; ) { | |
5184 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
5185 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
5186 | s.locked++; | |
5187 | } | |
5188 | } | |
f416885e | 5189 | |
3687c061 N |
5190 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
5191 | !sh->reconstruct_state) { | |
5192 | /* Need to write out all blocks after computing parity */ | |
5193 | sh->disks = conf->raid_disks; | |
5194 | stripe_set_idx(sh->sector, conf, 0, sh); | |
5195 | schedule_reconstruction(sh, &s, 1, 1); | |
5196 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
5197 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
5198 | atomic_dec(&conf->reshape_stripes); | |
5199 | wake_up(&conf->wait_for_overlap); | |
c911c46c | 5200 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), 1); |
3687c061 N |
5201 | } |
5202 | ||
5203 | if (s.expanding && s.locked == 0 && | |
5204 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
5205 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 5206 | |
3687c061 | 5207 | finish: |
6bfe0b49 | 5208 | /* wait for this device to become unblocked */ |
5f066c63 N |
5209 | if (unlikely(s.blocked_rdev)) { |
5210 | if (conf->mddev->external) | |
5211 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
5212 | conf->mddev); | |
5213 | else | |
5214 | /* Internal metadata will immediately | |
5215 | * be written by raid5d, so we don't | |
5216 | * need to wait here. | |
5217 | */ | |
5218 | rdev_dec_pending(s.blocked_rdev, | |
5219 | conf->mddev); | |
5220 | } | |
6bfe0b49 | 5221 | |
bc2607f3 N |
5222 | if (s.handle_bad_blocks) |
5223 | for (i = disks; i--; ) { | |
3cb03002 | 5224 | struct md_rdev *rdev; |
bc2607f3 N |
5225 | struct r5dev *dev = &sh->dev[i]; |
5226 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
5227 | /* We own a safe reference to the rdev */ | |
5228 | rdev = conf->disks[i].rdev; | |
5229 | if (!rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 5230 | RAID5_STRIPE_SECTORS(conf), 0)) |
bc2607f3 N |
5231 | md_error(conf->mddev, rdev); |
5232 | rdev_dec_pending(rdev, conf->mddev); | |
5233 | } | |
b84db560 N |
5234 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
5235 | rdev = conf->disks[i].rdev; | |
5236 | rdev_clear_badblocks(rdev, sh->sector, | |
c911c46c | 5237 | RAID5_STRIPE_SECTORS(conf), 0); |
b84db560 N |
5238 | rdev_dec_pending(rdev, conf->mddev); |
5239 | } | |
977df362 N |
5240 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
5241 | rdev = conf->disks[i].replacement; | |
dd054fce N |
5242 | if (!rdev) |
5243 | /* rdev have been moved down */ | |
5244 | rdev = conf->disks[i].rdev; | |
977df362 | 5245 | rdev_clear_badblocks(rdev, sh->sector, |
c911c46c | 5246 | RAID5_STRIPE_SECTORS(conf), 0); |
977df362 N |
5247 | rdev_dec_pending(rdev, conf->mddev); |
5248 | } | |
bc2607f3 N |
5249 | } |
5250 | ||
6c0069c0 YT |
5251 | if (s.ops_request) |
5252 | raid_run_ops(sh, s.ops_request); | |
5253 | ||
f0e43bcd | 5254 | ops_run_io(sh, &s); |
16a53ecc | 5255 | |
c5709ef6 | 5256 | if (s.dec_preread_active) { |
729a1866 | 5257 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 5258 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
5259 | * have actually been submitted. |
5260 | */ | |
5261 | atomic_dec(&conf->preread_active_stripes); | |
5262 | if (atomic_read(&conf->preread_active_stripes) < | |
5263 | IO_THRESHOLD) | |
5264 | md_wakeup_thread(conf->mddev->thread); | |
5265 | } | |
5266 | ||
257a4b42 | 5267 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
5268 | } |
5269 | ||
d1688a6d | 5270 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
5271 | { |
5272 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
5273 | while (!list_empty(&conf->delayed_list)) { | |
5274 | struct list_head *l = conf->delayed_list.next; | |
5275 | struct stripe_head *sh; | |
5276 | sh = list_entry(l, struct stripe_head, lru); | |
5277 | list_del_init(l); | |
5278 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5279 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5280 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 5281 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 5282 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 5283 | } |
482c0834 | 5284 | } |
16a53ecc N |
5285 | } |
5286 | ||
566c09c5 SL |
5287 | static void activate_bit_delay(struct r5conf *conf, |
5288 | struct list_head *temp_inactive_list) | |
16a53ecc N |
5289 | { |
5290 | /* device_lock is held */ | |
5291 | struct list_head head; | |
5292 | list_add(&head, &conf->bitmap_list); | |
5293 | list_del_init(&conf->bitmap_list); | |
5294 | while (!list_empty(&head)) { | |
5295 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 5296 | int hash; |
16a53ecc N |
5297 | list_del_init(&sh->lru); |
5298 | atomic_inc(&sh->count); | |
566c09c5 SL |
5299 | hash = sh->hash_lock_index; |
5300 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
5301 | } |
5302 | } | |
5303 | ||
fd01b88c | 5304 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 5305 | { |
3cb5edf4 | 5306 | struct r5conf *conf = mddev->private; |
10433d04 | 5307 | sector_t sector = bio->bi_iter.bi_sector; |
3cb5edf4 | 5308 | unsigned int chunk_sectors; |
aa8b57aa | 5309 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 5310 | |
3cb5edf4 | 5311 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
5312 | return chunk_sectors >= |
5313 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
5314 | } | |
5315 | ||
46031f9a RBJ |
5316 | /* |
5317 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
5318 | * later sampled by raid5d. | |
5319 | */ | |
d1688a6d | 5320 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
5321 | { |
5322 | unsigned long flags; | |
5323 | ||
5324 | spin_lock_irqsave(&conf->device_lock, flags); | |
5325 | ||
5326 | bi->bi_next = conf->retry_read_aligned_list; | |
5327 | conf->retry_read_aligned_list = bi; | |
5328 | ||
5329 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5330 | md_wakeup_thread(conf->mddev->thread); | |
5331 | } | |
5332 | ||
0472a42b N |
5333 | static struct bio *remove_bio_from_retry(struct r5conf *conf, |
5334 | unsigned int *offset) | |
46031f9a RBJ |
5335 | { |
5336 | struct bio *bi; | |
5337 | ||
5338 | bi = conf->retry_read_aligned; | |
5339 | if (bi) { | |
0472a42b | 5340 | *offset = conf->retry_read_offset; |
46031f9a RBJ |
5341 | conf->retry_read_aligned = NULL; |
5342 | return bi; | |
5343 | } | |
5344 | bi = conf->retry_read_aligned_list; | |
5345 | if(bi) { | |
387bb173 | 5346 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5347 | bi->bi_next = NULL; |
0472a42b | 5348 | *offset = 0; |
46031f9a RBJ |
5349 | } |
5350 | ||
5351 | return bi; | |
5352 | } | |
5353 | ||
f679623f RBJ |
5354 | /* |
5355 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5356 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5357 | * first). | |
5358 | * If the read failed.. | |
5359 | */ | |
4246a0b6 | 5360 | static void raid5_align_endio(struct bio *bi) |
f679623f | 5361 | { |
1147f58e GJ |
5362 | struct md_io_acct *md_io_acct = bi->bi_private; |
5363 | struct bio *raid_bi = md_io_acct->orig_bio; | |
fd01b88c | 5364 | struct mddev *mddev; |
d1688a6d | 5365 | struct r5conf *conf; |
3cb03002 | 5366 | struct md_rdev *rdev; |
4e4cbee9 | 5367 | blk_status_t error = bi->bi_status; |
1147f58e | 5368 | unsigned long start_time = md_io_acct->start_time; |
46031f9a | 5369 | |
f679623f | 5370 | bio_put(bi); |
46031f9a | 5371 | |
46031f9a RBJ |
5372 | rdev = (void*)raid_bi->bi_next; |
5373 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5374 | mddev = rdev->mddev; |
5375 | conf = mddev->private; | |
46031f9a RBJ |
5376 | |
5377 | rdev_dec_pending(rdev, conf->mddev); | |
5378 | ||
9b81c842 | 5379 | if (!error) { |
1147f58e GJ |
5380 | if (blk_queue_io_stat(raid_bi->bi_bdev->bd_disk->queue)) |
5381 | bio_end_io_acct(raid_bi, start_time); | |
4246a0b6 | 5382 | bio_endio(raid_bi); |
46031f9a | 5383 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5384 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5385 | return; |
46031f9a RBJ |
5386 | } |
5387 | ||
45b4233c | 5388 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5389 | |
5390 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5391 | } |
5392 | ||
7ef6b12a | 5393 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5394 | { |
d1688a6d | 5395 | struct r5conf *conf = mddev->private; |
e82ed3a4 | 5396 | struct bio *align_bio; |
3cb03002 | 5397 | struct md_rdev *rdev; |
e82ed3a4 CH |
5398 | sector_t sector, end_sector, first_bad; |
5399 | int bad_sectors, dd_idx; | |
1147f58e | 5400 | struct md_io_acct *md_io_acct; |
97ae2725 | 5401 | bool did_inc; |
f679623f RBJ |
5402 | |
5403 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5404 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5405 | return 0; |
5406 | } | |
f679623f | 5407 | |
e82ed3a4 CH |
5408 | sector = raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, 0, |
5409 | &dd_idx, NULL); | |
5410 | end_sector = bio_end_sector(raid_bio); | |
5411 | ||
f679623f | 5412 | rcu_read_lock(); |
e82ed3a4 CH |
5413 | if (r5c_big_stripe_cached(conf, sector)) |
5414 | goto out_rcu_unlock; | |
5415 | ||
671488cc N |
5416 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5417 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5418 | rdev->recovery_offset < end_sector) { | |
5419 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
e82ed3a4 CH |
5420 | if (!rdev) |
5421 | goto out_rcu_unlock; | |
5422 | if (test_bit(Faulty, &rdev->flags) || | |
671488cc | 5423 | !(test_bit(In_sync, &rdev->flags) || |
e82ed3a4 CH |
5424 | rdev->recovery_offset >= end_sector)) |
5425 | goto out_rcu_unlock; | |
671488cc | 5426 | } |
03b047f4 | 5427 | |
e82ed3a4 CH |
5428 | atomic_inc(&rdev->nr_pending); |
5429 | rcu_read_unlock(); | |
5430 | ||
c82aa1b7 | 5431 | if (is_badblock(rdev, sector, bio_sectors(raid_bio), &first_bad, |
e82ed3a4 | 5432 | &bad_sectors)) { |
c82aa1b7 | 5433 | bio_put(raid_bio); |
e82ed3a4 | 5434 | rdev_dec_pending(rdev, mddev); |
03b047f4 SL |
5435 | return 0; |
5436 | } | |
5437 | ||
1147f58e GJ |
5438 | align_bio = bio_clone_fast(raid_bio, GFP_NOIO, &mddev->io_acct_set); |
5439 | md_io_acct = container_of(align_bio, struct md_io_acct, bio_clone); | |
5440 | raid_bio->bi_next = (void *)rdev; | |
5441 | if (blk_queue_io_stat(raid_bio->bi_bdev->bd_disk->queue)) | |
5442 | md_io_acct->start_time = bio_start_io_acct(raid_bio); | |
5443 | md_io_acct->orig_bio = raid_bio; | |
5444 | ||
e82ed3a4 CH |
5445 | bio_set_dev(align_bio, rdev->bdev); |
5446 | align_bio->bi_end_io = raid5_align_endio; | |
1147f58e | 5447 | align_bio->bi_private = md_io_acct; |
e82ed3a4 CH |
5448 | align_bio->bi_iter.bi_sector = sector; |
5449 | ||
e82ed3a4 CH |
5450 | /* No reshape active, so we can trust rdev->data_offset */ |
5451 | align_bio->bi_iter.bi_sector += rdev->data_offset; | |
31c176ec | 5452 | |
97ae2725 GO |
5453 | did_inc = false; |
5454 | if (conf->quiesce == 0) { | |
5455 | atomic_inc(&conf->active_aligned_reads); | |
5456 | did_inc = true; | |
5457 | } | |
5458 | /* need a memory barrier to detect the race with raid5_quiesce() */ | |
5459 | if (!did_inc || smp_load_acquire(&conf->quiesce) != 0) { | |
5460 | /* quiesce is in progress, so we need to undo io activation and wait | |
5461 | * for it to finish | |
5462 | */ | |
5463 | if (did_inc && atomic_dec_and_test(&conf->active_aligned_reads)) | |
5464 | wake_up(&conf->wait_for_quiescent); | |
5465 | spin_lock_irq(&conf->device_lock); | |
5466 | wait_event_lock_irq(conf->wait_for_quiescent, conf->quiesce == 0, | |
5467 | conf->device_lock); | |
5468 | atomic_inc(&conf->active_aligned_reads); | |
5469 | spin_unlock_irq(&conf->device_lock); | |
5470 | } | |
387bb173 | 5471 | |
e82ed3a4 CH |
5472 | if (mddev->gendisk) |
5473 | trace_block_bio_remap(align_bio, disk_devt(mddev->gendisk), | |
5474 | raid_bio->bi_iter.bi_sector); | |
5475 | submit_bio_noacct(align_bio); | |
5476 | return 1; | |
6c0544e2 | 5477 | |
e82ed3a4 CH |
5478 | out_rcu_unlock: |
5479 | rcu_read_unlock(); | |
5480 | return 0; | |
f679623f RBJ |
5481 | } |
5482 | ||
7ef6b12a ML |
5483 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5484 | { | |
5485 | struct bio *split; | |
dd7a8f5d N |
5486 | sector_t sector = raid_bio->bi_iter.bi_sector; |
5487 | unsigned chunk_sects = mddev->chunk_sectors; | |
5488 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
7ef6b12a | 5489 | |
dd7a8f5d N |
5490 | if (sectors < bio_sectors(raid_bio)) { |
5491 | struct r5conf *conf = mddev->private; | |
afeee514 | 5492 | split = bio_split(raid_bio, sectors, GFP_NOIO, &conf->bio_split); |
dd7a8f5d | 5493 | bio_chain(split, raid_bio); |
ed00aabd | 5494 | submit_bio_noacct(raid_bio); |
dd7a8f5d N |
5495 | raid_bio = split; |
5496 | } | |
7ef6b12a | 5497 | |
dd7a8f5d N |
5498 | if (!raid5_read_one_chunk(mddev, raid_bio)) |
5499 | return raid_bio; | |
7ef6b12a ML |
5500 | |
5501 | return NULL; | |
5502 | } | |
5503 | ||
8b3e6cdc DW |
5504 | /* __get_priority_stripe - get the next stripe to process |
5505 | * | |
5506 | * Full stripe writes are allowed to pass preread active stripes up until | |
5507 | * the bypass_threshold is exceeded. In general the bypass_count | |
5508 | * increments when the handle_list is handled before the hold_list; however, it | |
5509 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5510 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5511 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5512 | * handle_list. | |
5513 | */ | |
851c30c9 | 5514 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5515 | { |
535ae4eb | 5516 | struct stripe_head *sh, *tmp; |
851c30c9 | 5517 | struct list_head *handle_list = NULL; |
535ae4eb | 5518 | struct r5worker_group *wg; |
70d466f7 SL |
5519 | bool second_try = !r5c_is_writeback(conf->log) && |
5520 | !r5l_log_disk_error(conf); | |
5521 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) || | |
5522 | r5l_log_disk_error(conf); | |
851c30c9 | 5523 | |
535ae4eb SL |
5524 | again: |
5525 | wg = NULL; | |
5526 | sh = NULL; | |
851c30c9 | 5527 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5528 | handle_list = try_loprio ? &conf->loprio_list : |
5529 | &conf->handle_list; | |
851c30c9 | 5530 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5531 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5532 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5533 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5534 | } else { |
5535 | int i; | |
5536 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5537 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5538 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5539 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5540 | if (!list_empty(handle_list)) |
5541 | break; | |
5542 | } | |
5543 | } | |
8b3e6cdc DW |
5544 | |
5545 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5546 | __func__, | |
851c30c9 | 5547 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5548 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5549 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5550 | ||
851c30c9 SL |
5551 | if (!list_empty(handle_list)) { |
5552 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5553 | |
5554 | if (list_empty(&conf->hold_list)) | |
5555 | conf->bypass_count = 0; | |
5556 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5557 | if (conf->hold_list.next == conf->last_hold) | |
5558 | conf->bypass_count++; | |
5559 | else { | |
5560 | conf->last_hold = conf->hold_list.next; | |
5561 | conf->bypass_count -= conf->bypass_threshold; | |
5562 | if (conf->bypass_count < 0) | |
5563 | conf->bypass_count = 0; | |
5564 | } | |
5565 | } | |
5566 | } else if (!list_empty(&conf->hold_list) && | |
5567 | ((conf->bypass_threshold && | |
5568 | conf->bypass_count > conf->bypass_threshold) || | |
5569 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5570 | |
5571 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5572 | if (conf->worker_cnt_per_group == 0 || | |
5573 | group == ANY_GROUP || | |
5574 | !cpu_online(tmp->cpu) || | |
5575 | cpu_to_group(tmp->cpu) == group) { | |
5576 | sh = tmp; | |
5577 | break; | |
5578 | } | |
5579 | } | |
5580 | ||
5581 | if (sh) { | |
5582 | conf->bypass_count -= conf->bypass_threshold; | |
5583 | if (conf->bypass_count < 0) | |
5584 | conf->bypass_count = 0; | |
5585 | } | |
bfc90cb0 | 5586 | wg = NULL; |
851c30c9 SL |
5587 | } |
5588 | ||
535ae4eb SL |
5589 | if (!sh) { |
5590 | if (second_try) | |
5591 | return NULL; | |
5592 | second_try = true; | |
5593 | try_loprio = !try_loprio; | |
5594 | goto again; | |
5595 | } | |
8b3e6cdc | 5596 | |
bfc90cb0 SL |
5597 | if (wg) { |
5598 | wg->stripes_cnt--; | |
5599 | sh->group = NULL; | |
5600 | } | |
8b3e6cdc | 5601 | list_del_init(&sh->lru); |
c7a6d35e | 5602 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5603 | return sh; |
5604 | } | |
f679623f | 5605 | |
8811b596 SL |
5606 | struct raid5_plug_cb { |
5607 | struct blk_plug_cb cb; | |
5608 | struct list_head list; | |
566c09c5 | 5609 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5610 | }; |
5611 | ||
5612 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5613 | { | |
5614 | struct raid5_plug_cb *cb = container_of( | |
5615 | blk_cb, struct raid5_plug_cb, cb); | |
5616 | struct stripe_head *sh; | |
5617 | struct mddev *mddev = cb->cb.data; | |
5618 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5619 | int cnt = 0; |
566c09c5 | 5620 | int hash; |
8811b596 SL |
5621 | |
5622 | if (cb->list.next && !list_empty(&cb->list)) { | |
5623 | spin_lock_irq(&conf->device_lock); | |
5624 | while (!list_empty(&cb->list)) { | |
5625 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5626 | list_del_init(&sh->lru); | |
5627 | /* | |
5628 | * avoid race release_stripe_plug() sees | |
5629 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5630 | * is still in our list | |
5631 | */ | |
4e857c58 | 5632 | smp_mb__before_atomic(); |
8811b596 | 5633 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5634 | /* |
5635 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5636 | * case, the count is always > 1 here | |
5637 | */ | |
566c09c5 SL |
5638 | hash = sh->hash_lock_index; |
5639 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5640 | cnt++; |
8811b596 SL |
5641 | } |
5642 | spin_unlock_irq(&conf->device_lock); | |
5643 | } | |
566c09c5 SL |
5644 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5645 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5646 | if (mddev->queue) |
5647 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5648 | kfree(cb); |
5649 | } | |
5650 | ||
5651 | static void release_stripe_plug(struct mddev *mddev, | |
5652 | struct stripe_head *sh) | |
5653 | { | |
5654 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5655 | raid5_unplug, mddev, | |
5656 | sizeof(struct raid5_plug_cb)); | |
5657 | struct raid5_plug_cb *cb; | |
5658 | ||
5659 | if (!blk_cb) { | |
6d036f7d | 5660 | raid5_release_stripe(sh); |
8811b596 SL |
5661 | return; |
5662 | } | |
5663 | ||
5664 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5665 | ||
566c09c5 SL |
5666 | if (cb->list.next == NULL) { |
5667 | int i; | |
8811b596 | 5668 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5669 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5670 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5671 | } | |
8811b596 SL |
5672 | |
5673 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5674 | list_add_tail(&sh->lru, &cb->list); | |
5675 | else | |
6d036f7d | 5676 | raid5_release_stripe(sh); |
8811b596 SL |
5677 | } |
5678 | ||
620125f2 SL |
5679 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5680 | { | |
5681 | struct r5conf *conf = mddev->private; | |
5682 | sector_t logical_sector, last_sector; | |
5683 | struct stripe_head *sh; | |
620125f2 SL |
5684 | int stripe_sectors; |
5685 | ||
5686 | if (mddev->reshape_position != MaxSector) | |
5687 | /* Skip discard while reshape is happening */ | |
5688 | return; | |
5689 | ||
c911c46c | 5690 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
b0f01ecf | 5691 | last_sector = bio_end_sector(bi); |
620125f2 SL |
5692 | |
5693 | bi->bi_next = NULL; | |
620125f2 SL |
5694 | |
5695 | stripe_sectors = conf->chunk_sectors * | |
5696 | (conf->raid_disks - conf->max_degraded); | |
5697 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5698 | stripe_sectors); | |
5699 | sector_div(last_sector, stripe_sectors); | |
5700 | ||
5701 | logical_sector *= conf->chunk_sectors; | |
5702 | last_sector *= conf->chunk_sectors; | |
5703 | ||
5704 | for (; logical_sector < last_sector; | |
c911c46c | 5705 | logical_sector += RAID5_STRIPE_SECTORS(conf)) { |
620125f2 SL |
5706 | DEFINE_WAIT(w); |
5707 | int d; | |
5708 | again: | |
6d036f7d | 5709 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5710 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5711 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5712 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5713 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5714 | raid5_release_stripe(sh); |
f8dfcffd N |
5715 | schedule(); |
5716 | goto again; | |
5717 | } | |
5718 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5719 | spin_lock_irq(&sh->stripe_lock); |
5720 | for (d = 0; d < conf->raid_disks; d++) { | |
5721 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5722 | continue; | |
5723 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5724 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5725 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5726 | raid5_release_stripe(sh); |
620125f2 SL |
5727 | schedule(); |
5728 | goto again; | |
5729 | } | |
5730 | } | |
f8dfcffd | 5731 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5732 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5733 | sh->overwrite_disks = 0; |
620125f2 SL |
5734 | for (d = 0; d < conf->raid_disks; d++) { |
5735 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5736 | continue; | |
5737 | sh->dev[d].towrite = bi; | |
5738 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
016c76ac | 5739 | bio_inc_remaining(bi); |
49728050 | 5740 | md_write_inc(mddev, bi); |
7a87f434 | 5741 | sh->overwrite_disks++; |
620125f2 SL |
5742 | } |
5743 | spin_unlock_irq(&sh->stripe_lock); | |
5744 | if (conf->mddev->bitmap) { | |
5745 | for (d = 0; | |
5746 | d < conf->raid_disks - conf->max_degraded; | |
5747 | d++) | |
e64e4018 AS |
5748 | md_bitmap_startwrite(mddev->bitmap, |
5749 | sh->sector, | |
c911c46c | 5750 | RAID5_STRIPE_SECTORS(conf), |
e64e4018 | 5751 | 0); |
620125f2 SL |
5752 | sh->bm_seq = conf->seq_flush + 1; |
5753 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5754 | } | |
5755 | ||
5756 | set_bit(STRIPE_HANDLE, &sh->state); | |
5757 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5758 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5759 | atomic_inc(&conf->preread_active_stripes); | |
5760 | release_stripe_plug(mddev, sh); | |
5761 | } | |
5762 | ||
016c76ac | 5763 | bio_endio(bi); |
620125f2 SL |
5764 | } |
5765 | ||
cc27b0c7 | 5766 | static bool raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5767 | { |
d1688a6d | 5768 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5769 | int dd_idx; |
1da177e4 LT |
5770 | sector_t new_sector; |
5771 | sector_t logical_sector, last_sector; | |
5772 | struct stripe_head *sh; | |
a362357b | 5773 | const int rw = bio_data_dir(bi); |
27c0f68f SL |
5774 | DEFINE_WAIT(w); |
5775 | bool do_prepare; | |
3bddb7f8 | 5776 | bool do_flush = false; |
1da177e4 | 5777 | |
1eff9d32 | 5778 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
1532d9e8 | 5779 | int ret = log_handle_flush_request(conf, bi); |
828cbe98 SL |
5780 | |
5781 | if (ret == 0) | |
cc27b0c7 | 5782 | return true; |
828cbe98 | 5783 | if (ret == -ENODEV) { |
775d7831 DJ |
5784 | if (md_flush_request(mddev, bi)) |
5785 | return true; | |
828cbe98 SL |
5786 | } |
5787 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5788 | /* |
5789 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5790 | * we need to flush journal device | |
5791 | */ | |
5792 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5793 | } |
5794 | ||
cc27b0c7 N |
5795 | if (!md_write_start(mddev, bi)) |
5796 | return false; | |
9ffc8f7c EM |
5797 | /* |
5798 | * If array is degraded, better not do chunk aligned read because | |
5799 | * later we might have to read it again in order to reconstruct | |
5800 | * data on failed drives. | |
5801 | */ | |
5802 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5803 | mddev->reshape_position == MaxSector) { |
5804 | bi = chunk_aligned_read(mddev, bi); | |
5805 | if (!bi) | |
cc27b0c7 | 5806 | return true; |
7ef6b12a | 5807 | } |
52488615 | 5808 | |
796a5cf0 | 5809 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 | 5810 | make_discard_request(mddev, bi); |
cc27b0c7 N |
5811 | md_write_end(mddev); |
5812 | return true; | |
620125f2 SL |
5813 | } |
5814 | ||
c911c46c | 5815 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
f73a1c7d | 5816 | last_sector = bio_end_sector(bi); |
1da177e4 | 5817 | bi->bi_next = NULL; |
06d91a5f | 5818 | |
1147f58e | 5819 | md_account_bio(mddev, &bi); |
27c0f68f | 5820 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
c911c46c | 5821 | for (; logical_sector < last_sector; logical_sector += RAID5_STRIPE_SECTORS(conf)) { |
b5663ba4 | 5822 | int previous; |
c46501b2 | 5823 | int seq; |
b578d55f | 5824 | |
27c0f68f | 5825 | do_prepare = false; |
7ecaa1e6 | 5826 | retry: |
c46501b2 | 5827 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5828 | previous = 0; |
27c0f68f SL |
5829 | if (do_prepare) |
5830 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5831 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5832 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5833 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5834 | * 64bit on a 32bit platform, and so it might be |
5835 | * possible to see a half-updated value | |
aeb878b0 | 5836 | * Of course reshape_progress could change after |
df8e7f76 N |
5837 | * the lock is dropped, so once we get a reference |
5838 | * to the stripe that we think it is, we will have | |
5839 | * to check again. | |
5840 | */ | |
7ecaa1e6 | 5841 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5842 | if (mddev->reshape_backwards |
fef9c61f N |
5843 | ? logical_sector < conf->reshape_progress |
5844 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5845 | previous = 1; |
5846 | } else { | |
2c810cdd | 5847 | if (mddev->reshape_backwards |
fef9c61f N |
5848 | ? logical_sector < conf->reshape_safe |
5849 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5850 | spin_unlock_irq(&conf->device_lock); |
5851 | schedule(); | |
27c0f68f | 5852 | do_prepare = true; |
b578d55f N |
5853 | goto retry; |
5854 | } | |
5855 | } | |
7ecaa1e6 N |
5856 | spin_unlock_irq(&conf->device_lock); |
5857 | } | |
16a53ecc | 5858 | |
112bf897 N |
5859 | new_sector = raid5_compute_sector(conf, logical_sector, |
5860 | previous, | |
911d4ee8 | 5861 | &dd_idx, NULL); |
849674e4 | 5862 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5863 | (unsigned long long)new_sector, |
1da177e4 LT |
5864 | (unsigned long long)logical_sector); |
5865 | ||
6d036f7d | 5866 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5867 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5868 | if (sh) { |
b0f9ec04 | 5869 | if (unlikely(previous)) { |
7ecaa1e6 | 5870 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5871 | * stripe, so we must do the range check again. |
5872 | * Expansion could still move past after this | |
5873 | * test, but as we are holding a reference to | |
5874 | * 'sh', we know that if that happens, | |
5875 | * STRIPE_EXPANDING will get set and the expansion | |
5876 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5877 | */ |
5878 | int must_retry = 0; | |
5879 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5880 | if (mddev->reshape_backwards |
b0f9ec04 N |
5881 | ? logical_sector >= conf->reshape_progress |
5882 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5883 | /* mismatch, need to try again */ |
5884 | must_retry = 1; | |
5885 | spin_unlock_irq(&conf->device_lock); | |
5886 | if (must_retry) { | |
6d036f7d | 5887 | raid5_release_stripe(sh); |
7a3ab908 | 5888 | schedule(); |
27c0f68f | 5889 | do_prepare = true; |
7ecaa1e6 N |
5890 | goto retry; |
5891 | } | |
5892 | } | |
c46501b2 N |
5893 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5894 | /* Might have got the wrong stripe_head | |
5895 | * by accident | |
5896 | */ | |
6d036f7d | 5897 | raid5_release_stripe(sh); |
c46501b2 N |
5898 | goto retry; |
5899 | } | |
e62e58a5 | 5900 | |
7ecaa1e6 | 5901 | if (test_bit(STRIPE_EXPANDING, &sh->state) || |
da41ba65 | 5902 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5903 | /* Stripe is busy expanding or |
5904 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5905 | * and wait a while |
5906 | */ | |
482c0834 | 5907 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5908 | raid5_release_stripe(sh); |
1da177e4 | 5909 | schedule(); |
27c0f68f | 5910 | do_prepare = true; |
1da177e4 LT |
5911 | goto retry; |
5912 | } | |
3bddb7f8 SL |
5913 | if (do_flush) { |
5914 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5915 | /* we only need flush for one stripe */ | |
5916 | do_flush = false; | |
5917 | } | |
5918 | ||
1684e975 | 5919 | set_bit(STRIPE_HANDLE, &sh->state); |
6ed3003c | 5920 | clear_bit(STRIPE_DELAYED, &sh->state); |
59fc630b | 5921 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5922 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5923 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5924 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5925 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5926 | } else { |
5927 | /* cannot get stripe for read-ahead, just give-up */ | |
4e4cbee9 | 5928 | bi->bi_status = BLK_STS_IOERR; |
1da177e4 LT |
5929 | break; |
5930 | } | |
1da177e4 | 5931 | } |
27c0f68f | 5932 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5933 | |
49728050 N |
5934 | if (rw == WRITE) |
5935 | md_write_end(mddev); | |
016c76ac | 5936 | bio_endio(bi); |
cc27b0c7 | 5937 | return true; |
1da177e4 LT |
5938 | } |
5939 | ||
fd01b88c | 5940 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5941 | |
fd01b88c | 5942 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5943 | { |
52c03291 N |
5944 | /* reshaping is quite different to recovery/resync so it is |
5945 | * handled quite separately ... here. | |
5946 | * | |
5947 | * On each call to sync_request, we gather one chunk worth of | |
5948 | * destination stripes and flag them as expanding. | |
5949 | * Then we find all the source stripes and request reads. | |
5950 | * As the reads complete, handle_stripe will copy the data | |
5951 | * into the destination stripe and release that stripe. | |
5952 | */ | |
d1688a6d | 5953 | struct r5conf *conf = mddev->private; |
1da177e4 | 5954 | struct stripe_head *sh; |
db0505d3 | 5955 | struct md_rdev *rdev; |
ccfcc3c1 | 5956 | sector_t first_sector, last_sector; |
f416885e N |
5957 | int raid_disks = conf->previous_raid_disks; |
5958 | int data_disks = raid_disks - conf->max_degraded; | |
5959 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5960 | int i; |
5961 | int dd_idx; | |
c8f517c4 | 5962 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5963 | sector_t stripe_addr; |
7a661381 | 5964 | int reshape_sectors; |
ab69ae12 | 5965 | struct list_head stripes; |
92140480 | 5966 | sector_t retn; |
52c03291 | 5967 | |
fef9c61f N |
5968 | if (sector_nr == 0) { |
5969 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5970 | if (mddev->reshape_backwards && |
fef9c61f N |
5971 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5972 | sector_nr = raid5_size(mddev, 0, 0) | |
5973 | - conf->reshape_progress; | |
6cbd8148 N |
5974 | } else if (mddev->reshape_backwards && |
5975 | conf->reshape_progress == MaxSector) { | |
5976 | /* shouldn't happen, but just in case, finish up.*/ | |
5977 | sector_nr = MaxSector; | |
2c810cdd | 5978 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5979 | conf->reshape_progress > 0) |
5980 | sector_nr = conf->reshape_progress; | |
f416885e | 5981 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5982 | if (sector_nr) { |
8dee7211 | 5983 | mddev->curr_resync_completed = sector_nr; |
e1a86dbb | 5984 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
fef9c61f | 5985 | *skipped = 1; |
92140480 N |
5986 | retn = sector_nr; |
5987 | goto finish; | |
fef9c61f | 5988 | } |
52c03291 N |
5989 | } |
5990 | ||
7a661381 N |
5991 | /* We need to process a full chunk at a time. |
5992 | * If old and new chunk sizes differ, we need to process the | |
5993 | * largest of these | |
5994 | */ | |
3cb5edf4 N |
5995 | |
5996 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5997 | |
b5254dd5 N |
5998 | /* We update the metadata at least every 10 seconds, or when |
5999 | * the data about to be copied would over-write the source of | |
6000 | * the data at the front of the range. i.e. one new_stripe | |
6001 | * along from reshape_progress new_maps to after where | |
6002 | * reshape_safe old_maps to | |
52c03291 | 6003 | */ |
fef9c61f | 6004 | writepos = conf->reshape_progress; |
f416885e | 6005 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
6006 | readpos = conf->reshape_progress; |
6007 | sector_div(readpos, data_disks); | |
fef9c61f | 6008 | safepos = conf->reshape_safe; |
f416885e | 6009 | sector_div(safepos, data_disks); |
2c810cdd | 6010 | if (mddev->reshape_backwards) { |
c74c0d76 N |
6011 | BUG_ON(writepos < reshape_sectors); |
6012 | writepos -= reshape_sectors; | |
c8f517c4 | 6013 | readpos += reshape_sectors; |
7a661381 | 6014 | safepos += reshape_sectors; |
fef9c61f | 6015 | } else { |
7a661381 | 6016 | writepos += reshape_sectors; |
c74c0d76 N |
6017 | /* readpos and safepos are worst-case calculations. |
6018 | * A negative number is overly pessimistic, and causes | |
6019 | * obvious problems for unsigned storage. So clip to 0. | |
6020 | */ | |
ed37d83e N |
6021 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
6022 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 6023 | } |
52c03291 | 6024 | |
b5254dd5 N |
6025 | /* Having calculated the 'writepos' possibly use it |
6026 | * to set 'stripe_addr' which is where we will write to. | |
6027 | */ | |
6028 | if (mddev->reshape_backwards) { | |
6029 | BUG_ON(conf->reshape_progress == 0); | |
6030 | stripe_addr = writepos; | |
6031 | BUG_ON((mddev->dev_sectors & | |
6032 | ~((sector_t)reshape_sectors - 1)) | |
6033 | - reshape_sectors - stripe_addr | |
6034 | != sector_nr); | |
6035 | } else { | |
6036 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
6037 | stripe_addr = sector_nr; | |
6038 | } | |
6039 | ||
c8f517c4 N |
6040 | /* 'writepos' is the most advanced device address we might write. |
6041 | * 'readpos' is the least advanced device address we might read. | |
6042 | * 'safepos' is the least address recorded in the metadata as having | |
6043 | * been reshaped. | |
b5254dd5 N |
6044 | * If there is a min_offset_diff, these are adjusted either by |
6045 | * increasing the safepos/readpos if diff is negative, or | |
6046 | * increasing writepos if diff is positive. | |
6047 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
6048 | * ensure safety in the face of a crash - that must be done by userspace |
6049 | * making a backup of the data. So in that case there is no particular | |
6050 | * rush to update metadata. | |
6051 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
6052 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
6053 | * we can be safe in the event of a crash. | |
6054 | * So we insist on updating metadata if safepos is behind writepos and | |
6055 | * readpos is beyond writepos. | |
6056 | * In any case, update the metadata every 10 seconds. | |
6057 | * Maybe that number should be configurable, but I'm not sure it is | |
6058 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
6059 | */ | |
b5254dd5 N |
6060 | if (conf->min_offset_diff < 0) { |
6061 | safepos += -conf->min_offset_diff; | |
6062 | readpos += -conf->min_offset_diff; | |
6063 | } else | |
6064 | writepos += conf->min_offset_diff; | |
6065 | ||
2c810cdd | 6066 | if ((mddev->reshape_backwards |
c8f517c4 N |
6067 | ? (safepos > writepos && readpos < writepos) |
6068 | : (safepos < writepos && readpos > writepos)) || | |
6069 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
6070 | /* Cannot proceed until we've updated the superblock... */ |
6071 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
6072 | atomic_read(&conf->reshape_stripes)==0 |
6073 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
6074 | if (atomic_read(&conf->reshape_stripes) != 0) | |
6075 | return 0; | |
fef9c61f | 6076 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 6077 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
6078 | if (!mddev->reshape_backwards) |
6079 | /* Can update recovery_offset */ | |
6080 | rdev_for_each(rdev, mddev) | |
6081 | if (rdev->raid_disk >= 0 && | |
6082 | !test_bit(Journal, &rdev->flags) && | |
6083 | !test_bit(In_sync, &rdev->flags) && | |
6084 | rdev->recovery_offset < sector_nr) | |
6085 | rdev->recovery_offset = sector_nr; | |
6086 | ||
c8f517c4 | 6087 | conf->reshape_checkpoint = jiffies; |
2953079c | 6088 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 6089 | md_wakeup_thread(mddev->thread); |
2953079c | 6090 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
6091 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
6092 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6093 | return 0; | |
52c03291 | 6094 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 6095 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
6096 | spin_unlock_irq(&conf->device_lock); |
6097 | wake_up(&conf->wait_for_overlap); | |
e1a86dbb | 6098 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
52c03291 N |
6099 | } |
6100 | ||
ab69ae12 | 6101 | INIT_LIST_HEAD(&stripes); |
c911c46c | 6102 | for (i = 0; i < reshape_sectors; i += RAID5_STRIPE_SECTORS(conf)) { |
52c03291 | 6103 | int j; |
a9f326eb | 6104 | int skipped_disk = 0; |
6d036f7d | 6105 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
6106 | set_bit(STRIPE_EXPANDING, &sh->state); |
6107 | atomic_inc(&conf->reshape_stripes); | |
6108 | /* If any of this stripe is beyond the end of the old | |
6109 | * array, then we need to zero those blocks | |
6110 | */ | |
6111 | for (j=sh->disks; j--;) { | |
6112 | sector_t s; | |
6113 | if (j == sh->pd_idx) | |
6114 | continue; | |
f416885e | 6115 | if (conf->level == 6 && |
d0dabf7e | 6116 | j == sh->qd_idx) |
f416885e | 6117 | continue; |
6d036f7d | 6118 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 6119 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 6120 | skipped_disk = 1; |
52c03291 N |
6121 | continue; |
6122 | } | |
c911c46c | 6123 | memset(page_address(sh->dev[j].page), 0, RAID5_STRIPE_SIZE(conf)); |
52c03291 N |
6124 | set_bit(R5_Expanded, &sh->dev[j].flags); |
6125 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
6126 | } | |
a9f326eb | 6127 | if (!skipped_disk) { |
52c03291 N |
6128 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
6129 | set_bit(STRIPE_HANDLE, &sh->state); | |
6130 | } | |
ab69ae12 | 6131 | list_add(&sh->lru, &stripes); |
52c03291 N |
6132 | } |
6133 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 6134 | if (mddev->reshape_backwards) |
7a661381 | 6135 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 6136 | else |
7a661381 | 6137 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
6138 | spin_unlock_irq(&conf->device_lock); |
6139 | /* Ok, those stripe are ready. We can start scheduling | |
6140 | * reads on the source stripes. | |
6141 | * The source stripes are determined by mapping the first and last | |
6142 | * block on the destination stripes. | |
6143 | */ | |
52c03291 | 6144 | first_sector = |
ec32a2bd | 6145 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 6146 | 1, &dd_idx, NULL); |
52c03291 | 6147 | last_sector = |
0e6e0271 | 6148 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 6149 | * new_data_disks - 1), |
911d4ee8 | 6150 | 1, &dd_idx, NULL); |
58c0fed4 AN |
6151 | if (last_sector >= mddev->dev_sectors) |
6152 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 6153 | while (first_sector <= last_sector) { |
6d036f7d | 6154 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
6155 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
6156 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 6157 | raid5_release_stripe(sh); |
c911c46c | 6158 | first_sector += RAID5_STRIPE_SECTORS(conf); |
52c03291 | 6159 | } |
ab69ae12 N |
6160 | /* Now that the sources are clearly marked, we can release |
6161 | * the destination stripes | |
6162 | */ | |
6163 | while (!list_empty(&stripes)) { | |
6164 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
6165 | list_del_init(&sh->lru); | |
6d036f7d | 6166 | raid5_release_stripe(sh); |
ab69ae12 | 6167 | } |
c6207277 N |
6168 | /* If this takes us to the resync_max point where we have to pause, |
6169 | * then we need to write out the superblock. | |
6170 | */ | |
7a661381 | 6171 | sector_nr += reshape_sectors; |
92140480 N |
6172 | retn = reshape_sectors; |
6173 | finish: | |
c5e19d90 N |
6174 | if (mddev->curr_resync_completed > mddev->resync_max || |
6175 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 6176 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
6177 | /* Cannot proceed until we've updated the superblock... */ |
6178 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
6179 | atomic_read(&conf->reshape_stripes) == 0 |
6180 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
6181 | if (atomic_read(&conf->reshape_stripes) != 0) | |
6182 | goto ret; | |
fef9c61f | 6183 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 6184 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
6185 | if (!mddev->reshape_backwards) |
6186 | /* Can update recovery_offset */ | |
6187 | rdev_for_each(rdev, mddev) | |
6188 | if (rdev->raid_disk >= 0 && | |
6189 | !test_bit(Journal, &rdev->flags) && | |
6190 | !test_bit(In_sync, &rdev->flags) && | |
6191 | rdev->recovery_offset < sector_nr) | |
6192 | rdev->recovery_offset = sector_nr; | |
c8f517c4 | 6193 | conf->reshape_checkpoint = jiffies; |
2953079c | 6194 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
6195 | md_wakeup_thread(mddev->thread); |
6196 | wait_event(mddev->sb_wait, | |
2953079c | 6197 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
6198 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
6199 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6200 | goto ret; | |
c6207277 | 6201 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 6202 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
6203 | spin_unlock_irq(&conf->device_lock); |
6204 | wake_up(&conf->wait_for_overlap); | |
e1a86dbb | 6205 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
c6207277 | 6206 | } |
c91abf5a | 6207 | ret: |
92140480 | 6208 | return retn; |
52c03291 N |
6209 | } |
6210 | ||
849674e4 SL |
6211 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
6212 | int *skipped) | |
52c03291 | 6213 | { |
d1688a6d | 6214 | struct r5conf *conf = mddev->private; |
52c03291 | 6215 | struct stripe_head *sh; |
58c0fed4 | 6216 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 6217 | sector_t sync_blocks; |
16a53ecc N |
6218 | int still_degraded = 0; |
6219 | int i; | |
1da177e4 | 6220 | |
72626685 | 6221 | if (sector_nr >= max_sector) { |
1da177e4 | 6222 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 6223 | |
29269553 N |
6224 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
6225 | end_reshape(conf); | |
6226 | return 0; | |
6227 | } | |
72626685 N |
6228 | |
6229 | if (mddev->curr_resync < max_sector) /* aborted */ | |
e64e4018 AS |
6230 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
6231 | &sync_blocks, 1); | |
16a53ecc | 6232 | else /* completed sync */ |
72626685 | 6233 | conf->fullsync = 0; |
e64e4018 | 6234 | md_bitmap_close_sync(mddev->bitmap); |
72626685 | 6235 | |
1da177e4 LT |
6236 | return 0; |
6237 | } | |
ccfcc3c1 | 6238 | |
64bd660b N |
6239 | /* Allow raid5_quiesce to complete */ |
6240 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
6241 | ||
52c03291 N |
6242 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
6243 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 6244 | |
c6207277 N |
6245 | /* No need to check resync_max as we never do more than one |
6246 | * stripe, and as resync_max will always be on a chunk boundary, | |
6247 | * if the check in md_do_sync didn't fire, there is no chance | |
6248 | * of overstepping resync_max here | |
6249 | */ | |
6250 | ||
16a53ecc | 6251 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
6252 | * to resync, then assert that we are finished, because there is |
6253 | * nothing we can do. | |
6254 | */ | |
3285edf1 | 6255 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 6256 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 6257 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 6258 | *skipped = 1; |
1da177e4 LT |
6259 | return rv; |
6260 | } | |
6f608040 | 6261 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
6262 | !conf->fullsync && | |
e64e4018 | 6263 | !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
c911c46c | 6264 | sync_blocks >= RAID5_STRIPE_SECTORS(conf)) { |
72626685 | 6265 | /* we can skip this block, and probably more */ |
83c3e5e1 | 6266 | do_div(sync_blocks, RAID5_STRIPE_SECTORS(conf)); |
72626685 | 6267 | *skipped = 1; |
c911c46c YY |
6268 | /* keep things rounded to whole stripes */ |
6269 | return sync_blocks * RAID5_STRIPE_SECTORS(conf); | |
72626685 | 6270 | } |
1da177e4 | 6271 | |
e64e4018 | 6272 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 6273 | |
6d036f7d | 6274 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 6275 | if (sh == NULL) { |
6d036f7d | 6276 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 6277 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 6278 | * is trying to get access |
1da177e4 | 6279 | */ |
66c006a5 | 6280 | schedule_timeout_uninterruptible(1); |
1da177e4 | 6281 | } |
16a53ecc | 6282 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
6283 | * Note in case of > 1 drive failures it's possible we're rebuilding |
6284 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 6285 | */ |
16d9cfab EM |
6286 | rcu_read_lock(); |
6287 | for (i = 0; i < conf->raid_disks; i++) { | |
6aa7de05 | 6288 | struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev); |
16d9cfab EM |
6289 | |
6290 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 6291 | still_degraded = 1; |
16d9cfab EM |
6292 | } |
6293 | rcu_read_unlock(); | |
16a53ecc | 6294 | |
e64e4018 | 6295 | md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); |
16a53ecc | 6296 | |
83206d66 | 6297 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 6298 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 6299 | |
6d036f7d | 6300 | raid5_release_stripe(sh); |
1da177e4 | 6301 | |
c911c46c | 6302 | return RAID5_STRIPE_SECTORS(conf); |
1da177e4 LT |
6303 | } |
6304 | ||
0472a42b N |
6305 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio, |
6306 | unsigned int offset) | |
46031f9a RBJ |
6307 | { |
6308 | /* We may not be able to submit a whole bio at once as there | |
6309 | * may not be enough stripe_heads available. | |
6310 | * We cannot pre-allocate enough stripe_heads as we may need | |
6311 | * more than exist in the cache (if we allow ever large chunks). | |
6312 | * So we do one stripe head at a time and record in | |
6313 | * ->bi_hw_segments how many have been done. | |
6314 | * | |
6315 | * We *know* that this entire raid_bio is in one chunk, so | |
6316 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
6317 | */ | |
6318 | struct stripe_head *sh; | |
911d4ee8 | 6319 | int dd_idx; |
46031f9a RBJ |
6320 | sector_t sector, logical_sector, last_sector; |
6321 | int scnt = 0; | |
46031f9a RBJ |
6322 | int handled = 0; |
6323 | ||
4f024f37 | 6324 | logical_sector = raid_bio->bi_iter.bi_sector & |
c911c46c | 6325 | ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
112bf897 | 6326 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6327 | 0, &dd_idx, NULL); |
f73a1c7d | 6328 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6329 | |
6330 | for (; logical_sector < last_sector; | |
c911c46c YY |
6331 | logical_sector += RAID5_STRIPE_SECTORS(conf), |
6332 | sector += RAID5_STRIPE_SECTORS(conf), | |
387bb173 | 6333 | scnt++) { |
46031f9a | 6334 | |
0472a42b | 6335 | if (scnt < offset) |
46031f9a RBJ |
6336 | /* already done this stripe */ |
6337 | continue; | |
6338 | ||
6d036f7d | 6339 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6340 | |
6341 | if (!sh) { | |
6342 | /* failed to get a stripe - must wait */ | |
46031f9a | 6343 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6344 | conf->retry_read_offset = scnt; |
46031f9a RBJ |
6345 | return handled; |
6346 | } | |
6347 | ||
da41ba65 | 6348 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6349 | raid5_release_stripe(sh); |
387bb173 | 6350 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6351 | conf->retry_read_offset = scnt; |
387bb173 NB |
6352 | return handled; |
6353 | } | |
6354 | ||
3f9e7c14 | 6355 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6356 | handle_stripe(sh); |
6d036f7d | 6357 | raid5_release_stripe(sh); |
46031f9a RBJ |
6358 | handled++; |
6359 | } | |
016c76ac N |
6360 | |
6361 | bio_endio(raid_bio); | |
6362 | ||
46031f9a | 6363 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6364 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6365 | return handled; |
6366 | } | |
6367 | ||
bfc90cb0 | 6368 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6369 | struct r5worker *worker, |
6370 | struct list_head *temp_inactive_list) | |
efcd487c CH |
6371 | __releases(&conf->device_lock) |
6372 | __acquires(&conf->device_lock) | |
46a06401 SL |
6373 | { |
6374 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6375 | int i, batch_size = 0, hash; |
6376 | bool release_inactive = false; | |
46a06401 SL |
6377 | |
6378 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6379 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6380 | batch[batch_size++] = sh; |
6381 | ||
566c09c5 SL |
6382 | if (batch_size == 0) { |
6383 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6384 | if (!list_empty(temp_inactive_list + i)) | |
6385 | break; | |
a8c34f91 SL |
6386 | if (i == NR_STRIPE_HASH_LOCKS) { |
6387 | spin_unlock_irq(&conf->device_lock); | |
1532d9e8 | 6388 | log_flush_stripe_to_raid(conf); |
a8c34f91 | 6389 | spin_lock_irq(&conf->device_lock); |
566c09c5 | 6390 | return batch_size; |
a8c34f91 | 6391 | } |
566c09c5 SL |
6392 | release_inactive = true; |
6393 | } | |
46a06401 SL |
6394 | spin_unlock_irq(&conf->device_lock); |
6395 | ||
566c09c5 SL |
6396 | release_inactive_stripe_list(conf, temp_inactive_list, |
6397 | NR_STRIPE_HASH_LOCKS); | |
6398 | ||
a8c34f91 | 6399 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6400 | if (release_inactive) { |
6401 | spin_lock_irq(&conf->device_lock); | |
6402 | return 0; | |
6403 | } | |
6404 | ||
46a06401 SL |
6405 | for (i = 0; i < batch_size; i++) |
6406 | handle_stripe(batch[i]); | |
ff875738 | 6407 | log_write_stripe_run(conf); |
46a06401 SL |
6408 | |
6409 | cond_resched(); | |
6410 | ||
6411 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6412 | for (i = 0; i < batch_size; i++) { |
6413 | hash = batch[i]->hash_lock_index; | |
6414 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6415 | } | |
46a06401 SL |
6416 | return batch_size; |
6417 | } | |
46031f9a | 6418 | |
851c30c9 SL |
6419 | static void raid5_do_work(struct work_struct *work) |
6420 | { | |
6421 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6422 | struct r5worker_group *group = worker->group; | |
6423 | struct r5conf *conf = group->conf; | |
16d997b7 | 6424 | struct mddev *mddev = conf->mddev; |
851c30c9 SL |
6425 | int group_id = group - conf->worker_groups; |
6426 | int handled; | |
6427 | struct blk_plug plug; | |
6428 | ||
6429 | pr_debug("+++ raid5worker active\n"); | |
6430 | ||
6431 | blk_start_plug(&plug); | |
6432 | handled = 0; | |
6433 | spin_lock_irq(&conf->device_lock); | |
6434 | while (1) { | |
6435 | int batch_size, released; | |
6436 | ||
566c09c5 | 6437 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6438 | |
566c09c5 SL |
6439 | batch_size = handle_active_stripes(conf, group_id, worker, |
6440 | worker->temp_inactive_list); | |
bfc90cb0 | 6441 | worker->working = false; |
851c30c9 SL |
6442 | if (!batch_size && !released) |
6443 | break; | |
6444 | handled += batch_size; | |
16d997b7 N |
6445 | wait_event_lock_irq(mddev->sb_wait, |
6446 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags), | |
6447 | conf->device_lock); | |
851c30c9 SL |
6448 | } |
6449 | pr_debug("%d stripes handled\n", handled); | |
6450 | ||
6451 | spin_unlock_irq(&conf->device_lock); | |
7e96d559 | 6452 | |
9c72a18e SL |
6453 | flush_deferred_bios(conf); |
6454 | ||
6455 | r5l_flush_stripe_to_raid(conf->log); | |
6456 | ||
7e96d559 | 6457 | async_tx_issue_pending_all(); |
851c30c9 SL |
6458 | blk_finish_plug(&plug); |
6459 | ||
6460 | pr_debug("--- raid5worker inactive\n"); | |
6461 | } | |
6462 | ||
1da177e4 LT |
6463 | /* |
6464 | * This is our raid5 kernel thread. | |
6465 | * | |
6466 | * We scan the hash table for stripes which can be handled now. | |
6467 | * During the scan, completed stripes are saved for us by the interrupt | |
6468 | * handler, so that they will not have to wait for our next wakeup. | |
6469 | */ | |
4ed8731d | 6470 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6471 | { |
4ed8731d | 6472 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6473 | struct r5conf *conf = mddev->private; |
1da177e4 | 6474 | int handled; |
e1dfa0a2 | 6475 | struct blk_plug plug; |
1da177e4 | 6476 | |
45b4233c | 6477 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6478 | |
6479 | md_check_recovery(mddev); | |
1da177e4 | 6480 | |
e1dfa0a2 | 6481 | blk_start_plug(&plug); |
1da177e4 LT |
6482 | handled = 0; |
6483 | spin_lock_irq(&conf->device_lock); | |
6484 | while (1) { | |
46031f9a | 6485 | struct bio *bio; |
773ca82f | 6486 | int batch_size, released; |
0472a42b | 6487 | unsigned int offset; |
773ca82f | 6488 | |
566c09c5 | 6489 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6490 | if (released) |
6491 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6492 | |
0021b7bc | 6493 | if ( |
7c13edc8 N |
6494 | !list_empty(&conf->bitmap_list)) { |
6495 | /* Now is a good time to flush some bitmap updates */ | |
6496 | conf->seq_flush++; | |
700e432d | 6497 | spin_unlock_irq(&conf->device_lock); |
e64e4018 | 6498 | md_bitmap_unplug(mddev->bitmap); |
700e432d | 6499 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6500 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6501 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6502 | } |
0021b7bc | 6503 | raid5_activate_delayed(conf); |
72626685 | 6504 | |
0472a42b | 6505 | while ((bio = remove_bio_from_retry(conf, &offset))) { |
46031f9a RBJ |
6506 | int ok; |
6507 | spin_unlock_irq(&conf->device_lock); | |
0472a42b | 6508 | ok = retry_aligned_read(conf, bio, offset); |
46031f9a RBJ |
6509 | spin_lock_irq(&conf->device_lock); |
6510 | if (!ok) | |
6511 | break; | |
6512 | handled++; | |
6513 | } | |
6514 | ||
566c09c5 SL |
6515 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6516 | conf->temp_inactive_list); | |
773ca82f | 6517 | if (!batch_size && !released) |
1da177e4 | 6518 | break; |
46a06401 | 6519 | handled += batch_size; |
1da177e4 | 6520 | |
2953079c | 6521 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6522 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6523 | md_check_recovery(mddev); |
46a06401 SL |
6524 | spin_lock_irq(&conf->device_lock); |
6525 | } | |
1da177e4 | 6526 | } |
45b4233c | 6527 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6528 | |
6529 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6530 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6531 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6532 | grow_one_stripe(conf, __GFP_NOWARN); |
6533 | /* Set flag even if allocation failed. This helps | |
6534 | * slow down allocation requests when mem is short | |
6535 | */ | |
6536 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6537 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6538 | } |
1da177e4 | 6539 | |
765d704d SL |
6540 | flush_deferred_bios(conf); |
6541 | ||
0576b1c6 SL |
6542 | r5l_flush_stripe_to_raid(conf->log); |
6543 | ||
c9f21aaf | 6544 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6545 | blk_finish_plug(&plug); |
1da177e4 | 6546 | |
45b4233c | 6547 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6548 | } |
6549 | ||
3f294f4f | 6550 | static ssize_t |
fd01b88c | 6551 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6552 | { |
7b1485ba N |
6553 | struct r5conf *conf; |
6554 | int ret = 0; | |
6555 | spin_lock(&mddev->lock); | |
6556 | conf = mddev->private; | |
96de1e66 | 6557 | if (conf) |
edbe83ab | 6558 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6559 | spin_unlock(&mddev->lock); |
6560 | return ret; | |
3f294f4f N |
6561 | } |
6562 | ||
c41d4ac4 | 6563 | int |
fd01b88c | 6564 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6565 | { |
483cbbed | 6566 | int result = 0; |
d1688a6d | 6567 | struct r5conf *conf = mddev->private; |
b5470dc5 | 6568 | |
c41d4ac4 | 6569 | if (size <= 16 || size > 32768) |
3f294f4f | 6570 | return -EINVAL; |
486f0644 | 6571 | |
edbe83ab | 6572 | conf->min_nr_stripes = size; |
2d5b569b | 6573 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6574 | while (size < conf->max_nr_stripes && |
6575 | drop_one_stripe(conf)) | |
6576 | ; | |
2d5b569b | 6577 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6578 | |
2214c260 | 6579 | md_allow_write(mddev); |
486f0644 | 6580 | |
2d5b569b | 6581 | mutex_lock(&conf->cache_size_mutex); |
486f0644 | 6582 | while (size > conf->max_nr_stripes) |
483cbbed AN |
6583 | if (!grow_one_stripe(conf, GFP_KERNEL)) { |
6584 | conf->min_nr_stripes = conf->max_nr_stripes; | |
6585 | result = -ENOMEM; | |
486f0644 | 6586 | break; |
483cbbed | 6587 | } |
2d5b569b | 6588 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6589 | |
483cbbed | 6590 | return result; |
c41d4ac4 N |
6591 | } |
6592 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6593 | ||
6594 | static ssize_t | |
fd01b88c | 6595 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6596 | { |
6791875e | 6597 | struct r5conf *conf; |
c41d4ac4 N |
6598 | unsigned long new; |
6599 | int err; | |
6600 | ||
6601 | if (len >= PAGE_SIZE) | |
6602 | return -EINVAL; | |
b29bebd6 | 6603 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6604 | return -EINVAL; |
6791875e | 6605 | err = mddev_lock(mddev); |
c41d4ac4 N |
6606 | if (err) |
6607 | return err; | |
6791875e N |
6608 | conf = mddev->private; |
6609 | if (!conf) | |
6610 | err = -ENODEV; | |
6611 | else | |
6612 | err = raid5_set_cache_size(mddev, new); | |
6613 | mddev_unlock(mddev); | |
6614 | ||
6615 | return err ?: len; | |
3f294f4f | 6616 | } |
007583c9 | 6617 | |
96de1e66 N |
6618 | static struct md_sysfs_entry |
6619 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6620 | raid5_show_stripe_cache_size, | |
6621 | raid5_store_stripe_cache_size); | |
3f294f4f | 6622 | |
d06f191f MS |
6623 | static ssize_t |
6624 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6625 | { | |
6626 | struct r5conf *conf = mddev->private; | |
6627 | if (conf) | |
6628 | return sprintf(page, "%d\n", conf->rmw_level); | |
6629 | else | |
6630 | return 0; | |
6631 | } | |
6632 | ||
6633 | static ssize_t | |
6634 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6635 | { | |
6636 | struct r5conf *conf = mddev->private; | |
6637 | unsigned long new; | |
6638 | ||
6639 | if (!conf) | |
6640 | return -ENODEV; | |
6641 | ||
6642 | if (len >= PAGE_SIZE) | |
6643 | return -EINVAL; | |
6644 | ||
6645 | if (kstrtoul(page, 10, &new)) | |
6646 | return -EINVAL; | |
6647 | ||
6648 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6649 | return -EINVAL; | |
6650 | ||
6651 | if (new != PARITY_DISABLE_RMW && | |
6652 | new != PARITY_ENABLE_RMW && | |
6653 | new != PARITY_PREFER_RMW) | |
6654 | return -EINVAL; | |
6655 | ||
6656 | conf->rmw_level = new; | |
6657 | return len; | |
6658 | } | |
6659 | ||
6660 | static struct md_sysfs_entry | |
6661 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6662 | raid5_show_rmw_level, | |
6663 | raid5_store_rmw_level); | |
6664 | ||
3b5408b9 YY |
6665 | static ssize_t |
6666 | raid5_show_stripe_size(struct mddev *mddev, char *page) | |
6667 | { | |
6668 | struct r5conf *conf; | |
6669 | int ret = 0; | |
6670 | ||
6671 | spin_lock(&mddev->lock); | |
6672 | conf = mddev->private; | |
6673 | if (conf) | |
6674 | ret = sprintf(page, "%lu\n", RAID5_STRIPE_SIZE(conf)); | |
6675 | spin_unlock(&mddev->lock); | |
6676 | return ret; | |
6677 | } | |
6678 | ||
6679 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE | |
6680 | static ssize_t | |
6681 | raid5_store_stripe_size(struct mddev *mddev, const char *page, size_t len) | |
6682 | { | |
6683 | struct r5conf *conf; | |
6684 | unsigned long new; | |
6685 | int err; | |
38912584 | 6686 | int size; |
3b5408b9 YY |
6687 | |
6688 | if (len >= PAGE_SIZE) | |
6689 | return -EINVAL; | |
6690 | if (kstrtoul(page, 10, &new)) | |
6691 | return -EINVAL; | |
6692 | ||
6693 | /* | |
6694 | * The value should not be bigger than PAGE_SIZE. It requires to | |
6af10a33 YY |
6695 | * be multiple of DEFAULT_STRIPE_SIZE and the value should be power |
6696 | * of two. | |
3b5408b9 | 6697 | */ |
6af10a33 YY |
6698 | if (new % DEFAULT_STRIPE_SIZE != 0 || |
6699 | new > PAGE_SIZE || new == 0 || | |
6700 | new != roundup_pow_of_two(new)) | |
3b5408b9 YY |
6701 | return -EINVAL; |
6702 | ||
6703 | err = mddev_lock(mddev); | |
6704 | if (err) | |
6705 | return err; | |
6706 | ||
6707 | conf = mddev->private; | |
6708 | if (!conf) { | |
6709 | err = -ENODEV; | |
6710 | goto out_unlock; | |
6711 | } | |
6712 | ||
6713 | if (new == conf->stripe_size) | |
6714 | goto out_unlock; | |
6715 | ||
6716 | pr_debug("md/raid: change stripe_size from %lu to %lu\n", | |
6717 | conf->stripe_size, new); | |
6718 | ||
38912584 YY |
6719 | if (mddev->sync_thread || |
6720 | test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) || | |
6721 | mddev->reshape_position != MaxSector || | |
6722 | mddev->sysfs_active) { | |
6723 | err = -EBUSY; | |
6724 | goto out_unlock; | |
6725 | } | |
6726 | ||
3b5408b9 | 6727 | mddev_suspend(mddev); |
38912584 YY |
6728 | mutex_lock(&conf->cache_size_mutex); |
6729 | size = conf->max_nr_stripes; | |
6730 | ||
6731 | shrink_stripes(conf); | |
6732 | ||
3b5408b9 YY |
6733 | conf->stripe_size = new; |
6734 | conf->stripe_shift = ilog2(new) - 9; | |
6735 | conf->stripe_sectors = new >> 9; | |
38912584 YY |
6736 | if (grow_stripes(conf, size)) { |
6737 | pr_warn("md/raid:%s: couldn't allocate buffers\n", | |
6738 | mdname(mddev)); | |
6739 | err = -ENOMEM; | |
6740 | } | |
6741 | mutex_unlock(&conf->cache_size_mutex); | |
3b5408b9 YY |
6742 | mddev_resume(mddev); |
6743 | ||
6744 | out_unlock: | |
6745 | mddev_unlock(mddev); | |
6746 | return err ?: len; | |
6747 | } | |
6748 | ||
6749 | static struct md_sysfs_entry | |
6750 | raid5_stripe_size = __ATTR(stripe_size, 0644, | |
6751 | raid5_show_stripe_size, | |
6752 | raid5_store_stripe_size); | |
6753 | #else | |
6754 | static struct md_sysfs_entry | |
6755 | raid5_stripe_size = __ATTR(stripe_size, 0444, | |
6756 | raid5_show_stripe_size, | |
6757 | NULL); | |
6758 | #endif | |
d06f191f | 6759 | |
8b3e6cdc | 6760 | static ssize_t |
fd01b88c | 6761 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6762 | { |
7b1485ba N |
6763 | struct r5conf *conf; |
6764 | int ret = 0; | |
6765 | spin_lock(&mddev->lock); | |
6766 | conf = mddev->private; | |
8b3e6cdc | 6767 | if (conf) |
7b1485ba N |
6768 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6769 | spin_unlock(&mddev->lock); | |
6770 | return ret; | |
8b3e6cdc DW |
6771 | } |
6772 | ||
6773 | static ssize_t | |
fd01b88c | 6774 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6775 | { |
6791875e | 6776 | struct r5conf *conf; |
4ef197d8 | 6777 | unsigned long new; |
6791875e N |
6778 | int err; |
6779 | ||
8b3e6cdc DW |
6780 | if (len >= PAGE_SIZE) |
6781 | return -EINVAL; | |
b29bebd6 | 6782 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6783 | return -EINVAL; |
6791875e N |
6784 | |
6785 | err = mddev_lock(mddev); | |
6786 | if (err) | |
6787 | return err; | |
6788 | conf = mddev->private; | |
6789 | if (!conf) | |
6790 | err = -ENODEV; | |
edbe83ab | 6791 | else if (new > conf->min_nr_stripes) |
6791875e N |
6792 | err = -EINVAL; |
6793 | else | |
6794 | conf->bypass_threshold = new; | |
6795 | mddev_unlock(mddev); | |
6796 | return err ?: len; | |
8b3e6cdc DW |
6797 | } |
6798 | ||
6799 | static struct md_sysfs_entry | |
6800 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6801 | S_IRUGO | S_IWUSR, | |
6802 | raid5_show_preread_threshold, | |
6803 | raid5_store_preread_threshold); | |
6804 | ||
d592a996 SL |
6805 | static ssize_t |
6806 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6807 | { | |
7b1485ba N |
6808 | struct r5conf *conf; |
6809 | int ret = 0; | |
6810 | spin_lock(&mddev->lock); | |
6811 | conf = mddev->private; | |
d592a996 | 6812 | if (conf) |
7b1485ba N |
6813 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6814 | spin_unlock(&mddev->lock); | |
6815 | return ret; | |
d592a996 SL |
6816 | } |
6817 | ||
6818 | static ssize_t | |
6819 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6820 | { | |
6791875e | 6821 | struct r5conf *conf; |
d592a996 | 6822 | unsigned long new; |
6791875e N |
6823 | int err; |
6824 | ||
d592a996 SL |
6825 | if (len >= PAGE_SIZE) |
6826 | return -EINVAL; | |
d592a996 SL |
6827 | if (kstrtoul(page, 10, &new)) |
6828 | return -EINVAL; | |
6829 | new = !!new; | |
6791875e N |
6830 | |
6831 | err = mddev_lock(mddev); | |
6832 | if (err) | |
6833 | return err; | |
6834 | conf = mddev->private; | |
6835 | if (!conf) | |
6836 | err = -ENODEV; | |
6837 | else if (new != conf->skip_copy) { | |
1cb039f3 CH |
6838 | struct request_queue *q = mddev->queue; |
6839 | ||
6791875e N |
6840 | mddev_suspend(mddev); |
6841 | conf->skip_copy = new; | |
6842 | if (new) | |
1cb039f3 | 6843 | blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q); |
6791875e | 6844 | else |
1cb039f3 | 6845 | blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q); |
6791875e N |
6846 | mddev_resume(mddev); |
6847 | } | |
6848 | mddev_unlock(mddev); | |
6849 | return err ?: len; | |
d592a996 SL |
6850 | } |
6851 | ||
6852 | static struct md_sysfs_entry | |
6853 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6854 | raid5_show_skip_copy, | |
6855 | raid5_store_skip_copy); | |
6856 | ||
3f294f4f | 6857 | static ssize_t |
fd01b88c | 6858 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6859 | { |
d1688a6d | 6860 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6861 | if (conf) |
6862 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6863 | else | |
6864 | return 0; | |
3f294f4f N |
6865 | } |
6866 | ||
96de1e66 N |
6867 | static struct md_sysfs_entry |
6868 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6869 | |
b721420e SL |
6870 | static ssize_t |
6871 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6872 | { | |
7b1485ba N |
6873 | struct r5conf *conf; |
6874 | int ret = 0; | |
6875 | spin_lock(&mddev->lock); | |
6876 | conf = mddev->private; | |
b721420e | 6877 | if (conf) |
7b1485ba N |
6878 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6879 | spin_unlock(&mddev->lock); | |
6880 | return ret; | |
b721420e SL |
6881 | } |
6882 | ||
60aaf933 | 6883 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6884 | int *group_cnt, | |
60aaf933 | 6885 | struct r5worker_group **worker_groups); |
b721420e SL |
6886 | static ssize_t |
6887 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6888 | { | |
6791875e | 6889 | struct r5conf *conf; |
7d5d7b50 | 6890 | unsigned int new; |
b721420e | 6891 | int err; |
60aaf933 | 6892 | struct r5worker_group *new_groups, *old_groups; |
d2c9ad41 | 6893 | int group_cnt; |
b721420e SL |
6894 | |
6895 | if (len >= PAGE_SIZE) | |
6896 | return -EINVAL; | |
7d5d7b50 SL |
6897 | if (kstrtouint(page, 10, &new)) |
6898 | return -EINVAL; | |
6899 | /* 8192 should be big enough */ | |
6900 | if (new > 8192) | |
b721420e SL |
6901 | return -EINVAL; |
6902 | ||
6791875e N |
6903 | err = mddev_lock(mddev); |
6904 | if (err) | |
6905 | return err; | |
6906 | conf = mddev->private; | |
6907 | if (!conf) | |
6908 | err = -ENODEV; | |
6909 | else if (new != conf->worker_cnt_per_group) { | |
6910 | mddev_suspend(mddev); | |
b721420e | 6911 | |
6791875e N |
6912 | old_groups = conf->worker_groups; |
6913 | if (old_groups) | |
6914 | flush_workqueue(raid5_wq); | |
d206dcfa | 6915 | |
d2c9ad41 | 6916 | err = alloc_thread_groups(conf, new, &group_cnt, &new_groups); |
6791875e N |
6917 | if (!err) { |
6918 | spin_lock_irq(&conf->device_lock); | |
6919 | conf->group_cnt = group_cnt; | |
d2c9ad41 | 6920 | conf->worker_cnt_per_group = new; |
6791875e N |
6921 | conf->worker_groups = new_groups; |
6922 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6923 | |
6791875e N |
6924 | if (old_groups) |
6925 | kfree(old_groups[0].workers); | |
6926 | kfree(old_groups); | |
6927 | } | |
6928 | mddev_resume(mddev); | |
b721420e | 6929 | } |
6791875e | 6930 | mddev_unlock(mddev); |
b721420e | 6931 | |
6791875e | 6932 | return err ?: len; |
b721420e SL |
6933 | } |
6934 | ||
6935 | static struct md_sysfs_entry | |
6936 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6937 | raid5_show_group_thread_cnt, | |
6938 | raid5_store_group_thread_cnt); | |
6939 | ||
007583c9 | 6940 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6941 | &raid5_stripecache_size.attr, |
6942 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6943 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6944 | &raid5_group_thread_cnt.attr, |
d592a996 | 6945 | &raid5_skip_copy.attr, |
d06f191f | 6946 | &raid5_rmw_level.attr, |
3b5408b9 | 6947 | &raid5_stripe_size.attr, |
2c7da14b | 6948 | &r5c_journal_mode.attr, |
a596d086 | 6949 | &ppl_write_hint.attr, |
3f294f4f N |
6950 | NULL, |
6951 | }; | |
c32dc040 | 6952 | static const struct attribute_group raid5_attrs_group = { |
007583c9 N |
6953 | .name = NULL, |
6954 | .attrs = raid5_attrs, | |
3f294f4f N |
6955 | }; |
6956 | ||
d2c9ad41 | 6957 | static int alloc_thread_groups(struct r5conf *conf, int cnt, int *group_cnt, |
60aaf933 | 6958 | struct r5worker_group **worker_groups) |
851c30c9 | 6959 | { |
566c09c5 | 6960 | int i, j, k; |
851c30c9 SL |
6961 | ssize_t size; |
6962 | struct r5worker *workers; | |
6963 | ||
851c30c9 | 6964 | if (cnt == 0) { |
60aaf933 | 6965 | *group_cnt = 0; |
6966 | *worker_groups = NULL; | |
851c30c9 SL |
6967 | return 0; |
6968 | } | |
60aaf933 | 6969 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6970 | size = sizeof(struct r5worker) * cnt; |
6396bb22 KC |
6971 | workers = kcalloc(size, *group_cnt, GFP_NOIO); |
6972 | *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group), | |
6973 | GFP_NOIO); | |
60aaf933 | 6974 | if (!*worker_groups || !workers) { |
851c30c9 | 6975 | kfree(workers); |
60aaf933 | 6976 | kfree(*worker_groups); |
851c30c9 SL |
6977 | return -ENOMEM; |
6978 | } | |
6979 | ||
60aaf933 | 6980 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6981 | struct r5worker_group *group; |
6982 | ||
0c775d52 | 6983 | group = &(*worker_groups)[i]; |
851c30c9 | 6984 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6985 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6986 | group->conf = conf; |
6987 | group->workers = workers + i * cnt; | |
6988 | ||
6989 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6990 | struct r5worker *worker = group->workers + j; |
6991 | worker->group = group; | |
6992 | INIT_WORK(&worker->work, raid5_do_work); | |
6993 | ||
6994 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6995 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6996 | } |
6997 | } | |
6998 | ||
6999 | return 0; | |
7000 | } | |
7001 | ||
7002 | static void free_thread_groups(struct r5conf *conf) | |
7003 | { | |
7004 | if (conf->worker_groups) | |
7005 | kfree(conf->worker_groups[0].workers); | |
7006 | kfree(conf->worker_groups); | |
7007 | conf->worker_groups = NULL; | |
7008 | } | |
7009 | ||
80c3a6ce | 7010 | static sector_t |
fd01b88c | 7011 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 7012 | { |
d1688a6d | 7013 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
7014 | |
7015 | if (!sectors) | |
7016 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 7017 | if (!raid_disks) |
7ec05478 | 7018 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 7019 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 7020 | |
3cb5edf4 N |
7021 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
7022 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
7023 | return sectors * (raid_disks - conf->max_degraded); |
7024 | } | |
7025 | ||
789b5e03 ON |
7026 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
7027 | { | |
7028 | safe_put_page(percpu->spare_page); | |
789b5e03 | 7029 | percpu->spare_page = NULL; |
b330e6a4 | 7030 | kvfree(percpu->scribble); |
789b5e03 ON |
7031 | percpu->scribble = NULL; |
7032 | } | |
7033 | ||
7034 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
7035 | { | |
b330e6a4 | 7036 | if (conf->level == 6 && !percpu->spare_page) { |
789b5e03 | 7037 | percpu->spare_page = alloc_page(GFP_KERNEL); |
b330e6a4 KO |
7038 | if (!percpu->spare_page) |
7039 | return -ENOMEM; | |
7040 | } | |
7041 | ||
7042 | if (scribble_alloc(percpu, | |
7043 | max(conf->raid_disks, | |
7044 | conf->previous_raid_disks), | |
7045 | max(conf->chunk_sectors, | |
7046 | conf->prev_chunk_sectors) | |
c911c46c | 7047 | / RAID5_STRIPE_SECTORS(conf))) { |
789b5e03 ON |
7048 | free_scratch_buffer(conf, percpu); |
7049 | return -ENOMEM; | |
7050 | } | |
7051 | ||
7052 | return 0; | |
7053 | } | |
7054 | ||
29c6d1bb | 7055 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 7056 | { |
29c6d1bb SAS |
7057 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
7058 | ||
7059 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
7060 | return 0; | |
7061 | } | |
36d1c647 | 7062 | |
29c6d1bb SAS |
7063 | static void raid5_free_percpu(struct r5conf *conf) |
7064 | { | |
36d1c647 DW |
7065 | if (!conf->percpu) |
7066 | return; | |
7067 | ||
29c6d1bb | 7068 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
7069 | free_percpu(conf->percpu); |
7070 | } | |
7071 | ||
d1688a6d | 7072 | static void free_conf(struct r5conf *conf) |
95fc17aa | 7073 | { |
d7bd398e SL |
7074 | int i; |
7075 | ||
ff875738 AP |
7076 | log_exit(conf); |
7077 | ||
565e0450 | 7078 | unregister_shrinker(&conf->shrinker); |
851c30c9 | 7079 | free_thread_groups(conf); |
95fc17aa | 7080 | shrink_stripes(conf); |
36d1c647 | 7081 | raid5_free_percpu(conf); |
d7bd398e SL |
7082 | for (i = 0; i < conf->pool_size; i++) |
7083 | if (conf->disks[i].extra_page) | |
7084 | put_page(conf->disks[i].extra_page); | |
95fc17aa | 7085 | kfree(conf->disks); |
afeee514 | 7086 | bioset_exit(&conf->bio_split); |
95fc17aa | 7087 | kfree(conf->stripe_hashtbl); |
aaf9f12e | 7088 | kfree(conf->pending_data); |
95fc17aa DW |
7089 | kfree(conf); |
7090 | } | |
7091 | ||
29c6d1bb | 7092 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 7093 | { |
29c6d1bb | 7094 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
7095 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
7096 | ||
29c6d1bb | 7097 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
7098 | pr_warn("%s: failed memory allocation for cpu%u\n", |
7099 | __func__, cpu); | |
29c6d1bb | 7100 | return -ENOMEM; |
36d1c647 | 7101 | } |
29c6d1bb | 7102 | return 0; |
36d1c647 | 7103 | } |
36d1c647 | 7104 | |
d1688a6d | 7105 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 7106 | { |
789b5e03 | 7107 | int err = 0; |
36d1c647 | 7108 | |
789b5e03 ON |
7109 | conf->percpu = alloc_percpu(struct raid5_percpu); |
7110 | if (!conf->percpu) | |
36d1c647 | 7111 | return -ENOMEM; |
789b5e03 | 7112 | |
29c6d1bb | 7113 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
7114 | if (!err) { |
7115 | conf->scribble_disks = max(conf->raid_disks, | |
7116 | conf->previous_raid_disks); | |
7117 | conf->scribble_sectors = max(conf->chunk_sectors, | |
7118 | conf->prev_chunk_sectors); | |
7119 | } | |
36d1c647 DW |
7120 | return err; |
7121 | } | |
7122 | ||
edbe83ab N |
7123 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
7124 | struct shrink_control *sc) | |
7125 | { | |
7126 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
7127 | unsigned long ret = SHRINK_STOP; |
7128 | ||
7129 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
7130 | ret= 0; | |
49895bcc N |
7131 | while (ret < sc->nr_to_scan && |
7132 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
7133 | if (drop_one_stripe(conf) == 0) { |
7134 | ret = SHRINK_STOP; | |
7135 | break; | |
7136 | } | |
7137 | ret++; | |
7138 | } | |
7139 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
7140 | } |
7141 | return ret; | |
7142 | } | |
7143 | ||
7144 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
7145 | struct shrink_control *sc) | |
7146 | { | |
7147 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
7148 | ||
7149 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
7150 | /* unlikely, but not impossible */ | |
7151 | return 0; | |
7152 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
7153 | } | |
7154 | ||
d1688a6d | 7155 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 7156 | { |
d1688a6d | 7157 | struct r5conf *conf; |
5e5e3e78 | 7158 | int raid_disk, memory, max_disks; |
3cb03002 | 7159 | struct md_rdev *rdev; |
1da177e4 | 7160 | struct disk_info *disk; |
0232605d | 7161 | char pers_name[6]; |
566c09c5 | 7162 | int i; |
d2c9ad41 | 7163 | int group_cnt; |
60aaf933 | 7164 | struct r5worker_group *new_group; |
afeee514 | 7165 | int ret; |
1da177e4 | 7166 | |
91adb564 N |
7167 | if (mddev->new_level != 5 |
7168 | && mddev->new_level != 4 | |
7169 | && mddev->new_level != 6) { | |
cc6167b4 N |
7170 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
7171 | mdname(mddev), mddev->new_level); | |
91adb564 | 7172 | return ERR_PTR(-EIO); |
1da177e4 | 7173 | } |
91adb564 N |
7174 | if ((mddev->new_level == 5 |
7175 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
7176 | (mddev->new_level == 6 | |
7177 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
7178 | pr_warn("md/raid:%s: layout %d not supported\n", |
7179 | mdname(mddev), mddev->new_layout); | |
91adb564 | 7180 | return ERR_PTR(-EIO); |
99c0fb5f | 7181 | } |
91adb564 | 7182 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
7183 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
7184 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 7185 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
7186 | } |
7187 | ||
664e7c41 AN |
7188 | if (!mddev->new_chunk_sectors || |
7189 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
7190 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
7191 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
7192 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 7193 | return ERR_PTR(-EINVAL); |
f6705578 N |
7194 | } |
7195 | ||
d1688a6d | 7196 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 7197 | if (conf == NULL) |
1da177e4 | 7198 | goto abort; |
c911c46c | 7199 | |
e2368582 YY |
7200 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
7201 | conf->stripe_size = DEFAULT_STRIPE_SIZE; | |
7202 | conf->stripe_shift = ilog2(DEFAULT_STRIPE_SIZE) - 9; | |
7203 | conf->stripe_sectors = DEFAULT_STRIPE_SIZE >> 9; | |
7204 | #endif | |
aaf9f12e SL |
7205 | INIT_LIST_HEAD(&conf->free_list); |
7206 | INIT_LIST_HEAD(&conf->pending_list); | |
6396bb22 KC |
7207 | conf->pending_data = kcalloc(PENDING_IO_MAX, |
7208 | sizeof(struct r5pending_data), | |
7209 | GFP_KERNEL); | |
aaf9f12e SL |
7210 | if (!conf->pending_data) |
7211 | goto abort; | |
7212 | for (i = 0; i < PENDING_IO_MAX; i++) | |
7213 | list_add(&conf->pending_data[i].sibling, &conf->free_list); | |
851c30c9 | 7214 | /* Don't enable multi-threading by default*/ |
d2c9ad41 | 7215 | if (!alloc_thread_groups(conf, 0, &group_cnt, &new_group)) { |
60aaf933 | 7216 | conf->group_cnt = group_cnt; |
d2c9ad41 | 7217 | conf->worker_cnt_per_group = 0; |
60aaf933 | 7218 | conf->worker_groups = new_group; |
7219 | } else | |
851c30c9 | 7220 | goto abort; |
f5efd45a | 7221 | spin_lock_init(&conf->device_lock); |
0a87b25f | 7222 | seqcount_spinlock_init(&conf->gen_lock, &conf->device_lock); |
2d5b569b | 7223 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 7224 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 7225 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
7226 | init_waitqueue_head(&conf->wait_for_overlap); |
7227 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 7228 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
7229 | INIT_LIST_HEAD(&conf->hold_list); |
7230 | INIT_LIST_HEAD(&conf->delayed_list); | |
7231 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 7232 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
7233 | atomic_set(&conf->active_stripes, 0); |
7234 | atomic_set(&conf->preread_active_stripes, 0); | |
7235 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
7236 | spin_lock_init(&conf->pending_bios_lock); |
7237 | conf->batch_bio_dispatch = true; | |
7238 | rdev_for_each(rdev, mddev) { | |
7239 | if (test_bit(Journal, &rdev->flags)) | |
7240 | continue; | |
7241 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
7242 | conf->batch_bio_dispatch = false; | |
7243 | break; | |
7244 | } | |
7245 | } | |
7246 | ||
f5efd45a | 7247 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 7248 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
7249 | |
7250 | conf->raid_disks = mddev->raid_disks; | |
7251 | if (mddev->reshape_position == MaxSector) | |
7252 | conf->previous_raid_disks = mddev->raid_disks; | |
7253 | else | |
f6705578 | 7254 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 7255 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 7256 | |
6396bb22 | 7257 | conf->disks = kcalloc(max_disks, sizeof(struct disk_info), |
b55e6bfc | 7258 | GFP_KERNEL); |
d7bd398e | 7259 | |
b55e6bfc N |
7260 | if (!conf->disks) |
7261 | goto abort; | |
9ffae0cf | 7262 | |
d7bd398e SL |
7263 | for (i = 0; i < max_disks; i++) { |
7264 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
7265 | if (!conf->disks[i].extra_page) | |
7266 | goto abort; | |
7267 | } | |
7268 | ||
afeee514 KO |
7269 | ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
7270 | if (ret) | |
dd7a8f5d | 7271 | goto abort; |
1da177e4 LT |
7272 | conf->mddev = mddev; |
7273 | ||
fccddba0 | 7274 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 7275 | goto abort; |
1da177e4 | 7276 | |
566c09c5 SL |
7277 | /* We init hash_locks[0] separately to that it can be used |
7278 | * as the reference lock in the spin_lock_nest_lock() call | |
7279 | * in lock_all_device_hash_locks_irq in order to convince | |
7280 | * lockdep that we know what we are doing. | |
7281 | */ | |
7282 | spin_lock_init(conf->hash_locks); | |
7283 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
7284 | spin_lock_init(conf->hash_locks + i); | |
7285 | ||
7286 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
7287 | INIT_LIST_HEAD(conf->inactive_list + i); | |
7288 | ||
7289 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
7290 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
7291 | ||
1e6d690b SL |
7292 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
7293 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
7294 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
7295 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
7296 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
7297 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 7298 | |
36d1c647 | 7299 | conf->level = mddev->new_level; |
46d5b785 | 7300 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
7301 | if (raid5_alloc_percpu(conf) != 0) |
7302 | goto abort; | |
7303 | ||
0c55e022 | 7304 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 7305 | |
dafb20fa | 7306 | rdev_for_each(rdev, mddev) { |
1da177e4 | 7307 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 7308 | if (raid_disk >= max_disks |
f2076e7d | 7309 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
7310 | continue; |
7311 | disk = conf->disks + raid_disk; | |
7312 | ||
17045f52 N |
7313 | if (test_bit(Replacement, &rdev->flags)) { |
7314 | if (disk->replacement) | |
7315 | goto abort; | |
7316 | disk->replacement = rdev; | |
7317 | } else { | |
7318 | if (disk->rdev) | |
7319 | goto abort; | |
7320 | disk->rdev = rdev; | |
7321 | } | |
1da177e4 | 7322 | |
b2d444d7 | 7323 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 7324 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
7325 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
7326 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 7327 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
7328 | /* Cannot rely on bitmap to complete recovery */ |
7329 | conf->fullsync = 1; | |
1da177e4 LT |
7330 | } |
7331 | ||
91adb564 | 7332 | conf->level = mddev->new_level; |
584acdd4 | 7333 | if (conf->level == 6) { |
16a53ecc | 7334 | conf->max_degraded = 2; |
584acdd4 MS |
7335 | if (raid6_call.xor_syndrome) |
7336 | conf->rmw_level = PARITY_ENABLE_RMW; | |
7337 | else | |
7338 | conf->rmw_level = PARITY_DISABLE_RMW; | |
7339 | } else { | |
16a53ecc | 7340 | conf->max_degraded = 1; |
584acdd4 MS |
7341 | conf->rmw_level = PARITY_ENABLE_RMW; |
7342 | } | |
91adb564 | 7343 | conf->algorithm = mddev->new_layout; |
fef9c61f | 7344 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 7345 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 7346 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 7347 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
7348 | } else { |
7349 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
7350 | conf->prev_algo = conf->algorithm; | |
e183eaed | 7351 | } |
1da177e4 | 7352 | |
edbe83ab | 7353 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
7354 | if (mddev->reshape_position != MaxSector) { |
7355 | int stripes = max_t(int, | |
c911c46c YY |
7356 | ((mddev->chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4, |
7357 | ((mddev->new_chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4); | |
ad5b0f76 SL |
7358 | conf->min_nr_stripes = max(NR_STRIPES, stripes); |
7359 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 7360 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
7361 | mdname(mddev), conf->min_nr_stripes); |
7362 | } | |
edbe83ab | 7363 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 7364 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 7365 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 7366 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
7367 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
7368 | mdname(mddev), memory); | |
91adb564 N |
7369 | goto abort; |
7370 | } else | |
cc6167b4 | 7371 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
7372 | /* |
7373 | * Losing a stripe head costs more than the time to refill it, | |
7374 | * it reduces the queue depth and so can hurt throughput. | |
7375 | * So set it rather large, scaled by number of devices. | |
7376 | */ | |
7377 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
7378 | conf->shrinker.scan_objects = raid5_cache_scan; | |
7379 | conf->shrinker.count_objects = raid5_cache_count; | |
7380 | conf->shrinker.batch = 128; | |
7381 | conf->shrinker.flags = 0; | |
6a0f53ff | 7382 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
7383 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
7384 | mdname(mddev)); | |
6a0f53ff CY |
7385 | goto abort; |
7386 | } | |
1da177e4 | 7387 | |
0232605d N |
7388 | sprintf(pers_name, "raid%d", mddev->new_level); |
7389 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 7390 | if (!conf->thread) { |
cc6167b4 N |
7391 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
7392 | mdname(mddev)); | |
16a53ecc N |
7393 | goto abort; |
7394 | } | |
91adb564 N |
7395 | |
7396 | return conf; | |
7397 | ||
7398 | abort: | |
7399 | if (conf) { | |
95fc17aa | 7400 | free_conf(conf); |
91adb564 N |
7401 | return ERR_PTR(-EIO); |
7402 | } else | |
7403 | return ERR_PTR(-ENOMEM); | |
7404 | } | |
7405 | ||
c148ffdc N |
7406 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
7407 | { | |
7408 | switch (algo) { | |
7409 | case ALGORITHM_PARITY_0: | |
7410 | if (raid_disk < max_degraded) | |
7411 | return 1; | |
7412 | break; | |
7413 | case ALGORITHM_PARITY_N: | |
7414 | if (raid_disk >= raid_disks - max_degraded) | |
7415 | return 1; | |
7416 | break; | |
7417 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 7418 | if (raid_disk == 0 || |
c148ffdc N |
7419 | raid_disk == raid_disks - 1) |
7420 | return 1; | |
7421 | break; | |
7422 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7423 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7424 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7425 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7426 | if (raid_disk == raid_disks - 1) | |
7427 | return 1; | |
7428 | } | |
7429 | return 0; | |
7430 | } | |
7431 | ||
16ef5101 CH |
7432 | static void raid5_set_io_opt(struct r5conf *conf) |
7433 | { | |
7434 | blk_queue_io_opt(conf->mddev->queue, (conf->chunk_sectors << 9) * | |
7435 | (conf->raid_disks - conf->max_degraded)); | |
7436 | } | |
7437 | ||
849674e4 | 7438 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7439 | { |
d1688a6d | 7440 | struct r5conf *conf; |
9f7c2220 | 7441 | int working_disks = 0; |
c148ffdc | 7442 | int dirty_parity_disks = 0; |
3cb03002 | 7443 | struct md_rdev *rdev; |
713cf5a6 | 7444 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7445 | sector_t reshape_offset = 0; |
306d8015 | 7446 | int i, ret = 0; |
b5254dd5 N |
7447 | long long min_offset_diff = 0; |
7448 | int first = 1; | |
91adb564 | 7449 | |
306d8015 XN |
7450 | if (acct_bioset_init(mddev)) { |
7451 | pr_err("md/raid456:%s: alloc acct bioset failed.\n", mdname(mddev)); | |
a415c0f1 | 7452 | return -ENOMEM; |
306d8015 XN |
7453 | } |
7454 | ||
7455 | if (mddev_init_writes_pending(mddev) < 0) { | |
7456 | ret = -ENOMEM; | |
7457 | goto exit_acct_set; | |
7458 | } | |
a415c0f1 | 7459 | |
8c6ac868 | 7460 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7461 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7462 | mdname(mddev)); | |
b5254dd5 N |
7463 | |
7464 | rdev_for_each(rdev, mddev) { | |
7465 | long long diff; | |
713cf5a6 | 7466 | |
f2076e7d | 7467 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7468 | journal_dev = rdev; |
f2076e7d SL |
7469 | continue; |
7470 | } | |
b5254dd5 N |
7471 | if (rdev->raid_disk < 0) |
7472 | continue; | |
7473 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7474 | if (first) { | |
7475 | min_offset_diff = diff; | |
7476 | first = 0; | |
7477 | } else if (mddev->reshape_backwards && | |
7478 | diff < min_offset_diff) | |
7479 | min_offset_diff = diff; | |
7480 | else if (!mddev->reshape_backwards && | |
7481 | diff > min_offset_diff) | |
7482 | min_offset_diff = diff; | |
7483 | } | |
7484 | ||
230b55fa N |
7485 | if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) && |
7486 | (mddev->bitmap_info.offset || mddev->bitmap_info.file)) { | |
7487 | pr_notice("md/raid:%s: array cannot have both journal and bitmap\n", | |
7488 | mdname(mddev)); | |
306d8015 XN |
7489 | ret = -EINVAL; |
7490 | goto exit_acct_set; | |
230b55fa N |
7491 | } |
7492 | ||
91adb564 N |
7493 | if (mddev->reshape_position != MaxSector) { |
7494 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7495 | * Difficulties arise if the stripe we would write to |
7496 | * next is at or after the stripe we would read from next. | |
7497 | * For a reshape that changes the number of devices, this | |
7498 | * is only possible for a very short time, and mdadm makes | |
7499 | * sure that time appears to have past before assembling | |
7500 | * the array. So we fail if that time hasn't passed. | |
7501 | * For a reshape that keeps the number of devices the same | |
7502 | * mdadm must be monitoring the reshape can keeping the | |
7503 | * critical areas read-only and backed up. It will start | |
7504 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7505 | */ |
7506 | sector_t here_new, here_old; | |
7507 | int old_disks; | |
18b00334 | 7508 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7509 | int chunk_sectors; |
7510 | int new_data_disks; | |
91adb564 | 7511 | |
713cf5a6 | 7512 | if (journal_dev) { |
cc6167b4 N |
7513 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7514 | mdname(mddev)); | |
306d8015 XN |
7515 | ret = -EINVAL; |
7516 | goto exit_acct_set; | |
713cf5a6 SL |
7517 | } |
7518 | ||
88ce4930 | 7519 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7520 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7521 | mdname(mddev)); | |
306d8015 XN |
7522 | ret = -EINVAL; |
7523 | goto exit_acct_set; | |
91adb564 | 7524 | } |
91adb564 N |
7525 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7526 | /* reshape_position must be on a new-stripe boundary, and one | |
7527 | * further up in new geometry must map after here in old | |
7528 | * geometry. | |
05256d98 N |
7529 | * If the chunk sizes are different, then as we perform reshape |
7530 | * in units of the largest of the two, reshape_position needs | |
7531 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7532 | */ |
7533 | here_new = mddev->reshape_position; | |
05256d98 N |
7534 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7535 | new_data_disks = mddev->raid_disks - max_degraded; | |
7536 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7537 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7538 | mdname(mddev)); | |
306d8015 XN |
7539 | ret = -EINVAL; |
7540 | goto exit_acct_set; | |
91adb564 | 7541 | } |
05256d98 | 7542 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7543 | /* here_new is the stripe we will write to */ |
7544 | here_old = mddev->reshape_position; | |
05256d98 | 7545 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7546 | /* here_old is the first stripe that we might need to read |
7547 | * from */ | |
67ac6011 N |
7548 | if (mddev->delta_disks == 0) { |
7549 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7550 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7551 | * and taking constant backups. |
7552 | * mdadm always starts a situation like this in | |
7553 | * readonly mode so it can take control before | |
7554 | * allowing any writes. So just check for that. | |
7555 | */ | |
b5254dd5 N |
7556 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7557 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7558 | /* not really in-place - so OK */; | |
7559 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7560 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7561 | mdname(mddev)); | |
306d8015 XN |
7562 | ret = -EINVAL; |
7563 | goto exit_acct_set; | |
67ac6011 | 7564 | } |
2c810cdd | 7565 | } else if (mddev->reshape_backwards |
05256d98 N |
7566 | ? (here_new * chunk_sectors + min_offset_diff <= |
7567 | here_old * chunk_sectors) | |
7568 | : (here_new * chunk_sectors >= | |
7569 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7570 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7571 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7572 | mdname(mddev)); | |
306d8015 XN |
7573 | ret = -EINVAL; |
7574 | goto exit_acct_set; | |
91adb564 | 7575 | } |
cc6167b4 | 7576 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7577 | /* OK, we should be able to continue; */ |
7578 | } else { | |
7579 | BUG_ON(mddev->level != mddev->new_level); | |
7580 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7581 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7582 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7583 | } |
91adb564 | 7584 | |
3418d036 AP |
7585 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && |
7586 | test_bit(MD_HAS_PPL, &mddev->flags)) { | |
7587 | pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n", | |
7588 | mdname(mddev)); | |
7589 | clear_bit(MD_HAS_PPL, &mddev->flags); | |
ddc08823 | 7590 | clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags); |
3418d036 AP |
7591 | } |
7592 | ||
245f46c2 N |
7593 | if (mddev->private == NULL) |
7594 | conf = setup_conf(mddev); | |
7595 | else | |
7596 | conf = mddev->private; | |
7597 | ||
306d8015 XN |
7598 | if (IS_ERR(conf)) { |
7599 | ret = PTR_ERR(conf); | |
7600 | goto exit_acct_set; | |
7601 | } | |
91adb564 | 7602 | |
486b0f7b SL |
7603 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7604 | if (!journal_dev) { | |
cc6167b4 N |
7605 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7606 | mdname(mddev)); | |
486b0f7b SL |
7607 | mddev->ro = 1; |
7608 | set_disk_ro(mddev->gendisk, 1); | |
7609 | } else if (mddev->recovery_cp == MaxSector) | |
7610 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7611 | } |
7612 | ||
b5254dd5 | 7613 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7614 | mddev->thread = conf->thread; |
7615 | conf->thread = NULL; | |
7616 | mddev->private = conf; | |
7617 | ||
17045f52 N |
7618 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7619 | i++) { | |
7620 | rdev = conf->disks[i].rdev; | |
7621 | if (!rdev && conf->disks[i].replacement) { | |
7622 | /* The replacement is all we have yet */ | |
7623 | rdev = conf->disks[i].replacement; | |
7624 | conf->disks[i].replacement = NULL; | |
7625 | clear_bit(Replacement, &rdev->flags); | |
7626 | conf->disks[i].rdev = rdev; | |
7627 | } | |
7628 | if (!rdev) | |
c148ffdc | 7629 | continue; |
17045f52 N |
7630 | if (conf->disks[i].replacement && |
7631 | conf->reshape_progress != MaxSector) { | |
7632 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7633 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7634 | goto abort; |
7635 | } | |
2f115882 | 7636 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7637 | working_disks++; |
2f115882 N |
7638 | continue; |
7639 | } | |
c148ffdc N |
7640 | /* This disc is not fully in-sync. However if it |
7641 | * just stored parity (beyond the recovery_offset), | |
7642 | * when we don't need to be concerned about the | |
7643 | * array being dirty. | |
7644 | * When reshape goes 'backwards', we never have | |
7645 | * partially completed devices, so we only need | |
7646 | * to worry about reshape going forwards. | |
7647 | */ | |
7648 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7649 | if (mddev->major_version == 0 && | |
7650 | mddev->minor_version > 90) | |
7651 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7652 | |
c148ffdc N |
7653 | if (rdev->recovery_offset < reshape_offset) { |
7654 | /* We need to check old and new layout */ | |
7655 | if (!only_parity(rdev->raid_disk, | |
7656 | conf->algorithm, | |
7657 | conf->raid_disks, | |
7658 | conf->max_degraded)) | |
7659 | continue; | |
7660 | } | |
7661 | if (!only_parity(rdev->raid_disk, | |
7662 | conf->prev_algo, | |
7663 | conf->previous_raid_disks, | |
7664 | conf->max_degraded)) | |
7665 | continue; | |
7666 | dirty_parity_disks++; | |
7667 | } | |
91adb564 | 7668 | |
17045f52 N |
7669 | /* |
7670 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7671 | */ | |
2e38a37f | 7672 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7673 | |
674806d6 | 7674 | if (has_failed(conf)) { |
cc6167b4 | 7675 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7676 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7677 | goto abort; |
7678 | } | |
7679 | ||
91adb564 | 7680 | /* device size must be a multiple of chunk size */ |
c5eec74f | 7681 | mddev->dev_sectors &= ~((sector_t)mddev->chunk_sectors - 1); |
91adb564 N |
7682 | mddev->resync_max_sectors = mddev->dev_sectors; |
7683 | ||
c148ffdc | 7684 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7685 | mddev->recovery_cp != MaxSector) { |
4536bf9b AP |
7686 | if (test_bit(MD_HAS_PPL, &mddev->flags)) |
7687 | pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n", | |
7688 | mdname(mddev)); | |
7689 | else if (mddev->ok_start_degraded) | |
cc6167b4 N |
7690 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7691 | mdname(mddev)); | |
6ff8d8ec | 7692 | else { |
cc6167b4 N |
7693 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7694 | mdname(mddev)); | |
6ff8d8ec N |
7695 | goto abort; |
7696 | } | |
1da177e4 LT |
7697 | } |
7698 | ||
cc6167b4 N |
7699 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7700 | mdname(mddev), conf->level, | |
7701 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7702 | mddev->new_layout); | |
1da177e4 LT |
7703 | |
7704 | print_raid5_conf(conf); | |
7705 | ||
fef9c61f | 7706 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7707 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7708 | atomic_set(&conf->reshape_stripes, 0); |
7709 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7710 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7711 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7712 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7713 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7714 | "reshape"); |
e406f12d AP |
7715 | if (!mddev->sync_thread) |
7716 | goto abort; | |
f6705578 N |
7717 | } |
7718 | ||
1da177e4 | 7719 | /* Ok, everything is just fine now */ |
a64c876f N |
7720 | if (mddev->to_remove == &raid5_attrs_group) |
7721 | mddev->to_remove = NULL; | |
00bcb4ac N |
7722 | else if (mddev->kobj.sd && |
7723 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7724 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7725 | mdname(mddev)); | |
4a5add49 | 7726 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7727 | |
4a5add49 | 7728 | if (mddev->queue) { |
9f7c2220 | 7729 | int chunk_size; |
4a5add49 N |
7730 | /* read-ahead size must cover two whole stripes, which |
7731 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7732 | * number of raid devices | |
7733 | */ | |
7734 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7735 | int stripe = data_disks * | |
7736 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
91adb564 | 7737 | |
9f7c2220 N |
7738 | chunk_size = mddev->chunk_sectors << 9; |
7739 | blk_queue_io_min(mddev->queue, chunk_size); | |
16ef5101 | 7740 | raid5_set_io_opt(conf); |
c78afc62 | 7741 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7742 | /* |
7743 | * We can only discard a whole stripe. It doesn't make sense to | |
7744 | * discard data disk but write parity disk | |
7745 | */ | |
7746 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7747 | /* Round up to power of 2, as discard handling |
7748 | * currently assumes that */ | |
7749 | while ((stripe-1) & stripe) | |
7750 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7751 | mddev->queue->limits.discard_alignment = stripe; |
7752 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 | 7753 | |
5026d7a9 | 7754 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 7755 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
5026d7a9 | 7756 | |
05616be5 | 7757 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7758 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7759 | rdev->data_offset << 9); | |
05616be5 N |
7760 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7761 | rdev->new_data_offset << 9); | |
7762 | } | |
620125f2 | 7763 | |
48920ff2 CH |
7764 | /* |
7765 | * zeroing is required, otherwise data | |
7766 | * could be lost. Consider a scenario: discard a stripe | |
7767 | * (the stripe could be inconsistent if | |
7768 | * discard_zeroes_data is 0); write one disk of the | |
7769 | * stripe (the stripe could be inconsistent again | |
7770 | * depending on which disks are used to calculate | |
7771 | * parity); the disk is broken; The stripe data of this | |
7772 | * disk is lost. | |
7773 | * | |
7774 | * We only allow DISCARD if the sysadmin has confirmed that | |
7775 | * only safe devices are in use by setting a module parameter. | |
7776 | * A better idea might be to turn DISCARD into WRITE_ZEROES | |
7777 | * requests, as that is required to be safe. | |
7778 | */ | |
7779 | if (devices_handle_discard_safely && | |
e7597e69 JS |
7780 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7781 | mddev->queue->limits.discard_granularity >= stripe) | |
8b904b5b | 7782 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
620125f2 SL |
7783 | mddev->queue); |
7784 | else | |
8b904b5b | 7785 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
620125f2 | 7786 | mddev->queue); |
1dffdddd SL |
7787 | |
7788 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7789 | } |
23032a0e | 7790 | |
845b9e22 | 7791 | if (log_init(conf, journal_dev, raid5_has_ppl(conf))) |
ff875738 | 7792 | goto abort; |
5c7e81c3 | 7793 | |
1da177e4 LT |
7794 | return 0; |
7795 | abort: | |
01f96c0a | 7796 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7797 | print_raid5_conf(conf); |
7798 | free_conf(conf); | |
1da177e4 | 7799 | mddev->private = NULL; |
cc6167b4 | 7800 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
306d8015 XN |
7801 | ret = -EIO; |
7802 | exit_acct_set: | |
7803 | acct_bioset_exit(mddev); | |
7804 | return ret; | |
1da177e4 LT |
7805 | } |
7806 | ||
afa0f557 | 7807 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7808 | { |
afa0f557 | 7809 | struct r5conf *conf = priv; |
1da177e4 | 7810 | |
95fc17aa | 7811 | free_conf(conf); |
306d8015 | 7812 | acct_bioset_exit(mddev); |
a64c876f | 7813 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7814 | } |
7815 | ||
849674e4 | 7816 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7817 | { |
d1688a6d | 7818 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7819 | int i; |
7820 | ||
9d8f0363 | 7821 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7822 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7823 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7824 | rcu_read_lock(); |
7825 | for (i = 0; i < conf->raid_disks; i++) { | |
7826 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7827 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7828 | } | |
7829 | rcu_read_unlock(); | |
1da177e4 | 7830 | seq_printf (seq, "]"); |
1da177e4 LT |
7831 | } |
7832 | ||
d1688a6d | 7833 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7834 | { |
7835 | int i; | |
7836 | struct disk_info *tmp; | |
7837 | ||
cc6167b4 | 7838 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7839 | if (!conf) { |
cc6167b4 | 7840 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7841 | return; |
7842 | } | |
cc6167b4 | 7843 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7844 | conf->raid_disks, |
7845 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7846 | |
7847 | for (i = 0; i < conf->raid_disks; i++) { | |
7848 | char b[BDEVNAME_SIZE]; | |
7849 | tmp = conf->disks + i; | |
7850 | if (tmp->rdev) | |
cc6167b4 | 7851 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7852 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7853 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7854 | } |
7855 | } | |
7856 | ||
fd01b88c | 7857 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7858 | { |
7859 | int i; | |
d1688a6d | 7860 | struct r5conf *conf = mddev->private; |
1da177e4 | 7861 | struct disk_info *tmp; |
6b965620 N |
7862 | int count = 0; |
7863 | unsigned long flags; | |
1da177e4 LT |
7864 | |
7865 | for (i = 0; i < conf->raid_disks; i++) { | |
7866 | tmp = conf->disks + i; | |
dd054fce N |
7867 | if (tmp->replacement |
7868 | && tmp->replacement->recovery_offset == MaxSector | |
7869 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7870 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7871 | /* Replacement has just become active. */ | |
7872 | if (!tmp->rdev | |
7873 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7874 | count++; | |
7875 | if (tmp->rdev) { | |
7876 | /* Replaced device not technically faulty, | |
7877 | * but we need to be sure it gets removed | |
7878 | * and never re-added. | |
7879 | */ | |
7880 | set_bit(Faulty, &tmp->rdev->flags); | |
7881 | sysfs_notify_dirent_safe( | |
7882 | tmp->rdev->sysfs_state); | |
7883 | } | |
7884 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7885 | } else if (tmp->rdev | |
70fffd0b | 7886 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7887 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7888 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7889 | count++; |
43c73ca4 | 7890 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7891 | } |
7892 | } | |
6b965620 | 7893 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7894 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7895 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7896 | print_raid5_conf(conf); |
6b965620 | 7897 | return count; |
1da177e4 LT |
7898 | } |
7899 | ||
b8321b68 | 7900 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7901 | { |
d1688a6d | 7902 | struct r5conf *conf = mddev->private; |
1da177e4 | 7903 | int err = 0; |
b8321b68 | 7904 | int number = rdev->raid_disk; |
657e3e4d | 7905 | struct md_rdev **rdevp; |
1da177e4 LT |
7906 | struct disk_info *p = conf->disks + number; |
7907 | ||
7908 | print_raid5_conf(conf); | |
f6b6ec5c | 7909 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
c2bb6242 | 7910 | /* |
f6b6ec5c SL |
7911 | * we can't wait pending write here, as this is called in |
7912 | * raid5d, wait will deadlock. | |
84dd97a6 N |
7913 | * neilb: there is no locking about new writes here, |
7914 | * so this cannot be safe. | |
c2bb6242 | 7915 | */ |
70d466f7 SL |
7916 | if (atomic_read(&conf->active_stripes) || |
7917 | atomic_read(&conf->r5c_cached_full_stripes) || | |
7918 | atomic_read(&conf->r5c_cached_partial_stripes)) { | |
f6b6ec5c | 7919 | return -EBUSY; |
84dd97a6 | 7920 | } |
ff875738 | 7921 | log_exit(conf); |
f6b6ec5c | 7922 | return 0; |
c2bb6242 | 7923 | } |
657e3e4d N |
7924 | if (rdev == p->rdev) |
7925 | rdevp = &p->rdev; | |
7926 | else if (rdev == p->replacement) | |
7927 | rdevp = &p->replacement; | |
7928 | else | |
7929 | return 0; | |
7930 | ||
7931 | if (number >= conf->raid_disks && | |
7932 | conf->reshape_progress == MaxSector) | |
7933 | clear_bit(In_sync, &rdev->flags); | |
7934 | ||
7935 | if (test_bit(In_sync, &rdev->flags) || | |
7936 | atomic_read(&rdev->nr_pending)) { | |
7937 | err = -EBUSY; | |
7938 | goto abort; | |
7939 | } | |
7940 | /* Only remove non-faulty devices if recovery | |
7941 | * isn't possible. | |
7942 | */ | |
7943 | if (!test_bit(Faulty, &rdev->flags) && | |
7944 | mddev->recovery_disabled != conf->recovery_disabled && | |
7945 | !has_failed(conf) && | |
dd054fce | 7946 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7947 | number < conf->raid_disks) { |
7948 | err = -EBUSY; | |
7949 | goto abort; | |
7950 | } | |
7951 | *rdevp = NULL; | |
d787be40 N |
7952 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7953 | synchronize_rcu(); | |
7954 | if (atomic_read(&rdev->nr_pending)) { | |
7955 | /* lost the race, try later */ | |
7956 | err = -EBUSY; | |
7957 | *rdevp = rdev; | |
7958 | } | |
7959 | } | |
6358c239 AP |
7960 | if (!err) { |
7961 | err = log_modify(conf, rdev, false); | |
7962 | if (err) | |
7963 | goto abort; | |
7964 | } | |
d787be40 | 7965 | if (p->replacement) { |
dd054fce N |
7966 | /* We must have just cleared 'rdev' */ |
7967 | p->rdev = p->replacement; | |
7968 | clear_bit(Replacement, &p->replacement->flags); | |
7969 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7970 | * but will never see neither - if they are careful | |
7971 | */ | |
7972 | p->replacement = NULL; | |
6358c239 AP |
7973 | |
7974 | if (!err) | |
7975 | err = log_modify(conf, p->rdev, true); | |
e5bc9c3c GJ |
7976 | } |
7977 | ||
7978 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7979 | abort: |
7980 | ||
7981 | print_raid5_conf(conf); | |
7982 | return err; | |
7983 | } | |
7984 | ||
fd01b88c | 7985 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7986 | { |
d1688a6d | 7987 | struct r5conf *conf = mddev->private; |
d9771f5e | 7988 | int ret, err = -EEXIST; |
1da177e4 LT |
7989 | int disk; |
7990 | struct disk_info *p; | |
6c2fce2e NB |
7991 | int first = 0; |
7992 | int last = conf->raid_disks - 1; | |
1da177e4 | 7993 | |
f6b6ec5c | 7994 | if (test_bit(Journal, &rdev->flags)) { |
f6b6ec5c SL |
7995 | if (conf->log) |
7996 | return -EBUSY; | |
7997 | ||
7998 | rdev->raid_disk = 0; | |
7999 | /* | |
8000 | * The array is in readonly mode if journal is missing, so no | |
8001 | * write requests running. We should be safe | |
8002 | */ | |
d9771f5e XN |
8003 | ret = log_init(conf, rdev, false); |
8004 | if (ret) | |
8005 | return ret; | |
8006 | ||
8007 | ret = r5l_start(conf->log); | |
8008 | if (ret) | |
8009 | return ret; | |
8010 | ||
f6b6ec5c SL |
8011 | return 0; |
8012 | } | |
7f0da59b N |
8013 | if (mddev->recovery_disabled == conf->recovery_disabled) |
8014 | return -EBUSY; | |
8015 | ||
dc10c643 | 8016 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 8017 | /* no point adding a device */ |
199050ea | 8018 | return -EINVAL; |
1da177e4 | 8019 | |
6c2fce2e NB |
8020 | if (rdev->raid_disk >= 0) |
8021 | first = last = rdev->raid_disk; | |
1da177e4 LT |
8022 | |
8023 | /* | |
16a53ecc N |
8024 | * find the disk ... but prefer rdev->saved_raid_disk |
8025 | * if possible. | |
1da177e4 | 8026 | */ |
16a53ecc | 8027 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 8028 | rdev->saved_raid_disk >= first && |
16a53ecc | 8029 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
8030 | first = rdev->saved_raid_disk; |
8031 | ||
8032 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
8033 | p = conf->disks + disk; |
8034 | if (p->rdev == NULL) { | |
b2d444d7 | 8035 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 8036 | rdev->raid_disk = disk; |
72626685 N |
8037 | if (rdev->saved_raid_disk != disk) |
8038 | conf->fullsync = 1; | |
d6065f7b | 8039 | rcu_assign_pointer(p->rdev, rdev); |
6358c239 AP |
8040 | |
8041 | err = log_modify(conf, rdev, true); | |
8042 | ||
5cfb22a1 | 8043 | goto out; |
1da177e4 | 8044 | } |
5cfb22a1 N |
8045 | } |
8046 | for (disk = first; disk <= last; disk++) { | |
8047 | p = conf->disks + disk; | |
7bfec5f3 N |
8048 | if (test_bit(WantReplacement, &p->rdev->flags) && |
8049 | p->replacement == NULL) { | |
8050 | clear_bit(In_sync, &rdev->flags); | |
8051 | set_bit(Replacement, &rdev->flags); | |
8052 | rdev->raid_disk = disk; | |
8053 | err = 0; | |
8054 | conf->fullsync = 1; | |
8055 | rcu_assign_pointer(p->replacement, rdev); | |
8056 | break; | |
8057 | } | |
8058 | } | |
5cfb22a1 | 8059 | out: |
1da177e4 | 8060 | print_raid5_conf(conf); |
199050ea | 8061 | return err; |
1da177e4 LT |
8062 | } |
8063 | ||
fd01b88c | 8064 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
8065 | { |
8066 | /* no resync is happening, and there is enough space | |
8067 | * on all devices, so we can resize. | |
8068 | * We need to make sure resync covers any new space. | |
8069 | * If the array is shrinking we should possibly wait until | |
8070 | * any io in the removed space completes, but it hardly seems | |
8071 | * worth it. | |
8072 | */ | |
a4a6125a | 8073 | sector_t newsize; |
3cb5edf4 N |
8074 | struct r5conf *conf = mddev->private; |
8075 | ||
e254de6b | 8076 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 8077 | return -EINVAL; |
3cb5edf4 | 8078 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
8079 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
8080 | if (mddev->external_size && | |
8081 | mddev->array_sectors > newsize) | |
b522adcd | 8082 | return -EINVAL; |
a4a6125a | 8083 | if (mddev->bitmap) { |
e64e4018 | 8084 | int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0); |
a4a6125a N |
8085 | if (ret) |
8086 | return ret; | |
8087 | } | |
8088 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
8089 | if (sectors > mddev->dev_sectors && |
8090 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 8091 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
8092 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
8093 | } | |
58c0fed4 | 8094 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 8095 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
8096 | return 0; |
8097 | } | |
8098 | ||
fd01b88c | 8099 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
8100 | { |
8101 | /* Can only proceed if there are plenty of stripe_heads. | |
8102 | * We need a minimum of one full stripe,, and for sensible progress | |
8103 | * it is best to have about 4 times that. | |
8104 | * If we require 4 times, then the default 256 4K stripe_heads will | |
8105 | * allow for chunk sizes up to 256K, which is probably OK. | |
8106 | * If the chunk size is greater, user-space should request more | |
8107 | * stripe_heads first. | |
8108 | */ | |
d1688a6d | 8109 | struct r5conf *conf = mddev->private; |
c911c46c | 8110 | if (((mddev->chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4 |
edbe83ab | 8111 | > conf->min_nr_stripes || |
c911c46c | 8112 | ((mddev->new_chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4 |
edbe83ab | 8113 | > conf->min_nr_stripes) { |
cc6167b4 N |
8114 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
8115 | mdname(mddev), | |
8116 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
c911c46c | 8117 | / RAID5_STRIPE_SIZE(conf))*4); |
01ee22b4 N |
8118 | return 0; |
8119 | } | |
8120 | return 1; | |
8121 | } | |
8122 | ||
fd01b88c | 8123 | static int check_reshape(struct mddev *mddev) |
29269553 | 8124 | { |
d1688a6d | 8125 | struct r5conf *conf = mddev->private; |
29269553 | 8126 | |
e254de6b | 8127 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 8128 | return -EINVAL; |
88ce4930 N |
8129 | if (mddev->delta_disks == 0 && |
8130 | mddev->new_layout == mddev->layout && | |
664e7c41 | 8131 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 8132 | return 0; /* nothing to do */ |
674806d6 | 8133 | if (has_failed(conf)) |
ec32a2bd | 8134 | return -EINVAL; |
fdcfbbb6 | 8135 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
8136 | /* We might be able to shrink, but the devices must |
8137 | * be made bigger first. | |
8138 | * For raid6, 4 is the minimum size. | |
8139 | * Otherwise 2 is the minimum | |
8140 | */ | |
8141 | int min = 2; | |
8142 | if (mddev->level == 6) | |
8143 | min = 4; | |
8144 | if (mddev->raid_disks + mddev->delta_disks < min) | |
8145 | return -EINVAL; | |
8146 | } | |
29269553 | 8147 | |
01ee22b4 | 8148 | if (!check_stripe_cache(mddev)) |
29269553 | 8149 | return -ENOSPC; |
29269553 | 8150 | |
738a2738 N |
8151 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
8152 | mddev->delta_disks > 0) | |
8153 | if (resize_chunks(conf, | |
8154 | conf->previous_raid_disks | |
8155 | + max(0, mddev->delta_disks), | |
8156 | max(mddev->new_chunk_sectors, | |
8157 | mddev->chunk_sectors) | |
8158 | ) < 0) | |
8159 | return -ENOMEM; | |
845b9e22 AP |
8160 | |
8161 | if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size) | |
8162 | return 0; /* never bother to shrink */ | |
e56108d6 N |
8163 | return resize_stripes(conf, (conf->previous_raid_disks |
8164 | + mddev->delta_disks)); | |
63c70c4f N |
8165 | } |
8166 | ||
fd01b88c | 8167 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 8168 | { |
d1688a6d | 8169 | struct r5conf *conf = mddev->private; |
3cb03002 | 8170 | struct md_rdev *rdev; |
63c70c4f | 8171 | int spares = 0; |
c04be0aa | 8172 | unsigned long flags; |
63c70c4f | 8173 | |
f416885e | 8174 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
8175 | return -EBUSY; |
8176 | ||
01ee22b4 N |
8177 | if (!check_stripe_cache(mddev)) |
8178 | return -ENOSPC; | |
8179 | ||
30b67645 N |
8180 | if (has_failed(conf)) |
8181 | return -EINVAL; | |
8182 | ||
c6563a8c | 8183 | rdev_for_each(rdev, mddev) { |
469518a3 N |
8184 | if (!test_bit(In_sync, &rdev->flags) |
8185 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 8186 | spares++; |
c6563a8c | 8187 | } |
63c70c4f | 8188 | |
f416885e | 8189 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
8190 | /* Not enough devices even to make a degraded array |
8191 | * of that size | |
8192 | */ | |
8193 | return -EINVAL; | |
8194 | ||
ec32a2bd N |
8195 | /* Refuse to reduce size of the array. Any reductions in |
8196 | * array size must be through explicit setting of array_size | |
8197 | * attribute. | |
8198 | */ | |
8199 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
8200 | < mddev->array_sectors) { | |
cc6167b4 N |
8201 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
8202 | mdname(mddev)); | |
ec32a2bd N |
8203 | return -EINVAL; |
8204 | } | |
8205 | ||
f6705578 | 8206 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 8207 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 8208 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 8209 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 8210 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
8211 | conf->prev_chunk_sectors = conf->chunk_sectors; |
8212 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
8213 | conf->prev_algo = conf->algorithm; |
8214 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
8215 | conf->generation++; |
8216 | /* Code that selects data_offset needs to see the generation update | |
8217 | * if reshape_progress has been set - so a memory barrier needed. | |
8218 | */ | |
8219 | smp_mb(); | |
2c810cdd | 8220 | if (mddev->reshape_backwards) |
fef9c61f N |
8221 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
8222 | else | |
8223 | conf->reshape_progress = 0; | |
8224 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 8225 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
8226 | spin_unlock_irq(&conf->device_lock); |
8227 | ||
4d77e3ba N |
8228 | /* Now make sure any requests that proceeded on the assumption |
8229 | * the reshape wasn't running - like Discard or Read - have | |
8230 | * completed. | |
8231 | */ | |
8232 | mddev_suspend(mddev); | |
8233 | mddev_resume(mddev); | |
8234 | ||
29269553 N |
8235 | /* Add some new drives, as many as will fit. |
8236 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
8237 | * Don't add devices if we are reducing the number of |
8238 | * devices in the array. This is because it is not possible | |
8239 | * to correctly record the "partially reconstructed" state of | |
8240 | * such devices during the reshape and confusion could result. | |
29269553 | 8241 | */ |
87a8dec9 | 8242 | if (mddev->delta_disks >= 0) { |
dafb20fa | 8243 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
8244 | if (rdev->raid_disk < 0 && |
8245 | !test_bit(Faulty, &rdev->flags)) { | |
8246 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 8247 | if (rdev->raid_disk |
9d4c7d87 | 8248 | >= conf->previous_raid_disks) |
87a8dec9 | 8249 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 8250 | else |
87a8dec9 | 8251 | rdev->recovery_offset = 0; |
36fad858 | 8252 | |
2aada5b1 DLM |
8253 | /* Failure here is OK */ |
8254 | sysfs_link_rdev(mddev, rdev); | |
50da0840 | 8255 | } |
87a8dec9 N |
8256 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
8257 | && !test_bit(Faulty, &rdev->flags)) { | |
8258 | /* This is a spare that was manually added */ | |
8259 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 8260 | } |
29269553 | 8261 | |
87a8dec9 N |
8262 | /* When a reshape changes the number of devices, |
8263 | * ->degraded is measured against the larger of the | |
8264 | * pre and post number of devices. | |
8265 | */ | |
ec32a2bd | 8266 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 8267 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
8268 | spin_unlock_irqrestore(&conf->device_lock, flags); |
8269 | } | |
63c70c4f | 8270 | mddev->raid_disks = conf->raid_disks; |
e516402c | 8271 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 8272 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 8273 | |
29269553 N |
8274 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8275 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 8276 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
8277 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
8278 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
8279 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 8280 | "reshape"); |
29269553 N |
8281 | if (!mddev->sync_thread) { |
8282 | mddev->recovery = 0; | |
8283 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 8284 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 8285 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
8286 | mddev->new_chunk_sectors = |
8287 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
8288 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
8289 | rdev_for_each(rdev, mddev) |
8290 | rdev->new_data_offset = rdev->data_offset; | |
8291 | smp_wmb(); | |
ba8805b9 | 8292 | conf->generation --; |
fef9c61f | 8293 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 8294 | mddev->reshape_position = MaxSector; |
ba8805b9 | 8295 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
8296 | spin_unlock_irq(&conf->device_lock); |
8297 | return -EAGAIN; | |
8298 | } | |
c8f517c4 | 8299 | conf->reshape_checkpoint = jiffies; |
29269553 N |
8300 | md_wakeup_thread(mddev->sync_thread); |
8301 | md_new_event(mddev); | |
8302 | return 0; | |
8303 | } | |
29269553 | 8304 | |
ec32a2bd N |
8305 | /* This is called from the reshape thread and should make any |
8306 | * changes needed in 'conf' | |
8307 | */ | |
d1688a6d | 8308 | static void end_reshape(struct r5conf *conf) |
29269553 | 8309 | { |
29269553 | 8310 | |
f6705578 | 8311 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
db0505d3 | 8312 | struct md_rdev *rdev; |
f6705578 | 8313 | |
f6705578 | 8314 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 8315 | conf->previous_raid_disks = conf->raid_disks; |
b5d27718 | 8316 | md_finish_reshape(conf->mddev); |
05616be5 | 8317 | smp_wmb(); |
fef9c61f | 8318 | conf->reshape_progress = MaxSector; |
6cbd8148 | 8319 | conf->mddev->reshape_position = MaxSector; |
db0505d3 N |
8320 | rdev_for_each(rdev, conf->mddev) |
8321 | if (rdev->raid_disk >= 0 && | |
8322 | !test_bit(Journal, &rdev->flags) && | |
8323 | !test_bit(In_sync, &rdev->flags)) | |
8324 | rdev->recovery_offset = MaxSector; | |
f6705578 | 8325 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 8326 | wake_up(&conf->wait_for_overlap); |
16a53ecc | 8327 | |
c2e4cd57 | 8328 | if (conf->mddev->queue) |
16ef5101 | 8329 | raid5_set_io_opt(conf); |
29269553 | 8330 | } |
29269553 N |
8331 | } |
8332 | ||
ec32a2bd N |
8333 | /* This is called from the raid5d thread with mddev_lock held. |
8334 | * It makes config changes to the device. | |
8335 | */ | |
fd01b88c | 8336 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 8337 | { |
d1688a6d | 8338 | struct r5conf *conf = mddev->private; |
cea9c228 N |
8339 | |
8340 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8341 | ||
8876391e | 8342 | if (mddev->delta_disks <= 0) { |
ec32a2bd | 8343 | int d; |
908f4fbd | 8344 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 8345 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 8346 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
8347 | for (d = conf->raid_disks ; |
8348 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 8349 | d++) { |
3cb03002 | 8350 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
8351 | if (rdev) |
8352 | clear_bit(In_sync, &rdev->flags); | |
8353 | rdev = conf->disks[d].replacement; | |
8354 | if (rdev) | |
8355 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 8356 | } |
cea9c228 | 8357 | } |
88ce4930 | 8358 | mddev->layout = conf->algorithm; |
09c9e5fa | 8359 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
8360 | mddev->reshape_position = MaxSector; |
8361 | mddev->delta_disks = 0; | |
2c810cdd | 8362 | mddev->reshape_backwards = 0; |
cea9c228 N |
8363 | } |
8364 | } | |
8365 | ||
b03e0ccb | 8366 | static void raid5_quiesce(struct mddev *mddev, int quiesce) |
72626685 | 8367 | { |
d1688a6d | 8368 | struct r5conf *conf = mddev->private; |
72626685 | 8369 | |
b03e0ccb N |
8370 | if (quiesce) { |
8371 | /* stop all writes */ | |
566c09c5 | 8372 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
8373 | /* '2' tells resync/reshape to pause so that all |
8374 | * active stripes can drain | |
8375 | */ | |
a39f7afd | 8376 | r5c_flush_cache(conf, INT_MAX); |
97ae2725 GO |
8377 | /* need a memory barrier to make sure read_one_chunk() sees |
8378 | * quiesce started and reverts to slow (locked) path. | |
8379 | */ | |
8380 | smp_store_release(&conf->quiesce, 2); | |
b1b46486 | 8381 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
8382 | atomic_read(&conf->active_stripes) == 0 && |
8383 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
8384 | unlock_all_device_hash_locks_irq(conf), |
8385 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 8386 | conf->quiesce = 1; |
566c09c5 | 8387 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
8388 | /* allow reshape to continue */ |
8389 | wake_up(&conf->wait_for_overlap); | |
b03e0ccb N |
8390 | } else { |
8391 | /* re-enable writes */ | |
566c09c5 | 8392 | lock_all_device_hash_locks_irq(conf); |
72626685 | 8393 | conf->quiesce = 0; |
b1b46486 | 8394 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 8395 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 8396 | unlock_all_device_hash_locks_irq(conf); |
72626685 | 8397 | } |
1532d9e8 | 8398 | log_quiesce(conf, quiesce); |
72626685 | 8399 | } |
b15c2e57 | 8400 | |
fd01b88c | 8401 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 8402 | { |
e373ab10 | 8403 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 8404 | sector_t sectors; |
54071b38 | 8405 | |
f1b29bca | 8406 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 8407 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
8408 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
8409 | mdname(mddev)); | |
f1b29bca DW |
8410 | return ERR_PTR(-EINVAL); |
8411 | } | |
8412 | ||
e373ab10 N |
8413 | sectors = raid0_conf->strip_zone[0].zone_end; |
8414 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 8415 | mddev->dev_sectors = sectors; |
f1b29bca | 8416 | mddev->new_level = level; |
54071b38 TM |
8417 | mddev->new_layout = ALGORITHM_PARITY_N; |
8418 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
8419 | mddev->raid_disks += 1; | |
8420 | mddev->delta_disks = 1; | |
8421 | /* make sure it will be not marked as dirty */ | |
8422 | mddev->recovery_cp = MaxSector; | |
8423 | ||
8424 | return setup_conf(mddev); | |
8425 | } | |
8426 | ||
fd01b88c | 8427 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
8428 | { |
8429 | int chunksect; | |
6995f0b2 | 8430 | void *ret; |
d562b0c4 N |
8431 | |
8432 | if (mddev->raid_disks != 2 || | |
8433 | mddev->degraded > 1) | |
8434 | return ERR_PTR(-EINVAL); | |
8435 | ||
8436 | /* Should check if there are write-behind devices? */ | |
8437 | ||
8438 | chunksect = 64*2; /* 64K by default */ | |
8439 | ||
8440 | /* The array must be an exact multiple of chunksize */ | |
8441 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8442 | chunksect >>= 1; | |
8443 | ||
c911c46c | 8444 | if ((chunksect<<9) < RAID5_STRIPE_SIZE((struct r5conf *)mddev->private)) |
d562b0c4 N |
8445 | /* array size does not allow a suitable chunk size */ |
8446 | return ERR_PTR(-EINVAL); | |
8447 | ||
8448 | mddev->new_level = 5; | |
8449 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8450 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8451 | |
6995f0b2 | 8452 | ret = setup_conf(mddev); |
32cd7cbb | 8453 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8454 | mddev_clear_unsupported_flags(mddev, |
8455 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8456 | return ret; |
d562b0c4 N |
8457 | } |
8458 | ||
fd01b88c | 8459 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8460 | { |
8461 | int new_layout; | |
8462 | ||
8463 | switch (mddev->layout) { | |
8464 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8465 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8466 | break; | |
8467 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8468 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8469 | break; | |
8470 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8471 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8472 | break; | |
8473 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8474 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8475 | break; | |
8476 | case ALGORITHM_PARITY_0_6: | |
8477 | new_layout = ALGORITHM_PARITY_0; | |
8478 | break; | |
8479 | case ALGORITHM_PARITY_N: | |
8480 | new_layout = ALGORITHM_PARITY_N; | |
8481 | break; | |
8482 | default: | |
8483 | return ERR_PTR(-EINVAL); | |
8484 | } | |
8485 | mddev->new_level = 5; | |
8486 | mddev->new_layout = new_layout; | |
8487 | mddev->delta_disks = -1; | |
8488 | mddev->raid_disks -= 1; | |
8489 | return setup_conf(mddev); | |
8490 | } | |
8491 | ||
fd01b88c | 8492 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8493 | { |
88ce4930 N |
8494 | /* For a 2-drive array, the layout and chunk size can be changed |
8495 | * immediately as not restriping is needed. | |
8496 | * For larger arrays we record the new value - after validation | |
8497 | * to be used by a reshape pass. | |
b3546035 | 8498 | */ |
d1688a6d | 8499 | struct r5conf *conf = mddev->private; |
597a711b | 8500 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8501 | |
597a711b | 8502 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8503 | return -EINVAL; |
8504 | if (new_chunk > 0) { | |
0ba459d2 | 8505 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8506 | return -EINVAL; |
597a711b | 8507 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8508 | return -EINVAL; |
597a711b | 8509 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8510 | /* not factor of array size */ |
8511 | return -EINVAL; | |
8512 | } | |
8513 | ||
8514 | /* They look valid */ | |
8515 | ||
88ce4930 | 8516 | if (mddev->raid_disks == 2) { |
597a711b N |
8517 | /* can make the change immediately */ |
8518 | if (mddev->new_layout >= 0) { | |
8519 | conf->algorithm = mddev->new_layout; | |
8520 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8521 | } |
8522 | if (new_chunk > 0) { | |
597a711b N |
8523 | conf->chunk_sectors = new_chunk ; |
8524 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8525 | } |
2953079c | 8526 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8527 | md_wakeup_thread(mddev->thread); |
b3546035 | 8528 | } |
50ac168a | 8529 | return check_reshape(mddev); |
88ce4930 N |
8530 | } |
8531 | ||
fd01b88c | 8532 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8533 | { |
597a711b | 8534 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8535 | |
597a711b | 8536 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8537 | return -EINVAL; |
b3546035 | 8538 | if (new_chunk > 0) { |
0ba459d2 | 8539 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8540 | return -EINVAL; |
597a711b | 8541 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8542 | return -EINVAL; |
597a711b | 8543 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8544 | /* not factor of array size */ |
8545 | return -EINVAL; | |
b3546035 | 8546 | } |
88ce4930 N |
8547 | |
8548 | /* They look valid */ | |
50ac168a | 8549 | return check_reshape(mddev); |
b3546035 N |
8550 | } |
8551 | ||
fd01b88c | 8552 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8553 | { |
8554 | /* raid5 can take over: | |
f1b29bca | 8555 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8556 | * raid1 - if there are two drives. We need to know the chunk size |
8557 | * raid4 - trivial - just use a raid4 layout. | |
8558 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8559 | */ |
f1b29bca DW |
8560 | if (mddev->level == 0) |
8561 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8562 | if (mddev->level == 1) |
8563 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8564 | if (mddev->level == 4) { |
8565 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8566 | mddev->new_level = 5; | |
8567 | return setup_conf(mddev); | |
8568 | } | |
fc9739c6 N |
8569 | if (mddev->level == 6) |
8570 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8571 | |
8572 | return ERR_PTR(-EINVAL); | |
8573 | } | |
8574 | ||
fd01b88c | 8575 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8576 | { |
f1b29bca DW |
8577 | /* raid4 can take over: |
8578 | * raid0 - if there is only one strip zone | |
8579 | * raid5 - if layout is right | |
a78d38a1 | 8580 | */ |
f1b29bca DW |
8581 | if (mddev->level == 0) |
8582 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8583 | if (mddev->level == 5 && |
8584 | mddev->layout == ALGORITHM_PARITY_N) { | |
8585 | mddev->new_layout = 0; | |
8586 | mddev->new_level = 4; | |
8587 | return setup_conf(mddev); | |
8588 | } | |
8589 | return ERR_PTR(-EINVAL); | |
8590 | } | |
d562b0c4 | 8591 | |
84fc4b56 | 8592 | static struct md_personality raid5_personality; |
245f46c2 | 8593 | |
fd01b88c | 8594 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8595 | { |
8596 | /* Currently can only take over a raid5. We map the | |
8597 | * personality to an equivalent raid6 personality | |
8598 | * with the Q block at the end. | |
8599 | */ | |
8600 | int new_layout; | |
8601 | ||
8602 | if (mddev->pers != &raid5_personality) | |
8603 | return ERR_PTR(-EINVAL); | |
8604 | if (mddev->degraded > 1) | |
8605 | return ERR_PTR(-EINVAL); | |
8606 | if (mddev->raid_disks > 253) | |
8607 | return ERR_PTR(-EINVAL); | |
8608 | if (mddev->raid_disks < 3) | |
8609 | return ERR_PTR(-EINVAL); | |
8610 | ||
8611 | switch (mddev->layout) { | |
8612 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8613 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8614 | break; | |
8615 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8616 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8617 | break; | |
8618 | case ALGORITHM_LEFT_SYMMETRIC: | |
8619 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8620 | break; | |
8621 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8622 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8623 | break; | |
8624 | case ALGORITHM_PARITY_0: | |
8625 | new_layout = ALGORITHM_PARITY_0_6; | |
8626 | break; | |
8627 | case ALGORITHM_PARITY_N: | |
8628 | new_layout = ALGORITHM_PARITY_N; | |
8629 | break; | |
8630 | default: | |
8631 | return ERR_PTR(-EINVAL); | |
8632 | } | |
8633 | mddev->new_level = 6; | |
8634 | mddev->new_layout = new_layout; | |
8635 | mddev->delta_disks = 1; | |
8636 | mddev->raid_disks += 1; | |
8637 | return setup_conf(mddev); | |
8638 | } | |
8639 | ||
ba903a3e AP |
8640 | static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf) |
8641 | { | |
8642 | struct r5conf *conf; | |
8643 | int err; | |
8644 | ||
8645 | err = mddev_lock(mddev); | |
8646 | if (err) | |
8647 | return err; | |
8648 | conf = mddev->private; | |
8649 | if (!conf) { | |
8650 | mddev_unlock(mddev); | |
8651 | return -ENODEV; | |
8652 | } | |
8653 | ||
845b9e22 | 8654 | if (strncmp(buf, "ppl", 3) == 0) { |
0bb0c105 | 8655 | /* ppl only works with RAID 5 */ |
845b9e22 AP |
8656 | if (!raid5_has_ppl(conf) && conf->level == 5) { |
8657 | err = log_init(conf, NULL, true); | |
8658 | if (!err) { | |
8659 | err = resize_stripes(conf, conf->pool_size); | |
8660 | if (err) | |
8661 | log_exit(conf); | |
8662 | } | |
0bb0c105 SL |
8663 | } else |
8664 | err = -EINVAL; | |
8665 | } else if (strncmp(buf, "resync", 6) == 0) { | |
8666 | if (raid5_has_ppl(conf)) { | |
8667 | mddev_suspend(mddev); | |
8668 | log_exit(conf); | |
0bb0c105 | 8669 | mddev_resume(mddev); |
845b9e22 | 8670 | err = resize_stripes(conf, conf->pool_size); |
0bb0c105 SL |
8671 | } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) && |
8672 | r5l_log_disk_error(conf)) { | |
8673 | bool journal_dev_exists = false; | |
8674 | struct md_rdev *rdev; | |
8675 | ||
8676 | rdev_for_each(rdev, mddev) | |
8677 | if (test_bit(Journal, &rdev->flags)) { | |
8678 | journal_dev_exists = true; | |
8679 | break; | |
8680 | } | |
8681 | ||
8682 | if (!journal_dev_exists) { | |
8683 | mddev_suspend(mddev); | |
8684 | clear_bit(MD_HAS_JOURNAL, &mddev->flags); | |
8685 | mddev_resume(mddev); | |
8686 | } else /* need remove journal device first */ | |
8687 | err = -EBUSY; | |
8688 | } else | |
8689 | err = -EINVAL; | |
ba903a3e AP |
8690 | } else { |
8691 | err = -EINVAL; | |
8692 | } | |
8693 | ||
8694 | if (!err) | |
8695 | md_update_sb(mddev, 1); | |
8696 | ||
8697 | mddev_unlock(mddev); | |
8698 | ||
8699 | return err; | |
8700 | } | |
8701 | ||
d5d885fd SL |
8702 | static int raid5_start(struct mddev *mddev) |
8703 | { | |
8704 | struct r5conf *conf = mddev->private; | |
8705 | ||
8706 | return r5l_start(conf->log); | |
8707 | } | |
8708 | ||
84fc4b56 | 8709 | static struct md_personality raid6_personality = |
16a53ecc N |
8710 | { |
8711 | .name = "raid6", | |
8712 | .level = 6, | |
8713 | .owner = THIS_MODULE, | |
849674e4 SL |
8714 | .make_request = raid5_make_request, |
8715 | .run = raid5_run, | |
d5d885fd | 8716 | .start = raid5_start, |
afa0f557 | 8717 | .free = raid5_free, |
849674e4 SL |
8718 | .status = raid5_status, |
8719 | .error_handler = raid5_error, | |
16a53ecc N |
8720 | .hot_add_disk = raid5_add_disk, |
8721 | .hot_remove_disk= raid5_remove_disk, | |
8722 | .spare_active = raid5_spare_active, | |
849674e4 | 8723 | .sync_request = raid5_sync_request, |
16a53ecc | 8724 | .resize = raid5_resize, |
80c3a6ce | 8725 | .size = raid5_size, |
50ac168a | 8726 | .check_reshape = raid6_check_reshape, |
f416885e | 8727 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8728 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8729 | .quiesce = raid5_quiesce, |
245f46c2 | 8730 | .takeover = raid6_takeover, |
0bb0c105 | 8731 | .change_consistency_policy = raid5_change_consistency_policy, |
16a53ecc | 8732 | }; |
84fc4b56 | 8733 | static struct md_personality raid5_personality = |
1da177e4 LT |
8734 | { |
8735 | .name = "raid5", | |
2604b703 | 8736 | .level = 5, |
1da177e4 | 8737 | .owner = THIS_MODULE, |
849674e4 SL |
8738 | .make_request = raid5_make_request, |
8739 | .run = raid5_run, | |
d5d885fd | 8740 | .start = raid5_start, |
afa0f557 | 8741 | .free = raid5_free, |
849674e4 SL |
8742 | .status = raid5_status, |
8743 | .error_handler = raid5_error, | |
1da177e4 LT |
8744 | .hot_add_disk = raid5_add_disk, |
8745 | .hot_remove_disk= raid5_remove_disk, | |
8746 | .spare_active = raid5_spare_active, | |
849674e4 | 8747 | .sync_request = raid5_sync_request, |
1da177e4 | 8748 | .resize = raid5_resize, |
80c3a6ce | 8749 | .size = raid5_size, |
63c70c4f N |
8750 | .check_reshape = raid5_check_reshape, |
8751 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8752 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8753 | .quiesce = raid5_quiesce, |
d562b0c4 | 8754 | .takeover = raid5_takeover, |
ba903a3e | 8755 | .change_consistency_policy = raid5_change_consistency_policy, |
1da177e4 LT |
8756 | }; |
8757 | ||
84fc4b56 | 8758 | static struct md_personality raid4_personality = |
1da177e4 | 8759 | { |
2604b703 N |
8760 | .name = "raid4", |
8761 | .level = 4, | |
8762 | .owner = THIS_MODULE, | |
849674e4 SL |
8763 | .make_request = raid5_make_request, |
8764 | .run = raid5_run, | |
d5d885fd | 8765 | .start = raid5_start, |
afa0f557 | 8766 | .free = raid5_free, |
849674e4 SL |
8767 | .status = raid5_status, |
8768 | .error_handler = raid5_error, | |
2604b703 N |
8769 | .hot_add_disk = raid5_add_disk, |
8770 | .hot_remove_disk= raid5_remove_disk, | |
8771 | .spare_active = raid5_spare_active, | |
849674e4 | 8772 | .sync_request = raid5_sync_request, |
2604b703 | 8773 | .resize = raid5_resize, |
80c3a6ce | 8774 | .size = raid5_size, |
3d37890b N |
8775 | .check_reshape = raid5_check_reshape, |
8776 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8777 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8778 | .quiesce = raid5_quiesce, |
a78d38a1 | 8779 | .takeover = raid4_takeover, |
0bb0c105 | 8780 | .change_consistency_policy = raid5_change_consistency_policy, |
2604b703 N |
8781 | }; |
8782 | ||
8783 | static int __init raid5_init(void) | |
8784 | { | |
29c6d1bb SAS |
8785 | int ret; |
8786 | ||
851c30c9 SL |
8787 | raid5_wq = alloc_workqueue("raid5wq", |
8788 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8789 | if (!raid5_wq) | |
8790 | return -ENOMEM; | |
29c6d1bb SAS |
8791 | |
8792 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8793 | "md/raid5:prepare", | |
8794 | raid456_cpu_up_prepare, | |
8795 | raid456_cpu_dead); | |
8796 | if (ret) { | |
8797 | destroy_workqueue(raid5_wq); | |
8798 | return ret; | |
8799 | } | |
16a53ecc | 8800 | register_md_personality(&raid6_personality); |
2604b703 N |
8801 | register_md_personality(&raid5_personality); |
8802 | register_md_personality(&raid4_personality); | |
8803 | return 0; | |
1da177e4 LT |
8804 | } |
8805 | ||
2604b703 | 8806 | static void raid5_exit(void) |
1da177e4 | 8807 | { |
16a53ecc | 8808 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8809 | unregister_md_personality(&raid5_personality); |
8810 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8811 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8812 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8813 | } |
8814 | ||
8815 | module_init(raid5_init); | |
8816 | module_exit(raid5_exit); | |
8817 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8818 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8819 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8820 | MODULE_ALIAS("md-raid5"); |
8821 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8822 | MODULE_ALIAS("md-level-5"); |
8823 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8824 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8825 | MODULE_ALIAS("md-raid6"); | |
8826 | MODULE_ALIAS("md-level-6"); | |
8827 | ||
8828 | /* This used to be two separate modules, they were: */ | |
8829 | MODULE_ALIAS("raid5"); | |
8830 | MODULE_ALIAS("raid6"); |