Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
851c30c9 | 56 | #include <linux/nodemask.h> |
46d5b785 | 57 | #include <linux/flex_array.h> |
3f07c014 | 58 | |
a9add5d9 | 59 | #include <trace/events/block.h> |
aaf9f12e | 60 | #include <linux/list_sort.h> |
a9add5d9 | 61 | |
43b2e5d8 | 62 | #include "md.h" |
bff61975 | 63 | #include "raid5.h" |
54071b38 | 64 | #include "raid0.h" |
935fe098 | 65 | #include "md-bitmap.h" |
ff875738 | 66 | #include "raid5-log.h" |
72626685 | 67 | |
394ed8e4 SL |
68 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
69 | ||
851c30c9 SL |
70 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
71 | #define ANY_GROUP NUMA_NO_NODE | |
72 | ||
8e0e99ba N |
73 | static bool devices_handle_discard_safely = false; |
74 | module_param(devices_handle_discard_safely, bool, 0644); | |
75 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
76 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 77 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 78 | |
d1688a6d | 79 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
80 | { |
81 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
82 | return &conf->stripe_hashtbl[hash]; | |
83 | } | |
1da177e4 | 84 | |
566c09c5 SL |
85 | static inline int stripe_hash_locks_hash(sector_t sect) |
86 | { | |
87 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
88 | } | |
89 | ||
90 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
91 | { | |
92 | spin_lock_irq(conf->hash_locks + hash); | |
93 | spin_lock(&conf->device_lock); | |
94 | } | |
95 | ||
96 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
97 | { | |
98 | spin_unlock(&conf->device_lock); | |
99 | spin_unlock_irq(conf->hash_locks + hash); | |
100 | } | |
101 | ||
102 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
103 | { | |
104 | int i; | |
3d05f3ae | 105 | spin_lock_irq(conf->hash_locks); |
566c09c5 SL |
106 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) |
107 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
108 | spin_lock(&conf->device_lock); | |
109 | } | |
110 | ||
111 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
112 | { | |
113 | int i; | |
114 | spin_unlock(&conf->device_lock); | |
3d05f3ae JC |
115 | for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--) |
116 | spin_unlock(conf->hash_locks + i); | |
117 | spin_unlock_irq(conf->hash_locks); | |
566c09c5 SL |
118 | } |
119 | ||
d0dabf7e N |
120 | /* Find first data disk in a raid6 stripe */ |
121 | static inline int raid6_d0(struct stripe_head *sh) | |
122 | { | |
67cc2b81 N |
123 | if (sh->ddf_layout) |
124 | /* ddf always start from first device */ | |
125 | return 0; | |
126 | /* md starts just after Q block */ | |
d0dabf7e N |
127 | if (sh->qd_idx == sh->disks - 1) |
128 | return 0; | |
129 | else | |
130 | return sh->qd_idx + 1; | |
131 | } | |
16a53ecc N |
132 | static inline int raid6_next_disk(int disk, int raid_disks) |
133 | { | |
134 | disk++; | |
135 | return (disk < raid_disks) ? disk : 0; | |
136 | } | |
a4456856 | 137 | |
d0dabf7e N |
138 | /* When walking through the disks in a raid5, starting at raid6_d0, |
139 | * We need to map each disk to a 'slot', where the data disks are slot | |
140 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
141 | * is raid_disks-1. This help does that mapping. | |
142 | */ | |
67cc2b81 N |
143 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
144 | int *count, int syndrome_disks) | |
d0dabf7e | 145 | { |
6629542e | 146 | int slot = *count; |
67cc2b81 | 147 | |
e4424fee | 148 | if (sh->ddf_layout) |
6629542e | 149 | (*count)++; |
d0dabf7e | 150 | if (idx == sh->pd_idx) |
67cc2b81 | 151 | return syndrome_disks; |
d0dabf7e | 152 | if (idx == sh->qd_idx) |
67cc2b81 | 153 | return syndrome_disks + 1; |
e4424fee | 154 | if (!sh->ddf_layout) |
6629542e | 155 | (*count)++; |
d0dabf7e N |
156 | return slot; |
157 | } | |
158 | ||
d1688a6d | 159 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 160 | |
600aa109 DW |
161 | static int stripe_operations_active(struct stripe_head *sh) |
162 | { | |
163 | return sh->check_state || sh->reconstruct_state || | |
164 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
165 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
166 | } | |
167 | ||
535ae4eb SL |
168 | static bool stripe_is_lowprio(struct stripe_head *sh) |
169 | { | |
170 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
171 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
172 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
173 | } | |
174 | ||
851c30c9 SL |
175 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
176 | { | |
177 | struct r5conf *conf = sh->raid_conf; | |
178 | struct r5worker_group *group; | |
bfc90cb0 | 179 | int thread_cnt; |
851c30c9 SL |
180 | int i, cpu = sh->cpu; |
181 | ||
182 | if (!cpu_online(cpu)) { | |
183 | cpu = cpumask_any(cpu_online_mask); | |
184 | sh->cpu = cpu; | |
185 | } | |
186 | ||
187 | if (list_empty(&sh->lru)) { | |
188 | struct r5worker_group *group; | |
189 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
190 | if (stripe_is_lowprio(sh)) |
191 | list_add_tail(&sh->lru, &group->loprio_list); | |
192 | else | |
193 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
194 | group->stripes_cnt++; |
195 | sh->group = group; | |
851c30c9 SL |
196 | } |
197 | ||
198 | if (conf->worker_cnt_per_group == 0) { | |
199 | md_wakeup_thread(conf->mddev->thread); | |
200 | return; | |
201 | } | |
202 | ||
203 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
204 | ||
bfc90cb0 SL |
205 | group->workers[0].working = true; |
206 | /* at least one worker should run to avoid race */ | |
207 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
208 | ||
209 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
210 | /* wakeup more workers */ | |
211 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
212 | if (group->workers[i].working == false) { | |
213 | group->workers[i].working = true; | |
214 | queue_work_on(sh->cpu, raid5_wq, | |
215 | &group->workers[i].work); | |
216 | thread_cnt--; | |
217 | } | |
218 | } | |
851c30c9 SL |
219 | } |
220 | ||
566c09c5 SL |
221 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
222 | struct list_head *temp_inactive_list) | |
1da177e4 | 223 | { |
1e6d690b SL |
224 | int i; |
225 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
226 | ||
4eb788df SL |
227 | BUG_ON(!list_empty(&sh->lru)); |
228 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
229 | |
230 | if (r5c_is_writeback(conf->log)) | |
231 | for (i = sh->disks; i--; ) | |
232 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
233 | injournal++; | |
a39f7afd | 234 | /* |
5ddf0440 SL |
235 | * In the following cases, the stripe cannot be released to cached |
236 | * lists. Therefore, we make the stripe write out and set | |
237 | * STRIPE_HANDLE: | |
238 | * 1. when quiesce in r5c write back; | |
239 | * 2. when resync is requested fot the stripe. | |
a39f7afd | 240 | */ |
5ddf0440 SL |
241 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) || |
242 | (conf->quiesce && r5c_is_writeback(conf->log) && | |
243 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) { | |
a39f7afd SL |
244 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) |
245 | r5c_make_stripe_write_out(sh); | |
246 | set_bit(STRIPE_HANDLE, &sh->state); | |
247 | } | |
1e6d690b | 248 | |
4eb788df SL |
249 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
250 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 251 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 252 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 253 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
254 | sh->bm_seq - conf->seq_write > 0) |
255 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
256 | else { | |
257 | clear_bit(STRIPE_DELAYED, &sh->state); | |
258 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 259 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
260 | if (stripe_is_lowprio(sh)) |
261 | list_add_tail(&sh->lru, | |
262 | &conf->loprio_list); | |
263 | else | |
264 | list_add_tail(&sh->lru, | |
265 | &conf->handle_list); | |
851c30c9 SL |
266 | } else { |
267 | raid5_wakeup_stripe_thread(sh); | |
268 | return; | |
269 | } | |
4eb788df SL |
270 | } |
271 | md_wakeup_thread(conf->mddev->thread); | |
272 | } else { | |
273 | BUG_ON(stripe_operations_active(sh)); | |
274 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
275 | if (atomic_dec_return(&conf->preread_active_stripes) | |
276 | < IO_THRESHOLD) | |
277 | md_wakeup_thread(conf->mddev->thread); | |
278 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
279 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
280 | if (!r5c_is_writeback(conf->log)) | |
281 | list_add_tail(&sh->lru, temp_inactive_list); | |
282 | else { | |
283 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
284 | if (injournal == 0) | |
285 | list_add_tail(&sh->lru, temp_inactive_list); | |
286 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
287 | /* full stripe */ | |
288 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
289 | atomic_inc(&conf->r5c_cached_full_stripes); | |
290 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
291 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
292 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 293 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
294 | } else |
295 | /* | |
296 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
297 | * r5c_try_caching_write(). No need to | |
298 | * set it again. | |
299 | */ | |
1e6d690b | 300 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
301 | } |
302 | } | |
1da177e4 LT |
303 | } |
304 | } | |
d0dabf7e | 305 | |
566c09c5 SL |
306 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
307 | struct list_head *temp_inactive_list) | |
4eb788df SL |
308 | { |
309 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
310 | do_release_stripe(conf, sh, temp_inactive_list); |
311 | } | |
312 | ||
313 | /* | |
314 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
315 | * | |
316 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
317 | * given time. Adding stripes only takes device lock, while deleting stripes | |
318 | * only takes hash lock. | |
319 | */ | |
320 | static void release_inactive_stripe_list(struct r5conf *conf, | |
321 | struct list_head *temp_inactive_list, | |
322 | int hash) | |
323 | { | |
324 | int size; | |
6ab2a4b8 | 325 | bool do_wakeup = false; |
566c09c5 SL |
326 | unsigned long flags; |
327 | ||
328 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
329 | size = NR_STRIPE_HASH_LOCKS; | |
330 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
331 | } else | |
332 | size = 1; | |
333 | while (size) { | |
334 | struct list_head *list = &temp_inactive_list[size - 1]; | |
335 | ||
336 | /* | |
6d036f7d | 337 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
338 | * remove stripes from the list |
339 | */ | |
340 | if (!list_empty_careful(list)) { | |
341 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
342 | if (list_empty(conf->inactive_list + hash) && |
343 | !list_empty(list)) | |
344 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 345 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 346 | do_wakeup = true; |
566c09c5 SL |
347 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
348 | } | |
349 | size--; | |
350 | hash--; | |
351 | } | |
352 | ||
353 | if (do_wakeup) { | |
6ab2a4b8 | 354 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
355 | if (atomic_read(&conf->active_stripes) == 0) |
356 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
357 | if (conf->retry_read_aligned) |
358 | md_wakeup_thread(conf->mddev->thread); | |
359 | } | |
4eb788df SL |
360 | } |
361 | ||
773ca82f | 362 | /* should hold conf->device_lock already */ |
566c09c5 SL |
363 | static int release_stripe_list(struct r5conf *conf, |
364 | struct list_head *temp_inactive_list) | |
773ca82f | 365 | { |
eae8263f | 366 | struct stripe_head *sh, *t; |
773ca82f SL |
367 | int count = 0; |
368 | struct llist_node *head; | |
369 | ||
370 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 371 | head = llist_reverse_order(head); |
eae8263f | 372 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
373 | int hash; |
374 | ||
773ca82f SL |
375 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
376 | smp_mb(); | |
377 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
378 | /* | |
379 | * Don't worry the bit is set here, because if the bit is set | |
380 | * again, the count is always > 1. This is true for | |
381 | * STRIPE_ON_UNPLUG_LIST bit too. | |
382 | */ | |
566c09c5 SL |
383 | hash = sh->hash_lock_index; |
384 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
385 | count++; |
386 | } | |
387 | ||
388 | return count; | |
389 | } | |
390 | ||
6d036f7d | 391 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 392 | { |
d1688a6d | 393 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 394 | unsigned long flags; |
566c09c5 SL |
395 | struct list_head list; |
396 | int hash; | |
773ca82f | 397 | bool wakeup; |
16a53ecc | 398 | |
cf170f3f ES |
399 | /* Avoid release_list until the last reference. |
400 | */ | |
401 | if (atomic_add_unless(&sh->count, -1, 1)) | |
402 | return; | |
403 | ||
ad4068de | 404 | if (unlikely(!conf->mddev->thread) || |
405 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
406 | goto slow_path; |
407 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
408 | if (wakeup) | |
409 | md_wakeup_thread(conf->mddev->thread); | |
410 | return; | |
411 | slow_path: | |
4eb788df | 412 | local_irq_save(flags); |
773ca82f | 413 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 414 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
415 | INIT_LIST_HEAD(&list); |
416 | hash = sh->hash_lock_index; | |
417 | do_release_stripe(conf, sh, &list); | |
4eb788df | 418 | spin_unlock(&conf->device_lock); |
566c09c5 | 419 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
420 | } |
421 | local_irq_restore(flags); | |
1da177e4 LT |
422 | } |
423 | ||
fccddba0 | 424 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 425 | { |
45b4233c DW |
426 | pr_debug("remove_hash(), stripe %llu\n", |
427 | (unsigned long long)sh->sector); | |
1da177e4 | 428 | |
fccddba0 | 429 | hlist_del_init(&sh->hash); |
1da177e4 LT |
430 | } |
431 | ||
d1688a6d | 432 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 433 | { |
fccddba0 | 434 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 435 | |
45b4233c DW |
436 | pr_debug("insert_hash(), stripe %llu\n", |
437 | (unsigned long long)sh->sector); | |
1da177e4 | 438 | |
fccddba0 | 439 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
440 | } |
441 | ||
1da177e4 | 442 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 443 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
444 | { |
445 | struct stripe_head *sh = NULL; | |
446 | struct list_head *first; | |
447 | ||
566c09c5 | 448 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 449 | goto out; |
566c09c5 | 450 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
451 | sh = list_entry(first, struct stripe_head, lru); |
452 | list_del_init(first); | |
453 | remove_hash(sh); | |
454 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 455 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
456 | if (list_empty(conf->inactive_list + hash)) |
457 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
458 | out: |
459 | return sh; | |
460 | } | |
461 | ||
e4e11e38 | 462 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
463 | { |
464 | struct page *p; | |
465 | int i; | |
e4e11e38 | 466 | int num = sh->raid_conf->pool_size; |
1da177e4 | 467 | |
e4e11e38 | 468 | for (i = 0; i < num ; i++) { |
d592a996 | 469 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
470 | p = sh->dev[i].page; |
471 | if (!p) | |
472 | continue; | |
473 | sh->dev[i].page = NULL; | |
2d1f3b5d | 474 | put_page(p); |
1da177e4 LT |
475 | } |
476 | } | |
477 | ||
a9683a79 | 478 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
479 | { |
480 | int i; | |
e4e11e38 | 481 | int num = sh->raid_conf->pool_size; |
1da177e4 | 482 | |
e4e11e38 | 483 | for (i = 0; i < num; i++) { |
1da177e4 LT |
484 | struct page *page; |
485 | ||
a9683a79 | 486 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
487 | return 1; |
488 | } | |
489 | sh->dev[i].page = page; | |
d592a996 | 490 | sh->dev[i].orig_page = page; |
1da177e4 | 491 | } |
3418d036 | 492 | |
1da177e4 LT |
493 | return 0; |
494 | } | |
495 | ||
d1688a6d | 496 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 497 | struct stripe_head *sh); |
1da177e4 | 498 | |
b5663ba4 | 499 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 500 | { |
d1688a6d | 501 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 502 | int i, seq; |
1da177e4 | 503 | |
78bafebd ES |
504 | BUG_ON(atomic_read(&sh->count) != 0); |
505 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 506 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 507 | BUG_ON(sh->batch_head); |
d84e0f10 | 508 | |
45b4233c | 509 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 510 | (unsigned long long)sector); |
566c09c5 SL |
511 | retry: |
512 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 513 | sh->generation = conf->generation - previous; |
b5663ba4 | 514 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 515 | sh->sector = sector; |
911d4ee8 | 516 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
517 | sh->state = 0; |
518 | ||
7ecaa1e6 | 519 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
520 | struct r5dev *dev = &sh->dev[i]; |
521 | ||
d84e0f10 | 522 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 523 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 524 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 525 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 526 | dev->read, dev->towrite, dev->written, |
1da177e4 | 527 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 528 | WARN_ON(1); |
1da177e4 LT |
529 | } |
530 | dev->flags = 0; | |
27a4ff8f | 531 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 | 532 | } |
566c09c5 SL |
533 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
534 | goto retry; | |
7a87f434 | 535 | sh->overwrite_disks = 0; |
1da177e4 | 536 | insert_hash(conf, sh); |
851c30c9 | 537 | sh->cpu = smp_processor_id(); |
da41ba65 | 538 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
539 | } |
540 | ||
d1688a6d | 541 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 542 | short generation) |
1da177e4 LT |
543 | { |
544 | struct stripe_head *sh; | |
545 | ||
45b4233c | 546 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 547 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 548 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 549 | return sh; |
45b4233c | 550 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
551 | return NULL; |
552 | } | |
553 | ||
674806d6 N |
554 | /* |
555 | * Need to check if array has failed when deciding whether to: | |
556 | * - start an array | |
557 | * - remove non-faulty devices | |
558 | * - add a spare | |
559 | * - allow a reshape | |
560 | * This determination is simple when no reshape is happening. | |
561 | * However if there is a reshape, we need to carefully check | |
562 | * both the before and after sections. | |
563 | * This is because some failed devices may only affect one | |
564 | * of the two sections, and some non-in_sync devices may | |
565 | * be insync in the section most affected by failed devices. | |
566 | */ | |
2e38a37f | 567 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 568 | { |
908f4fbd | 569 | int degraded, degraded2; |
674806d6 | 570 | int i; |
674806d6 N |
571 | |
572 | rcu_read_lock(); | |
573 | degraded = 0; | |
574 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 575 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
576 | if (rdev && test_bit(Faulty, &rdev->flags)) |
577 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
578 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
579 | degraded++; | |
580 | else if (test_bit(In_sync, &rdev->flags)) | |
581 | ; | |
582 | else | |
583 | /* not in-sync or faulty. | |
584 | * If the reshape increases the number of devices, | |
585 | * this is being recovered by the reshape, so | |
586 | * this 'previous' section is not in_sync. | |
587 | * If the number of devices is being reduced however, | |
588 | * the device can only be part of the array if | |
589 | * we are reverting a reshape, so this section will | |
590 | * be in-sync. | |
591 | */ | |
592 | if (conf->raid_disks >= conf->previous_raid_disks) | |
593 | degraded++; | |
594 | } | |
595 | rcu_read_unlock(); | |
908f4fbd N |
596 | if (conf->raid_disks == conf->previous_raid_disks) |
597 | return degraded; | |
674806d6 | 598 | rcu_read_lock(); |
908f4fbd | 599 | degraded2 = 0; |
674806d6 | 600 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 601 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
602 | if (rdev && test_bit(Faulty, &rdev->flags)) |
603 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 604 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 605 | degraded2++; |
674806d6 N |
606 | else if (test_bit(In_sync, &rdev->flags)) |
607 | ; | |
608 | else | |
609 | /* not in-sync or faulty. | |
610 | * If reshape increases the number of devices, this | |
611 | * section has already been recovered, else it | |
612 | * almost certainly hasn't. | |
613 | */ | |
614 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 615 | degraded2++; |
674806d6 N |
616 | } |
617 | rcu_read_unlock(); | |
908f4fbd N |
618 | if (degraded2 > degraded) |
619 | return degraded2; | |
620 | return degraded; | |
621 | } | |
622 | ||
623 | static int has_failed(struct r5conf *conf) | |
624 | { | |
625 | int degraded; | |
626 | ||
627 | if (conf->mddev->reshape_position == MaxSector) | |
628 | return conf->mddev->degraded > conf->max_degraded; | |
629 | ||
2e38a37f | 630 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
631 | if (degraded > conf->max_degraded) |
632 | return 1; | |
633 | return 0; | |
634 | } | |
635 | ||
6d036f7d SL |
636 | struct stripe_head * |
637 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
638 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
639 | { |
640 | struct stripe_head *sh; | |
566c09c5 | 641 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 642 | int inc_empty_inactive_list_flag; |
1da177e4 | 643 | |
45b4233c | 644 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 645 | |
566c09c5 | 646 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
647 | |
648 | do { | |
b1b46486 | 649 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 650 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 651 | *(conf->hash_locks + hash)); |
86b42c71 | 652 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 653 | if (!sh) { |
edbe83ab | 654 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 655 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
656 | if (!sh && !test_bit(R5_DID_ALLOC, |
657 | &conf->cache_state)) | |
edbe83ab N |
658 | set_bit(R5_ALLOC_MORE, |
659 | &conf->cache_state); | |
660 | } | |
1da177e4 LT |
661 | if (noblock && sh == NULL) |
662 | break; | |
a39f7afd SL |
663 | |
664 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 665 | if (!sh) { |
5423399a N |
666 | set_bit(R5_INACTIVE_BLOCKED, |
667 | &conf->cache_state); | |
a39f7afd | 668 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
669 | wait_event_lock_irq( |
670 | conf->wait_for_stripe, | |
566c09c5 SL |
671 | !list_empty(conf->inactive_list + hash) && |
672 | (atomic_read(&conf->active_stripes) | |
673 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
674 | || !test_bit(R5_INACTIVE_BLOCKED, |
675 | &conf->cache_state)), | |
6ab2a4b8 | 676 | *(conf->hash_locks + hash)); |
5423399a N |
677 | clear_bit(R5_INACTIVE_BLOCKED, |
678 | &conf->cache_state); | |
7da9d450 | 679 | } else { |
b5663ba4 | 680 | init_stripe(sh, sector, previous); |
7da9d450 N |
681 | atomic_inc(&sh->count); |
682 | } | |
e240c183 | 683 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 684 | spin_lock(&conf->device_lock); |
e240c183 | 685 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
686 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
687 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
688 | BUG_ON(list_empty(&sh->lru) && |
689 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
690 | inc_empty_inactive_list_flag = 0; |
691 | if (!list_empty(conf->inactive_list + hash)) | |
692 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 693 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
694 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
695 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
696 | if (sh->group) { |
697 | sh->group->stripes_cnt--; | |
698 | sh->group = NULL; | |
699 | } | |
1da177e4 | 700 | } |
7da9d450 | 701 | atomic_inc(&sh->count); |
6d183de4 | 702 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
703 | } |
704 | } while (sh == NULL); | |
705 | ||
566c09c5 | 706 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
707 | return sh; |
708 | } | |
709 | ||
7a87f434 | 710 | static bool is_full_stripe_write(struct stripe_head *sh) |
711 | { | |
712 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
713 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
714 | } | |
715 | ||
59fc630b | 716 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
717 | { | |
59fc630b | 718 | if (sh1 > sh2) { |
3d05f3ae | 719 | spin_lock_irq(&sh2->stripe_lock); |
59fc630b | 720 | spin_lock_nested(&sh1->stripe_lock, 1); |
721 | } else { | |
3d05f3ae | 722 | spin_lock_irq(&sh1->stripe_lock); |
59fc630b | 723 | spin_lock_nested(&sh2->stripe_lock, 1); |
724 | } | |
725 | } | |
726 | ||
727 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
728 | { | |
729 | spin_unlock(&sh1->stripe_lock); | |
3d05f3ae | 730 | spin_unlock_irq(&sh2->stripe_lock); |
59fc630b | 731 | } |
732 | ||
733 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
734 | static bool stripe_can_batch(struct stripe_head *sh) | |
735 | { | |
9c3e333d SL |
736 | struct r5conf *conf = sh->raid_conf; |
737 | ||
3418d036 | 738 | if (conf->log || raid5_has_ppl(conf)) |
9c3e333d | 739 | return false; |
59fc630b | 740 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 741 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 742 | is_full_stripe_write(sh); |
743 | } | |
744 | ||
745 | /* we only do back search */ | |
746 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
747 | { | |
748 | struct stripe_head *head; | |
749 | sector_t head_sector, tmp_sec; | |
750 | int hash; | |
751 | int dd_idx; | |
ff00d3b4 | 752 | int inc_empty_inactive_list_flag; |
59fc630b | 753 | |
59fc630b | 754 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
755 | tmp_sec = sh->sector; | |
756 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
757 | return; | |
758 | head_sector = sh->sector - STRIPE_SECTORS; | |
759 | ||
760 | hash = stripe_hash_locks_hash(head_sector); | |
761 | spin_lock_irq(conf->hash_locks + hash); | |
762 | head = __find_stripe(conf, head_sector, conf->generation); | |
763 | if (head && !atomic_inc_not_zero(&head->count)) { | |
764 | spin_lock(&conf->device_lock); | |
765 | if (!atomic_read(&head->count)) { | |
766 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
767 | atomic_inc(&conf->active_stripes); | |
768 | BUG_ON(list_empty(&head->lru) && | |
769 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
770 | inc_empty_inactive_list_flag = 0; |
771 | if (!list_empty(conf->inactive_list + hash)) | |
772 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 773 | list_del_init(&head->lru); |
ff00d3b4 ZL |
774 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
775 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 776 | if (head->group) { |
777 | head->group->stripes_cnt--; | |
778 | head->group = NULL; | |
779 | } | |
780 | } | |
781 | atomic_inc(&head->count); | |
782 | spin_unlock(&conf->device_lock); | |
783 | } | |
784 | spin_unlock_irq(conf->hash_locks + hash); | |
785 | ||
786 | if (!head) | |
787 | return; | |
788 | if (!stripe_can_batch(head)) | |
789 | goto out; | |
790 | ||
791 | lock_two_stripes(head, sh); | |
792 | /* clear_batch_ready clear the flag */ | |
793 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
794 | goto unlock_out; | |
795 | ||
796 | if (sh->batch_head) | |
797 | goto unlock_out; | |
798 | ||
799 | dd_idx = 0; | |
800 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
801 | dd_idx++; | |
1eff9d32 | 802 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 803 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 804 | goto unlock_out; |
805 | ||
806 | if (head->batch_head) { | |
807 | spin_lock(&head->batch_head->batch_lock); | |
808 | /* This batch list is already running */ | |
809 | if (!stripe_can_batch(head)) { | |
810 | spin_unlock(&head->batch_head->batch_lock); | |
811 | goto unlock_out; | |
812 | } | |
3664847d SL |
813 | /* |
814 | * We must assign batch_head of this stripe within the | |
815 | * batch_lock, otherwise clear_batch_ready of batch head | |
816 | * stripe could clear BATCH_READY bit of this stripe and | |
817 | * this stripe->batch_head doesn't get assigned, which | |
818 | * could confuse clear_batch_ready for this stripe | |
819 | */ | |
820 | sh->batch_head = head->batch_head; | |
59fc630b | 821 | |
822 | /* | |
823 | * at this point, head's BATCH_READY could be cleared, but we | |
824 | * can still add the stripe to batch list | |
825 | */ | |
826 | list_add(&sh->batch_list, &head->batch_list); | |
827 | spin_unlock(&head->batch_head->batch_lock); | |
59fc630b | 828 | } else { |
829 | head->batch_head = head; | |
830 | sh->batch_head = head->batch_head; | |
831 | spin_lock(&head->batch_lock); | |
832 | list_add_tail(&sh->batch_list, &head->batch_list); | |
833 | spin_unlock(&head->batch_lock); | |
834 | } | |
835 | ||
836 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
837 | if (atomic_dec_return(&conf->preread_active_stripes) | |
838 | < IO_THRESHOLD) | |
839 | md_wakeup_thread(conf->mddev->thread); | |
840 | ||
2b6b2457 N |
841 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
842 | int seq = sh->bm_seq; | |
843 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
844 | sh->batch_head->bm_seq > seq) | |
845 | seq = sh->batch_head->bm_seq; | |
846 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
847 | sh->batch_head->bm_seq = seq; | |
848 | } | |
849 | ||
59fc630b | 850 | atomic_inc(&sh->count); |
851 | unlock_out: | |
852 | unlock_two_stripes(head, sh); | |
853 | out: | |
6d036f7d | 854 | raid5_release_stripe(head); |
59fc630b | 855 | } |
856 | ||
05616be5 N |
857 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
858 | * in this stripe_head. | |
859 | */ | |
860 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
861 | { | |
862 | sector_t progress = conf->reshape_progress; | |
863 | /* Need a memory barrier to make sure we see the value | |
864 | * of conf->generation, or ->data_offset that was set before | |
865 | * reshape_progress was updated. | |
866 | */ | |
867 | smp_rmb(); | |
868 | if (progress == MaxSector) | |
869 | return 0; | |
870 | if (sh->generation == conf->generation - 1) | |
871 | return 0; | |
872 | /* We are in a reshape, and this is a new-generation stripe, | |
873 | * so use new_data_offset. | |
874 | */ | |
875 | return 1; | |
876 | } | |
877 | ||
aaf9f12e | 878 | static void dispatch_bio_list(struct bio_list *tmp) |
765d704d | 879 | { |
765d704d SL |
880 | struct bio *bio; |
881 | ||
aaf9f12e SL |
882 | while ((bio = bio_list_pop(tmp))) |
883 | generic_make_request(bio); | |
884 | } | |
885 | ||
886 | static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b) | |
887 | { | |
888 | const struct r5pending_data *da = list_entry(a, | |
889 | struct r5pending_data, sibling); | |
890 | const struct r5pending_data *db = list_entry(b, | |
891 | struct r5pending_data, sibling); | |
892 | if (da->sector > db->sector) | |
893 | return 1; | |
894 | if (da->sector < db->sector) | |
895 | return -1; | |
896 | return 0; | |
897 | } | |
898 | ||
899 | static void dispatch_defer_bios(struct r5conf *conf, int target, | |
900 | struct bio_list *list) | |
901 | { | |
902 | struct r5pending_data *data; | |
903 | struct list_head *first, *next = NULL; | |
904 | int cnt = 0; | |
905 | ||
906 | if (conf->pending_data_cnt == 0) | |
907 | return; | |
908 | ||
909 | list_sort(NULL, &conf->pending_list, cmp_stripe); | |
910 | ||
911 | first = conf->pending_list.next; | |
912 | ||
913 | /* temporarily move the head */ | |
914 | if (conf->next_pending_data) | |
915 | list_move_tail(&conf->pending_list, | |
916 | &conf->next_pending_data->sibling); | |
917 | ||
918 | while (!list_empty(&conf->pending_list)) { | |
919 | data = list_first_entry(&conf->pending_list, | |
920 | struct r5pending_data, sibling); | |
921 | if (&data->sibling == first) | |
922 | first = data->sibling.next; | |
923 | next = data->sibling.next; | |
924 | ||
925 | bio_list_merge(list, &data->bios); | |
926 | list_move(&data->sibling, &conf->free_list); | |
927 | cnt++; | |
928 | if (cnt >= target) | |
929 | break; | |
930 | } | |
931 | conf->pending_data_cnt -= cnt; | |
932 | BUG_ON(conf->pending_data_cnt < 0 || cnt < target); | |
933 | ||
934 | if (next != &conf->pending_list) | |
935 | conf->next_pending_data = list_entry(next, | |
936 | struct r5pending_data, sibling); | |
937 | else | |
938 | conf->next_pending_data = NULL; | |
939 | /* list isn't empty */ | |
940 | if (first != &conf->pending_list) | |
941 | list_move_tail(&conf->pending_list, first); | |
942 | } | |
943 | ||
944 | static void flush_deferred_bios(struct r5conf *conf) | |
945 | { | |
946 | struct bio_list tmp = BIO_EMPTY_LIST; | |
947 | ||
948 | if (conf->pending_data_cnt == 0) | |
765d704d SL |
949 | return; |
950 | ||
765d704d | 951 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
952 | dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp); |
953 | BUG_ON(conf->pending_data_cnt != 0); | |
765d704d SL |
954 | spin_unlock(&conf->pending_bios_lock); |
955 | ||
aaf9f12e | 956 | dispatch_bio_list(&tmp); |
765d704d SL |
957 | } |
958 | ||
aaf9f12e SL |
959 | static void defer_issue_bios(struct r5conf *conf, sector_t sector, |
960 | struct bio_list *bios) | |
765d704d | 961 | { |
aaf9f12e SL |
962 | struct bio_list tmp = BIO_EMPTY_LIST; |
963 | struct r5pending_data *ent; | |
964 | ||
765d704d | 965 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
966 | ent = list_first_entry(&conf->free_list, struct r5pending_data, |
967 | sibling); | |
968 | list_move_tail(&ent->sibling, &conf->pending_list); | |
969 | ent->sector = sector; | |
970 | bio_list_init(&ent->bios); | |
971 | bio_list_merge(&ent->bios, bios); | |
972 | conf->pending_data_cnt++; | |
973 | if (conf->pending_data_cnt >= PENDING_IO_MAX) | |
974 | dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp); | |
975 | ||
765d704d | 976 | spin_unlock(&conf->pending_bios_lock); |
aaf9f12e SL |
977 | |
978 | dispatch_bio_list(&tmp); | |
765d704d SL |
979 | } |
980 | ||
6712ecf8 | 981 | static void |
4246a0b6 | 982 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 983 | static void |
4246a0b6 | 984 | raid5_end_write_request(struct bio *bi); |
91c00924 | 985 | |
c4e5ac0a | 986 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 987 | { |
d1688a6d | 988 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 989 | int i, disks = sh->disks; |
59fc630b | 990 | struct stripe_head *head_sh = sh; |
aaf9f12e SL |
991 | struct bio_list pending_bios = BIO_EMPTY_LIST; |
992 | bool should_defer; | |
91c00924 DW |
993 | |
994 | might_sleep(); | |
995 | ||
ff875738 AP |
996 | if (log_stripe(sh, s) == 0) |
997 | return; | |
1e6d690b | 998 | |
aaf9f12e | 999 | should_defer = conf->batch_bio_dispatch && conf->group_cnt; |
1e6d690b | 1000 | |
91c00924 | 1001 | for (i = disks; i--; ) { |
796a5cf0 | 1002 | int op, op_flags = 0; |
9a3e1101 | 1003 | int replace_only = 0; |
977df362 N |
1004 | struct bio *bi, *rbi; |
1005 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 1006 | |
1007 | sh = head_sh; | |
e9c7469b | 1008 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 1009 | op = REQ_OP_WRITE; |
e9c7469b | 1010 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 1011 | op_flags = REQ_FUA; |
9e444768 | 1012 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 1013 | op = REQ_OP_DISCARD; |
e9c7469b | 1014 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 1015 | op = REQ_OP_READ; |
9a3e1101 N |
1016 | else if (test_and_clear_bit(R5_WantReplace, |
1017 | &sh->dev[i].flags)) { | |
796a5cf0 | 1018 | op = REQ_OP_WRITE; |
9a3e1101 N |
1019 | replace_only = 1; |
1020 | } else | |
91c00924 | 1021 | continue; |
bc0934f0 | 1022 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 1023 | op_flags |= REQ_SYNC; |
91c00924 | 1024 | |
59fc630b | 1025 | again: |
91c00924 | 1026 | bi = &sh->dev[i].req; |
977df362 | 1027 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 1028 | |
91c00924 | 1029 | rcu_read_lock(); |
9a3e1101 | 1030 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
1031 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
1032 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1033 | if (!rdev) { | |
1034 | rdev = rrdev; | |
1035 | rrdev = NULL; | |
1036 | } | |
796a5cf0 | 1037 | if (op_is_write(op)) { |
9a3e1101 N |
1038 | if (replace_only) |
1039 | rdev = NULL; | |
dd054fce N |
1040 | if (rdev == rrdev) |
1041 | /* We raced and saw duplicates */ | |
1042 | rrdev = NULL; | |
9a3e1101 | 1043 | } else { |
59fc630b | 1044 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
1045 | rdev = rrdev; |
1046 | rrdev = NULL; | |
1047 | } | |
977df362 | 1048 | |
91c00924 DW |
1049 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1050 | rdev = NULL; | |
1051 | if (rdev) | |
1052 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
1053 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
1054 | rrdev = NULL; | |
1055 | if (rrdev) | |
1056 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1057 | rcu_read_unlock(); |
1058 | ||
73e92e51 | 1059 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1060 | * need to check for writes. We never accept write errors |
1061 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1062 | */ |
796a5cf0 | 1063 | while (op_is_write(op) && rdev && |
73e92e51 N |
1064 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1065 | sector_t first_bad; | |
1066 | int bad_sectors; | |
1067 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
1068 | &first_bad, &bad_sectors); | |
1069 | if (!bad) | |
1070 | break; | |
1071 | ||
1072 | if (bad < 0) { | |
1073 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1074 | if (!conf->mddev->external && | |
2953079c | 1075 | conf->mddev->sb_flags) { |
73e92e51 N |
1076 | /* It is very unlikely, but we might |
1077 | * still need to write out the | |
1078 | * bad block log - better give it | |
1079 | * a chance*/ | |
1080 | md_check_recovery(conf->mddev); | |
1081 | } | |
1850753d | 1082 | /* |
1083 | * Because md_wait_for_blocked_rdev | |
1084 | * will dec nr_pending, we must | |
1085 | * increment it first. | |
1086 | */ | |
1087 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1088 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1089 | } else { | |
1090 | /* Acknowledged bad block - skip the write */ | |
1091 | rdev_dec_pending(rdev, conf->mddev); | |
1092 | rdev = NULL; | |
1093 | } | |
1094 | } | |
1095 | ||
91c00924 | 1096 | if (rdev) { |
9a3e1101 N |
1097 | if (s->syncing || s->expanding || s->expanded |
1098 | || s->replacing) | |
91c00924 DW |
1099 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1100 | ||
2b7497f0 DW |
1101 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1102 | ||
74d46992 | 1103 | bio_set_dev(bi, rdev->bdev); |
796a5cf0 MC |
1104 | bio_set_op_attrs(bi, op, op_flags); |
1105 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1106 | ? raid5_end_write_request |
1107 | : raid5_end_read_request; | |
1108 | bi->bi_private = sh; | |
1109 | ||
6296b960 | 1110 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1111 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1112 | bi->bi_opf, i); |
91c00924 | 1113 | atomic_inc(&sh->count); |
59fc630b | 1114 | if (sh != head_sh) |
1115 | atomic_inc(&head_sh->count); | |
05616be5 | 1116 | if (use_new_offset(conf, sh)) |
4f024f37 | 1117 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1118 | + rdev->new_data_offset); |
1119 | else | |
4f024f37 | 1120 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1121 | + rdev->data_offset); |
59fc630b | 1122 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1123 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1124 | |
d592a996 SL |
1125 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1126 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1127 | |
1128 | if (!op_is_write(op) && | |
1129 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1130 | /* | |
1131 | * issuing read for a page in journal, this | |
1132 | * must be preparing for prexor in rmw; read | |
1133 | * the data into orig_page | |
1134 | */ | |
1135 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1136 | else | |
1137 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1138 | bi->bi_vcnt = 1; |
91c00924 DW |
1139 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1140 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1141 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1142 | /* |
1143 | * If this is discard request, set bi_vcnt 0. We don't | |
1144 | * want to confuse SCSI because SCSI will replace payload | |
1145 | */ | |
796a5cf0 | 1146 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1147 | bi->bi_vcnt = 0; |
977df362 N |
1148 | if (rrdev) |
1149 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1150 | |
1151 | if (conf->mddev->gendisk) | |
74d46992 | 1152 | trace_block_bio_remap(bi->bi_disk->queue, |
e3620a3a JB |
1153 | bi, disk_devt(conf->mddev->gendisk), |
1154 | sh->dev[i].sector); | |
aaf9f12e SL |
1155 | if (should_defer && op_is_write(op)) |
1156 | bio_list_add(&pending_bios, bi); | |
1157 | else | |
1158 | generic_make_request(bi); | |
977df362 N |
1159 | } |
1160 | if (rrdev) { | |
9a3e1101 N |
1161 | if (s->syncing || s->expanding || s->expanded |
1162 | || s->replacing) | |
977df362 N |
1163 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1164 | ||
1165 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1166 | ||
74d46992 | 1167 | bio_set_dev(rbi, rrdev->bdev); |
796a5cf0 MC |
1168 | bio_set_op_attrs(rbi, op, op_flags); |
1169 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1170 | rbi->bi_end_io = raid5_end_write_request; |
1171 | rbi->bi_private = sh; | |
1172 | ||
6296b960 | 1173 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1174 | "replacement disc %d\n", |
1175 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1176 | rbi->bi_opf, i); |
977df362 | 1177 | atomic_inc(&sh->count); |
59fc630b | 1178 | if (sh != head_sh) |
1179 | atomic_inc(&head_sh->count); | |
05616be5 | 1180 | if (use_new_offset(conf, sh)) |
4f024f37 | 1181 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1182 | + rrdev->new_data_offset); |
1183 | else | |
4f024f37 | 1184 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1185 | + rrdev->data_offset); |
d592a996 SL |
1186 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1187 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1188 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1189 | rbi->bi_vcnt = 1; |
977df362 N |
1190 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1191 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1192 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1193 | /* |
1194 | * If this is discard request, set bi_vcnt 0. We don't | |
1195 | * want to confuse SCSI because SCSI will replace payload | |
1196 | */ | |
796a5cf0 | 1197 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1198 | rbi->bi_vcnt = 0; |
e3620a3a | 1199 | if (conf->mddev->gendisk) |
74d46992 | 1200 | trace_block_bio_remap(rbi->bi_disk->queue, |
e3620a3a JB |
1201 | rbi, disk_devt(conf->mddev->gendisk), |
1202 | sh->dev[i].sector); | |
aaf9f12e SL |
1203 | if (should_defer && op_is_write(op)) |
1204 | bio_list_add(&pending_bios, rbi); | |
1205 | else | |
1206 | generic_make_request(rbi); | |
977df362 N |
1207 | } |
1208 | if (!rdev && !rrdev) { | |
796a5cf0 | 1209 | if (op_is_write(op)) |
91c00924 | 1210 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1211 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1212 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1213 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1214 | set_bit(STRIPE_HANDLE, &sh->state); | |
1215 | } | |
59fc630b | 1216 | |
1217 | if (!head_sh->batch_head) | |
1218 | continue; | |
1219 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1220 | batch_list); | |
1221 | if (sh != head_sh) | |
1222 | goto again; | |
91c00924 | 1223 | } |
aaf9f12e SL |
1224 | |
1225 | if (should_defer && !bio_list_empty(&pending_bios)) | |
1226 | defer_issue_bios(conf, head_sh->sector, &pending_bios); | |
91c00924 DW |
1227 | } |
1228 | ||
1229 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1230 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1231 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1232 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1233 | { |
7988613b KO |
1234 | struct bio_vec bvl; |
1235 | struct bvec_iter iter; | |
91c00924 | 1236 | struct page *bio_page; |
91c00924 | 1237 | int page_offset; |
a08abd8c | 1238 | struct async_submit_ctl submit; |
0403e382 | 1239 | enum async_tx_flags flags = 0; |
91c00924 | 1240 | |
4f024f37 KO |
1241 | if (bio->bi_iter.bi_sector >= sector) |
1242 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1243 | else |
4f024f37 | 1244 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1245 | |
0403e382 DW |
1246 | if (frombio) |
1247 | flags |= ASYNC_TX_FENCE; | |
1248 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1249 | ||
7988613b KO |
1250 | bio_for_each_segment(bvl, bio, iter) { |
1251 | int len = bvl.bv_len; | |
91c00924 DW |
1252 | int clen; |
1253 | int b_offset = 0; | |
1254 | ||
1255 | if (page_offset < 0) { | |
1256 | b_offset = -page_offset; | |
1257 | page_offset += b_offset; | |
1258 | len -= b_offset; | |
1259 | } | |
1260 | ||
1261 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1262 | clen = STRIPE_SIZE - page_offset; | |
1263 | else | |
1264 | clen = len; | |
1265 | ||
1266 | if (clen > 0) { | |
7988613b KO |
1267 | b_offset += bvl.bv_offset; |
1268 | bio_page = bvl.bv_page; | |
d592a996 SL |
1269 | if (frombio) { |
1270 | if (sh->raid_conf->skip_copy && | |
1271 | b_offset == 0 && page_offset == 0 && | |
1e6d690b SL |
1272 | clen == STRIPE_SIZE && |
1273 | !no_skipcopy) | |
d592a996 SL |
1274 | *page = bio_page; |
1275 | else | |
1276 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1277 | b_offset, clen, &submit); |
d592a996 SL |
1278 | } else |
1279 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1280 | page_offset, clen, &submit); |
91c00924 | 1281 | } |
a08abd8c DW |
1282 | /* chain the operations */ |
1283 | submit.depend_tx = tx; | |
1284 | ||
91c00924 DW |
1285 | if (clen < len) /* hit end of page */ |
1286 | break; | |
1287 | page_offset += len; | |
1288 | } | |
1289 | ||
1290 | return tx; | |
1291 | } | |
1292 | ||
1293 | static void ops_complete_biofill(void *stripe_head_ref) | |
1294 | { | |
1295 | struct stripe_head *sh = stripe_head_ref; | |
e4d84909 | 1296 | int i; |
91c00924 | 1297 | |
e46b272b | 1298 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1299 | (unsigned long long)sh->sector); |
1300 | ||
1301 | /* clear completed biofills */ | |
1302 | for (i = sh->disks; i--; ) { | |
1303 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1304 | |
1305 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1306 | /* and check if we need to reply to a read request, |
1307 | * new R5_Wantfill requests are held off until | |
83de75cc | 1308 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1309 | */ |
1310 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1311 | struct bio *rbi, *rbi2; |
91c00924 | 1312 | |
91c00924 DW |
1313 | BUG_ON(!dev->read); |
1314 | rbi = dev->read; | |
1315 | dev->read = NULL; | |
4f024f37 | 1316 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1317 | dev->sector + STRIPE_SECTORS) { |
1318 | rbi2 = r5_next_bio(rbi, dev->sector); | |
016c76ac | 1319 | bio_endio(rbi); |
91c00924 DW |
1320 | rbi = rbi2; |
1321 | } | |
1322 | } | |
1323 | } | |
83de75cc | 1324 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1325 | |
e4d84909 | 1326 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1327 | raid5_release_stripe(sh); |
91c00924 DW |
1328 | } |
1329 | ||
1330 | static void ops_run_biofill(struct stripe_head *sh) | |
1331 | { | |
1332 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1333 | struct async_submit_ctl submit; |
91c00924 DW |
1334 | int i; |
1335 | ||
59fc630b | 1336 | BUG_ON(sh->batch_head); |
e46b272b | 1337 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1338 | (unsigned long long)sh->sector); |
1339 | ||
1340 | for (i = sh->disks; i--; ) { | |
1341 | struct r5dev *dev = &sh->dev[i]; | |
1342 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1343 | struct bio *rbi; | |
b17459c0 | 1344 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1345 | dev->read = rbi = dev->toread; |
1346 | dev->toread = NULL; | |
b17459c0 | 1347 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1348 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1349 | dev->sector + STRIPE_SECTORS) { |
d592a996 | 1350 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1351 | dev->sector, tx, sh, 0); |
91c00924 DW |
1352 | rbi = r5_next_bio(rbi, dev->sector); |
1353 | } | |
1354 | } | |
1355 | } | |
1356 | ||
1357 | atomic_inc(&sh->count); | |
a08abd8c DW |
1358 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1359 | async_trigger_callback(&submit); | |
91c00924 DW |
1360 | } |
1361 | ||
4e7d2c0a | 1362 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1363 | { |
4e7d2c0a | 1364 | struct r5dev *tgt; |
91c00924 | 1365 | |
4e7d2c0a DW |
1366 | if (target < 0) |
1367 | return; | |
91c00924 | 1368 | |
4e7d2c0a | 1369 | tgt = &sh->dev[target]; |
91c00924 DW |
1370 | set_bit(R5_UPTODATE, &tgt->flags); |
1371 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1372 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1373 | } |
1374 | ||
ac6b53b6 | 1375 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1376 | { |
1377 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1378 | |
e46b272b | 1379 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1380 | (unsigned long long)sh->sector); |
1381 | ||
ac6b53b6 | 1382 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1383 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1384 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1385 | |
ecc65c9b DW |
1386 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1387 | if (sh->check_state == check_state_compute_run) | |
1388 | sh->check_state = check_state_compute_result; | |
91c00924 | 1389 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1390 | raid5_release_stripe(sh); |
91c00924 DW |
1391 | } |
1392 | ||
d6f38f31 DW |
1393 | /* return a pointer to the address conversion region of the scribble buffer */ |
1394 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1395 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1396 | { |
46d5b785 | 1397 | void *addr; |
1398 | ||
1399 | addr = flex_array_get(percpu->scribble, i); | |
1400 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1401 | } | |
1402 | ||
1403 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1404 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1405 | { | |
1406 | void *addr; | |
1407 | ||
1408 | addr = flex_array_get(percpu->scribble, i); | |
1409 | return addr; | |
d6f38f31 DW |
1410 | } |
1411 | ||
1412 | static struct dma_async_tx_descriptor * | |
1413 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1414 | { |
91c00924 | 1415 | int disks = sh->disks; |
46d5b785 | 1416 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1417 | int target = sh->ops.target; |
1418 | struct r5dev *tgt = &sh->dev[target]; | |
1419 | struct page *xor_dest = tgt->page; | |
1420 | int count = 0; | |
1421 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1422 | struct async_submit_ctl submit; |
91c00924 DW |
1423 | int i; |
1424 | ||
59fc630b | 1425 | BUG_ON(sh->batch_head); |
1426 | ||
91c00924 | 1427 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1428 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1429 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1430 | ||
1431 | for (i = disks; i--; ) | |
1432 | if (i != target) | |
1433 | xor_srcs[count++] = sh->dev[i].page; | |
1434 | ||
1435 | atomic_inc(&sh->count); | |
1436 | ||
0403e382 | 1437 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1438 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1439 | if (unlikely(count == 1)) |
a08abd8c | 1440 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1441 | else |
a08abd8c | 1442 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1443 | |
91c00924 DW |
1444 | return tx; |
1445 | } | |
1446 | ||
ac6b53b6 DW |
1447 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1448 | * @srcs - (struct page *) array of size sh->disks | |
1449 | * @sh - stripe_head to parse | |
1450 | * | |
1451 | * Populates srcs in proper layout order for the stripe and returns the | |
1452 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1453 | * destination buffer is recorded in srcs[count] and the Q destination | |
1454 | * is recorded in srcs[count+1]]. | |
1455 | */ | |
584acdd4 MS |
1456 | static int set_syndrome_sources(struct page **srcs, |
1457 | struct stripe_head *sh, | |
1458 | int srctype) | |
ac6b53b6 DW |
1459 | { |
1460 | int disks = sh->disks; | |
1461 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1462 | int d0_idx = raid6_d0(sh); | |
1463 | int count; | |
1464 | int i; | |
1465 | ||
1466 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1467 | srcs[i] = NULL; |
ac6b53b6 DW |
1468 | |
1469 | count = 0; | |
1470 | i = d0_idx; | |
1471 | do { | |
1472 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1473 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1474 | |
584acdd4 MS |
1475 | if (i == sh->qd_idx || i == sh->pd_idx || |
1476 | (srctype == SYNDROME_SRC_ALL) || | |
1477 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1478 | (test_bit(R5_Wantdrain, &dev->flags) || |
1479 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1480 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1481 | (dev->written || |
1482 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1483 | if (test_bit(R5_InJournal, &dev->flags)) |
1484 | srcs[slot] = sh->dev[i].orig_page; | |
1485 | else | |
1486 | srcs[slot] = sh->dev[i].page; | |
1487 | } | |
ac6b53b6 DW |
1488 | i = raid6_next_disk(i, disks); |
1489 | } while (i != d0_idx); | |
ac6b53b6 | 1490 | |
e4424fee | 1491 | return syndrome_disks; |
ac6b53b6 DW |
1492 | } |
1493 | ||
1494 | static struct dma_async_tx_descriptor * | |
1495 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1496 | { | |
1497 | int disks = sh->disks; | |
46d5b785 | 1498 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1499 | int target; |
1500 | int qd_idx = sh->qd_idx; | |
1501 | struct dma_async_tx_descriptor *tx; | |
1502 | struct async_submit_ctl submit; | |
1503 | struct r5dev *tgt; | |
1504 | struct page *dest; | |
1505 | int i; | |
1506 | int count; | |
1507 | ||
59fc630b | 1508 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1509 | if (sh->ops.target < 0) |
1510 | target = sh->ops.target2; | |
1511 | else if (sh->ops.target2 < 0) | |
1512 | target = sh->ops.target; | |
91c00924 | 1513 | else |
ac6b53b6 DW |
1514 | /* we should only have one valid target */ |
1515 | BUG(); | |
1516 | BUG_ON(target < 0); | |
1517 | pr_debug("%s: stripe %llu block: %d\n", | |
1518 | __func__, (unsigned long long)sh->sector, target); | |
1519 | ||
1520 | tgt = &sh->dev[target]; | |
1521 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1522 | dest = tgt->page; | |
1523 | ||
1524 | atomic_inc(&sh->count); | |
1525 | ||
1526 | if (target == qd_idx) { | |
584acdd4 | 1527 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1528 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1529 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1530 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1531 | ops_complete_compute, sh, | |
46d5b785 | 1532 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1533 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1534 | } else { | |
1535 | /* Compute any data- or p-drive using XOR */ | |
1536 | count = 0; | |
1537 | for (i = disks; i-- ; ) { | |
1538 | if (i == target || i == qd_idx) | |
1539 | continue; | |
1540 | blocks[count++] = sh->dev[i].page; | |
1541 | } | |
1542 | ||
0403e382 DW |
1543 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1544 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1545 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1546 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1547 | } | |
91c00924 | 1548 | |
91c00924 DW |
1549 | return tx; |
1550 | } | |
1551 | ||
ac6b53b6 DW |
1552 | static struct dma_async_tx_descriptor * |
1553 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1554 | { | |
1555 | int i, count, disks = sh->disks; | |
1556 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1557 | int d0_idx = raid6_d0(sh); | |
1558 | int faila = -1, failb = -1; | |
1559 | int target = sh->ops.target; | |
1560 | int target2 = sh->ops.target2; | |
1561 | struct r5dev *tgt = &sh->dev[target]; | |
1562 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1563 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1564 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1565 | struct async_submit_ctl submit; |
1566 | ||
59fc630b | 1567 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1568 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1569 | __func__, (unsigned long long)sh->sector, target, target2); | |
1570 | BUG_ON(target < 0 || target2 < 0); | |
1571 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1572 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1573 | ||
6c910a78 | 1574 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1575 | * slot number conversion for 'faila' and 'failb' |
1576 | */ | |
1577 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1578 | blocks[i] = NULL; |
ac6b53b6 DW |
1579 | count = 0; |
1580 | i = d0_idx; | |
1581 | do { | |
1582 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1583 | ||
1584 | blocks[slot] = sh->dev[i].page; | |
1585 | ||
1586 | if (i == target) | |
1587 | faila = slot; | |
1588 | if (i == target2) | |
1589 | failb = slot; | |
1590 | i = raid6_next_disk(i, disks); | |
1591 | } while (i != d0_idx); | |
ac6b53b6 DW |
1592 | |
1593 | BUG_ON(faila == failb); | |
1594 | if (failb < faila) | |
1595 | swap(faila, failb); | |
1596 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1597 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1598 | ||
1599 | atomic_inc(&sh->count); | |
1600 | ||
1601 | if (failb == syndrome_disks+1) { | |
1602 | /* Q disk is one of the missing disks */ | |
1603 | if (faila == syndrome_disks) { | |
1604 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1605 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1606 | ops_complete_compute, sh, | |
46d5b785 | 1607 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1608 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1609 | STRIPE_SIZE, &submit); |
1610 | } else { | |
1611 | struct page *dest; | |
1612 | int data_target; | |
1613 | int qd_idx = sh->qd_idx; | |
1614 | ||
1615 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1616 | if (target == qd_idx) | |
1617 | data_target = target2; | |
1618 | else | |
1619 | data_target = target; | |
1620 | ||
1621 | count = 0; | |
1622 | for (i = disks; i-- ; ) { | |
1623 | if (i == data_target || i == qd_idx) | |
1624 | continue; | |
1625 | blocks[count++] = sh->dev[i].page; | |
1626 | } | |
1627 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1628 | init_async_submit(&submit, |
1629 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1630 | NULL, NULL, NULL, | |
46d5b785 | 1631 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1632 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1633 | &submit); | |
1634 | ||
584acdd4 | 1635 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1636 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1637 | ops_complete_compute, sh, | |
46d5b785 | 1638 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1639 | return async_gen_syndrome(blocks, 0, count+2, |
1640 | STRIPE_SIZE, &submit); | |
1641 | } | |
ac6b53b6 | 1642 | } else { |
6c910a78 DW |
1643 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1644 | ops_complete_compute, sh, | |
46d5b785 | 1645 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1646 | if (failb == syndrome_disks) { |
1647 | /* We're missing D+P. */ | |
1648 | return async_raid6_datap_recov(syndrome_disks+2, | |
1649 | STRIPE_SIZE, faila, | |
1650 | blocks, &submit); | |
1651 | } else { | |
1652 | /* We're missing D+D. */ | |
1653 | return async_raid6_2data_recov(syndrome_disks+2, | |
1654 | STRIPE_SIZE, faila, failb, | |
1655 | blocks, &submit); | |
1656 | } | |
ac6b53b6 DW |
1657 | } |
1658 | } | |
1659 | ||
91c00924 DW |
1660 | static void ops_complete_prexor(void *stripe_head_ref) |
1661 | { | |
1662 | struct stripe_head *sh = stripe_head_ref; | |
1663 | ||
e46b272b | 1664 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1665 | (unsigned long long)sh->sector); |
1e6d690b SL |
1666 | |
1667 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1668 | /* | |
1669 | * raid5-cache write back uses orig_page during prexor. | |
1670 | * After prexor, it is time to free orig_page | |
1671 | */ | |
1672 | r5c_release_extra_page(sh); | |
91c00924 DW |
1673 | } |
1674 | ||
1675 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1676 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1677 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1678 | { |
91c00924 | 1679 | int disks = sh->disks; |
46d5b785 | 1680 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1681 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1682 | struct async_submit_ctl submit; |
91c00924 DW |
1683 | |
1684 | /* existing parity data subtracted */ | |
1685 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1686 | ||
59fc630b | 1687 | BUG_ON(sh->batch_head); |
e46b272b | 1688 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1689 | (unsigned long long)sh->sector); |
1690 | ||
1691 | for (i = disks; i--; ) { | |
1692 | struct r5dev *dev = &sh->dev[i]; | |
1693 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1694 | if (test_bit(R5_InJournal, &dev->flags)) |
1695 | xor_srcs[count++] = dev->orig_page; | |
1696 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1697 | xor_srcs[count++] = dev->page; |
1698 | } | |
1699 | ||
0403e382 | 1700 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1701 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1702 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1703 | |
1704 | return tx; | |
1705 | } | |
1706 | ||
584acdd4 MS |
1707 | static struct dma_async_tx_descriptor * |
1708 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1709 | struct dma_async_tx_descriptor *tx) | |
1710 | { | |
1711 | struct page **blocks = to_addr_page(percpu, 0); | |
1712 | int count; | |
1713 | struct async_submit_ctl submit; | |
1714 | ||
1715 | pr_debug("%s: stripe %llu\n", __func__, | |
1716 | (unsigned long long)sh->sector); | |
1717 | ||
1718 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1719 | ||
1720 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1721 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1722 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1723 | ||
1724 | return tx; | |
1725 | } | |
1726 | ||
91c00924 | 1727 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1728 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1729 | { |
1e6d690b | 1730 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1731 | int disks = sh->disks; |
d8ee0728 | 1732 | int i; |
59fc630b | 1733 | struct stripe_head *head_sh = sh; |
91c00924 | 1734 | |
e46b272b | 1735 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1736 | (unsigned long long)sh->sector); |
1737 | ||
1738 | for (i = disks; i--; ) { | |
59fc630b | 1739 | struct r5dev *dev; |
91c00924 | 1740 | struct bio *chosen; |
91c00924 | 1741 | |
59fc630b | 1742 | sh = head_sh; |
1743 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1744 | struct bio *wbi; |
1745 | ||
59fc630b | 1746 | again: |
1747 | dev = &sh->dev[i]; | |
1e6d690b SL |
1748 | /* |
1749 | * clear R5_InJournal, so when rewriting a page in | |
1750 | * journal, it is not skipped by r5l_log_stripe() | |
1751 | */ | |
1752 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1753 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1754 | chosen = dev->towrite; |
1755 | dev->towrite = NULL; | |
7a87f434 | 1756 | sh->overwrite_disks = 0; |
91c00924 DW |
1757 | BUG_ON(dev->written); |
1758 | wbi = dev->written = chosen; | |
b17459c0 | 1759 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1760 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1761 | |
4f024f37 | 1762 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1763 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1764 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1765 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1766 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1767 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1768 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1769 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1770 | else { |
1771 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1772 | dev->sector, tx, sh, |
1773 | r5c_is_writeback(conf->log)); | |
1774 | if (dev->page != dev->orig_page && | |
1775 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1776 | set_bit(R5_SkipCopy, &dev->flags); |
1777 | clear_bit(R5_UPTODATE, &dev->flags); | |
1778 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1779 | } | |
1780 | } | |
91c00924 DW |
1781 | wbi = r5_next_bio(wbi, dev->sector); |
1782 | } | |
59fc630b | 1783 | |
1784 | if (head_sh->batch_head) { | |
1785 | sh = list_first_entry(&sh->batch_list, | |
1786 | struct stripe_head, | |
1787 | batch_list); | |
1788 | if (sh == head_sh) | |
1789 | continue; | |
1790 | goto again; | |
1791 | } | |
91c00924 DW |
1792 | } |
1793 | } | |
1794 | ||
1795 | return tx; | |
1796 | } | |
1797 | ||
ac6b53b6 | 1798 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1799 | { |
1800 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1801 | int disks = sh->disks; |
1802 | int pd_idx = sh->pd_idx; | |
1803 | int qd_idx = sh->qd_idx; | |
1804 | int i; | |
9e444768 | 1805 | bool fua = false, sync = false, discard = false; |
91c00924 | 1806 | |
e46b272b | 1807 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1808 | (unsigned long long)sh->sector); |
1809 | ||
bc0934f0 | 1810 | for (i = disks; i--; ) { |
e9c7469b | 1811 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1812 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1813 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1814 | } |
e9c7469b | 1815 | |
91c00924 DW |
1816 | for (i = disks; i--; ) { |
1817 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1818 | |
e9c7469b | 1819 | if (dev->written || i == pd_idx || i == qd_idx) { |
235b6003 | 1820 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) { |
9e444768 | 1821 | set_bit(R5_UPTODATE, &dev->flags); |
235b6003 N |
1822 | if (test_bit(STRIPE_EXPAND_READY, &sh->state)) |
1823 | set_bit(R5_Expanded, &dev->flags); | |
1824 | } | |
e9c7469b TH |
1825 | if (fua) |
1826 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1827 | if (sync) |
1828 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1829 | } |
91c00924 DW |
1830 | } |
1831 | ||
d8ee0728 DW |
1832 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1833 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1834 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1835 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1836 | else { | |
1837 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1838 | sh->reconstruct_state = reconstruct_state_result; | |
1839 | } | |
91c00924 DW |
1840 | |
1841 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1842 | raid5_release_stripe(sh); |
91c00924 DW |
1843 | } |
1844 | ||
1845 | static void | |
ac6b53b6 DW |
1846 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1847 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1848 | { |
91c00924 | 1849 | int disks = sh->disks; |
59fc630b | 1850 | struct page **xor_srcs; |
a08abd8c | 1851 | struct async_submit_ctl submit; |
59fc630b | 1852 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1853 | struct page *xor_dest; |
d8ee0728 | 1854 | int prexor = 0; |
91c00924 | 1855 | unsigned long flags; |
59fc630b | 1856 | int j = 0; |
1857 | struct stripe_head *head_sh = sh; | |
1858 | int last_stripe; | |
91c00924 | 1859 | |
e46b272b | 1860 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1861 | (unsigned long long)sh->sector); |
1862 | ||
620125f2 SL |
1863 | for (i = 0; i < sh->disks; i++) { |
1864 | if (pd_idx == i) | |
1865 | continue; | |
1866 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1867 | break; | |
1868 | } | |
1869 | if (i >= sh->disks) { | |
1870 | atomic_inc(&sh->count); | |
620125f2 SL |
1871 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1872 | ops_complete_reconstruct(sh); | |
1873 | return; | |
1874 | } | |
59fc630b | 1875 | again: |
1876 | count = 0; | |
1877 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1878 | /* check if prexor is active which means only process blocks |
1879 | * that are part of a read-modify-write (written) | |
1880 | */ | |
59fc630b | 1881 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1882 | prexor = 1; |
91c00924 DW |
1883 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1884 | for (i = disks; i--; ) { | |
1885 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1886 | if (head_sh->dev[i].written || |
1887 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1888 | xor_srcs[count++] = dev->page; |
1889 | } | |
1890 | } else { | |
1891 | xor_dest = sh->dev[pd_idx].page; | |
1892 | for (i = disks; i--; ) { | |
1893 | struct r5dev *dev = &sh->dev[i]; | |
1894 | if (i != pd_idx) | |
1895 | xor_srcs[count++] = dev->page; | |
1896 | } | |
1897 | } | |
1898 | ||
91c00924 DW |
1899 | /* 1/ if we prexor'd then the dest is reused as a source |
1900 | * 2/ if we did not prexor then we are redoing the parity | |
1901 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1902 | * for the synchronous xor case | |
1903 | */ | |
59fc630b | 1904 | last_stripe = !head_sh->batch_head || |
1905 | list_first_entry(&sh->batch_list, | |
1906 | struct stripe_head, batch_list) == head_sh; | |
1907 | if (last_stripe) { | |
1908 | flags = ASYNC_TX_ACK | | |
1909 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1910 | ||
1911 | atomic_inc(&head_sh->count); | |
1912 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1913 | to_addr_conv(sh, percpu, j)); | |
1914 | } else { | |
1915 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1916 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1917 | to_addr_conv(sh, percpu, j)); | |
1918 | } | |
91c00924 | 1919 | |
a08abd8c DW |
1920 | if (unlikely(count == 1)) |
1921 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1922 | else | |
1923 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1924 | if (!last_stripe) { |
1925 | j++; | |
1926 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1927 | batch_list); | |
1928 | goto again; | |
1929 | } | |
91c00924 DW |
1930 | } |
1931 | ||
ac6b53b6 DW |
1932 | static void |
1933 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1934 | struct dma_async_tx_descriptor *tx) | |
1935 | { | |
1936 | struct async_submit_ctl submit; | |
59fc630b | 1937 | struct page **blocks; |
1938 | int count, i, j = 0; | |
1939 | struct stripe_head *head_sh = sh; | |
1940 | int last_stripe; | |
584acdd4 MS |
1941 | int synflags; |
1942 | unsigned long txflags; | |
ac6b53b6 DW |
1943 | |
1944 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1945 | ||
620125f2 SL |
1946 | for (i = 0; i < sh->disks; i++) { |
1947 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1948 | continue; | |
1949 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1950 | break; | |
1951 | } | |
1952 | if (i >= sh->disks) { | |
1953 | atomic_inc(&sh->count); | |
620125f2 SL |
1954 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1955 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1956 | ops_complete_reconstruct(sh); | |
1957 | return; | |
1958 | } | |
1959 | ||
59fc630b | 1960 | again: |
1961 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1962 | |
1963 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1964 | synflags = SYNDROME_SRC_WRITTEN; | |
1965 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1966 | } else { | |
1967 | synflags = SYNDROME_SRC_ALL; | |
1968 | txflags = ASYNC_TX_ACK; | |
1969 | } | |
1970 | ||
1971 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1972 | last_stripe = !head_sh->batch_head || |
1973 | list_first_entry(&sh->batch_list, | |
1974 | struct stripe_head, batch_list) == head_sh; | |
1975 | ||
1976 | if (last_stripe) { | |
1977 | atomic_inc(&head_sh->count); | |
584acdd4 | 1978 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1979 | head_sh, to_addr_conv(sh, percpu, j)); |
1980 | } else | |
1981 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1982 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1983 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1984 | if (!last_stripe) { |
1985 | j++; | |
1986 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1987 | batch_list); | |
1988 | goto again; | |
1989 | } | |
91c00924 DW |
1990 | } |
1991 | ||
1992 | static void ops_complete_check(void *stripe_head_ref) | |
1993 | { | |
1994 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1995 | |
e46b272b | 1996 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1997 | (unsigned long long)sh->sector); |
1998 | ||
ecc65c9b | 1999 | sh->check_state = check_state_check_result; |
91c00924 | 2000 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2001 | raid5_release_stripe(sh); |
91c00924 DW |
2002 | } |
2003 | ||
ac6b53b6 | 2004 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 2005 | { |
91c00924 | 2006 | int disks = sh->disks; |
ac6b53b6 DW |
2007 | int pd_idx = sh->pd_idx; |
2008 | int qd_idx = sh->qd_idx; | |
2009 | struct page *xor_dest; | |
46d5b785 | 2010 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 2011 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 2012 | struct async_submit_ctl submit; |
ac6b53b6 DW |
2013 | int count; |
2014 | int i; | |
91c00924 | 2015 | |
e46b272b | 2016 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2017 | (unsigned long long)sh->sector); |
2018 | ||
59fc630b | 2019 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
2020 | count = 0; |
2021 | xor_dest = sh->dev[pd_idx].page; | |
2022 | xor_srcs[count++] = xor_dest; | |
91c00924 | 2023 | for (i = disks; i--; ) { |
ac6b53b6 DW |
2024 | if (i == pd_idx || i == qd_idx) |
2025 | continue; | |
2026 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
2027 | } |
2028 | ||
d6f38f31 | 2029 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 2030 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 2031 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 2032 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 2033 | |
91c00924 | 2034 | atomic_inc(&sh->count); |
a08abd8c DW |
2035 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
2036 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
2037 | } |
2038 | ||
ac6b53b6 DW |
2039 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
2040 | { | |
46d5b785 | 2041 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
2042 | struct async_submit_ctl submit; |
2043 | int count; | |
2044 | ||
2045 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
2046 | (unsigned long long)sh->sector, checkp); | |
2047 | ||
59fc630b | 2048 | BUG_ON(sh->batch_head); |
584acdd4 | 2049 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
2050 | if (!checkp) |
2051 | srcs[count] = NULL; | |
91c00924 | 2052 | |
91c00924 | 2053 | atomic_inc(&sh->count); |
ac6b53b6 | 2054 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 2055 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
2056 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
2057 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
2058 | } |
2059 | ||
51acbcec | 2060 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
2061 | { |
2062 | int overlap_clear = 0, i, disks = sh->disks; | |
2063 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 2064 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2065 | int level = conf->level; |
d6f38f31 DW |
2066 | struct raid5_percpu *percpu; |
2067 | unsigned long cpu; | |
91c00924 | 2068 | |
d6f38f31 DW |
2069 | cpu = get_cpu(); |
2070 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2071 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2072 | ops_run_biofill(sh); |
2073 | overlap_clear++; | |
2074 | } | |
2075 | ||
7b3a871e | 2076 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2077 | if (level < 6) |
2078 | tx = ops_run_compute5(sh, percpu); | |
2079 | else { | |
2080 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2081 | tx = ops_run_compute6_1(sh, percpu); | |
2082 | else | |
2083 | tx = ops_run_compute6_2(sh, percpu); | |
2084 | } | |
2085 | /* terminate the chain if reconstruct is not set to be run */ | |
2086 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2087 | async_tx_ack(tx); |
2088 | } | |
91c00924 | 2089 | |
584acdd4 MS |
2090 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2091 | if (level < 6) | |
2092 | tx = ops_run_prexor5(sh, percpu, tx); | |
2093 | else | |
2094 | tx = ops_run_prexor6(sh, percpu, tx); | |
2095 | } | |
91c00924 | 2096 | |
ae1713e2 AP |
2097 | if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request)) |
2098 | tx = ops_run_partial_parity(sh, percpu, tx); | |
2099 | ||
600aa109 | 2100 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2101 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2102 | overlap_clear++; |
2103 | } | |
2104 | ||
ac6b53b6 DW |
2105 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2106 | if (level < 6) | |
2107 | ops_run_reconstruct5(sh, percpu, tx); | |
2108 | else | |
2109 | ops_run_reconstruct6(sh, percpu, tx); | |
2110 | } | |
91c00924 | 2111 | |
ac6b53b6 DW |
2112 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2113 | if (sh->check_state == check_state_run) | |
2114 | ops_run_check_p(sh, percpu); | |
2115 | else if (sh->check_state == check_state_run_q) | |
2116 | ops_run_check_pq(sh, percpu, 0); | |
2117 | else if (sh->check_state == check_state_run_pq) | |
2118 | ops_run_check_pq(sh, percpu, 1); | |
2119 | else | |
2120 | BUG(); | |
2121 | } | |
91c00924 | 2122 | |
59fc630b | 2123 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2124 | for (i = disks; i--; ) { |
2125 | struct r5dev *dev = &sh->dev[i]; | |
2126 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2127 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2128 | } | |
d6f38f31 | 2129 | put_cpu(); |
91c00924 DW |
2130 | } |
2131 | ||
845b9e22 AP |
2132 | static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh) |
2133 | { | |
2134 | if (sh->ppl_page) | |
2135 | __free_page(sh->ppl_page); | |
2136 | kmem_cache_free(sc, sh); | |
2137 | } | |
2138 | ||
5f9d1fde | 2139 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
845b9e22 | 2140 | int disks, struct r5conf *conf) |
f18c1a35 N |
2141 | { |
2142 | struct stripe_head *sh; | |
5f9d1fde | 2143 | int i; |
f18c1a35 N |
2144 | |
2145 | sh = kmem_cache_zalloc(sc, gfp); | |
2146 | if (sh) { | |
2147 | spin_lock_init(&sh->stripe_lock); | |
2148 | spin_lock_init(&sh->batch_lock); | |
2149 | INIT_LIST_HEAD(&sh->batch_list); | |
2150 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2151 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2152 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2153 | atomic_set(&sh->count, 1); |
845b9e22 | 2154 | sh->raid_conf = conf; |
a39f7afd | 2155 | sh->log_start = MaxSector; |
5f9d1fde SL |
2156 | for (i = 0; i < disks; i++) { |
2157 | struct r5dev *dev = &sh->dev[i]; | |
2158 | ||
3a83f467 ML |
2159 | bio_init(&dev->req, &dev->vec, 1); |
2160 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2161 | } |
845b9e22 AP |
2162 | |
2163 | if (raid5_has_ppl(conf)) { | |
2164 | sh->ppl_page = alloc_page(gfp); | |
2165 | if (!sh->ppl_page) { | |
2166 | free_stripe(sc, sh); | |
2167 | sh = NULL; | |
2168 | } | |
2169 | } | |
f18c1a35 N |
2170 | } |
2171 | return sh; | |
2172 | } | |
486f0644 | 2173 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2174 | { |
2175 | struct stripe_head *sh; | |
f18c1a35 | 2176 | |
845b9e22 | 2177 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf); |
3f294f4f N |
2178 | if (!sh) |
2179 | return 0; | |
6ce32846 | 2180 | |
a9683a79 | 2181 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2182 | shrink_buffers(sh); |
845b9e22 | 2183 | free_stripe(conf->slab_cache, sh); |
3f294f4f N |
2184 | return 0; |
2185 | } | |
486f0644 N |
2186 | sh->hash_lock_index = |
2187 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2188 | /* we just created an active stripe so... */ |
3f294f4f | 2189 | atomic_inc(&conf->active_stripes); |
59fc630b | 2190 | |
6d036f7d | 2191 | raid5_release_stripe(sh); |
486f0644 | 2192 | conf->max_nr_stripes++; |
3f294f4f N |
2193 | return 1; |
2194 | } | |
2195 | ||
d1688a6d | 2196 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2197 | { |
e18b890b | 2198 | struct kmem_cache *sc; |
5e5e3e78 | 2199 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2200 | |
f4be6b43 N |
2201 | if (conf->mddev->gendisk) |
2202 | sprintf(conf->cache_name[0], | |
2203 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2204 | else | |
2205 | sprintf(conf->cache_name[0], | |
2206 | "raid%d-%p", conf->level, conf->mddev); | |
2207 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2208 | ||
ad01c9e3 N |
2209 | conf->active_name = 0; |
2210 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2211 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2212 | 0, 0, NULL); |
1da177e4 LT |
2213 | if (!sc) |
2214 | return 1; | |
2215 | conf->slab_cache = sc; | |
ad01c9e3 | 2216 | conf->pool_size = devs; |
486f0644 N |
2217 | while (num--) |
2218 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2219 | return 1; |
486f0644 | 2220 | |
1da177e4 LT |
2221 | return 0; |
2222 | } | |
29269553 | 2223 | |
d6f38f31 DW |
2224 | /** |
2225 | * scribble_len - return the required size of the scribble region | |
2226 | * @num - total number of disks in the array | |
2227 | * | |
2228 | * The size must be enough to contain: | |
2229 | * 1/ a struct page pointer for each device in the array +2 | |
2230 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2231 | * (dma_map_page()) or page (page_address()) address. | |
2232 | * | |
2233 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2234 | * calculate over all devices (not just the data blocks), using zeros in place | |
2235 | * of the P and Q blocks. | |
2236 | */ | |
46d5b785 | 2237 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2238 | { |
46d5b785 | 2239 | struct flex_array *ret; |
d6f38f31 DW |
2240 | size_t len; |
2241 | ||
2242 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2243 | ret = flex_array_alloc(len, cnt, flags); |
2244 | if (!ret) | |
2245 | return NULL; | |
2246 | /* always prealloc all elements, so no locking is required */ | |
2247 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2248 | flex_array_free(ret); | |
2249 | return NULL; | |
2250 | } | |
2251 | return ret; | |
d6f38f31 DW |
2252 | } |
2253 | ||
738a2738 N |
2254 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2255 | { | |
2256 | unsigned long cpu; | |
2257 | int err = 0; | |
2258 | ||
27a353c0 SL |
2259 | /* |
2260 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2261 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2262 | * should equal to new_disks and new_sectors | |
2263 | */ | |
2264 | if (conf->scribble_disks >= new_disks && | |
2265 | conf->scribble_sectors >= new_sectors) | |
2266 | return 0; | |
738a2738 N |
2267 | mddev_suspend(conf->mddev); |
2268 | get_online_cpus(); | |
2269 | for_each_present_cpu(cpu) { | |
2270 | struct raid5_percpu *percpu; | |
2271 | struct flex_array *scribble; | |
2272 | ||
2273 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2274 | scribble = scribble_alloc(new_disks, | |
2275 | new_sectors / STRIPE_SECTORS, | |
2276 | GFP_NOIO); | |
2277 | ||
2278 | if (scribble) { | |
2279 | flex_array_free(percpu->scribble); | |
2280 | percpu->scribble = scribble; | |
2281 | } else { | |
2282 | err = -ENOMEM; | |
2283 | break; | |
2284 | } | |
2285 | } | |
2286 | put_online_cpus(); | |
2287 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2288 | if (!err) { |
2289 | conf->scribble_disks = new_disks; | |
2290 | conf->scribble_sectors = new_sectors; | |
2291 | } | |
738a2738 N |
2292 | return err; |
2293 | } | |
2294 | ||
d1688a6d | 2295 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2296 | { |
2297 | /* Make all the stripes able to hold 'newsize' devices. | |
2298 | * New slots in each stripe get 'page' set to a new page. | |
2299 | * | |
2300 | * This happens in stages: | |
2301 | * 1/ create a new kmem_cache and allocate the required number of | |
2302 | * stripe_heads. | |
83f0d77a | 2303 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2304 | * to the new stripe_heads. This will have the side effect of |
2305 | * freezing the array as once all stripe_heads have been collected, | |
2306 | * no IO will be possible. Old stripe heads are freed once their | |
2307 | * pages have been transferred over, and the old kmem_cache is | |
2308 | * freed when all stripes are done. | |
2309 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
3560741e | 2310 | * we simple return a failure status - no need to clean anything up. |
ad01c9e3 N |
2311 | * 4/ allocate new pages for the new slots in the new stripe_heads. |
2312 | * If this fails, we don't bother trying the shrink the | |
2313 | * stripe_heads down again, we just leave them as they are. | |
2314 | * As each stripe_head is processed the new one is released into | |
2315 | * active service. | |
2316 | * | |
2317 | * Once step2 is started, we cannot afford to wait for a write, | |
2318 | * so we use GFP_NOIO allocations. | |
2319 | */ | |
2320 | struct stripe_head *osh, *nsh; | |
2321 | LIST_HEAD(newstripes); | |
2322 | struct disk_info *ndisks; | |
2214c260 | 2323 | int err = 0; |
e18b890b | 2324 | struct kmem_cache *sc; |
ad01c9e3 | 2325 | int i; |
566c09c5 | 2326 | int hash, cnt; |
ad01c9e3 | 2327 | |
2214c260 | 2328 | md_allow_write(conf->mddev); |
2a2275d6 | 2329 | |
ad01c9e3 N |
2330 | /* Step 1 */ |
2331 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2332 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2333 | 0, 0, NULL); |
ad01c9e3 N |
2334 | if (!sc) |
2335 | return -ENOMEM; | |
2336 | ||
2d5b569b N |
2337 | /* Need to ensure auto-resizing doesn't interfere */ |
2338 | mutex_lock(&conf->cache_size_mutex); | |
2339 | ||
ad01c9e3 | 2340 | for (i = conf->max_nr_stripes; i; i--) { |
845b9e22 | 2341 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf); |
ad01c9e3 N |
2342 | if (!nsh) |
2343 | break; | |
2344 | ||
ad01c9e3 N |
2345 | list_add(&nsh->lru, &newstripes); |
2346 | } | |
2347 | if (i) { | |
2348 | /* didn't get enough, give up */ | |
2349 | while (!list_empty(&newstripes)) { | |
2350 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2351 | list_del(&nsh->lru); | |
845b9e22 | 2352 | free_stripe(sc, nsh); |
ad01c9e3 N |
2353 | } |
2354 | kmem_cache_destroy(sc); | |
2d5b569b | 2355 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2356 | return -ENOMEM; |
2357 | } | |
2358 | /* Step 2 - Must use GFP_NOIO now. | |
2359 | * OK, we have enough stripes, start collecting inactive | |
2360 | * stripes and copying them over | |
2361 | */ | |
566c09c5 SL |
2362 | hash = 0; |
2363 | cnt = 0; | |
ad01c9e3 | 2364 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2365 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2366 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2367 | !list_empty(conf->inactive_list + hash), |
2368 | unlock_device_hash_lock(conf, hash), | |
2369 | lock_device_hash_lock(conf, hash)); | |
2370 | osh = get_free_stripe(conf, hash); | |
2371 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2372 | |
d592a996 | 2373 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2374 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2375 | nsh->dev[i].orig_page = osh->dev[i].page; |
2376 | } | |
566c09c5 | 2377 | nsh->hash_lock_index = hash; |
845b9e22 | 2378 | free_stripe(conf->slab_cache, osh); |
566c09c5 SL |
2379 | cnt++; |
2380 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2381 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2382 | hash++; | |
2383 | cnt = 0; | |
2384 | } | |
ad01c9e3 N |
2385 | } |
2386 | kmem_cache_destroy(conf->slab_cache); | |
2387 | ||
2388 | /* Step 3. | |
2389 | * At this point, we are holding all the stripes so the array | |
2390 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2391 | * conf->disks and the scribble region |
ad01c9e3 N |
2392 | */ |
2393 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2394 | if (ndisks) { | |
d7bd398e | 2395 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2396 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2397 | |
2398 | for (i = conf->pool_size; i < newsize; i++) { | |
2399 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2400 | if (!ndisks[i].extra_page) | |
2401 | err = -ENOMEM; | |
2402 | } | |
2403 | ||
2404 | if (err) { | |
2405 | for (i = conf->pool_size; i < newsize; i++) | |
2406 | if (ndisks[i].extra_page) | |
2407 | put_page(ndisks[i].extra_page); | |
2408 | kfree(ndisks); | |
2409 | } else { | |
2410 | kfree(conf->disks); | |
2411 | conf->disks = ndisks; | |
2412 | } | |
ad01c9e3 N |
2413 | } else |
2414 | err = -ENOMEM; | |
2415 | ||
2d5b569b | 2416 | mutex_unlock(&conf->cache_size_mutex); |
583da48e DY |
2417 | |
2418 | conf->slab_cache = sc; | |
2419 | conf->active_name = 1-conf->active_name; | |
2420 | ||
ad01c9e3 N |
2421 | /* Step 4, return new stripes to service */ |
2422 | while(!list_empty(&newstripes)) { | |
2423 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2424 | list_del_init(&nsh->lru); | |
d6f38f31 | 2425 | |
ad01c9e3 N |
2426 | for (i=conf->raid_disks; i < newsize; i++) |
2427 | if (nsh->dev[i].page == NULL) { | |
2428 | struct page *p = alloc_page(GFP_NOIO); | |
2429 | nsh->dev[i].page = p; | |
d592a996 | 2430 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2431 | if (!p) |
2432 | err = -ENOMEM; | |
2433 | } | |
6d036f7d | 2434 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2435 | } |
2436 | /* critical section pass, GFP_NOIO no longer needed */ | |
2437 | ||
6e9eac2d N |
2438 | if (!err) |
2439 | conf->pool_size = newsize; | |
ad01c9e3 N |
2440 | return err; |
2441 | } | |
1da177e4 | 2442 | |
486f0644 | 2443 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2444 | { |
2445 | struct stripe_head *sh; | |
49895bcc | 2446 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2447 | |
566c09c5 SL |
2448 | spin_lock_irq(conf->hash_locks + hash); |
2449 | sh = get_free_stripe(conf, hash); | |
2450 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2451 | if (!sh) |
2452 | return 0; | |
78bafebd | 2453 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2454 | shrink_buffers(sh); |
845b9e22 | 2455 | free_stripe(conf->slab_cache, sh); |
3f294f4f | 2456 | atomic_dec(&conf->active_stripes); |
486f0644 | 2457 | conf->max_nr_stripes--; |
3f294f4f N |
2458 | return 1; |
2459 | } | |
2460 | ||
d1688a6d | 2461 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2462 | { |
486f0644 N |
2463 | while (conf->max_nr_stripes && |
2464 | drop_one_stripe(conf)) | |
2465 | ; | |
3f294f4f | 2466 | |
644df1a8 | 2467 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2468 | conf->slab_cache = NULL; |
2469 | } | |
2470 | ||
4246a0b6 | 2471 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2472 | { |
99c0fb5f | 2473 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2474 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2475 | int disks = sh->disks, i; |
d6950432 | 2476 | char b[BDEVNAME_SIZE]; |
dd054fce | 2477 | struct md_rdev *rdev = NULL; |
05616be5 | 2478 | sector_t s; |
1da177e4 LT |
2479 | |
2480 | for (i=0 ; i<disks; i++) | |
2481 | if (bi == &sh->dev[i].req) | |
2482 | break; | |
2483 | ||
4246a0b6 | 2484 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2485 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2486 | bi->bi_status); |
1da177e4 | 2487 | if (i == disks) { |
5f9d1fde | 2488 | bio_reset(bi); |
1da177e4 | 2489 | BUG(); |
6712ecf8 | 2490 | return; |
1da177e4 | 2491 | } |
14a75d3e | 2492 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2493 | /* If replacement finished while this request was outstanding, |
2494 | * 'replacement' might be NULL already. | |
2495 | * In that case it moved down to 'rdev'. | |
2496 | * rdev is not removed until all requests are finished. | |
2497 | */ | |
14a75d3e | 2498 | rdev = conf->disks[i].replacement; |
dd054fce | 2499 | if (!rdev) |
14a75d3e | 2500 | rdev = conf->disks[i].rdev; |
1da177e4 | 2501 | |
05616be5 N |
2502 | if (use_new_offset(conf, sh)) |
2503 | s = sh->sector + rdev->new_data_offset; | |
2504 | else | |
2505 | s = sh->sector + rdev->data_offset; | |
4e4cbee9 | 2506 | if (!bi->bi_status) { |
1da177e4 | 2507 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2508 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2509 | /* Note that this cannot happen on a |
2510 | * replacement device. We just fail those on | |
2511 | * any error | |
2512 | */ | |
cc6167b4 N |
2513 | pr_info_ratelimited( |
2514 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
8bda470e | 2515 | mdname(conf->mddev), STRIPE_SECTORS, |
05616be5 | 2516 | (unsigned long long)s, |
8bda470e | 2517 | bdevname(rdev->bdev, b)); |
ddd5115f | 2518 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2519 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2520 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2521 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2522 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2523 | ||
86aa1397 SL |
2524 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2525 | /* | |
2526 | * end read for a page in journal, this | |
2527 | * must be preparing for prexor in rmw | |
2528 | */ | |
2529 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2530 | ||
14a75d3e N |
2531 | if (atomic_read(&rdev->read_errors)) |
2532 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2533 | } else { |
14a75d3e | 2534 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2535 | int retry = 0; |
2e8ac303 | 2536 | int set_bad = 0; |
d6950432 | 2537 | |
1da177e4 | 2538 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2539 | atomic_inc(&rdev->read_errors); |
14a75d3e | 2540 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2541 | pr_warn_ratelimited( |
2542 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2543 | mdname(conf->mddev), |
05616be5 | 2544 | (unsigned long long)s, |
14a75d3e | 2545 | bdn); |
2e8ac303 | 2546 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2547 | set_bad = 1; | |
cc6167b4 N |
2548 | pr_warn_ratelimited( |
2549 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2550 | mdname(conf->mddev), |
05616be5 | 2551 | (unsigned long long)s, |
8bda470e | 2552 | bdn); |
2e8ac303 | 2553 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2554 | /* Oh, no!!! */ |
2e8ac303 | 2555 | set_bad = 1; |
cc6167b4 N |
2556 | pr_warn_ratelimited( |
2557 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2558 | mdname(conf->mddev), |
05616be5 | 2559 | (unsigned long long)s, |
8bda470e | 2560 | bdn); |
2e8ac303 | 2561 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2562 | > conf->max_nr_stripes) |
cc6167b4 | 2563 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2564 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2565 | else |
2566 | retry = 1; | |
edfa1f65 BY |
2567 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2568 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2569 | retry = 1; | |
ba22dcbf | 2570 | if (retry) |
3f9e7c14 | 2571 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2572 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2573 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2574 | } else | |
2575 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2576 | else { |
4e5314b5 N |
2577 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2578 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2579 | if (!(set_bad |
2580 | && test_bit(In_sync, &rdev->flags) | |
2581 | && rdev_set_badblocks( | |
2582 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2583 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2584 | } |
1da177e4 | 2585 | } |
14a75d3e | 2586 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2587 | bio_reset(bi); |
1da177e4 LT |
2588 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2589 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2590 | raid5_release_stripe(sh); |
1da177e4 LT |
2591 | } |
2592 | ||
4246a0b6 | 2593 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2594 | { |
99c0fb5f | 2595 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2596 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2597 | int disks = sh->disks, i; |
977df362 | 2598 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2599 | sector_t first_bad; |
2600 | int bad_sectors; | |
977df362 | 2601 | int replacement = 0; |
1da177e4 | 2602 | |
977df362 N |
2603 | for (i = 0 ; i < disks; i++) { |
2604 | if (bi == &sh->dev[i].req) { | |
2605 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2606 | break; |
977df362 N |
2607 | } |
2608 | if (bi == &sh->dev[i].rreq) { | |
2609 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2610 | if (rdev) |
2611 | replacement = 1; | |
2612 | else | |
2613 | /* rdev was removed and 'replacement' | |
2614 | * replaced it. rdev is not removed | |
2615 | * until all requests are finished. | |
2616 | */ | |
2617 | rdev = conf->disks[i].rdev; | |
977df362 N |
2618 | break; |
2619 | } | |
2620 | } | |
4246a0b6 | 2621 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2622 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2623 | bi->bi_status); |
1da177e4 | 2624 | if (i == disks) { |
5f9d1fde | 2625 | bio_reset(bi); |
1da177e4 | 2626 | BUG(); |
6712ecf8 | 2627 | return; |
1da177e4 LT |
2628 | } |
2629 | ||
977df362 | 2630 | if (replacement) { |
4e4cbee9 | 2631 | if (bi->bi_status) |
977df362 N |
2632 | md_error(conf->mddev, rdev); |
2633 | else if (is_badblock(rdev, sh->sector, | |
2634 | STRIPE_SECTORS, | |
2635 | &first_bad, &bad_sectors)) | |
2636 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2637 | } else { | |
4e4cbee9 | 2638 | if (bi->bi_status) { |
9f97e4b1 | 2639 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2640 | set_bit(WriteErrorSeen, &rdev->flags); |
2641 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2642 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2643 | set_bit(MD_RECOVERY_NEEDED, | |
2644 | &rdev->mddev->recovery); | |
977df362 N |
2645 | } else if (is_badblock(rdev, sh->sector, |
2646 | STRIPE_SECTORS, | |
c0b32972 | 2647 | &first_bad, &bad_sectors)) { |
977df362 | 2648 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2649 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2650 | /* That was a successful write so make | |
2651 | * sure it looks like we already did | |
2652 | * a re-write. | |
2653 | */ | |
2654 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2655 | } | |
977df362 N |
2656 | } |
2657 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2658 | |
4e4cbee9 | 2659 | if (sh->batch_head && bi->bi_status && !replacement) |
72ac7330 | 2660 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2661 | ||
c9445555 | 2662 | bio_reset(bi); |
977df362 N |
2663 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2664 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2665 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2666 | raid5_release_stripe(sh); |
59fc630b | 2667 | |
2668 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2669 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2670 | } |
2671 | ||
849674e4 | 2672 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2673 | { |
2674 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2675 | struct r5conf *conf = mddev->private; |
908f4fbd | 2676 | unsigned long flags; |
0c55e022 | 2677 | pr_debug("raid456: error called\n"); |
1da177e4 | 2678 | |
908f4fbd N |
2679 | spin_lock_irqsave(&conf->device_lock, flags); |
2680 | clear_bit(In_sync, &rdev->flags); | |
2e38a37f | 2681 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2682 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2683 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2684 | ||
de393cde | 2685 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 | 2686 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
2687 | set_mask_bits(&mddev->sb_flags, 0, |
2688 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2689 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2690 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2691 | mdname(mddev), | |
2692 | bdevname(rdev->bdev, b), | |
2693 | mdname(mddev), | |
2694 | conf->raid_disks - mddev->degraded); | |
70d466f7 | 2695 | r5c_update_on_rdev_error(mddev, rdev); |
16a53ecc | 2696 | } |
1da177e4 LT |
2697 | |
2698 | /* | |
2699 | * Input: a 'big' sector number, | |
2700 | * Output: index of the data and parity disk, and the sector # in them. | |
2701 | */ | |
6d036f7d SL |
2702 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2703 | int previous, int *dd_idx, | |
2704 | struct stripe_head *sh) | |
1da177e4 | 2705 | { |
6e3b96ed | 2706 | sector_t stripe, stripe2; |
35f2a591 | 2707 | sector_t chunk_number; |
1da177e4 | 2708 | unsigned int chunk_offset; |
911d4ee8 | 2709 | int pd_idx, qd_idx; |
67cc2b81 | 2710 | int ddf_layout = 0; |
1da177e4 | 2711 | sector_t new_sector; |
e183eaed N |
2712 | int algorithm = previous ? conf->prev_algo |
2713 | : conf->algorithm; | |
09c9e5fa AN |
2714 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2715 | : conf->chunk_sectors; | |
112bf897 N |
2716 | int raid_disks = previous ? conf->previous_raid_disks |
2717 | : conf->raid_disks; | |
2718 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2719 | |
2720 | /* First compute the information on this sector */ | |
2721 | ||
2722 | /* | |
2723 | * Compute the chunk number and the sector offset inside the chunk | |
2724 | */ | |
2725 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2726 | chunk_number = r_sector; | |
1da177e4 LT |
2727 | |
2728 | /* | |
2729 | * Compute the stripe number | |
2730 | */ | |
35f2a591 N |
2731 | stripe = chunk_number; |
2732 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2733 | stripe2 = stripe; |
1da177e4 LT |
2734 | /* |
2735 | * Select the parity disk based on the user selected algorithm. | |
2736 | */ | |
84789554 | 2737 | pd_idx = qd_idx = -1; |
16a53ecc N |
2738 | switch(conf->level) { |
2739 | case 4: | |
911d4ee8 | 2740 | pd_idx = data_disks; |
16a53ecc N |
2741 | break; |
2742 | case 5: | |
e183eaed | 2743 | switch (algorithm) { |
1da177e4 | 2744 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2745 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2746 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2747 | (*dd_idx)++; |
2748 | break; | |
2749 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2750 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2751 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2752 | (*dd_idx)++; |
2753 | break; | |
2754 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2755 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2756 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2757 | break; |
2758 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2759 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2760 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2761 | break; |
99c0fb5f N |
2762 | case ALGORITHM_PARITY_0: |
2763 | pd_idx = 0; | |
2764 | (*dd_idx)++; | |
2765 | break; | |
2766 | case ALGORITHM_PARITY_N: | |
2767 | pd_idx = data_disks; | |
2768 | break; | |
1da177e4 | 2769 | default: |
99c0fb5f | 2770 | BUG(); |
16a53ecc N |
2771 | } |
2772 | break; | |
2773 | case 6: | |
2774 | ||
e183eaed | 2775 | switch (algorithm) { |
16a53ecc | 2776 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2777 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2778 | qd_idx = pd_idx + 1; |
2779 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2780 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2781 | qd_idx = 0; |
2782 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2783 | (*dd_idx) += 2; /* D D P Q D */ |
2784 | break; | |
2785 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2786 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2787 | qd_idx = pd_idx + 1; |
2788 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2789 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2790 | qd_idx = 0; |
2791 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2792 | (*dd_idx) += 2; /* D D P Q D */ |
2793 | break; | |
2794 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2795 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2796 | qd_idx = (pd_idx + 1) % raid_disks; |
2797 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2798 | break; |
2799 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2800 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2801 | qd_idx = (pd_idx + 1) % raid_disks; |
2802 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2803 | break; |
99c0fb5f N |
2804 | |
2805 | case ALGORITHM_PARITY_0: | |
2806 | pd_idx = 0; | |
2807 | qd_idx = 1; | |
2808 | (*dd_idx) += 2; | |
2809 | break; | |
2810 | case ALGORITHM_PARITY_N: | |
2811 | pd_idx = data_disks; | |
2812 | qd_idx = data_disks + 1; | |
2813 | break; | |
2814 | ||
2815 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2816 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2817 | * of blocks for computing Q is different. | |
2818 | */ | |
6e3b96ed | 2819 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2820 | qd_idx = pd_idx + 1; |
2821 | if (pd_idx == raid_disks-1) { | |
2822 | (*dd_idx)++; /* Q D D D P */ | |
2823 | qd_idx = 0; | |
2824 | } else if (*dd_idx >= pd_idx) | |
2825 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2826 | ddf_layout = 1; |
99c0fb5f N |
2827 | break; |
2828 | ||
2829 | case ALGORITHM_ROTATING_N_RESTART: | |
2830 | /* Same a left_asymmetric, by first stripe is | |
2831 | * D D D P Q rather than | |
2832 | * Q D D D P | |
2833 | */ | |
6e3b96ed N |
2834 | stripe2 += 1; |
2835 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2836 | qd_idx = pd_idx + 1; |
2837 | if (pd_idx == raid_disks-1) { | |
2838 | (*dd_idx)++; /* Q D D D P */ | |
2839 | qd_idx = 0; | |
2840 | } else if (*dd_idx >= pd_idx) | |
2841 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2842 | ddf_layout = 1; |
99c0fb5f N |
2843 | break; |
2844 | ||
2845 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2846 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2847 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2848 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2849 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2850 | ddf_layout = 1; |
99c0fb5f N |
2851 | break; |
2852 | ||
2853 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2854 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2855 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2856 | if (*dd_idx >= pd_idx) |
2857 | (*dd_idx)++; | |
2858 | qd_idx = raid_disks - 1; | |
2859 | break; | |
2860 | ||
2861 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2862 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2863 | if (*dd_idx >= pd_idx) |
2864 | (*dd_idx)++; | |
2865 | qd_idx = raid_disks - 1; | |
2866 | break; | |
2867 | ||
2868 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2869 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2870 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2871 | qd_idx = raid_disks - 1; | |
2872 | break; | |
2873 | ||
2874 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2875 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2876 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2877 | qd_idx = raid_disks - 1; | |
2878 | break; | |
2879 | ||
2880 | case ALGORITHM_PARITY_0_6: | |
2881 | pd_idx = 0; | |
2882 | (*dd_idx)++; | |
2883 | qd_idx = raid_disks - 1; | |
2884 | break; | |
2885 | ||
16a53ecc | 2886 | default: |
99c0fb5f | 2887 | BUG(); |
16a53ecc N |
2888 | } |
2889 | break; | |
1da177e4 LT |
2890 | } |
2891 | ||
911d4ee8 N |
2892 | if (sh) { |
2893 | sh->pd_idx = pd_idx; | |
2894 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2895 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2896 | } |
1da177e4 LT |
2897 | /* |
2898 | * Finally, compute the new sector number | |
2899 | */ | |
2900 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2901 | return new_sector; | |
2902 | } | |
2903 | ||
6d036f7d | 2904 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2905 | { |
d1688a6d | 2906 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2907 | int raid_disks = sh->disks; |
2908 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2909 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2910 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2911 | : conf->chunk_sectors; | |
e183eaed N |
2912 | int algorithm = previous ? conf->prev_algo |
2913 | : conf->algorithm; | |
1da177e4 LT |
2914 | sector_t stripe; |
2915 | int chunk_offset; | |
35f2a591 N |
2916 | sector_t chunk_number; |
2917 | int dummy1, dd_idx = i; | |
1da177e4 | 2918 | sector_t r_sector; |
911d4ee8 | 2919 | struct stripe_head sh2; |
1da177e4 LT |
2920 | |
2921 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2922 | stripe = new_sector; | |
1da177e4 | 2923 | |
16a53ecc N |
2924 | if (i == sh->pd_idx) |
2925 | return 0; | |
2926 | switch(conf->level) { | |
2927 | case 4: break; | |
2928 | case 5: | |
e183eaed | 2929 | switch (algorithm) { |
1da177e4 LT |
2930 | case ALGORITHM_LEFT_ASYMMETRIC: |
2931 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2932 | if (i > sh->pd_idx) | |
2933 | i--; | |
2934 | break; | |
2935 | case ALGORITHM_LEFT_SYMMETRIC: | |
2936 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2937 | if (i < sh->pd_idx) | |
2938 | i += raid_disks; | |
2939 | i -= (sh->pd_idx + 1); | |
2940 | break; | |
99c0fb5f N |
2941 | case ALGORITHM_PARITY_0: |
2942 | i -= 1; | |
2943 | break; | |
2944 | case ALGORITHM_PARITY_N: | |
2945 | break; | |
1da177e4 | 2946 | default: |
99c0fb5f | 2947 | BUG(); |
16a53ecc N |
2948 | } |
2949 | break; | |
2950 | case 6: | |
d0dabf7e | 2951 | if (i == sh->qd_idx) |
16a53ecc | 2952 | return 0; /* It is the Q disk */ |
e183eaed | 2953 | switch (algorithm) { |
16a53ecc N |
2954 | case ALGORITHM_LEFT_ASYMMETRIC: |
2955 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2956 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2957 | case ALGORITHM_ROTATING_N_RESTART: | |
2958 | if (sh->pd_idx == raid_disks-1) | |
2959 | i--; /* Q D D D P */ | |
16a53ecc N |
2960 | else if (i > sh->pd_idx) |
2961 | i -= 2; /* D D P Q D */ | |
2962 | break; | |
2963 | case ALGORITHM_LEFT_SYMMETRIC: | |
2964 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2965 | if (sh->pd_idx == raid_disks-1) | |
2966 | i--; /* Q D D D P */ | |
2967 | else { | |
2968 | /* D D P Q D */ | |
2969 | if (i < sh->pd_idx) | |
2970 | i += raid_disks; | |
2971 | i -= (sh->pd_idx + 2); | |
2972 | } | |
2973 | break; | |
99c0fb5f N |
2974 | case ALGORITHM_PARITY_0: |
2975 | i -= 2; | |
2976 | break; | |
2977 | case ALGORITHM_PARITY_N: | |
2978 | break; | |
2979 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2980 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2981 | if (sh->pd_idx == 0) |
2982 | i--; /* P D D D Q */ | |
e4424fee N |
2983 | else { |
2984 | /* D D Q P D */ | |
2985 | if (i < sh->pd_idx) | |
2986 | i += raid_disks; | |
2987 | i -= (sh->pd_idx + 1); | |
2988 | } | |
99c0fb5f N |
2989 | break; |
2990 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2991 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2992 | if (i > sh->pd_idx) | |
2993 | i--; | |
2994 | break; | |
2995 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2996 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2997 | if (i < sh->pd_idx) | |
2998 | i += data_disks + 1; | |
2999 | i -= (sh->pd_idx + 1); | |
3000 | break; | |
3001 | case ALGORITHM_PARITY_0_6: | |
3002 | i -= 1; | |
3003 | break; | |
16a53ecc | 3004 | default: |
99c0fb5f | 3005 | BUG(); |
16a53ecc N |
3006 | } |
3007 | break; | |
1da177e4 LT |
3008 | } |
3009 | ||
3010 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 3011 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 3012 | |
112bf897 | 3013 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 3014 | previous, &dummy1, &sh2); |
911d4ee8 N |
3015 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
3016 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
3017 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
3018 | mdname(conf->mddev)); | |
1da177e4 LT |
3019 | return 0; |
3020 | } | |
3021 | return r_sector; | |
3022 | } | |
3023 | ||
07e83364 SL |
3024 | /* |
3025 | * There are cases where we want handle_stripe_dirtying() and | |
3026 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
3027 | * | |
3028 | * This function checks whether we want to delay the towrite. Specifically, | |
3029 | * we delay the towrite when: | |
3030 | * | |
3031 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
3032 | * stripe has data in journal (for other devices). | |
3033 | * | |
3034 | * In this case, when reading data for the non-overwrite dev, it is | |
3035 | * necessary to handle complex rmw of write back cache (prexor with | |
3036 | * orig_page, and xor with page). To keep read path simple, we would | |
3037 | * like to flush data in journal to RAID disks first, so complex rmw | |
3038 | * is handled in the write patch (handle_stripe_dirtying). | |
3039 | * | |
39b99586 SL |
3040 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
3041 | * | |
3042 | * It is important to be able to flush all stripes in raid5-cache. | |
3043 | * Therefore, we need reserve some space on the journal device for | |
3044 | * these flushes. If flush operation includes pending writes to the | |
3045 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
3046 | * for the flush out. If we exclude these pending writes from flush | |
3047 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
3048 | * Therefore, excluding pending writes in these cases enables more | |
3049 | * efficient use of the journal device. | |
3050 | * | |
3051 | * Note: To make sure the stripe makes progress, we only delay | |
3052 | * towrite for stripes with data already in journal (injournal > 0). | |
3053 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
3054 | * no_space_stripes list. | |
3055 | * | |
70d466f7 SL |
3056 | * 3. during journal failure |
3057 | * In journal failure, we try to flush all cached data to raid disks | |
3058 | * based on data in stripe cache. The array is read-only to upper | |
3059 | * layers, so we would skip all pending writes. | |
3060 | * | |
07e83364 | 3061 | */ |
39b99586 SL |
3062 | static inline bool delay_towrite(struct r5conf *conf, |
3063 | struct r5dev *dev, | |
3064 | struct stripe_head_state *s) | |
07e83364 | 3065 | { |
39b99586 SL |
3066 | /* case 1 above */ |
3067 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
3068 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
3069 | return true; | |
3070 | /* case 2 above */ | |
3071 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
3072 | s->injournal > 0) | |
3073 | return true; | |
70d466f7 SL |
3074 | /* case 3 above */ |
3075 | if (s->log_failed && s->injournal) | |
3076 | return true; | |
39b99586 | 3077 | return false; |
07e83364 SL |
3078 | } |
3079 | ||
600aa109 | 3080 | static void |
c0f7bddb | 3081 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3082 | int rcw, int expand) |
e33129d8 | 3083 | { |
584acdd4 | 3084 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3085 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3086 | int level = conf->level; |
e33129d8 DW |
3087 | |
3088 | if (rcw) { | |
1e6d690b SL |
3089 | /* |
3090 | * In some cases, handle_stripe_dirtying initially decided to | |
3091 | * run rmw and allocates extra page for prexor. However, rcw is | |
3092 | * cheaper later on. We need to free the extra page now, | |
3093 | * because we won't be able to do that in ops_complete_prexor(). | |
3094 | */ | |
3095 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3096 | |
3097 | for (i = disks; i--; ) { | |
3098 | struct r5dev *dev = &sh->dev[i]; | |
3099 | ||
39b99586 | 3100 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3101 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3102 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3103 | if (!expand) |
3104 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3105 | s->locked++; |
1e6d690b SL |
3106 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3107 | set_bit(R5_LOCKED, &dev->flags); | |
3108 | s->locked++; | |
e33129d8 DW |
3109 | } |
3110 | } | |
ce7d363a N |
3111 | /* if we are not expanding this is a proper write request, and |
3112 | * there will be bios with new data to be drained into the | |
3113 | * stripe cache | |
3114 | */ | |
3115 | if (!expand) { | |
3116 | if (!s->locked) | |
3117 | /* False alarm, nothing to do */ | |
3118 | return; | |
3119 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3120 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3121 | } else | |
3122 | sh->reconstruct_state = reconstruct_state_run; | |
3123 | ||
3124 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3125 | ||
c0f7bddb | 3126 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3127 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3128 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3129 | } else { |
3130 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3131 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3132 | BUG_ON(level == 6 && |
3133 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3134 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3135 | |
e33129d8 DW |
3136 | for (i = disks; i--; ) { |
3137 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3138 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3139 | continue; |
3140 | ||
e33129d8 DW |
3141 | if (dev->towrite && |
3142 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3143 | test_bit(R5_Wantcompute, &dev->flags))) { |
3144 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3145 | set_bit(R5_LOCKED, &dev->flags); |
3146 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3147 | s->locked++; |
1e6d690b SL |
3148 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3149 | set_bit(R5_LOCKED, &dev->flags); | |
3150 | s->locked++; | |
e33129d8 DW |
3151 | } |
3152 | } | |
ce7d363a N |
3153 | if (!s->locked) |
3154 | /* False alarm - nothing to do */ | |
3155 | return; | |
3156 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3157 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3158 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3159 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3160 | } |
3161 | ||
c0f7bddb | 3162 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3163 | * are in flight |
3164 | */ | |
3165 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3166 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3167 | s->locked++; |
e33129d8 | 3168 | |
c0f7bddb YT |
3169 | if (level == 6) { |
3170 | int qd_idx = sh->qd_idx; | |
3171 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3172 | ||
3173 | set_bit(R5_LOCKED, &dev->flags); | |
3174 | clear_bit(R5_UPTODATE, &dev->flags); | |
3175 | s->locked++; | |
3176 | } | |
3177 | ||
845b9e22 | 3178 | if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page && |
3418d036 AP |
3179 | test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) && |
3180 | !test_bit(STRIPE_FULL_WRITE, &sh->state) && | |
3181 | test_bit(R5_Insync, &sh->dev[pd_idx].flags)) | |
3182 | set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request); | |
3183 | ||
600aa109 | 3184 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3185 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3186 | s->locked, s->ops_request); |
e33129d8 | 3187 | } |
16a53ecc | 3188 | |
1da177e4 LT |
3189 | /* |
3190 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3191 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3192 | * The bi_next chain must be in order. |
3193 | */ | |
da41ba65 | 3194 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3195 | int forwrite, int previous) | |
1da177e4 LT |
3196 | { |
3197 | struct bio **bip; | |
d1688a6d | 3198 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3199 | int firstwrite=0; |
1da177e4 | 3200 | |
cbe47ec5 | 3201 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3202 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3203 | (unsigned long long)sh->sector); |
3204 | ||
b17459c0 | 3205 | spin_lock_irq(&sh->stripe_lock); |
59fc630b | 3206 | /* Don't allow new IO added to stripes in batch list */ |
3207 | if (sh->batch_head) | |
3208 | goto overlap; | |
72626685 | 3209 | if (forwrite) { |
1da177e4 | 3210 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3211 | if (*bip == NULL) |
72626685 N |
3212 | firstwrite = 1; |
3213 | } else | |
1da177e4 | 3214 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3215 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3216 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3217 | goto overlap; |
3218 | bip = & (*bip)->bi_next; | |
3219 | } | |
4f024f37 | 3220 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3221 | goto overlap; |
3222 | ||
3418d036 AP |
3223 | if (forwrite && raid5_has_ppl(conf)) { |
3224 | /* | |
3225 | * With PPL only writes to consecutive data chunks within a | |
3226 | * stripe are allowed because for a single stripe_head we can | |
3227 | * only have one PPL entry at a time, which describes one data | |
3228 | * range. Not really an overlap, but wait_for_overlap can be | |
3229 | * used to handle this. | |
3230 | */ | |
3231 | sector_t sector; | |
3232 | sector_t first = 0; | |
3233 | sector_t last = 0; | |
3234 | int count = 0; | |
3235 | int i; | |
3236 | ||
3237 | for (i = 0; i < sh->disks; i++) { | |
3238 | if (i != sh->pd_idx && | |
3239 | (i == dd_idx || sh->dev[i].towrite)) { | |
3240 | sector = sh->dev[i].sector; | |
3241 | if (count == 0 || sector < first) | |
3242 | first = sector; | |
3243 | if (sector > last) | |
3244 | last = sector; | |
3245 | count++; | |
3246 | } | |
3247 | } | |
3248 | ||
3249 | if (first + conf->chunk_sectors * (count - 1) != last) | |
3250 | goto overlap; | |
3251 | } | |
3252 | ||
da41ba65 | 3253 | if (!forwrite || previous) |
3254 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3255 | ||
78bafebd | 3256 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3257 | if (*bip) |
3258 | bi->bi_next = *bip; | |
3259 | *bip = bi; | |
016c76ac | 3260 | bio_inc_remaining(bi); |
49728050 | 3261 | md_write_inc(conf->mddev, bi); |
72626685 | 3262 | |
1da177e4 LT |
3263 | if (forwrite) { |
3264 | /* check if page is covered */ | |
3265 | sector_t sector = sh->dev[dd_idx].sector; | |
3266 | for (bi=sh->dev[dd_idx].towrite; | |
3267 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3268 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3269 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3270 | if (bio_end_sector(bi) >= sector) |
3271 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3272 | } |
3273 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3274 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3275 | sh->overwrite_disks++; | |
1da177e4 | 3276 | } |
cbe47ec5 N |
3277 | |
3278 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3279 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3280 | (unsigned long long)sh->sector, dd_idx); |
3281 | ||
3282 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3283 | /* Cannot hold spinlock over bitmap_startwrite, |
3284 | * but must ensure this isn't added to a batch until | |
3285 | * we have added to the bitmap and set bm_seq. | |
3286 | * So set STRIPE_BITMAP_PENDING to prevent | |
3287 | * batching. | |
3288 | * If multiple add_stripe_bio() calls race here they | |
3289 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3290 | * to complete "bitmap_startwrite" gets to set | |
3291 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3292 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3293 | * any more. | |
3294 | */ | |
3295 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3296 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3297 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3298 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3299 | spin_lock_irq(&sh->stripe_lock); |
3300 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3301 | if (!sh->batch_head) { | |
3302 | sh->bm_seq = conf->seq_flush+1; | |
3303 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3304 | } | |
cbe47ec5 | 3305 | } |
d0852df5 | 3306 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3307 | |
3308 | if (stripe_can_batch(sh)) | |
3309 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3310 | return 1; |
3311 | ||
3312 | overlap: | |
3313 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3314 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3315 | return 0; |
3316 | } | |
3317 | ||
d1688a6d | 3318 | static void end_reshape(struct r5conf *conf); |
29269553 | 3319 | |
d1688a6d | 3320 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3321 | struct stripe_head *sh) |
ccfcc3c1 | 3322 | { |
784052ec | 3323 | int sectors_per_chunk = |
09c9e5fa | 3324 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3325 | int dd_idx; |
2d2063ce | 3326 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3327 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3328 | |
112bf897 N |
3329 | raid5_compute_sector(conf, |
3330 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3331 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3332 | previous, |
911d4ee8 | 3333 | &dd_idx, sh); |
ccfcc3c1 N |
3334 | } |
3335 | ||
a4456856 | 3336 | static void |
d1688a6d | 3337 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
bd83d0a2 | 3338 | struct stripe_head_state *s, int disks) |
a4456856 DW |
3339 | { |
3340 | int i; | |
59fc630b | 3341 | BUG_ON(sh->batch_head); |
a4456856 DW |
3342 | for (i = disks; i--; ) { |
3343 | struct bio *bi; | |
3344 | int bitmap_end = 0; | |
3345 | ||
3346 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3347 | struct md_rdev *rdev; |
a4456856 DW |
3348 | rcu_read_lock(); |
3349 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3350 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3351 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3352 | atomic_inc(&rdev->nr_pending); |
3353 | else | |
3354 | rdev = NULL; | |
a4456856 | 3355 | rcu_read_unlock(); |
7f0da59b N |
3356 | if (rdev) { |
3357 | if (!rdev_set_badblocks( | |
3358 | rdev, | |
3359 | sh->sector, | |
3360 | STRIPE_SECTORS, 0)) | |
3361 | md_error(conf->mddev, rdev); | |
3362 | rdev_dec_pending(rdev, conf->mddev); | |
3363 | } | |
a4456856 | 3364 | } |
b17459c0 | 3365 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3366 | /* fail all writes first */ |
3367 | bi = sh->dev[i].towrite; | |
3368 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3369 | sh->overwrite_disks = 0; |
b17459c0 | 3370 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3371 | if (bi) |
a4456856 | 3372 | bitmap_end = 1; |
a4456856 | 3373 | |
ff875738 | 3374 | log_stripe_write_finished(sh); |
0576b1c6 | 3375 | |
a4456856 DW |
3376 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3377 | wake_up(&conf->wait_for_overlap); | |
3378 | ||
4f024f37 | 3379 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3380 | sh->dev[i].sector + STRIPE_SECTORS) { |
3381 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3382 | |
49728050 | 3383 | md_write_end(conf->mddev); |
6308d8e3 | 3384 | bio_io_error(bi); |
a4456856 DW |
3385 | bi = nextbi; |
3386 | } | |
7eaf7e8e SL |
3387 | if (bitmap_end) |
3388 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3389 | STRIPE_SECTORS, 0, 0); | |
3390 | bitmap_end = 0; | |
a4456856 DW |
3391 | /* and fail all 'written' */ |
3392 | bi = sh->dev[i].written; | |
3393 | sh->dev[i].written = NULL; | |
d592a996 SL |
3394 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3395 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3396 | sh->dev[i].page = sh->dev[i].orig_page; | |
3397 | } | |
3398 | ||
a4456856 | 3399 | if (bi) bitmap_end = 1; |
4f024f37 | 3400 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3401 | sh->dev[i].sector + STRIPE_SECTORS) { |
3402 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3403 | |
49728050 | 3404 | md_write_end(conf->mddev); |
6308d8e3 | 3405 | bio_io_error(bi); |
a4456856 DW |
3406 | bi = bi2; |
3407 | } | |
3408 | ||
b5e98d65 DW |
3409 | /* fail any reads if this device is non-operational and |
3410 | * the data has not reached the cache yet. | |
3411 | */ | |
3412 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3413 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3414 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3415 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3416 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3417 | bi = sh->dev[i].toread; |
3418 | sh->dev[i].toread = NULL; | |
143c4d05 | 3419 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3420 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3421 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3422 | if (bi) |
3423 | s->to_read--; | |
4f024f37 | 3424 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3425 | sh->dev[i].sector + STRIPE_SECTORS) { |
3426 | struct bio *nextbi = | |
3427 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3428 | |
6308d8e3 | 3429 | bio_io_error(bi); |
a4456856 DW |
3430 | bi = nextbi; |
3431 | } | |
3432 | } | |
a4456856 DW |
3433 | if (bitmap_end) |
3434 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3435 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3436 | /* If we were in the middle of a write the parity block might |
3437 | * still be locked - so just clear all R5_LOCKED flags | |
3438 | */ | |
3439 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3440 | } |
ebda780b SL |
3441 | s->to_write = 0; |
3442 | s->written = 0; | |
a4456856 | 3443 | |
8b3e6cdc DW |
3444 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3445 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3446 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3447 | } |
3448 | ||
7f0da59b | 3449 | static void |
d1688a6d | 3450 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3451 | struct stripe_head_state *s) |
3452 | { | |
3453 | int abort = 0; | |
3454 | int i; | |
3455 | ||
59fc630b | 3456 | BUG_ON(sh->batch_head); |
7f0da59b | 3457 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3458 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3459 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3460 | s->syncing = 0; |
9a3e1101 | 3461 | s->replacing = 0; |
7f0da59b | 3462 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3463 | * Don't even need to abort as that is handled elsewhere |
3464 | * if needed, and not always wanted e.g. if there is a known | |
3465 | * bad block here. | |
9a3e1101 | 3466 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3467 | * non-sync devices, or abort the recovery |
3468 | */ | |
18b9837e N |
3469 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3470 | /* During recovery devices cannot be removed, so | |
3471 | * locking and refcounting of rdevs is not needed | |
3472 | */ | |
e50d3992 | 3473 | rcu_read_lock(); |
18b9837e | 3474 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3475 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3476 | if (rdev |
3477 | && !test_bit(Faulty, &rdev->flags) | |
3478 | && !test_bit(In_sync, &rdev->flags) | |
3479 | && !rdev_set_badblocks(rdev, sh->sector, | |
3480 | STRIPE_SECTORS, 0)) | |
3481 | abort = 1; | |
e50d3992 | 3482 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3483 | if (rdev |
3484 | && !test_bit(Faulty, &rdev->flags) | |
3485 | && !test_bit(In_sync, &rdev->flags) | |
3486 | && !rdev_set_badblocks(rdev, sh->sector, | |
3487 | STRIPE_SECTORS, 0)) | |
3488 | abort = 1; | |
3489 | } | |
e50d3992 | 3490 | rcu_read_unlock(); |
18b9837e N |
3491 | if (abort) |
3492 | conf->recovery_disabled = | |
3493 | conf->mddev->recovery_disabled; | |
7f0da59b | 3494 | } |
18b9837e | 3495 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3496 | } |
3497 | ||
9a3e1101 N |
3498 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3499 | { | |
3500 | struct md_rdev *rdev; | |
3501 | int rv = 0; | |
3f232d6a N |
3502 | |
3503 | rcu_read_lock(); | |
3504 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3505 | if (rdev |
3506 | && !test_bit(Faulty, &rdev->flags) | |
3507 | && !test_bit(In_sync, &rdev->flags) | |
3508 | && (rdev->recovery_offset <= sh->sector | |
3509 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3510 | rv = 1; | |
3f232d6a | 3511 | rcu_read_unlock(); |
9a3e1101 N |
3512 | return rv; |
3513 | } | |
3514 | ||
2c58f06e N |
3515 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3516 | int disk_idx, int disks) | |
a4456856 | 3517 | { |
5599becc | 3518 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3519 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3520 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3521 | int i; |
5599becc | 3522 | |
a79cfe12 N |
3523 | |
3524 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3525 | test_bit(R5_UPTODATE, &dev->flags)) | |
3526 | /* No point reading this as we already have it or have | |
3527 | * decided to get it. | |
3528 | */ | |
3529 | return 0; | |
3530 | ||
3531 | if (dev->toread || | |
3532 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3533 | /* We need this block to directly satisfy a request */ | |
3534 | return 1; | |
3535 | ||
3536 | if (s->syncing || s->expanding || | |
3537 | (s->replacing && want_replace(sh, disk_idx))) | |
3538 | /* When syncing, or expanding we read everything. | |
3539 | * When replacing, we need the replaced block. | |
3540 | */ | |
3541 | return 1; | |
3542 | ||
3543 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3544 | (s->failed >= 2 && fdev[1]->toread)) | |
3545 | /* If we want to read from a failed device, then | |
3546 | * we need to actually read every other device. | |
3547 | */ | |
3548 | return 1; | |
3549 | ||
a9d56950 N |
3550 | /* Sometimes neither read-modify-write nor reconstruct-write |
3551 | * cycles can work. In those cases we read every block we | |
3552 | * can. Then the parity-update is certain to have enough to | |
3553 | * work with. | |
3554 | * This can only be a problem when we need to write something, | |
3555 | * and some device has failed. If either of those tests | |
3556 | * fail we need look no further. | |
3557 | */ | |
3558 | if (!s->failed || !s->to_write) | |
3559 | return 0; | |
3560 | ||
3561 | if (test_bit(R5_Insync, &dev->flags) && | |
3562 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3563 | /* Pre-reads at not permitted until after short delay | |
3564 | * to gather multiple requests. However if this | |
3560741e | 3565 | * device is no Insync, the block could only be computed |
a9d56950 N |
3566 | * and there is no need to delay that. |
3567 | */ | |
3568 | return 0; | |
ea664c82 | 3569 | |
36707bb2 | 3570 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3571 | if (fdev[i]->towrite && |
3572 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3573 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3574 | /* If we have a partial write to a failed | |
3575 | * device, then we will need to reconstruct | |
3576 | * the content of that device, so all other | |
3577 | * devices must be read. | |
3578 | */ | |
3579 | return 1; | |
3580 | } | |
3581 | ||
3582 | /* If we are forced to do a reconstruct-write, either because | |
3583 | * the current RAID6 implementation only supports that, or | |
3560741e | 3584 | * because parity cannot be trusted and we are currently |
ea664c82 N |
3585 | * recovering it, there is extra need to be careful. |
3586 | * If one of the devices that we would need to read, because | |
3587 | * it is not being overwritten (and maybe not written at all) | |
3588 | * is missing/faulty, then we need to read everything we can. | |
3589 | */ | |
3590 | if (sh->raid_conf->level != 6 && | |
3591 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3592 | /* reconstruct-write isn't being forced */ | |
3593 | return 0; | |
36707bb2 | 3594 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3595 | if (s->failed_num[i] != sh->pd_idx && |
3596 | s->failed_num[i] != sh->qd_idx && | |
3597 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3598 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3599 | return 1; | |
3600 | } | |
3601 | ||
2c58f06e N |
3602 | return 0; |
3603 | } | |
3604 | ||
ba02684d SL |
3605 | /* fetch_block - checks the given member device to see if its data needs |
3606 | * to be read or computed to satisfy a request. | |
3607 | * | |
3608 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3609 | * 0 to tell the loop in handle_stripe_fill to continue | |
3610 | */ | |
2c58f06e N |
3611 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3612 | int disk_idx, int disks) | |
3613 | { | |
3614 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3615 | ||
3616 | /* is the data in this block needed, and can we get it? */ | |
3617 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3618 | /* we would like to get this block, possibly by computing it, |
3619 | * otherwise read it if the backing disk is insync | |
3620 | */ | |
3621 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3622 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3623 | BUG_ON(sh->batch_head); |
7471fb77 N |
3624 | |
3625 | /* | |
3626 | * In the raid6 case if the only non-uptodate disk is P | |
3627 | * then we already trusted P to compute the other failed | |
3628 | * drives. It is safe to compute rather than re-read P. | |
3629 | * In other cases we only compute blocks from failed | |
3630 | * devices, otherwise check/repair might fail to detect | |
3631 | * a real inconsistency. | |
3632 | */ | |
3633 | ||
5599becc | 3634 | if ((s->uptodate == disks - 1) && |
7471fb77 | 3635 | ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) || |
f2b3b44d | 3636 | (s->failed && (disk_idx == s->failed_num[0] || |
7471fb77 | 3637 | disk_idx == s->failed_num[1])))) { |
5599becc YT |
3638 | /* have disk failed, and we're requested to fetch it; |
3639 | * do compute it | |
a4456856 | 3640 | */ |
5599becc YT |
3641 | pr_debug("Computing stripe %llu block %d\n", |
3642 | (unsigned long long)sh->sector, disk_idx); | |
3643 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3644 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3645 | set_bit(R5_Wantcompute, &dev->flags); | |
3646 | sh->ops.target = disk_idx; | |
3647 | sh->ops.target2 = -1; /* no 2nd target */ | |
3648 | s->req_compute = 1; | |
93b3dbce N |
3649 | /* Careful: from this point on 'uptodate' is in the eye |
3650 | * of raid_run_ops which services 'compute' operations | |
3651 | * before writes. R5_Wantcompute flags a block that will | |
3652 | * be R5_UPTODATE by the time it is needed for a | |
3653 | * subsequent operation. | |
3654 | */ | |
5599becc YT |
3655 | s->uptodate++; |
3656 | return 1; | |
3657 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3658 | /* Computing 2-failure is *very* expensive; only | |
3659 | * do it if failed >= 2 | |
3660 | */ | |
3661 | int other; | |
3662 | for (other = disks; other--; ) { | |
3663 | if (other == disk_idx) | |
3664 | continue; | |
3665 | if (!test_bit(R5_UPTODATE, | |
3666 | &sh->dev[other].flags)) | |
3667 | break; | |
a4456856 | 3668 | } |
5599becc YT |
3669 | BUG_ON(other < 0); |
3670 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3671 | (unsigned long long)sh->sector, | |
3672 | disk_idx, other); | |
3673 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3674 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3675 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3676 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3677 | sh->ops.target = disk_idx; | |
3678 | sh->ops.target2 = other; | |
3679 | s->uptodate += 2; | |
3680 | s->req_compute = 1; | |
3681 | return 1; | |
3682 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3683 | set_bit(R5_LOCKED, &dev->flags); | |
3684 | set_bit(R5_Wantread, &dev->flags); | |
3685 | s->locked++; | |
3686 | pr_debug("Reading block %d (sync=%d)\n", | |
3687 | disk_idx, s->syncing); | |
a4456856 DW |
3688 | } |
3689 | } | |
5599becc YT |
3690 | |
3691 | return 0; | |
3692 | } | |
3693 | ||
3694 | /** | |
93b3dbce | 3695 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3696 | */ |
93b3dbce N |
3697 | static void handle_stripe_fill(struct stripe_head *sh, |
3698 | struct stripe_head_state *s, | |
3699 | int disks) | |
5599becc YT |
3700 | { |
3701 | int i; | |
3702 | ||
3703 | /* look for blocks to read/compute, skip this if a compute | |
3704 | * is already in flight, or if the stripe contents are in the | |
3705 | * midst of changing due to a write | |
3706 | */ | |
3707 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3708 | !sh->reconstruct_state) { |
3709 | ||
3710 | /* | |
3711 | * For degraded stripe with data in journal, do not handle | |
3712 | * read requests yet, instead, flush the stripe to raid | |
3713 | * disks first, this avoids handling complex rmw of write | |
3714 | * back cache (prexor with orig_page, and then xor with | |
3715 | * page) in the read path | |
3716 | */ | |
3717 | if (s->injournal && s->failed) { | |
3718 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3719 | r5c_make_stripe_write_out(sh); | |
3720 | goto out; | |
3721 | } | |
3722 | ||
5599becc | 3723 | for (i = disks; i--; ) |
93b3dbce | 3724 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3725 | break; |
07e83364 SL |
3726 | } |
3727 | out: | |
a4456856 DW |
3728 | set_bit(STRIPE_HANDLE, &sh->state); |
3729 | } | |
3730 | ||
787b76fa N |
3731 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3732 | unsigned long handle_flags); | |
1fe797e6 | 3733 | /* handle_stripe_clean_event |
a4456856 DW |
3734 | * any written block on an uptodate or failed drive can be returned. |
3735 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3736 | * never LOCKED, so we don't need to test 'failed' directly. | |
3737 | */ | |
d1688a6d | 3738 | static void handle_stripe_clean_event(struct r5conf *conf, |
bd83d0a2 | 3739 | struct stripe_head *sh, int disks) |
a4456856 DW |
3740 | { |
3741 | int i; | |
3742 | struct r5dev *dev; | |
f8dfcffd | 3743 | int discard_pending = 0; |
59fc630b | 3744 | struct stripe_head *head_sh = sh; |
3745 | bool do_endio = false; | |
a4456856 DW |
3746 | |
3747 | for (i = disks; i--; ) | |
3748 | if (sh->dev[i].written) { | |
3749 | dev = &sh->dev[i]; | |
3750 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3751 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3752 | test_bit(R5_Discard, &dev->flags) || |
3753 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3754 | /* We can return any write requests */ |
3755 | struct bio *wbi, *wbi2; | |
45b4233c | 3756 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3757 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3758 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3759 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3760 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3761 | } |
59fc630b | 3762 | do_endio = true; |
3763 | ||
3764 | returnbi: | |
3765 | dev->page = dev->orig_page; | |
a4456856 DW |
3766 | wbi = dev->written; |
3767 | dev->written = NULL; | |
4f024f37 | 3768 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3769 | dev->sector + STRIPE_SECTORS) { |
3770 | wbi2 = r5_next_bio(wbi, dev->sector); | |
49728050 | 3771 | md_write_end(conf->mddev); |
016c76ac | 3772 | bio_endio(wbi); |
a4456856 DW |
3773 | wbi = wbi2; |
3774 | } | |
7eaf7e8e SL |
3775 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3776 | STRIPE_SECTORS, | |
a4456856 | 3777 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3778 | 0); |
59fc630b | 3779 | if (head_sh->batch_head) { |
3780 | sh = list_first_entry(&sh->batch_list, | |
3781 | struct stripe_head, | |
3782 | batch_list); | |
3783 | if (sh != head_sh) { | |
3784 | dev = &sh->dev[i]; | |
3785 | goto returnbi; | |
3786 | } | |
3787 | } | |
3788 | sh = head_sh; | |
3789 | dev = &sh->dev[i]; | |
f8dfcffd N |
3790 | } else if (test_bit(R5_Discard, &dev->flags)) |
3791 | discard_pending = 1; | |
3792 | } | |
f6bed0ef | 3793 | |
ff875738 | 3794 | log_stripe_write_finished(sh); |
0576b1c6 | 3795 | |
f8dfcffd N |
3796 | if (!discard_pending && |
3797 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3798 | int hash; |
f8dfcffd N |
3799 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3800 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3801 | if (sh->qd_idx >= 0) { | |
3802 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3803 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3804 | } | |
3805 | /* now that discard is done we can proceed with any sync */ | |
3806 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3807 | /* |
3808 | * SCSI discard will change some bio fields and the stripe has | |
3809 | * no updated data, so remove it from hash list and the stripe | |
3810 | * will be reinitialized | |
3811 | */ | |
59fc630b | 3812 | unhash: |
b8a9d66d RG |
3813 | hash = sh->hash_lock_index; |
3814 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3815 | remove_hash(sh); |
b8a9d66d | 3816 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3817 | if (head_sh->batch_head) { |
3818 | sh = list_first_entry(&sh->batch_list, | |
3819 | struct stripe_head, batch_list); | |
3820 | if (sh != head_sh) | |
3821 | goto unhash; | |
3822 | } | |
59fc630b | 3823 | sh = head_sh; |
3824 | ||
f8dfcffd N |
3825 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3826 | set_bit(STRIPE_HANDLE, &sh->state); | |
3827 | ||
3828 | } | |
8b3e6cdc DW |
3829 | |
3830 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3831 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3832 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3833 | |
787b76fa N |
3834 | if (head_sh->batch_head && do_endio) |
3835 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3836 | } |
3837 | ||
86aa1397 SL |
3838 | /* |
3839 | * For RMW in write back cache, we need extra page in prexor to store the | |
3840 | * old data. This page is stored in dev->orig_page. | |
3841 | * | |
3842 | * This function checks whether we have data for prexor. The exact logic | |
3843 | * is: | |
3844 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3845 | */ | |
3846 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3847 | { | |
3848 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3849 | (!test_bit(R5_InJournal, &dev->flags) || | |
3850 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3851 | } | |
3852 | ||
d7bd398e SL |
3853 | static int handle_stripe_dirtying(struct r5conf *conf, |
3854 | struct stripe_head *sh, | |
3855 | struct stripe_head_state *s, | |
3856 | int disks) | |
a4456856 DW |
3857 | { |
3858 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3859 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3860 | ||
584acdd4 | 3861 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3862 | * If yes, then the array is dirty (after unclean shutdown or |
3863 | * initial creation), so parity in some stripes might be inconsistent. | |
3864 | * In this case, we need to always do reconstruct-write, to ensure | |
3865 | * that in case of drive failure or read-error correction, we | |
3866 | * generate correct data from the parity. | |
3867 | */ | |
584acdd4 | 3868 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3869 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3870 | s->failed == 0)) { | |
a7854487 | 3871 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3872 | * look like rcw is cheaper |
3873 | */ | |
3874 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3875 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3876 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3877 | (unsigned long long)sh->sector); |
c8ac1803 | 3878 | } else for (i = disks; i--; ) { |
a4456856 DW |
3879 | /* would I have to read this buffer for read_modify_write */ |
3880 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3881 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3882 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3883 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3884 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3885 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3886 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3887 | if (test_bit(R5_Insync, &dev->flags)) |
3888 | rmw++; | |
3889 | else | |
3890 | rmw += 2*disks; /* cannot read it */ | |
3891 | } | |
3892 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3893 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3894 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3895 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3896 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3897 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3898 | if (test_bit(R5_Insync, &dev->flags)) |
3899 | rcw++; | |
a4456856 DW |
3900 | else |
3901 | rcw += 2*disks; | |
3902 | } | |
3903 | } | |
1e6d690b | 3904 | |
39b99586 SL |
3905 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3906 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3907 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3908 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3909 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3910 | if (conf->mddev->queue) |
3911 | blk_add_trace_msg(conf->mddev->queue, | |
3912 | "raid5 rmw %llu %d", | |
3913 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3914 | for (i = disks; i--; ) { |
3915 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3916 | if (test_bit(R5_InJournal, &dev->flags) && |
3917 | dev->page == dev->orig_page && | |
3918 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3919 | /* alloc page for prexor */ | |
d7bd398e SL |
3920 | struct page *p = alloc_page(GFP_NOIO); |
3921 | ||
3922 | if (p) { | |
3923 | dev->orig_page = p; | |
3924 | continue; | |
3925 | } | |
3926 | ||
3927 | /* | |
3928 | * alloc_page() failed, try use | |
3929 | * disk_info->extra_page | |
3930 | */ | |
3931 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3932 | &conf->cache_state)) { | |
3933 | r5c_use_extra_page(sh); | |
3934 | break; | |
3935 | } | |
1e6d690b | 3936 | |
d7bd398e SL |
3937 | /* extra_page in use, add to delayed_list */ |
3938 | set_bit(STRIPE_DELAYED, &sh->state); | |
3939 | s->waiting_extra_page = 1; | |
3940 | return -EAGAIN; | |
1e6d690b | 3941 | } |
d7bd398e | 3942 | } |
1e6d690b | 3943 | |
d7bd398e SL |
3944 | for (i = disks; i--; ) { |
3945 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3946 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3947 | i == sh->pd_idx || i == sh->qd_idx || |
3948 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3949 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3950 | !(uptodate_for_rmw(dev) || |
1e6d690b | 3951 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 3952 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3953 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3954 | &sh->state)) { | |
3955 | pr_debug("Read_old block %d for r-m-w\n", | |
3956 | i); | |
a4456856 DW |
3957 | set_bit(R5_LOCKED, &dev->flags); |
3958 | set_bit(R5_Wantread, &dev->flags); | |
3959 | s->locked++; | |
3960 | } else { | |
3961 | set_bit(STRIPE_DELAYED, &sh->state); | |
3962 | set_bit(STRIPE_HANDLE, &sh->state); | |
3963 | } | |
3964 | } | |
3965 | } | |
a9add5d9 | 3966 | } |
41257580 | 3967 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3968 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3969 | int qread =0; |
c8ac1803 | 3970 | rcw = 0; |
a4456856 DW |
3971 | for (i = disks; i--; ) { |
3972 | struct r5dev *dev = &sh->dev[i]; | |
3973 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3974 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3975 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3976 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3977 | test_bit(R5_Wantcompute, &dev->flags))) { |
3978 | rcw++; | |
67f45548 N |
3979 | if (test_bit(R5_Insync, &dev->flags) && |
3980 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3981 | &sh->state)) { | |
45b4233c | 3982 | pr_debug("Read_old block " |
a4456856 DW |
3983 | "%d for Reconstruct\n", i); |
3984 | set_bit(R5_LOCKED, &dev->flags); | |
3985 | set_bit(R5_Wantread, &dev->flags); | |
3986 | s->locked++; | |
a9add5d9 | 3987 | qread++; |
a4456856 DW |
3988 | } else { |
3989 | set_bit(STRIPE_DELAYED, &sh->state); | |
3990 | set_bit(STRIPE_HANDLE, &sh->state); | |
3991 | } | |
3992 | } | |
3993 | } | |
e3620a3a | 3994 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3995 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3996 | (unsigned long long)sh->sector, | |
3997 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3998 | } |
b1b02fe9 N |
3999 | |
4000 | if (rcw > disks && rmw > disks && | |
4001 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4002 | set_bit(STRIPE_DELAYED, &sh->state); | |
4003 | ||
a4456856 DW |
4004 | /* now if nothing is locked, and if we have enough data, |
4005 | * we can start a write request | |
4006 | */ | |
f38e1219 DW |
4007 | /* since handle_stripe can be called at any time we need to handle the |
4008 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
4009 | * subsequent call wants to start a write request. raid_run_ops only |
4010 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
4011 | * simultaneously. If this is not the case then new writes need to be |
4012 | * held off until the compute completes. | |
4013 | */ | |
976ea8d4 DW |
4014 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
4015 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 4016 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 4017 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 4018 | return 0; |
a4456856 DW |
4019 | } |
4020 | ||
d1688a6d | 4021 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
4022 | struct stripe_head_state *s, int disks) |
4023 | { | |
ecc65c9b | 4024 | struct r5dev *dev = NULL; |
bd2ab670 | 4025 | |
59fc630b | 4026 | BUG_ON(sh->batch_head); |
a4456856 | 4027 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 4028 | |
ecc65c9b DW |
4029 | switch (sh->check_state) { |
4030 | case check_state_idle: | |
4031 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 4032 | if (s->failed == 0) { |
bd2ab670 | 4033 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
4034 | sh->check_state = check_state_run; |
4035 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 4036 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 4037 | s->uptodate--; |
ecc65c9b | 4038 | break; |
bd2ab670 | 4039 | } |
f2b3b44d | 4040 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
4041 | /* fall through */ |
4042 | case check_state_compute_result: | |
4043 | sh->check_state = check_state_idle; | |
4044 | if (!dev) | |
4045 | dev = &sh->dev[sh->pd_idx]; | |
4046 | ||
4047 | /* check that a write has not made the stripe insync */ | |
4048 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4049 | break; | |
c8894419 | 4050 | |
a4456856 | 4051 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
4052 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
4053 | BUG_ON(s->uptodate != disks); | |
4054 | ||
4055 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 4056 | s->locked++; |
a4456856 | 4057 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 4058 | |
a4456856 | 4059 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 4060 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
4061 | break; |
4062 | case check_state_run: | |
4063 | break; /* we will be called again upon completion */ | |
4064 | case check_state_check_result: | |
4065 | sh->check_state = check_state_idle; | |
4066 | ||
4067 | /* if a failure occurred during the check operation, leave | |
4068 | * STRIPE_INSYNC not set and let the stripe be handled again | |
4069 | */ | |
4070 | if (s->failed) | |
4071 | break; | |
4072 | ||
4073 | /* handle a successful check operation, if parity is correct | |
4074 | * we are done. Otherwise update the mismatch count and repair | |
4075 | * parity if !MD_RECOVERY_CHECK | |
4076 | */ | |
ad283ea4 | 4077 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
4078 | /* parity is correct (on disc, |
4079 | * not in buffer any more) | |
4080 | */ | |
4081 | set_bit(STRIPE_INSYNC, &sh->state); | |
4082 | else { | |
7f7583d4 | 4083 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
e1539036 | 4084 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
ecc65c9b DW |
4085 | /* don't try to repair!! */ |
4086 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4087 | pr_warn_ratelimited("%s: mismatch sector in range " |
4088 | "%llu-%llu\n", mdname(conf->mddev), | |
4089 | (unsigned long long) sh->sector, | |
4090 | (unsigned long long) sh->sector + | |
4091 | STRIPE_SECTORS); | |
4092 | } else { | |
ecc65c9b | 4093 | sh->check_state = check_state_compute_run; |
976ea8d4 | 4094 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
4095 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
4096 | set_bit(R5_Wantcompute, | |
4097 | &sh->dev[sh->pd_idx].flags); | |
4098 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 4099 | sh->ops.target2 = -1; |
ecc65c9b DW |
4100 | s->uptodate++; |
4101 | } | |
4102 | } | |
4103 | break; | |
4104 | case check_state_compute_run: | |
4105 | break; | |
4106 | default: | |
cc6167b4 | 4107 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
4108 | __func__, sh->check_state, |
4109 | (unsigned long long) sh->sector); | |
4110 | BUG(); | |
a4456856 DW |
4111 | } |
4112 | } | |
4113 | ||
d1688a6d | 4114 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4115 | struct stripe_head_state *s, |
f2b3b44d | 4116 | int disks) |
a4456856 | 4117 | { |
a4456856 | 4118 | int pd_idx = sh->pd_idx; |
34e04e87 | 4119 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4120 | struct r5dev *dev; |
a4456856 | 4121 | |
59fc630b | 4122 | BUG_ON(sh->batch_head); |
a4456856 DW |
4123 | set_bit(STRIPE_HANDLE, &sh->state); |
4124 | ||
4125 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4126 | |
a4456856 DW |
4127 | /* Want to check and possibly repair P and Q. |
4128 | * However there could be one 'failed' device, in which | |
4129 | * case we can only check one of them, possibly using the | |
4130 | * other to generate missing data | |
4131 | */ | |
4132 | ||
d82dfee0 DW |
4133 | switch (sh->check_state) { |
4134 | case check_state_idle: | |
4135 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4136 | if (s->failed == s->q_failed) { |
d82dfee0 | 4137 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4138 | * makes sense to check P (If anything else were failed, |
4139 | * we would have used P to recreate it). | |
4140 | */ | |
d82dfee0 | 4141 | sh->check_state = check_state_run; |
a4456856 | 4142 | } |
f2b3b44d | 4143 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4144 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4145 | * anything, so it makes sense to check it |
4146 | */ | |
d82dfee0 DW |
4147 | if (sh->check_state == check_state_run) |
4148 | sh->check_state = check_state_run_pq; | |
4149 | else | |
4150 | sh->check_state = check_state_run_q; | |
a4456856 | 4151 | } |
a4456856 | 4152 | |
d82dfee0 DW |
4153 | /* discard potentially stale zero_sum_result */ |
4154 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4155 | |
d82dfee0 DW |
4156 | if (sh->check_state == check_state_run) { |
4157 | /* async_xor_zero_sum destroys the contents of P */ | |
4158 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4159 | s->uptodate--; | |
a4456856 | 4160 | } |
d82dfee0 DW |
4161 | if (sh->check_state >= check_state_run && |
4162 | sh->check_state <= check_state_run_pq) { | |
4163 | /* async_syndrome_zero_sum preserves P and Q, so | |
4164 | * no need to mark them !uptodate here | |
4165 | */ | |
4166 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4167 | break; | |
a4456856 DW |
4168 | } |
4169 | ||
d82dfee0 DW |
4170 | /* we have 2-disk failure */ |
4171 | BUG_ON(s->failed != 2); | |
4172 | /* fall through */ | |
4173 | case check_state_compute_result: | |
4174 | sh->check_state = check_state_idle; | |
a4456856 | 4175 | |
d82dfee0 DW |
4176 | /* check that a write has not made the stripe insync */ |
4177 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4178 | break; | |
a4456856 DW |
4179 | |
4180 | /* now write out any block on a failed drive, | |
d82dfee0 | 4181 | * or P or Q if they were recomputed |
a4456856 | 4182 | */ |
d82dfee0 | 4183 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 4184 | if (s->failed == 2) { |
f2b3b44d | 4185 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4186 | s->locked++; |
4187 | set_bit(R5_LOCKED, &dev->flags); | |
4188 | set_bit(R5_Wantwrite, &dev->flags); | |
4189 | } | |
4190 | if (s->failed >= 1) { | |
f2b3b44d | 4191 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4192 | s->locked++; |
4193 | set_bit(R5_LOCKED, &dev->flags); | |
4194 | set_bit(R5_Wantwrite, &dev->flags); | |
4195 | } | |
d82dfee0 | 4196 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4197 | dev = &sh->dev[pd_idx]; |
4198 | s->locked++; | |
4199 | set_bit(R5_LOCKED, &dev->flags); | |
4200 | set_bit(R5_Wantwrite, &dev->flags); | |
4201 | } | |
d82dfee0 | 4202 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4203 | dev = &sh->dev[qd_idx]; |
4204 | s->locked++; | |
4205 | set_bit(R5_LOCKED, &dev->flags); | |
4206 | set_bit(R5_Wantwrite, &dev->flags); | |
4207 | } | |
4208 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
4209 | ||
4210 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4211 | break; |
4212 | case check_state_run: | |
4213 | case check_state_run_q: | |
4214 | case check_state_run_pq: | |
4215 | break; /* we will be called again upon completion */ | |
4216 | case check_state_check_result: | |
4217 | sh->check_state = check_state_idle; | |
4218 | ||
4219 | /* handle a successful check operation, if parity is correct | |
4220 | * we are done. Otherwise update the mismatch count and repair | |
4221 | * parity if !MD_RECOVERY_CHECK | |
4222 | */ | |
4223 | if (sh->ops.zero_sum_result == 0) { | |
4224 | /* both parities are correct */ | |
4225 | if (!s->failed) | |
4226 | set_bit(STRIPE_INSYNC, &sh->state); | |
4227 | else { | |
4228 | /* in contrast to the raid5 case we can validate | |
4229 | * parity, but still have a failure to write | |
4230 | * back | |
4231 | */ | |
4232 | sh->check_state = check_state_compute_result; | |
4233 | /* Returning at this point means that we may go | |
4234 | * off and bring p and/or q uptodate again so | |
4235 | * we make sure to check zero_sum_result again | |
4236 | * to verify if p or q need writeback | |
4237 | */ | |
4238 | } | |
4239 | } else { | |
7f7583d4 | 4240 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
e1539036 | 4241 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
d82dfee0 DW |
4242 | /* don't try to repair!! */ |
4243 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4244 | pr_warn_ratelimited("%s: mismatch sector in range " |
4245 | "%llu-%llu\n", mdname(conf->mddev), | |
4246 | (unsigned long long) sh->sector, | |
4247 | (unsigned long long) sh->sector + | |
4248 | STRIPE_SECTORS); | |
4249 | } else { | |
d82dfee0 DW |
4250 | int *target = &sh->ops.target; |
4251 | ||
4252 | sh->ops.target = -1; | |
4253 | sh->ops.target2 = -1; | |
4254 | sh->check_state = check_state_compute_run; | |
4255 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4256 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4257 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4258 | set_bit(R5_Wantcompute, | |
4259 | &sh->dev[pd_idx].flags); | |
4260 | *target = pd_idx; | |
4261 | target = &sh->ops.target2; | |
4262 | s->uptodate++; | |
4263 | } | |
4264 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4265 | set_bit(R5_Wantcompute, | |
4266 | &sh->dev[qd_idx].flags); | |
4267 | *target = qd_idx; | |
4268 | s->uptodate++; | |
4269 | } | |
4270 | } | |
4271 | } | |
4272 | break; | |
4273 | case check_state_compute_run: | |
4274 | break; | |
4275 | default: | |
cc6167b4 N |
4276 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4277 | __func__, sh->check_state, | |
4278 | (unsigned long long) sh->sector); | |
d82dfee0 | 4279 | BUG(); |
a4456856 DW |
4280 | } |
4281 | } | |
4282 | ||
d1688a6d | 4283 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4284 | { |
4285 | int i; | |
4286 | ||
4287 | /* We have read all the blocks in this stripe and now we need to | |
4288 | * copy some of them into a target stripe for expand. | |
4289 | */ | |
f0a50d37 | 4290 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4291 | BUG_ON(sh->batch_head); |
a4456856 DW |
4292 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4293 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4294 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4295 | int dd_idx, j; |
a4456856 | 4296 | struct stripe_head *sh2; |
a08abd8c | 4297 | struct async_submit_ctl submit; |
a4456856 | 4298 | |
6d036f7d | 4299 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4300 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4301 | &dd_idx, NULL); | |
6d036f7d | 4302 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4303 | if (sh2 == NULL) |
4304 | /* so far only the early blocks of this stripe | |
4305 | * have been requested. When later blocks | |
4306 | * get requested, we will try again | |
4307 | */ | |
4308 | continue; | |
4309 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4310 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4311 | /* must have already done this block */ | |
6d036f7d | 4312 | raid5_release_stripe(sh2); |
a4456856 DW |
4313 | continue; |
4314 | } | |
f0a50d37 DW |
4315 | |
4316 | /* place all the copies on one channel */ | |
a08abd8c | 4317 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4318 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 4319 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 4320 | &submit); |
f0a50d37 | 4321 | |
a4456856 DW |
4322 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4323 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4324 | for (j = 0; j < conf->raid_disks; j++) | |
4325 | if (j != sh2->pd_idx && | |
86c374ba | 4326 | j != sh2->qd_idx && |
a4456856 DW |
4327 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4328 | break; | |
4329 | if (j == conf->raid_disks) { | |
4330 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4331 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4332 | } | |
6d036f7d | 4333 | raid5_release_stripe(sh2); |
f0a50d37 | 4334 | |
a4456856 | 4335 | } |
a2e08551 | 4336 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4337 | async_tx_quiesce(&tx); |
a4456856 | 4338 | } |
1da177e4 LT |
4339 | |
4340 | /* | |
4341 | * handle_stripe - do things to a stripe. | |
4342 | * | |
9a3e1101 N |
4343 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4344 | * state of various bits to see what needs to be done. | |
1da177e4 | 4345 | * Possible results: |
9a3e1101 N |
4346 | * return some read requests which now have data |
4347 | * return some write requests which are safely on storage | |
1da177e4 LT |
4348 | * schedule a read on some buffers |
4349 | * schedule a write of some buffers | |
4350 | * return confirmation of parity correctness | |
4351 | * | |
1da177e4 | 4352 | */ |
a4456856 | 4353 | |
acfe726b | 4354 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4355 | { |
d1688a6d | 4356 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4357 | int disks = sh->disks; |
474af965 N |
4358 | struct r5dev *dev; |
4359 | int i; | |
9a3e1101 | 4360 | int do_recovery = 0; |
1da177e4 | 4361 | |
acfe726b N |
4362 | memset(s, 0, sizeof(*s)); |
4363 | ||
dabc4ec6 | 4364 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4365 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4366 | s->failed_num[0] = -1; |
4367 | s->failed_num[1] = -1; | |
6e74a9cf | 4368 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4369 | |
acfe726b | 4370 | /* Now to look around and see what can be done */ |
1da177e4 | 4371 | rcu_read_lock(); |
16a53ecc | 4372 | for (i=disks; i--; ) { |
3cb03002 | 4373 | struct md_rdev *rdev; |
31c176ec N |
4374 | sector_t first_bad; |
4375 | int bad_sectors; | |
4376 | int is_bad = 0; | |
acfe726b | 4377 | |
16a53ecc | 4378 | dev = &sh->dev[i]; |
1da177e4 | 4379 | |
45b4233c | 4380 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4381 | i, dev->flags, |
4382 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4383 | /* maybe we can reply to a read |
4384 | * | |
4385 | * new wantfill requests are only permitted while | |
4386 | * ops_complete_biofill is guaranteed to be inactive | |
4387 | */ | |
4388 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4389 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4390 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4391 | |
16a53ecc | 4392 | /* now count some things */ |
cc94015a N |
4393 | if (test_bit(R5_LOCKED, &dev->flags)) |
4394 | s->locked++; | |
4395 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4396 | s->uptodate++; | |
2d6e4ecc | 4397 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4398 | s->compute++; |
4399 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4400 | } |
1da177e4 | 4401 | |
acfe726b | 4402 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4403 | s->to_fill++; |
acfe726b | 4404 | else if (dev->toread) |
cc94015a | 4405 | s->to_read++; |
16a53ecc | 4406 | if (dev->towrite) { |
cc94015a | 4407 | s->to_write++; |
16a53ecc | 4408 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4409 | s->non_overwrite++; |
16a53ecc | 4410 | } |
a4456856 | 4411 | if (dev->written) |
cc94015a | 4412 | s->written++; |
14a75d3e N |
4413 | /* Prefer to use the replacement for reads, but only |
4414 | * if it is recovered enough and has no bad blocks. | |
4415 | */ | |
4416 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4417 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4418 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4419 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4420 | &first_bad, &bad_sectors)) | |
4421 | set_bit(R5_ReadRepl, &dev->flags); | |
4422 | else { | |
e6030cb0 | 4423 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4424 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4425 | else |
4426 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4427 | rdev = rcu_dereference(conf->disks[i].rdev); |
4428 | clear_bit(R5_ReadRepl, &dev->flags); | |
4429 | } | |
9283d8c5 N |
4430 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4431 | rdev = NULL; | |
31c176ec N |
4432 | if (rdev) { |
4433 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4434 | &first_bad, &bad_sectors); | |
4435 | if (s->blocked_rdev == NULL | |
4436 | && (test_bit(Blocked, &rdev->flags) | |
4437 | || is_bad < 0)) { | |
4438 | if (is_bad < 0) | |
4439 | set_bit(BlockedBadBlocks, | |
4440 | &rdev->flags); | |
4441 | s->blocked_rdev = rdev; | |
4442 | atomic_inc(&rdev->nr_pending); | |
4443 | } | |
6bfe0b49 | 4444 | } |
415e72d0 N |
4445 | clear_bit(R5_Insync, &dev->flags); |
4446 | if (!rdev) | |
4447 | /* Not in-sync */; | |
31c176ec N |
4448 | else if (is_bad) { |
4449 | /* also not in-sync */ | |
18b9837e N |
4450 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4451 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4452 | /* treat as in-sync, but with a read error |
4453 | * which we can now try to correct | |
4454 | */ | |
4455 | set_bit(R5_Insync, &dev->flags); | |
4456 | set_bit(R5_ReadError, &dev->flags); | |
4457 | } | |
4458 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4459 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4460 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4461 | /* in sync if before recovery_offset */ |
30d7a483 N |
4462 | set_bit(R5_Insync, &dev->flags); |
4463 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4464 | test_bit(R5_Expanded, &dev->flags)) | |
4465 | /* If we've reshaped into here, we assume it is Insync. | |
4466 | * We will shortly update recovery_offset to make | |
4467 | * it official. | |
4468 | */ | |
4469 | set_bit(R5_Insync, &dev->flags); | |
4470 | ||
1cc03eb9 | 4471 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4472 | /* This flag does not apply to '.replacement' |
4473 | * only to .rdev, so make sure to check that*/ | |
4474 | struct md_rdev *rdev2 = rcu_dereference( | |
4475 | conf->disks[i].rdev); | |
4476 | if (rdev2 == rdev) | |
4477 | clear_bit(R5_Insync, &dev->flags); | |
4478 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4479 | s->handle_bad_blocks = 1; |
14a75d3e | 4480 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4481 | } else |
4482 | clear_bit(R5_WriteError, &dev->flags); | |
4483 | } | |
1cc03eb9 | 4484 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4485 | /* This flag does not apply to '.replacement' |
4486 | * only to .rdev, so make sure to check that*/ | |
4487 | struct md_rdev *rdev2 = rcu_dereference( | |
4488 | conf->disks[i].rdev); | |
4489 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4490 | s->handle_bad_blocks = 1; |
14a75d3e | 4491 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4492 | } else |
4493 | clear_bit(R5_MadeGood, &dev->flags); | |
4494 | } | |
977df362 N |
4495 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4496 | struct md_rdev *rdev2 = rcu_dereference( | |
4497 | conf->disks[i].replacement); | |
4498 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4499 | s->handle_bad_blocks = 1; | |
4500 | atomic_inc(&rdev2->nr_pending); | |
4501 | } else | |
4502 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4503 | } | |
415e72d0 | 4504 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4505 | /* The ReadError flag will just be confusing now */ |
4506 | clear_bit(R5_ReadError, &dev->flags); | |
4507 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4508 | } |
415e72d0 N |
4509 | if (test_bit(R5_ReadError, &dev->flags)) |
4510 | clear_bit(R5_Insync, &dev->flags); | |
4511 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4512 | if (s->failed < 2) |
4513 | s->failed_num[s->failed] = i; | |
4514 | s->failed++; | |
9a3e1101 N |
4515 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4516 | do_recovery = 1; | |
415e72d0 | 4517 | } |
2ded3703 SL |
4518 | |
4519 | if (test_bit(R5_InJournal, &dev->flags)) | |
4520 | s->injournal++; | |
1e6d690b SL |
4521 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4522 | s->just_cached++; | |
1da177e4 | 4523 | } |
9a3e1101 N |
4524 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4525 | /* If there is a failed device being replaced, | |
4526 | * we must be recovering. | |
4527 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4528 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4529 | * else we can only be replacing |
4530 | * sync and recovery both need to read all devices, and so | |
4531 | * use the same flag. | |
4532 | */ | |
4533 | if (do_recovery || | |
c6d2e084 | 4534 | sh->sector >= conf->mddev->recovery_cp || |
4535 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4536 | s->syncing = 1; |
4537 | else | |
4538 | s->replacing = 1; | |
4539 | } | |
1da177e4 | 4540 | rcu_read_unlock(); |
cc94015a N |
4541 | } |
4542 | ||
59fc630b | 4543 | static int clear_batch_ready(struct stripe_head *sh) |
4544 | { | |
b15a9dbd N |
4545 | /* Return '1' if this is a member of batch, or |
4546 | * '0' if it is a lone stripe or a head which can now be | |
4547 | * handled. | |
4548 | */ | |
59fc630b | 4549 | struct stripe_head *tmp; |
4550 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4551 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4552 | spin_lock(&sh->stripe_lock); |
4553 | if (!sh->batch_head) { | |
4554 | spin_unlock(&sh->stripe_lock); | |
4555 | return 0; | |
4556 | } | |
4557 | ||
4558 | /* | |
4559 | * this stripe could be added to a batch list before we check | |
4560 | * BATCH_READY, skips it | |
4561 | */ | |
4562 | if (sh->batch_head != sh) { | |
4563 | spin_unlock(&sh->stripe_lock); | |
4564 | return 1; | |
4565 | } | |
4566 | spin_lock(&sh->batch_lock); | |
4567 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4568 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4569 | spin_unlock(&sh->batch_lock); | |
4570 | spin_unlock(&sh->stripe_lock); | |
4571 | ||
4572 | /* | |
4573 | * BATCH_READY is cleared, no new stripes can be added. | |
4574 | * batch_list can be accessed without lock | |
4575 | */ | |
4576 | return 0; | |
4577 | } | |
4578 | ||
3960ce79 N |
4579 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4580 | unsigned long handle_flags) | |
72ac7330 | 4581 | { |
4e3d62ff | 4582 | struct stripe_head *sh, *next; |
72ac7330 | 4583 | int i; |
fb642b92 | 4584 | int do_wakeup = 0; |
72ac7330 | 4585 | |
bb27051f N |
4586 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4587 | ||
72ac7330 | 4588 | list_del_init(&sh->batch_list); |
4589 | ||
fb3229d5 | 4590 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4591 | (1 << STRIPE_SYNCING) | |
4592 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4593 | (1 << STRIPE_DELAYED) | |
4594 | (1 << STRIPE_BIT_DELAY) | | |
4595 | (1 << STRIPE_FULL_WRITE) | | |
4596 | (1 << STRIPE_BIOFILL_RUN) | | |
4597 | (1 << STRIPE_COMPUTE_RUN) | | |
4598 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4599 | (1 << STRIPE_DISCARD) | | |
4600 | (1 << STRIPE_BATCH_READY) | | |
4601 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4602 | (1 << STRIPE_BITMAP_PENDING)), |
4603 | "stripe state: %lx\n", sh->state); | |
4604 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4605 | (1 << STRIPE_REPLACED)), | |
4606 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4607 | |
4608 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4609 | (1 << STRIPE_PREREAD_ACTIVE) | |
184a09eb DY |
4610 | (1 << STRIPE_DEGRADED) | |
4611 | (1 << STRIPE_ON_UNPLUG_LIST)), | |
1b956f7a N |
4612 | head_sh->state & (1 << STRIPE_INSYNC)); |
4613 | ||
72ac7330 | 4614 | sh->check_state = head_sh->check_state; |
4615 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4616 | for (i = 0; i < sh->disks; i++) { |
4617 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4618 | do_wakeup = 1; | |
72ac7330 | 4619 | sh->dev[i].flags = head_sh->dev[i].flags & |
4620 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4621 | } |
72ac7330 | 4622 | spin_lock_irq(&sh->stripe_lock); |
4623 | sh->batch_head = NULL; | |
4624 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4625 | if (handle_flags == 0 || |
4626 | sh->state & handle_flags) | |
4627 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4628 | raid5_release_stripe(sh); |
72ac7330 | 4629 | } |
fb642b92 N |
4630 | spin_lock_irq(&head_sh->stripe_lock); |
4631 | head_sh->batch_head = NULL; | |
4632 | spin_unlock_irq(&head_sh->stripe_lock); | |
4633 | for (i = 0; i < head_sh->disks; i++) | |
4634 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4635 | do_wakeup = 1; | |
3960ce79 N |
4636 | if (head_sh->state & handle_flags) |
4637 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4638 | |
4639 | if (do_wakeup) | |
4640 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4641 | } |
4642 | ||
cc94015a N |
4643 | static void handle_stripe(struct stripe_head *sh) |
4644 | { | |
4645 | struct stripe_head_state s; | |
d1688a6d | 4646 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4647 | int i; |
84789554 N |
4648 | int prexor; |
4649 | int disks = sh->disks; | |
474af965 | 4650 | struct r5dev *pdev, *qdev; |
cc94015a N |
4651 | |
4652 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4653 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4654 | /* already being handled, ensure it gets handled |
4655 | * again when current action finishes */ | |
4656 | set_bit(STRIPE_HANDLE, &sh->state); | |
4657 | return; | |
4658 | } | |
4659 | ||
59fc630b | 4660 | if (clear_batch_ready(sh) ) { |
4661 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4662 | return; | |
4663 | } | |
4664 | ||
4e3d62ff | 4665 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4666 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4667 | |
dabc4ec6 | 4668 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd | 4669 | spin_lock(&sh->stripe_lock); |
5ddf0440 SL |
4670 | /* |
4671 | * Cannot process 'sync' concurrently with 'discard'. | |
4672 | * Flush data in r5cache before 'sync'. | |
4673 | */ | |
4674 | if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) && | |
4675 | !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) && | |
4676 | !test_bit(STRIPE_DISCARD, &sh->state) && | |
f8dfcffd N |
4677 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
4678 | set_bit(STRIPE_SYNCING, &sh->state); | |
4679 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4680 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4681 | } |
4682 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4683 | } |
4684 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4685 | ||
4686 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4687 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4688 | (unsigned long long)sh->sector, sh->state, | |
4689 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4690 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4691 | |
acfe726b | 4692 | analyse_stripe(sh, &s); |
c5a31000 | 4693 | |
b70abcb2 SL |
4694 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4695 | goto finish; | |
4696 | ||
16d997b7 N |
4697 | if (s.handle_bad_blocks || |
4698 | test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { | |
bc2607f3 N |
4699 | set_bit(STRIPE_HANDLE, &sh->state); |
4700 | goto finish; | |
4701 | } | |
4702 | ||
474af965 N |
4703 | if (unlikely(s.blocked_rdev)) { |
4704 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4705 | s.replacing || s.to_write || s.written) { |
474af965 N |
4706 | set_bit(STRIPE_HANDLE, &sh->state); |
4707 | goto finish; | |
4708 | } | |
4709 | /* There is nothing for the blocked_rdev to block */ | |
4710 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4711 | s.blocked_rdev = NULL; | |
4712 | } | |
4713 | ||
4714 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4715 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4716 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4717 | } | |
4718 | ||
4719 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4720 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4721 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4722 | s.failed_num[0], s.failed_num[1]); | |
70d466f7 SL |
4723 | /* |
4724 | * check if the array has lost more than max_degraded devices and, | |
474af965 | 4725 | * if so, some requests might need to be failed. |
70d466f7 SL |
4726 | * |
4727 | * When journal device failed (log_failed), we will only process | |
4728 | * the stripe if there is data need write to raid disks | |
474af965 | 4729 | */ |
70d466f7 SL |
4730 | if (s.failed > conf->max_degraded || |
4731 | (s.log_failed && s.injournal == 0)) { | |
9a3f530f N |
4732 | sh->check_state = 0; |
4733 | sh->reconstruct_state = 0; | |
626f2092 | 4734 | break_stripe_batch_list(sh, 0); |
9a3f530f | 4735 | if (s.to_read+s.to_write+s.written) |
bd83d0a2 | 4736 | handle_failed_stripe(conf, sh, &s, disks); |
9a3e1101 | 4737 | if (s.syncing + s.replacing) |
9a3f530f N |
4738 | handle_failed_sync(conf, sh, &s); |
4739 | } | |
474af965 | 4740 | |
84789554 N |
4741 | /* Now we check to see if any write operations have recently |
4742 | * completed | |
4743 | */ | |
4744 | prexor = 0; | |
4745 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4746 | prexor = 1; | |
4747 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4748 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4749 | sh->reconstruct_state = reconstruct_state_idle; | |
4750 | ||
4751 | /* All the 'written' buffers and the parity block are ready to | |
4752 | * be written back to disk | |
4753 | */ | |
9e444768 SL |
4754 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4755 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4756 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4757 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4758 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4759 | for (i = disks; i--; ) { |
4760 | struct r5dev *dev = &sh->dev[i]; | |
4761 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4762 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4763 | dev->written || test_bit(R5_InJournal, |
4764 | &dev->flags))) { | |
84789554 N |
4765 | pr_debug("Writing block %d\n", i); |
4766 | set_bit(R5_Wantwrite, &dev->flags); | |
4767 | if (prexor) | |
4768 | continue; | |
9c4bdf69 N |
4769 | if (s.failed > 1) |
4770 | continue; | |
84789554 N |
4771 | if (!test_bit(R5_Insync, &dev->flags) || |
4772 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4773 | s.failed == 0)) | |
4774 | set_bit(STRIPE_INSYNC, &sh->state); | |
4775 | } | |
4776 | } | |
4777 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4778 | s.dec_preread_active = 1; | |
4779 | } | |
4780 | ||
ef5b7c69 N |
4781 | /* |
4782 | * might be able to return some write requests if the parity blocks | |
4783 | * are safe, or on a failed drive | |
4784 | */ | |
4785 | pdev = &sh->dev[sh->pd_idx]; | |
4786 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4787 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4788 | qdev = &sh->dev[sh->qd_idx]; | |
4789 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4790 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4791 | || conf->level < 6; | |
4792 | ||
4793 | if (s.written && | |
4794 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4795 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4796 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4797 | test_bit(R5_Discard, &pdev->flags))))) && | |
4798 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4799 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4800 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4801 | test_bit(R5_Discard, &qdev->flags)))))) | |
bd83d0a2 | 4802 | handle_stripe_clean_event(conf, sh, disks); |
ef5b7c69 | 4803 | |
1e6d690b | 4804 | if (s.just_cached) |
bd83d0a2 | 4805 | r5c_handle_cached_data_endio(conf, sh, disks); |
ff875738 | 4806 | log_stripe_write_finished(sh); |
1e6d690b | 4807 | |
ef5b7c69 N |
4808 | /* Now we might consider reading some blocks, either to check/generate |
4809 | * parity, or to satisfy requests | |
4810 | * or to load a block that is being partially written. | |
4811 | */ | |
4812 | if (s.to_read || s.non_overwrite | |
4813 | || (conf->level == 6 && s.to_write && s.failed) | |
4814 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4815 | || s.replacing | |
4816 | || s.expanding) | |
4817 | handle_stripe_fill(sh, &s, disks); | |
4818 | ||
2ded3703 SL |
4819 | /* |
4820 | * When the stripe finishes full journal write cycle (write to journal | |
4821 | * and raid disk), this is the clean up procedure so it is ready for | |
4822 | * next operation. | |
4823 | */ | |
4824 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4825 | ||
4826 | /* | |
4827 | * Now to consider new write requests, cache write back and what else, | |
4828 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4829 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4830 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4831 | * block. | |
2ded3703 | 4832 | * 3/ A r5c cache log write is in flight. |
84789554 | 4833 | */ |
2ded3703 SL |
4834 | |
4835 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4836 | if (!r5c_is_writeback(conf->log)) { | |
4837 | if (s.to_write) | |
4838 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4839 | } else { /* write back cache */ | |
4840 | int ret = 0; | |
4841 | ||
4842 | /* First, try handle writes in caching phase */ | |
4843 | if (s.to_write) | |
4844 | ret = r5c_try_caching_write(conf, sh, &s, | |
4845 | disks); | |
4846 | /* | |
4847 | * If caching phase failed: ret == -EAGAIN | |
4848 | * OR | |
4849 | * stripe under reclaim: !caching && injournal | |
4850 | * | |
4851 | * fall back to handle_stripe_dirtying() | |
4852 | */ | |
4853 | if (ret == -EAGAIN || | |
4854 | /* stripe under reclaim: !caching && injournal */ | |
4855 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4856 | s.injournal > 0)) { |
4857 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4858 | disks); | |
4859 | if (ret == -EAGAIN) | |
4860 | goto finish; | |
4861 | } | |
2ded3703 SL |
4862 | } |
4863 | } | |
84789554 N |
4864 | |
4865 | /* maybe we need to check and possibly fix the parity for this stripe | |
4866 | * Any reads will already have been scheduled, so we just see if enough | |
4867 | * data is available. The parity check is held off while parity | |
4868 | * dependent operations are in flight. | |
4869 | */ | |
4870 | if (sh->check_state || | |
4871 | (s.syncing && s.locked == 0 && | |
4872 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4873 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4874 | if (conf->level == 6) | |
4875 | handle_parity_checks6(conf, sh, &s, disks); | |
4876 | else | |
4877 | handle_parity_checks5(conf, sh, &s, disks); | |
4878 | } | |
c5a31000 | 4879 | |
f94c0b66 N |
4880 | if ((s.replacing || s.syncing) && s.locked == 0 |
4881 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4882 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4883 | /* Write out to replacement devices where possible */ |
4884 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4885 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4886 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4887 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4888 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4889 | s.locked++; | |
4890 | } | |
f94c0b66 N |
4891 | if (s.replacing) |
4892 | set_bit(STRIPE_INSYNC, &sh->state); | |
4893 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4894 | } |
4895 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4896 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4897 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4898 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4899 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4900 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4901 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4902 | } |
4903 | ||
4904 | /* If the failed drives are just a ReadError, then we might need | |
4905 | * to progress the repair/check process | |
4906 | */ | |
4907 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4908 | for (i = 0; i < s.failed; i++) { | |
4909 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4910 | if (test_bit(R5_ReadError, &dev->flags) | |
4911 | && !test_bit(R5_LOCKED, &dev->flags) | |
4912 | && test_bit(R5_UPTODATE, &dev->flags) | |
4913 | ) { | |
4914 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4915 | set_bit(R5_Wantwrite, &dev->flags); | |
4916 | set_bit(R5_ReWrite, &dev->flags); | |
4917 | set_bit(R5_LOCKED, &dev->flags); | |
4918 | s.locked++; | |
4919 | } else { | |
4920 | /* let's read it back */ | |
4921 | set_bit(R5_Wantread, &dev->flags); | |
4922 | set_bit(R5_LOCKED, &dev->flags); | |
4923 | s.locked++; | |
4924 | } | |
4925 | } | |
4926 | } | |
4927 | ||
3687c061 N |
4928 | /* Finish reconstruct operations initiated by the expansion process */ |
4929 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4930 | struct stripe_head *sh_src | |
6d036f7d | 4931 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4932 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4933 | /* sh cannot be written until sh_src has been read. | |
4934 | * so arrange for sh to be delayed a little | |
4935 | */ | |
4936 | set_bit(STRIPE_DELAYED, &sh->state); | |
4937 | set_bit(STRIPE_HANDLE, &sh->state); | |
4938 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4939 | &sh_src->state)) | |
4940 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4941 | raid5_release_stripe(sh_src); |
3687c061 N |
4942 | goto finish; |
4943 | } | |
4944 | if (sh_src) | |
6d036f7d | 4945 | raid5_release_stripe(sh_src); |
3687c061 N |
4946 | |
4947 | sh->reconstruct_state = reconstruct_state_idle; | |
4948 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4949 | for (i = conf->raid_disks; i--; ) { | |
4950 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4951 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4952 | s.locked++; | |
4953 | } | |
4954 | } | |
f416885e | 4955 | |
3687c061 N |
4956 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4957 | !sh->reconstruct_state) { | |
4958 | /* Need to write out all blocks after computing parity */ | |
4959 | sh->disks = conf->raid_disks; | |
4960 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4961 | schedule_reconstruction(sh, &s, 1, 1); | |
4962 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4963 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4964 | atomic_dec(&conf->reshape_stripes); | |
4965 | wake_up(&conf->wait_for_overlap); | |
4966 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4967 | } | |
4968 | ||
4969 | if (s.expanding && s.locked == 0 && | |
4970 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4971 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4972 | |
3687c061 | 4973 | finish: |
6bfe0b49 | 4974 | /* wait for this device to become unblocked */ |
5f066c63 N |
4975 | if (unlikely(s.blocked_rdev)) { |
4976 | if (conf->mddev->external) | |
4977 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4978 | conf->mddev); | |
4979 | else | |
4980 | /* Internal metadata will immediately | |
4981 | * be written by raid5d, so we don't | |
4982 | * need to wait here. | |
4983 | */ | |
4984 | rdev_dec_pending(s.blocked_rdev, | |
4985 | conf->mddev); | |
4986 | } | |
6bfe0b49 | 4987 | |
bc2607f3 N |
4988 | if (s.handle_bad_blocks) |
4989 | for (i = disks; i--; ) { | |
3cb03002 | 4990 | struct md_rdev *rdev; |
bc2607f3 N |
4991 | struct r5dev *dev = &sh->dev[i]; |
4992 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4993 | /* We own a safe reference to the rdev */ | |
4994 | rdev = conf->disks[i].rdev; | |
4995 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4996 | STRIPE_SECTORS, 0)) | |
4997 | md_error(conf->mddev, rdev); | |
4998 | rdev_dec_pending(rdev, conf->mddev); | |
4999 | } | |
b84db560 N |
5000 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
5001 | rdev = conf->disks[i].rdev; | |
5002 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 5003 | STRIPE_SECTORS, 0); |
b84db560 N |
5004 | rdev_dec_pending(rdev, conf->mddev); |
5005 | } | |
977df362 N |
5006 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
5007 | rdev = conf->disks[i].replacement; | |
dd054fce N |
5008 | if (!rdev) |
5009 | /* rdev have been moved down */ | |
5010 | rdev = conf->disks[i].rdev; | |
977df362 | 5011 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 5012 | STRIPE_SECTORS, 0); |
977df362 N |
5013 | rdev_dec_pending(rdev, conf->mddev); |
5014 | } | |
bc2607f3 N |
5015 | } |
5016 | ||
6c0069c0 YT |
5017 | if (s.ops_request) |
5018 | raid_run_ops(sh, s.ops_request); | |
5019 | ||
f0e43bcd | 5020 | ops_run_io(sh, &s); |
16a53ecc | 5021 | |
c5709ef6 | 5022 | if (s.dec_preread_active) { |
729a1866 | 5023 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 5024 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
5025 | * have actually been submitted. |
5026 | */ | |
5027 | atomic_dec(&conf->preread_active_stripes); | |
5028 | if (atomic_read(&conf->preread_active_stripes) < | |
5029 | IO_THRESHOLD) | |
5030 | md_wakeup_thread(conf->mddev->thread); | |
5031 | } | |
5032 | ||
257a4b42 | 5033 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
5034 | } |
5035 | ||
d1688a6d | 5036 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
5037 | { |
5038 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
5039 | while (!list_empty(&conf->delayed_list)) { | |
5040 | struct list_head *l = conf->delayed_list.next; | |
5041 | struct stripe_head *sh; | |
5042 | sh = list_entry(l, struct stripe_head, lru); | |
5043 | list_del_init(l); | |
5044 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5045 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5046 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 5047 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 5048 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 5049 | } |
482c0834 | 5050 | } |
16a53ecc N |
5051 | } |
5052 | ||
566c09c5 SL |
5053 | static void activate_bit_delay(struct r5conf *conf, |
5054 | struct list_head *temp_inactive_list) | |
16a53ecc N |
5055 | { |
5056 | /* device_lock is held */ | |
5057 | struct list_head head; | |
5058 | list_add(&head, &conf->bitmap_list); | |
5059 | list_del_init(&conf->bitmap_list); | |
5060 | while (!list_empty(&head)) { | |
5061 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 5062 | int hash; |
16a53ecc N |
5063 | list_del_init(&sh->lru); |
5064 | atomic_inc(&sh->count); | |
566c09c5 SL |
5065 | hash = sh->hash_lock_index; |
5066 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
5067 | } |
5068 | } | |
5069 | ||
5c675f83 | 5070 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 5071 | { |
d1688a6d | 5072 | struct r5conf *conf = mddev->private; |
f022b2fd N |
5073 | |
5074 | /* No difference between reads and writes. Just check | |
5075 | * how busy the stripe_cache is | |
5076 | */ | |
3fa841d7 | 5077 | |
5423399a | 5078 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd | 5079 | return 1; |
a39f7afd SL |
5080 | |
5081 | /* Also checks whether there is pressure on r5cache log space */ | |
5082 | if (test_bit(R5C_LOG_TIGHT, &conf->cache_state)) | |
5083 | return 1; | |
f022b2fd N |
5084 | if (conf->quiesce) |
5085 | return 1; | |
4bda556a | 5086 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
5087 | return 1; |
5088 | ||
5089 | return 0; | |
5090 | } | |
5091 | ||
fd01b88c | 5092 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 5093 | { |
3cb5edf4 | 5094 | struct r5conf *conf = mddev->private; |
10433d04 | 5095 | sector_t sector = bio->bi_iter.bi_sector; |
3cb5edf4 | 5096 | unsigned int chunk_sectors; |
aa8b57aa | 5097 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 5098 | |
10433d04 CH |
5099 | WARN_ON_ONCE(bio->bi_partno); |
5100 | ||
3cb5edf4 | 5101 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
5102 | return chunk_sectors >= |
5103 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
5104 | } | |
5105 | ||
46031f9a RBJ |
5106 | /* |
5107 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
5108 | * later sampled by raid5d. | |
5109 | */ | |
d1688a6d | 5110 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
5111 | { |
5112 | unsigned long flags; | |
5113 | ||
5114 | spin_lock_irqsave(&conf->device_lock, flags); | |
5115 | ||
5116 | bi->bi_next = conf->retry_read_aligned_list; | |
5117 | conf->retry_read_aligned_list = bi; | |
5118 | ||
5119 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5120 | md_wakeup_thread(conf->mddev->thread); | |
5121 | } | |
5122 | ||
0472a42b N |
5123 | static struct bio *remove_bio_from_retry(struct r5conf *conf, |
5124 | unsigned int *offset) | |
46031f9a RBJ |
5125 | { |
5126 | struct bio *bi; | |
5127 | ||
5128 | bi = conf->retry_read_aligned; | |
5129 | if (bi) { | |
0472a42b | 5130 | *offset = conf->retry_read_offset; |
46031f9a RBJ |
5131 | conf->retry_read_aligned = NULL; |
5132 | return bi; | |
5133 | } | |
5134 | bi = conf->retry_read_aligned_list; | |
5135 | if(bi) { | |
387bb173 | 5136 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5137 | bi->bi_next = NULL; |
0472a42b | 5138 | *offset = 0; |
46031f9a RBJ |
5139 | } |
5140 | ||
5141 | return bi; | |
5142 | } | |
5143 | ||
f679623f RBJ |
5144 | /* |
5145 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5146 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5147 | * first). | |
5148 | * If the read failed.. | |
5149 | */ | |
4246a0b6 | 5150 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5151 | { |
5152 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5153 | struct mddev *mddev; |
d1688a6d | 5154 | struct r5conf *conf; |
3cb03002 | 5155 | struct md_rdev *rdev; |
4e4cbee9 | 5156 | blk_status_t error = bi->bi_status; |
46031f9a | 5157 | |
f679623f | 5158 | bio_put(bi); |
46031f9a | 5159 | |
46031f9a RBJ |
5160 | rdev = (void*)raid_bi->bi_next; |
5161 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5162 | mddev = rdev->mddev; |
5163 | conf = mddev->private; | |
46031f9a RBJ |
5164 | |
5165 | rdev_dec_pending(rdev, conf->mddev); | |
5166 | ||
9b81c842 | 5167 | if (!error) { |
4246a0b6 | 5168 | bio_endio(raid_bi); |
46031f9a | 5169 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5170 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5171 | return; |
46031f9a RBJ |
5172 | } |
5173 | ||
45b4233c | 5174 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5175 | |
5176 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5177 | } |
5178 | ||
7ef6b12a | 5179 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5180 | { |
d1688a6d | 5181 | struct r5conf *conf = mddev->private; |
8553fe7e | 5182 | int dd_idx; |
f679623f | 5183 | struct bio* align_bi; |
3cb03002 | 5184 | struct md_rdev *rdev; |
671488cc | 5185 | sector_t end_sector; |
f679623f RBJ |
5186 | |
5187 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5188 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5189 | return 0; |
5190 | } | |
5191 | /* | |
d7a10308 | 5192 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5193 | */ |
d7a10308 | 5194 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, mddev->bio_set); |
f679623f RBJ |
5195 | if (!align_bi) |
5196 | return 0; | |
5197 | /* | |
5198 | * set bi_end_io to a new function, and set bi_private to the | |
5199 | * original bio. | |
5200 | */ | |
5201 | align_bi->bi_end_io = raid5_align_endio; | |
5202 | align_bi->bi_private = raid_bio; | |
5203 | /* | |
5204 | * compute position | |
5205 | */ | |
4f024f37 KO |
5206 | align_bi->bi_iter.bi_sector = |
5207 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5208 | 0, &dd_idx, NULL); | |
f679623f | 5209 | |
f73a1c7d | 5210 | end_sector = bio_end_sector(align_bi); |
f679623f | 5211 | rcu_read_lock(); |
671488cc N |
5212 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5213 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5214 | rdev->recovery_offset < end_sector) { | |
5215 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5216 | if (rdev && | |
5217 | (test_bit(Faulty, &rdev->flags) || | |
5218 | !(test_bit(In_sync, &rdev->flags) || | |
5219 | rdev->recovery_offset >= end_sector))) | |
5220 | rdev = NULL; | |
5221 | } | |
03b047f4 SL |
5222 | |
5223 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5224 | rcu_read_unlock(); | |
5225 | bio_put(align_bi); | |
5226 | return 0; | |
5227 | } | |
5228 | ||
671488cc | 5229 | if (rdev) { |
31c176ec N |
5230 | sector_t first_bad; |
5231 | int bad_sectors; | |
5232 | ||
f679623f RBJ |
5233 | atomic_inc(&rdev->nr_pending); |
5234 | rcu_read_unlock(); | |
46031f9a | 5235 | raid_bio->bi_next = (void*)rdev; |
74d46992 | 5236 | bio_set_dev(align_bi, rdev->bdev); |
b7c44ed9 | 5237 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 5238 | |
7140aafc | 5239 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5240 | bio_sectors(align_bi), |
31c176ec | 5241 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5242 | bio_put(align_bi); |
5243 | rdev_dec_pending(rdev, mddev); | |
5244 | return 0; | |
5245 | } | |
5246 | ||
6c0544e2 | 5247 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5248 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5249 | |
46031f9a | 5250 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5251 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5252 | conf->quiesce == 0, |
eed8c02e | 5253 | conf->device_lock); |
46031f9a RBJ |
5254 | atomic_inc(&conf->active_aligned_reads); |
5255 | spin_unlock_irq(&conf->device_lock); | |
5256 | ||
e3620a3a | 5257 | if (mddev->gendisk) |
74d46992 | 5258 | trace_block_bio_remap(align_bi->bi_disk->queue, |
e3620a3a | 5259 | align_bi, disk_devt(mddev->gendisk), |
4f024f37 | 5260 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
5261 | generic_make_request(align_bi); |
5262 | return 1; | |
5263 | } else { | |
5264 | rcu_read_unlock(); | |
46031f9a | 5265 | bio_put(align_bi); |
f679623f RBJ |
5266 | return 0; |
5267 | } | |
5268 | } | |
5269 | ||
7ef6b12a ML |
5270 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5271 | { | |
5272 | struct bio *split; | |
dd7a8f5d N |
5273 | sector_t sector = raid_bio->bi_iter.bi_sector; |
5274 | unsigned chunk_sects = mddev->chunk_sectors; | |
5275 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
7ef6b12a | 5276 | |
dd7a8f5d N |
5277 | if (sectors < bio_sectors(raid_bio)) { |
5278 | struct r5conf *conf = mddev->private; | |
5279 | split = bio_split(raid_bio, sectors, GFP_NOIO, conf->bio_split); | |
5280 | bio_chain(split, raid_bio); | |
5281 | generic_make_request(raid_bio); | |
5282 | raid_bio = split; | |
5283 | } | |
7ef6b12a | 5284 | |
dd7a8f5d N |
5285 | if (!raid5_read_one_chunk(mddev, raid_bio)) |
5286 | return raid_bio; | |
7ef6b12a ML |
5287 | |
5288 | return NULL; | |
5289 | } | |
5290 | ||
8b3e6cdc DW |
5291 | /* __get_priority_stripe - get the next stripe to process |
5292 | * | |
5293 | * Full stripe writes are allowed to pass preread active stripes up until | |
5294 | * the bypass_threshold is exceeded. In general the bypass_count | |
5295 | * increments when the handle_list is handled before the hold_list; however, it | |
5296 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5297 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5298 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5299 | * handle_list. | |
5300 | */ | |
851c30c9 | 5301 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5302 | { |
535ae4eb | 5303 | struct stripe_head *sh, *tmp; |
851c30c9 | 5304 | struct list_head *handle_list = NULL; |
535ae4eb | 5305 | struct r5worker_group *wg; |
70d466f7 SL |
5306 | bool second_try = !r5c_is_writeback(conf->log) && |
5307 | !r5l_log_disk_error(conf); | |
5308 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) || | |
5309 | r5l_log_disk_error(conf); | |
851c30c9 | 5310 | |
535ae4eb SL |
5311 | again: |
5312 | wg = NULL; | |
5313 | sh = NULL; | |
851c30c9 | 5314 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5315 | handle_list = try_loprio ? &conf->loprio_list : |
5316 | &conf->handle_list; | |
851c30c9 | 5317 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5318 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5319 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5320 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5321 | } else { |
5322 | int i; | |
5323 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5324 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5325 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5326 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5327 | if (!list_empty(handle_list)) |
5328 | break; | |
5329 | } | |
5330 | } | |
8b3e6cdc DW |
5331 | |
5332 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5333 | __func__, | |
851c30c9 | 5334 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5335 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5336 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5337 | ||
851c30c9 SL |
5338 | if (!list_empty(handle_list)) { |
5339 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5340 | |
5341 | if (list_empty(&conf->hold_list)) | |
5342 | conf->bypass_count = 0; | |
5343 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5344 | if (conf->hold_list.next == conf->last_hold) | |
5345 | conf->bypass_count++; | |
5346 | else { | |
5347 | conf->last_hold = conf->hold_list.next; | |
5348 | conf->bypass_count -= conf->bypass_threshold; | |
5349 | if (conf->bypass_count < 0) | |
5350 | conf->bypass_count = 0; | |
5351 | } | |
5352 | } | |
5353 | } else if (!list_empty(&conf->hold_list) && | |
5354 | ((conf->bypass_threshold && | |
5355 | conf->bypass_count > conf->bypass_threshold) || | |
5356 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5357 | |
5358 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5359 | if (conf->worker_cnt_per_group == 0 || | |
5360 | group == ANY_GROUP || | |
5361 | !cpu_online(tmp->cpu) || | |
5362 | cpu_to_group(tmp->cpu) == group) { | |
5363 | sh = tmp; | |
5364 | break; | |
5365 | } | |
5366 | } | |
5367 | ||
5368 | if (sh) { | |
5369 | conf->bypass_count -= conf->bypass_threshold; | |
5370 | if (conf->bypass_count < 0) | |
5371 | conf->bypass_count = 0; | |
5372 | } | |
bfc90cb0 | 5373 | wg = NULL; |
851c30c9 SL |
5374 | } |
5375 | ||
535ae4eb SL |
5376 | if (!sh) { |
5377 | if (second_try) | |
5378 | return NULL; | |
5379 | second_try = true; | |
5380 | try_loprio = !try_loprio; | |
5381 | goto again; | |
5382 | } | |
8b3e6cdc | 5383 | |
bfc90cb0 SL |
5384 | if (wg) { |
5385 | wg->stripes_cnt--; | |
5386 | sh->group = NULL; | |
5387 | } | |
8b3e6cdc | 5388 | list_del_init(&sh->lru); |
c7a6d35e | 5389 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5390 | return sh; |
5391 | } | |
f679623f | 5392 | |
8811b596 SL |
5393 | struct raid5_plug_cb { |
5394 | struct blk_plug_cb cb; | |
5395 | struct list_head list; | |
566c09c5 | 5396 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5397 | }; |
5398 | ||
5399 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5400 | { | |
5401 | struct raid5_plug_cb *cb = container_of( | |
5402 | blk_cb, struct raid5_plug_cb, cb); | |
5403 | struct stripe_head *sh; | |
5404 | struct mddev *mddev = cb->cb.data; | |
5405 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5406 | int cnt = 0; |
566c09c5 | 5407 | int hash; |
8811b596 SL |
5408 | |
5409 | if (cb->list.next && !list_empty(&cb->list)) { | |
5410 | spin_lock_irq(&conf->device_lock); | |
5411 | while (!list_empty(&cb->list)) { | |
5412 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5413 | list_del_init(&sh->lru); | |
5414 | /* | |
5415 | * avoid race release_stripe_plug() sees | |
5416 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5417 | * is still in our list | |
5418 | */ | |
4e857c58 | 5419 | smp_mb__before_atomic(); |
8811b596 | 5420 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5421 | /* |
5422 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5423 | * case, the count is always > 1 here | |
5424 | */ | |
566c09c5 SL |
5425 | hash = sh->hash_lock_index; |
5426 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5427 | cnt++; |
8811b596 SL |
5428 | } |
5429 | spin_unlock_irq(&conf->device_lock); | |
5430 | } | |
566c09c5 SL |
5431 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5432 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5433 | if (mddev->queue) |
5434 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5435 | kfree(cb); |
5436 | } | |
5437 | ||
5438 | static void release_stripe_plug(struct mddev *mddev, | |
5439 | struct stripe_head *sh) | |
5440 | { | |
5441 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5442 | raid5_unplug, mddev, | |
5443 | sizeof(struct raid5_plug_cb)); | |
5444 | struct raid5_plug_cb *cb; | |
5445 | ||
5446 | if (!blk_cb) { | |
6d036f7d | 5447 | raid5_release_stripe(sh); |
8811b596 SL |
5448 | return; |
5449 | } | |
5450 | ||
5451 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5452 | ||
566c09c5 SL |
5453 | if (cb->list.next == NULL) { |
5454 | int i; | |
8811b596 | 5455 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5456 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5457 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5458 | } | |
8811b596 SL |
5459 | |
5460 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5461 | list_add_tail(&sh->lru, &cb->list); | |
5462 | else | |
6d036f7d | 5463 | raid5_release_stripe(sh); |
8811b596 SL |
5464 | } |
5465 | ||
620125f2 SL |
5466 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5467 | { | |
5468 | struct r5conf *conf = mddev->private; | |
5469 | sector_t logical_sector, last_sector; | |
5470 | struct stripe_head *sh; | |
620125f2 SL |
5471 | int stripe_sectors; |
5472 | ||
5473 | if (mddev->reshape_position != MaxSector) | |
5474 | /* Skip discard while reshape is happening */ | |
5475 | return; | |
5476 | ||
4f024f37 KO |
5477 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5478 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5479 | |
5480 | bi->bi_next = NULL; | |
620125f2 SL |
5481 | |
5482 | stripe_sectors = conf->chunk_sectors * | |
5483 | (conf->raid_disks - conf->max_degraded); | |
5484 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5485 | stripe_sectors); | |
5486 | sector_div(last_sector, stripe_sectors); | |
5487 | ||
5488 | logical_sector *= conf->chunk_sectors; | |
5489 | last_sector *= conf->chunk_sectors; | |
5490 | ||
5491 | for (; logical_sector < last_sector; | |
5492 | logical_sector += STRIPE_SECTORS) { | |
5493 | DEFINE_WAIT(w); | |
5494 | int d; | |
5495 | again: | |
6d036f7d | 5496 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5497 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5498 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5499 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5500 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5501 | raid5_release_stripe(sh); |
f8dfcffd N |
5502 | schedule(); |
5503 | goto again; | |
5504 | } | |
5505 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5506 | spin_lock_irq(&sh->stripe_lock); |
5507 | for (d = 0; d < conf->raid_disks; d++) { | |
5508 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5509 | continue; | |
5510 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5511 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5512 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5513 | raid5_release_stripe(sh); |
620125f2 SL |
5514 | schedule(); |
5515 | goto again; | |
5516 | } | |
5517 | } | |
f8dfcffd | 5518 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5519 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5520 | sh->overwrite_disks = 0; |
620125f2 SL |
5521 | for (d = 0; d < conf->raid_disks; d++) { |
5522 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5523 | continue; | |
5524 | sh->dev[d].towrite = bi; | |
5525 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
016c76ac | 5526 | bio_inc_remaining(bi); |
49728050 | 5527 | md_write_inc(mddev, bi); |
7a87f434 | 5528 | sh->overwrite_disks++; |
620125f2 SL |
5529 | } |
5530 | spin_unlock_irq(&sh->stripe_lock); | |
5531 | if (conf->mddev->bitmap) { | |
5532 | for (d = 0; | |
5533 | d < conf->raid_disks - conf->max_degraded; | |
5534 | d++) | |
5535 | bitmap_startwrite(mddev->bitmap, | |
5536 | sh->sector, | |
5537 | STRIPE_SECTORS, | |
5538 | 0); | |
5539 | sh->bm_seq = conf->seq_flush + 1; | |
5540 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5541 | } | |
5542 | ||
5543 | set_bit(STRIPE_HANDLE, &sh->state); | |
5544 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5545 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5546 | atomic_inc(&conf->preread_active_stripes); | |
5547 | release_stripe_plug(mddev, sh); | |
5548 | } | |
5549 | ||
016c76ac | 5550 | bio_endio(bi); |
620125f2 SL |
5551 | } |
5552 | ||
cc27b0c7 | 5553 | static bool raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5554 | { |
d1688a6d | 5555 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5556 | int dd_idx; |
1da177e4 LT |
5557 | sector_t new_sector; |
5558 | sector_t logical_sector, last_sector; | |
5559 | struct stripe_head *sh; | |
a362357b | 5560 | const int rw = bio_data_dir(bi); |
27c0f68f SL |
5561 | DEFINE_WAIT(w); |
5562 | bool do_prepare; | |
3bddb7f8 | 5563 | bool do_flush = false; |
1da177e4 | 5564 | |
1eff9d32 | 5565 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
828cbe98 SL |
5566 | int ret = r5l_handle_flush_request(conf->log, bi); |
5567 | ||
5568 | if (ret == 0) | |
cc27b0c7 | 5569 | return true; |
828cbe98 SL |
5570 | if (ret == -ENODEV) { |
5571 | md_flush_request(mddev, bi); | |
cc27b0c7 | 5572 | return true; |
828cbe98 SL |
5573 | } |
5574 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5575 | /* |
5576 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5577 | * we need to flush journal device | |
5578 | */ | |
5579 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5580 | } |
5581 | ||
cc27b0c7 N |
5582 | if (!md_write_start(mddev, bi)) |
5583 | return false; | |
9ffc8f7c EM |
5584 | /* |
5585 | * If array is degraded, better not do chunk aligned read because | |
5586 | * later we might have to read it again in order to reconstruct | |
5587 | * data on failed drives. | |
5588 | */ | |
5589 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5590 | mddev->reshape_position == MaxSector) { |
5591 | bi = chunk_aligned_read(mddev, bi); | |
5592 | if (!bi) | |
cc27b0c7 | 5593 | return true; |
7ef6b12a | 5594 | } |
52488615 | 5595 | |
796a5cf0 | 5596 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 | 5597 | make_discard_request(mddev, bi); |
cc27b0c7 N |
5598 | md_write_end(mddev); |
5599 | return true; | |
620125f2 SL |
5600 | } |
5601 | ||
4f024f37 | 5602 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5603 | last_sector = bio_end_sector(bi); |
1da177e4 | 5604 | bi->bi_next = NULL; |
06d91a5f | 5605 | |
27c0f68f | 5606 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5607 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5608 | int previous; |
c46501b2 | 5609 | int seq; |
b578d55f | 5610 | |
27c0f68f | 5611 | do_prepare = false; |
7ecaa1e6 | 5612 | retry: |
c46501b2 | 5613 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5614 | previous = 0; |
27c0f68f SL |
5615 | if (do_prepare) |
5616 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5617 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5618 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5619 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5620 | * 64bit on a 32bit platform, and so it might be |
5621 | * possible to see a half-updated value | |
aeb878b0 | 5622 | * Of course reshape_progress could change after |
df8e7f76 N |
5623 | * the lock is dropped, so once we get a reference |
5624 | * to the stripe that we think it is, we will have | |
5625 | * to check again. | |
5626 | */ | |
7ecaa1e6 | 5627 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5628 | if (mddev->reshape_backwards |
fef9c61f N |
5629 | ? logical_sector < conf->reshape_progress |
5630 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5631 | previous = 1; |
5632 | } else { | |
2c810cdd | 5633 | if (mddev->reshape_backwards |
fef9c61f N |
5634 | ? logical_sector < conf->reshape_safe |
5635 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5636 | spin_unlock_irq(&conf->device_lock); |
5637 | schedule(); | |
27c0f68f | 5638 | do_prepare = true; |
b578d55f N |
5639 | goto retry; |
5640 | } | |
5641 | } | |
7ecaa1e6 N |
5642 | spin_unlock_irq(&conf->device_lock); |
5643 | } | |
16a53ecc | 5644 | |
112bf897 N |
5645 | new_sector = raid5_compute_sector(conf, logical_sector, |
5646 | previous, | |
911d4ee8 | 5647 | &dd_idx, NULL); |
849674e4 | 5648 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5649 | (unsigned long long)new_sector, |
1da177e4 LT |
5650 | (unsigned long long)logical_sector); |
5651 | ||
6d036f7d | 5652 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5653 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5654 | if (sh) { |
b0f9ec04 | 5655 | if (unlikely(previous)) { |
7ecaa1e6 | 5656 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5657 | * stripe, so we must do the range check again. |
5658 | * Expansion could still move past after this | |
5659 | * test, but as we are holding a reference to | |
5660 | * 'sh', we know that if that happens, | |
5661 | * STRIPE_EXPANDING will get set and the expansion | |
5662 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5663 | */ |
5664 | int must_retry = 0; | |
5665 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5666 | if (mddev->reshape_backwards |
b0f9ec04 N |
5667 | ? logical_sector >= conf->reshape_progress |
5668 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5669 | /* mismatch, need to try again */ |
5670 | must_retry = 1; | |
5671 | spin_unlock_irq(&conf->device_lock); | |
5672 | if (must_retry) { | |
6d036f7d | 5673 | raid5_release_stripe(sh); |
7a3ab908 | 5674 | schedule(); |
27c0f68f | 5675 | do_prepare = true; |
7ecaa1e6 N |
5676 | goto retry; |
5677 | } | |
5678 | } | |
c46501b2 N |
5679 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5680 | /* Might have got the wrong stripe_head | |
5681 | * by accident | |
5682 | */ | |
6d036f7d | 5683 | raid5_release_stripe(sh); |
c46501b2 N |
5684 | goto retry; |
5685 | } | |
e62e58a5 | 5686 | |
7ecaa1e6 | 5687 | if (test_bit(STRIPE_EXPANDING, &sh->state) || |
da41ba65 | 5688 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5689 | /* Stripe is busy expanding or |
5690 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5691 | * and wait a while |
5692 | */ | |
482c0834 | 5693 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5694 | raid5_release_stripe(sh); |
1da177e4 | 5695 | schedule(); |
27c0f68f | 5696 | do_prepare = true; |
1da177e4 LT |
5697 | goto retry; |
5698 | } | |
3bddb7f8 SL |
5699 | if (do_flush) { |
5700 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5701 | /* we only need flush for one stripe */ | |
5702 | do_flush = false; | |
5703 | } | |
5704 | ||
6ed3003c N |
5705 | set_bit(STRIPE_HANDLE, &sh->state); |
5706 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5707 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5708 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5709 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5710 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5711 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5712 | } else { |
5713 | /* cannot get stripe for read-ahead, just give-up */ | |
4e4cbee9 | 5714 | bi->bi_status = BLK_STS_IOERR; |
1da177e4 LT |
5715 | break; |
5716 | } | |
1da177e4 | 5717 | } |
27c0f68f | 5718 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5719 | |
49728050 N |
5720 | if (rw == WRITE) |
5721 | md_write_end(mddev); | |
016c76ac | 5722 | bio_endio(bi); |
cc27b0c7 | 5723 | return true; |
1da177e4 LT |
5724 | } |
5725 | ||
fd01b88c | 5726 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5727 | |
fd01b88c | 5728 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5729 | { |
52c03291 N |
5730 | /* reshaping is quite different to recovery/resync so it is |
5731 | * handled quite separately ... here. | |
5732 | * | |
5733 | * On each call to sync_request, we gather one chunk worth of | |
5734 | * destination stripes and flag them as expanding. | |
5735 | * Then we find all the source stripes and request reads. | |
5736 | * As the reads complete, handle_stripe will copy the data | |
5737 | * into the destination stripe and release that stripe. | |
5738 | */ | |
d1688a6d | 5739 | struct r5conf *conf = mddev->private; |
1da177e4 | 5740 | struct stripe_head *sh; |
ccfcc3c1 | 5741 | sector_t first_sector, last_sector; |
f416885e N |
5742 | int raid_disks = conf->previous_raid_disks; |
5743 | int data_disks = raid_disks - conf->max_degraded; | |
5744 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5745 | int i; |
5746 | int dd_idx; | |
c8f517c4 | 5747 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5748 | sector_t stripe_addr; |
7a661381 | 5749 | int reshape_sectors; |
ab69ae12 | 5750 | struct list_head stripes; |
92140480 | 5751 | sector_t retn; |
52c03291 | 5752 | |
fef9c61f N |
5753 | if (sector_nr == 0) { |
5754 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5755 | if (mddev->reshape_backwards && |
fef9c61f N |
5756 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5757 | sector_nr = raid5_size(mddev, 0, 0) | |
5758 | - conf->reshape_progress; | |
6cbd8148 N |
5759 | } else if (mddev->reshape_backwards && |
5760 | conf->reshape_progress == MaxSector) { | |
5761 | /* shouldn't happen, but just in case, finish up.*/ | |
5762 | sector_nr = MaxSector; | |
2c810cdd | 5763 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5764 | conf->reshape_progress > 0) |
5765 | sector_nr = conf->reshape_progress; | |
f416885e | 5766 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5767 | if (sector_nr) { |
8dee7211 N |
5768 | mddev->curr_resync_completed = sector_nr; |
5769 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5770 | *skipped = 1; |
92140480 N |
5771 | retn = sector_nr; |
5772 | goto finish; | |
fef9c61f | 5773 | } |
52c03291 N |
5774 | } |
5775 | ||
7a661381 N |
5776 | /* We need to process a full chunk at a time. |
5777 | * If old and new chunk sizes differ, we need to process the | |
5778 | * largest of these | |
5779 | */ | |
3cb5edf4 N |
5780 | |
5781 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5782 | |
b5254dd5 N |
5783 | /* We update the metadata at least every 10 seconds, or when |
5784 | * the data about to be copied would over-write the source of | |
5785 | * the data at the front of the range. i.e. one new_stripe | |
5786 | * along from reshape_progress new_maps to after where | |
5787 | * reshape_safe old_maps to | |
52c03291 | 5788 | */ |
fef9c61f | 5789 | writepos = conf->reshape_progress; |
f416885e | 5790 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5791 | readpos = conf->reshape_progress; |
5792 | sector_div(readpos, data_disks); | |
fef9c61f | 5793 | safepos = conf->reshape_safe; |
f416885e | 5794 | sector_div(safepos, data_disks); |
2c810cdd | 5795 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5796 | BUG_ON(writepos < reshape_sectors); |
5797 | writepos -= reshape_sectors; | |
c8f517c4 | 5798 | readpos += reshape_sectors; |
7a661381 | 5799 | safepos += reshape_sectors; |
fef9c61f | 5800 | } else { |
7a661381 | 5801 | writepos += reshape_sectors; |
c74c0d76 N |
5802 | /* readpos and safepos are worst-case calculations. |
5803 | * A negative number is overly pessimistic, and causes | |
5804 | * obvious problems for unsigned storage. So clip to 0. | |
5805 | */ | |
ed37d83e N |
5806 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5807 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5808 | } |
52c03291 | 5809 | |
b5254dd5 N |
5810 | /* Having calculated the 'writepos' possibly use it |
5811 | * to set 'stripe_addr' which is where we will write to. | |
5812 | */ | |
5813 | if (mddev->reshape_backwards) { | |
5814 | BUG_ON(conf->reshape_progress == 0); | |
5815 | stripe_addr = writepos; | |
5816 | BUG_ON((mddev->dev_sectors & | |
5817 | ~((sector_t)reshape_sectors - 1)) | |
5818 | - reshape_sectors - stripe_addr | |
5819 | != sector_nr); | |
5820 | } else { | |
5821 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5822 | stripe_addr = sector_nr; | |
5823 | } | |
5824 | ||
c8f517c4 N |
5825 | /* 'writepos' is the most advanced device address we might write. |
5826 | * 'readpos' is the least advanced device address we might read. | |
5827 | * 'safepos' is the least address recorded in the metadata as having | |
5828 | * been reshaped. | |
b5254dd5 N |
5829 | * If there is a min_offset_diff, these are adjusted either by |
5830 | * increasing the safepos/readpos if diff is negative, or | |
5831 | * increasing writepos if diff is positive. | |
5832 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5833 | * ensure safety in the face of a crash - that must be done by userspace |
5834 | * making a backup of the data. So in that case there is no particular | |
5835 | * rush to update metadata. | |
5836 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5837 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5838 | * we can be safe in the event of a crash. | |
5839 | * So we insist on updating metadata if safepos is behind writepos and | |
5840 | * readpos is beyond writepos. | |
5841 | * In any case, update the metadata every 10 seconds. | |
5842 | * Maybe that number should be configurable, but I'm not sure it is | |
5843 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5844 | */ | |
b5254dd5 N |
5845 | if (conf->min_offset_diff < 0) { |
5846 | safepos += -conf->min_offset_diff; | |
5847 | readpos += -conf->min_offset_diff; | |
5848 | } else | |
5849 | writepos += conf->min_offset_diff; | |
5850 | ||
2c810cdd | 5851 | if ((mddev->reshape_backwards |
c8f517c4 N |
5852 | ? (safepos > writepos && readpos < writepos) |
5853 | : (safepos < writepos && readpos > writepos)) || | |
5854 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5855 | /* Cannot proceed until we've updated the superblock... */ |
5856 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5857 | atomic_read(&conf->reshape_stripes)==0 |
5858 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5859 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5860 | return 0; | |
fef9c61f | 5861 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5862 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5863 | conf->reshape_checkpoint = jiffies; |
2953079c | 5864 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5865 | md_wakeup_thread(mddev->thread); |
2953079c | 5866 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5867 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5868 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5869 | return 0; | |
52c03291 | 5870 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5871 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5872 | spin_unlock_irq(&conf->device_lock); |
5873 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5874 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5875 | } |
5876 | ||
ab69ae12 | 5877 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5878 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5879 | int j; |
a9f326eb | 5880 | int skipped_disk = 0; |
6d036f7d | 5881 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5882 | set_bit(STRIPE_EXPANDING, &sh->state); |
5883 | atomic_inc(&conf->reshape_stripes); | |
5884 | /* If any of this stripe is beyond the end of the old | |
5885 | * array, then we need to zero those blocks | |
5886 | */ | |
5887 | for (j=sh->disks; j--;) { | |
5888 | sector_t s; | |
5889 | if (j == sh->pd_idx) | |
5890 | continue; | |
f416885e | 5891 | if (conf->level == 6 && |
d0dabf7e | 5892 | j == sh->qd_idx) |
f416885e | 5893 | continue; |
6d036f7d | 5894 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5895 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5896 | skipped_disk = 1; |
52c03291 N |
5897 | continue; |
5898 | } | |
5899 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5900 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5901 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5902 | } | |
a9f326eb | 5903 | if (!skipped_disk) { |
52c03291 N |
5904 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5905 | set_bit(STRIPE_HANDLE, &sh->state); | |
5906 | } | |
ab69ae12 | 5907 | list_add(&sh->lru, &stripes); |
52c03291 N |
5908 | } |
5909 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5910 | if (mddev->reshape_backwards) |
7a661381 | 5911 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5912 | else |
7a661381 | 5913 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5914 | spin_unlock_irq(&conf->device_lock); |
5915 | /* Ok, those stripe are ready. We can start scheduling | |
5916 | * reads on the source stripes. | |
5917 | * The source stripes are determined by mapping the first and last | |
5918 | * block on the destination stripes. | |
5919 | */ | |
52c03291 | 5920 | first_sector = |
ec32a2bd | 5921 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5922 | 1, &dd_idx, NULL); |
52c03291 | 5923 | last_sector = |
0e6e0271 | 5924 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5925 | * new_data_disks - 1), |
911d4ee8 | 5926 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5927 | if (last_sector >= mddev->dev_sectors) |
5928 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5929 | while (first_sector <= last_sector) { |
6d036f7d | 5930 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5931 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5932 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5933 | raid5_release_stripe(sh); |
52c03291 N |
5934 | first_sector += STRIPE_SECTORS; |
5935 | } | |
ab69ae12 N |
5936 | /* Now that the sources are clearly marked, we can release |
5937 | * the destination stripes | |
5938 | */ | |
5939 | while (!list_empty(&stripes)) { | |
5940 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5941 | list_del_init(&sh->lru); | |
6d036f7d | 5942 | raid5_release_stripe(sh); |
ab69ae12 | 5943 | } |
c6207277 N |
5944 | /* If this takes us to the resync_max point where we have to pause, |
5945 | * then we need to write out the superblock. | |
5946 | */ | |
7a661381 | 5947 | sector_nr += reshape_sectors; |
92140480 N |
5948 | retn = reshape_sectors; |
5949 | finish: | |
c5e19d90 N |
5950 | if (mddev->curr_resync_completed > mddev->resync_max || |
5951 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5952 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5953 | /* Cannot proceed until we've updated the superblock... */ |
5954 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5955 | atomic_read(&conf->reshape_stripes) == 0 |
5956 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5957 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5958 | goto ret; | |
fef9c61f | 5959 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5960 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5961 | conf->reshape_checkpoint = jiffies; |
2953079c | 5962 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
5963 | md_wakeup_thread(mddev->thread); |
5964 | wait_event(mddev->sb_wait, | |
2953079c | 5965 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
5966 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5967 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5968 | goto ret; | |
c6207277 | 5969 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5970 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5971 | spin_unlock_irq(&conf->device_lock); |
5972 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5973 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5974 | } |
c91abf5a | 5975 | ret: |
92140480 | 5976 | return retn; |
52c03291 N |
5977 | } |
5978 | ||
849674e4 SL |
5979 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
5980 | int *skipped) | |
52c03291 | 5981 | { |
d1688a6d | 5982 | struct r5conf *conf = mddev->private; |
52c03291 | 5983 | struct stripe_head *sh; |
58c0fed4 | 5984 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5985 | sector_t sync_blocks; |
16a53ecc N |
5986 | int still_degraded = 0; |
5987 | int i; | |
1da177e4 | 5988 | |
72626685 | 5989 | if (sector_nr >= max_sector) { |
1da177e4 | 5990 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5991 | |
29269553 N |
5992 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5993 | end_reshape(conf); | |
5994 | return 0; | |
5995 | } | |
72626685 N |
5996 | |
5997 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5998 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5999 | &sync_blocks, 1); | |
16a53ecc | 6000 | else /* completed sync */ |
72626685 N |
6001 | conf->fullsync = 0; |
6002 | bitmap_close_sync(mddev->bitmap); | |
6003 | ||
1da177e4 LT |
6004 | return 0; |
6005 | } | |
ccfcc3c1 | 6006 | |
64bd660b N |
6007 | /* Allow raid5_quiesce to complete */ |
6008 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
6009 | ||
52c03291 N |
6010 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
6011 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 6012 | |
c6207277 N |
6013 | /* No need to check resync_max as we never do more than one |
6014 | * stripe, and as resync_max will always be on a chunk boundary, | |
6015 | * if the check in md_do_sync didn't fire, there is no chance | |
6016 | * of overstepping resync_max here | |
6017 | */ | |
6018 | ||
16a53ecc | 6019 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
6020 | * to resync, then assert that we are finished, because there is |
6021 | * nothing we can do. | |
6022 | */ | |
3285edf1 | 6023 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 6024 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 6025 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 6026 | *skipped = 1; |
1da177e4 LT |
6027 | return rv; |
6028 | } | |
6f608040 | 6029 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
6030 | !conf->fullsync && | |
6031 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
6032 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
6033 | /* we can skip this block, and probably more */ |
6034 | sync_blocks /= STRIPE_SECTORS; | |
6035 | *skipped = 1; | |
6036 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
6037 | } | |
1da177e4 | 6038 | |
c40f341f | 6039 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 6040 | |
6d036f7d | 6041 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 6042 | if (sh == NULL) { |
6d036f7d | 6043 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 6044 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 6045 | * is trying to get access |
1da177e4 | 6046 | */ |
66c006a5 | 6047 | schedule_timeout_uninterruptible(1); |
1da177e4 | 6048 | } |
16a53ecc | 6049 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
6050 | * Note in case of > 1 drive failures it's possible we're rebuilding |
6051 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 6052 | */ |
16d9cfab EM |
6053 | rcu_read_lock(); |
6054 | for (i = 0; i < conf->raid_disks; i++) { | |
6055 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
6056 | ||
6057 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 6058 | still_degraded = 1; |
16d9cfab EM |
6059 | } |
6060 | rcu_read_unlock(); | |
16a53ecc N |
6061 | |
6062 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
6063 | ||
83206d66 | 6064 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 6065 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 6066 | |
6d036f7d | 6067 | raid5_release_stripe(sh); |
1da177e4 LT |
6068 | |
6069 | return STRIPE_SECTORS; | |
6070 | } | |
6071 | ||
0472a42b N |
6072 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio, |
6073 | unsigned int offset) | |
46031f9a RBJ |
6074 | { |
6075 | /* We may not be able to submit a whole bio at once as there | |
6076 | * may not be enough stripe_heads available. | |
6077 | * We cannot pre-allocate enough stripe_heads as we may need | |
6078 | * more than exist in the cache (if we allow ever large chunks). | |
6079 | * So we do one stripe head at a time and record in | |
6080 | * ->bi_hw_segments how many have been done. | |
6081 | * | |
6082 | * We *know* that this entire raid_bio is in one chunk, so | |
6083 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
6084 | */ | |
6085 | struct stripe_head *sh; | |
911d4ee8 | 6086 | int dd_idx; |
46031f9a RBJ |
6087 | sector_t sector, logical_sector, last_sector; |
6088 | int scnt = 0; | |
46031f9a RBJ |
6089 | int handled = 0; |
6090 | ||
4f024f37 KO |
6091 | logical_sector = raid_bio->bi_iter.bi_sector & |
6092 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 6093 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6094 | 0, &dd_idx, NULL); |
f73a1c7d | 6095 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6096 | |
6097 | for (; logical_sector < last_sector; | |
387bb173 NB |
6098 | logical_sector += STRIPE_SECTORS, |
6099 | sector += STRIPE_SECTORS, | |
6100 | scnt++) { | |
46031f9a | 6101 | |
0472a42b | 6102 | if (scnt < offset) |
46031f9a RBJ |
6103 | /* already done this stripe */ |
6104 | continue; | |
6105 | ||
6d036f7d | 6106 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6107 | |
6108 | if (!sh) { | |
6109 | /* failed to get a stripe - must wait */ | |
46031f9a | 6110 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6111 | conf->retry_read_offset = scnt; |
46031f9a RBJ |
6112 | return handled; |
6113 | } | |
6114 | ||
da41ba65 | 6115 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6116 | raid5_release_stripe(sh); |
387bb173 | 6117 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6118 | conf->retry_read_offset = scnt; |
387bb173 NB |
6119 | return handled; |
6120 | } | |
6121 | ||
3f9e7c14 | 6122 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6123 | handle_stripe(sh); |
6d036f7d | 6124 | raid5_release_stripe(sh); |
46031f9a RBJ |
6125 | handled++; |
6126 | } | |
016c76ac N |
6127 | |
6128 | bio_endio(raid_bio); | |
6129 | ||
46031f9a | 6130 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6131 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6132 | return handled; |
6133 | } | |
6134 | ||
bfc90cb0 | 6135 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6136 | struct r5worker *worker, |
6137 | struct list_head *temp_inactive_list) | |
46a06401 SL |
6138 | { |
6139 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6140 | int i, batch_size = 0, hash; |
6141 | bool release_inactive = false; | |
46a06401 SL |
6142 | |
6143 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6144 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6145 | batch[batch_size++] = sh; |
6146 | ||
566c09c5 SL |
6147 | if (batch_size == 0) { |
6148 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6149 | if (!list_empty(temp_inactive_list + i)) | |
6150 | break; | |
a8c34f91 SL |
6151 | if (i == NR_STRIPE_HASH_LOCKS) { |
6152 | spin_unlock_irq(&conf->device_lock); | |
6153 | r5l_flush_stripe_to_raid(conf->log); | |
6154 | spin_lock_irq(&conf->device_lock); | |
566c09c5 | 6155 | return batch_size; |
a8c34f91 | 6156 | } |
566c09c5 SL |
6157 | release_inactive = true; |
6158 | } | |
46a06401 SL |
6159 | spin_unlock_irq(&conf->device_lock); |
6160 | ||
566c09c5 SL |
6161 | release_inactive_stripe_list(conf, temp_inactive_list, |
6162 | NR_STRIPE_HASH_LOCKS); | |
6163 | ||
a8c34f91 | 6164 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6165 | if (release_inactive) { |
6166 | spin_lock_irq(&conf->device_lock); | |
6167 | return 0; | |
6168 | } | |
6169 | ||
46a06401 SL |
6170 | for (i = 0; i < batch_size; i++) |
6171 | handle_stripe(batch[i]); | |
ff875738 | 6172 | log_write_stripe_run(conf); |
46a06401 SL |
6173 | |
6174 | cond_resched(); | |
6175 | ||
6176 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6177 | for (i = 0; i < batch_size; i++) { |
6178 | hash = batch[i]->hash_lock_index; | |
6179 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6180 | } | |
46a06401 SL |
6181 | return batch_size; |
6182 | } | |
46031f9a | 6183 | |
851c30c9 SL |
6184 | static void raid5_do_work(struct work_struct *work) |
6185 | { | |
6186 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6187 | struct r5worker_group *group = worker->group; | |
6188 | struct r5conf *conf = group->conf; | |
16d997b7 | 6189 | struct mddev *mddev = conf->mddev; |
851c30c9 SL |
6190 | int group_id = group - conf->worker_groups; |
6191 | int handled; | |
6192 | struct blk_plug plug; | |
6193 | ||
6194 | pr_debug("+++ raid5worker active\n"); | |
6195 | ||
6196 | blk_start_plug(&plug); | |
6197 | handled = 0; | |
6198 | spin_lock_irq(&conf->device_lock); | |
6199 | while (1) { | |
6200 | int batch_size, released; | |
6201 | ||
566c09c5 | 6202 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6203 | |
566c09c5 SL |
6204 | batch_size = handle_active_stripes(conf, group_id, worker, |
6205 | worker->temp_inactive_list); | |
bfc90cb0 | 6206 | worker->working = false; |
851c30c9 SL |
6207 | if (!batch_size && !released) |
6208 | break; | |
6209 | handled += batch_size; | |
16d997b7 N |
6210 | wait_event_lock_irq(mddev->sb_wait, |
6211 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags), | |
6212 | conf->device_lock); | |
851c30c9 SL |
6213 | } |
6214 | pr_debug("%d stripes handled\n", handled); | |
6215 | ||
6216 | spin_unlock_irq(&conf->device_lock); | |
7e96d559 | 6217 | |
9c72a18e SL |
6218 | flush_deferred_bios(conf); |
6219 | ||
6220 | r5l_flush_stripe_to_raid(conf->log); | |
6221 | ||
7e96d559 | 6222 | async_tx_issue_pending_all(); |
851c30c9 SL |
6223 | blk_finish_plug(&plug); |
6224 | ||
6225 | pr_debug("--- raid5worker inactive\n"); | |
6226 | } | |
6227 | ||
1da177e4 LT |
6228 | /* |
6229 | * This is our raid5 kernel thread. | |
6230 | * | |
6231 | * We scan the hash table for stripes which can be handled now. | |
6232 | * During the scan, completed stripes are saved for us by the interrupt | |
6233 | * handler, so that they will not have to wait for our next wakeup. | |
6234 | */ | |
4ed8731d | 6235 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6236 | { |
4ed8731d | 6237 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6238 | struct r5conf *conf = mddev->private; |
1da177e4 | 6239 | int handled; |
e1dfa0a2 | 6240 | struct blk_plug plug; |
1da177e4 | 6241 | |
45b4233c | 6242 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6243 | |
6244 | md_check_recovery(mddev); | |
1da177e4 | 6245 | |
e1dfa0a2 | 6246 | blk_start_plug(&plug); |
1da177e4 LT |
6247 | handled = 0; |
6248 | spin_lock_irq(&conf->device_lock); | |
6249 | while (1) { | |
46031f9a | 6250 | struct bio *bio; |
773ca82f | 6251 | int batch_size, released; |
0472a42b | 6252 | unsigned int offset; |
773ca82f | 6253 | |
566c09c5 | 6254 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6255 | if (released) |
6256 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6257 | |
0021b7bc | 6258 | if ( |
7c13edc8 N |
6259 | !list_empty(&conf->bitmap_list)) { |
6260 | /* Now is a good time to flush some bitmap updates */ | |
6261 | conf->seq_flush++; | |
700e432d | 6262 | spin_unlock_irq(&conf->device_lock); |
72626685 | 6263 | bitmap_unplug(mddev->bitmap); |
700e432d | 6264 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6265 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6266 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6267 | } |
0021b7bc | 6268 | raid5_activate_delayed(conf); |
72626685 | 6269 | |
0472a42b | 6270 | while ((bio = remove_bio_from_retry(conf, &offset))) { |
46031f9a RBJ |
6271 | int ok; |
6272 | spin_unlock_irq(&conf->device_lock); | |
0472a42b | 6273 | ok = retry_aligned_read(conf, bio, offset); |
46031f9a RBJ |
6274 | spin_lock_irq(&conf->device_lock); |
6275 | if (!ok) | |
6276 | break; | |
6277 | handled++; | |
6278 | } | |
6279 | ||
566c09c5 SL |
6280 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6281 | conf->temp_inactive_list); | |
773ca82f | 6282 | if (!batch_size && !released) |
1da177e4 | 6283 | break; |
46a06401 | 6284 | handled += batch_size; |
1da177e4 | 6285 | |
2953079c | 6286 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6287 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6288 | md_check_recovery(mddev); |
46a06401 SL |
6289 | spin_lock_irq(&conf->device_lock); |
6290 | } | |
1da177e4 | 6291 | } |
45b4233c | 6292 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6293 | |
6294 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6295 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6296 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6297 | grow_one_stripe(conf, __GFP_NOWARN); |
6298 | /* Set flag even if allocation failed. This helps | |
6299 | * slow down allocation requests when mem is short | |
6300 | */ | |
6301 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6302 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6303 | } |
1da177e4 | 6304 | |
765d704d SL |
6305 | flush_deferred_bios(conf); |
6306 | ||
0576b1c6 SL |
6307 | r5l_flush_stripe_to_raid(conf->log); |
6308 | ||
c9f21aaf | 6309 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6310 | blk_finish_plug(&plug); |
1da177e4 | 6311 | |
45b4233c | 6312 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6313 | } |
6314 | ||
3f294f4f | 6315 | static ssize_t |
fd01b88c | 6316 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6317 | { |
7b1485ba N |
6318 | struct r5conf *conf; |
6319 | int ret = 0; | |
6320 | spin_lock(&mddev->lock); | |
6321 | conf = mddev->private; | |
96de1e66 | 6322 | if (conf) |
edbe83ab | 6323 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6324 | spin_unlock(&mddev->lock); |
6325 | return ret; | |
3f294f4f N |
6326 | } |
6327 | ||
c41d4ac4 | 6328 | int |
fd01b88c | 6329 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6330 | { |
d1688a6d | 6331 | struct r5conf *conf = mddev->private; |
b5470dc5 | 6332 | |
c41d4ac4 | 6333 | if (size <= 16 || size > 32768) |
3f294f4f | 6334 | return -EINVAL; |
486f0644 | 6335 | |
edbe83ab | 6336 | conf->min_nr_stripes = size; |
2d5b569b | 6337 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6338 | while (size < conf->max_nr_stripes && |
6339 | drop_one_stripe(conf)) | |
6340 | ; | |
2d5b569b | 6341 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6342 | |
2214c260 | 6343 | md_allow_write(mddev); |
486f0644 | 6344 | |
2d5b569b | 6345 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6346 | while (size > conf->max_nr_stripes) |
6347 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
6348 | break; | |
2d5b569b | 6349 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6350 | |
c41d4ac4 N |
6351 | return 0; |
6352 | } | |
6353 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6354 | ||
6355 | static ssize_t | |
fd01b88c | 6356 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6357 | { |
6791875e | 6358 | struct r5conf *conf; |
c41d4ac4 N |
6359 | unsigned long new; |
6360 | int err; | |
6361 | ||
6362 | if (len >= PAGE_SIZE) | |
6363 | return -EINVAL; | |
b29bebd6 | 6364 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6365 | return -EINVAL; |
6791875e | 6366 | err = mddev_lock(mddev); |
c41d4ac4 N |
6367 | if (err) |
6368 | return err; | |
6791875e N |
6369 | conf = mddev->private; |
6370 | if (!conf) | |
6371 | err = -ENODEV; | |
6372 | else | |
6373 | err = raid5_set_cache_size(mddev, new); | |
6374 | mddev_unlock(mddev); | |
6375 | ||
6376 | return err ?: len; | |
3f294f4f | 6377 | } |
007583c9 | 6378 | |
96de1e66 N |
6379 | static struct md_sysfs_entry |
6380 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6381 | raid5_show_stripe_cache_size, | |
6382 | raid5_store_stripe_cache_size); | |
3f294f4f | 6383 | |
d06f191f MS |
6384 | static ssize_t |
6385 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6386 | { | |
6387 | struct r5conf *conf = mddev->private; | |
6388 | if (conf) | |
6389 | return sprintf(page, "%d\n", conf->rmw_level); | |
6390 | else | |
6391 | return 0; | |
6392 | } | |
6393 | ||
6394 | static ssize_t | |
6395 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6396 | { | |
6397 | struct r5conf *conf = mddev->private; | |
6398 | unsigned long new; | |
6399 | ||
6400 | if (!conf) | |
6401 | return -ENODEV; | |
6402 | ||
6403 | if (len >= PAGE_SIZE) | |
6404 | return -EINVAL; | |
6405 | ||
6406 | if (kstrtoul(page, 10, &new)) | |
6407 | return -EINVAL; | |
6408 | ||
6409 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6410 | return -EINVAL; | |
6411 | ||
6412 | if (new != PARITY_DISABLE_RMW && | |
6413 | new != PARITY_ENABLE_RMW && | |
6414 | new != PARITY_PREFER_RMW) | |
6415 | return -EINVAL; | |
6416 | ||
6417 | conf->rmw_level = new; | |
6418 | return len; | |
6419 | } | |
6420 | ||
6421 | static struct md_sysfs_entry | |
6422 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6423 | raid5_show_rmw_level, | |
6424 | raid5_store_rmw_level); | |
6425 | ||
6426 | ||
8b3e6cdc | 6427 | static ssize_t |
fd01b88c | 6428 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6429 | { |
7b1485ba N |
6430 | struct r5conf *conf; |
6431 | int ret = 0; | |
6432 | spin_lock(&mddev->lock); | |
6433 | conf = mddev->private; | |
8b3e6cdc | 6434 | if (conf) |
7b1485ba N |
6435 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6436 | spin_unlock(&mddev->lock); | |
6437 | return ret; | |
8b3e6cdc DW |
6438 | } |
6439 | ||
6440 | static ssize_t | |
fd01b88c | 6441 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6442 | { |
6791875e | 6443 | struct r5conf *conf; |
4ef197d8 | 6444 | unsigned long new; |
6791875e N |
6445 | int err; |
6446 | ||
8b3e6cdc DW |
6447 | if (len >= PAGE_SIZE) |
6448 | return -EINVAL; | |
b29bebd6 | 6449 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6450 | return -EINVAL; |
6791875e N |
6451 | |
6452 | err = mddev_lock(mddev); | |
6453 | if (err) | |
6454 | return err; | |
6455 | conf = mddev->private; | |
6456 | if (!conf) | |
6457 | err = -ENODEV; | |
edbe83ab | 6458 | else if (new > conf->min_nr_stripes) |
6791875e N |
6459 | err = -EINVAL; |
6460 | else | |
6461 | conf->bypass_threshold = new; | |
6462 | mddev_unlock(mddev); | |
6463 | return err ?: len; | |
8b3e6cdc DW |
6464 | } |
6465 | ||
6466 | static struct md_sysfs_entry | |
6467 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6468 | S_IRUGO | S_IWUSR, | |
6469 | raid5_show_preread_threshold, | |
6470 | raid5_store_preread_threshold); | |
6471 | ||
d592a996 SL |
6472 | static ssize_t |
6473 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6474 | { | |
7b1485ba N |
6475 | struct r5conf *conf; |
6476 | int ret = 0; | |
6477 | spin_lock(&mddev->lock); | |
6478 | conf = mddev->private; | |
d592a996 | 6479 | if (conf) |
7b1485ba N |
6480 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6481 | spin_unlock(&mddev->lock); | |
6482 | return ret; | |
d592a996 SL |
6483 | } |
6484 | ||
6485 | static ssize_t | |
6486 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6487 | { | |
6791875e | 6488 | struct r5conf *conf; |
d592a996 | 6489 | unsigned long new; |
6791875e N |
6490 | int err; |
6491 | ||
d592a996 SL |
6492 | if (len >= PAGE_SIZE) |
6493 | return -EINVAL; | |
d592a996 SL |
6494 | if (kstrtoul(page, 10, &new)) |
6495 | return -EINVAL; | |
6496 | new = !!new; | |
6791875e N |
6497 | |
6498 | err = mddev_lock(mddev); | |
6499 | if (err) | |
6500 | return err; | |
6501 | conf = mddev->private; | |
6502 | if (!conf) | |
6503 | err = -ENODEV; | |
6504 | else if (new != conf->skip_copy) { | |
6505 | mddev_suspend(mddev); | |
6506 | conf->skip_copy = new; | |
6507 | if (new) | |
dc3b17cc | 6508 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6509 | BDI_CAP_STABLE_WRITES; |
6510 | else | |
dc3b17cc | 6511 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6512 | ~BDI_CAP_STABLE_WRITES; |
6513 | mddev_resume(mddev); | |
6514 | } | |
6515 | mddev_unlock(mddev); | |
6516 | return err ?: len; | |
d592a996 SL |
6517 | } |
6518 | ||
6519 | static struct md_sysfs_entry | |
6520 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6521 | raid5_show_skip_copy, | |
6522 | raid5_store_skip_copy); | |
6523 | ||
3f294f4f | 6524 | static ssize_t |
fd01b88c | 6525 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6526 | { |
d1688a6d | 6527 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6528 | if (conf) |
6529 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6530 | else | |
6531 | return 0; | |
3f294f4f N |
6532 | } |
6533 | ||
96de1e66 N |
6534 | static struct md_sysfs_entry |
6535 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6536 | |
b721420e SL |
6537 | static ssize_t |
6538 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6539 | { | |
7b1485ba N |
6540 | struct r5conf *conf; |
6541 | int ret = 0; | |
6542 | spin_lock(&mddev->lock); | |
6543 | conf = mddev->private; | |
b721420e | 6544 | if (conf) |
7b1485ba N |
6545 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6546 | spin_unlock(&mddev->lock); | |
6547 | return ret; | |
b721420e SL |
6548 | } |
6549 | ||
60aaf933 | 6550 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6551 | int *group_cnt, | |
6552 | int *worker_cnt_per_group, | |
6553 | struct r5worker_group **worker_groups); | |
b721420e SL |
6554 | static ssize_t |
6555 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6556 | { | |
6791875e | 6557 | struct r5conf *conf; |
7d5d7b50 | 6558 | unsigned int new; |
b721420e | 6559 | int err; |
60aaf933 | 6560 | struct r5worker_group *new_groups, *old_groups; |
6561 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6562 | |
6563 | if (len >= PAGE_SIZE) | |
6564 | return -EINVAL; | |
7d5d7b50 SL |
6565 | if (kstrtouint(page, 10, &new)) |
6566 | return -EINVAL; | |
6567 | /* 8192 should be big enough */ | |
6568 | if (new > 8192) | |
b721420e SL |
6569 | return -EINVAL; |
6570 | ||
6791875e N |
6571 | err = mddev_lock(mddev); |
6572 | if (err) | |
6573 | return err; | |
6574 | conf = mddev->private; | |
6575 | if (!conf) | |
6576 | err = -ENODEV; | |
6577 | else if (new != conf->worker_cnt_per_group) { | |
6578 | mddev_suspend(mddev); | |
b721420e | 6579 | |
6791875e N |
6580 | old_groups = conf->worker_groups; |
6581 | if (old_groups) | |
6582 | flush_workqueue(raid5_wq); | |
d206dcfa | 6583 | |
6791875e N |
6584 | err = alloc_thread_groups(conf, new, |
6585 | &group_cnt, &worker_cnt_per_group, | |
6586 | &new_groups); | |
6587 | if (!err) { | |
6588 | spin_lock_irq(&conf->device_lock); | |
6589 | conf->group_cnt = group_cnt; | |
6590 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6591 | conf->worker_groups = new_groups; | |
6592 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6593 | |
6791875e N |
6594 | if (old_groups) |
6595 | kfree(old_groups[0].workers); | |
6596 | kfree(old_groups); | |
6597 | } | |
6598 | mddev_resume(mddev); | |
b721420e | 6599 | } |
6791875e | 6600 | mddev_unlock(mddev); |
b721420e | 6601 | |
6791875e | 6602 | return err ?: len; |
b721420e SL |
6603 | } |
6604 | ||
6605 | static struct md_sysfs_entry | |
6606 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6607 | raid5_show_group_thread_cnt, | |
6608 | raid5_store_group_thread_cnt); | |
6609 | ||
007583c9 | 6610 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6611 | &raid5_stripecache_size.attr, |
6612 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6613 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6614 | &raid5_group_thread_cnt.attr, |
d592a996 | 6615 | &raid5_skip_copy.attr, |
d06f191f | 6616 | &raid5_rmw_level.attr, |
2c7da14b | 6617 | &r5c_journal_mode.attr, |
3f294f4f N |
6618 | NULL, |
6619 | }; | |
007583c9 N |
6620 | static struct attribute_group raid5_attrs_group = { |
6621 | .name = NULL, | |
6622 | .attrs = raid5_attrs, | |
3f294f4f N |
6623 | }; |
6624 | ||
60aaf933 | 6625 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6626 | int *group_cnt, | |
6627 | int *worker_cnt_per_group, | |
6628 | struct r5worker_group **worker_groups) | |
851c30c9 | 6629 | { |
566c09c5 | 6630 | int i, j, k; |
851c30c9 SL |
6631 | ssize_t size; |
6632 | struct r5worker *workers; | |
6633 | ||
60aaf933 | 6634 | *worker_cnt_per_group = cnt; |
851c30c9 | 6635 | if (cnt == 0) { |
60aaf933 | 6636 | *group_cnt = 0; |
6637 | *worker_groups = NULL; | |
851c30c9 SL |
6638 | return 0; |
6639 | } | |
60aaf933 | 6640 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6641 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6642 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6643 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6644 | *group_cnt, GFP_NOIO); | |
6645 | if (!*worker_groups || !workers) { | |
851c30c9 | 6646 | kfree(workers); |
60aaf933 | 6647 | kfree(*worker_groups); |
851c30c9 SL |
6648 | return -ENOMEM; |
6649 | } | |
6650 | ||
60aaf933 | 6651 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6652 | struct r5worker_group *group; |
6653 | ||
0c775d52 | 6654 | group = &(*worker_groups)[i]; |
851c30c9 | 6655 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6656 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6657 | group->conf = conf; |
6658 | group->workers = workers + i * cnt; | |
6659 | ||
6660 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6661 | struct r5worker *worker = group->workers + j; |
6662 | worker->group = group; | |
6663 | INIT_WORK(&worker->work, raid5_do_work); | |
6664 | ||
6665 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6666 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6667 | } |
6668 | } | |
6669 | ||
6670 | return 0; | |
6671 | } | |
6672 | ||
6673 | static void free_thread_groups(struct r5conf *conf) | |
6674 | { | |
6675 | if (conf->worker_groups) | |
6676 | kfree(conf->worker_groups[0].workers); | |
6677 | kfree(conf->worker_groups); | |
6678 | conf->worker_groups = NULL; | |
6679 | } | |
6680 | ||
80c3a6ce | 6681 | static sector_t |
fd01b88c | 6682 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6683 | { |
d1688a6d | 6684 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6685 | |
6686 | if (!sectors) | |
6687 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6688 | if (!raid_disks) |
7ec05478 | 6689 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6690 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6691 | |
3cb5edf4 N |
6692 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6693 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6694 | return sectors * (raid_disks - conf->max_degraded); |
6695 | } | |
6696 | ||
789b5e03 ON |
6697 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6698 | { | |
6699 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6700 | if (percpu->scribble) |
6701 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6702 | percpu->spare_page = NULL; |
6703 | percpu->scribble = NULL; | |
6704 | } | |
6705 | ||
6706 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6707 | { | |
6708 | if (conf->level == 6 && !percpu->spare_page) | |
6709 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6710 | if (!percpu->scribble) | |
46d5b785 | 6711 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6712 | conf->previous_raid_disks), |
6713 | max(conf->chunk_sectors, | |
6714 | conf->prev_chunk_sectors) | |
6715 | / STRIPE_SECTORS, | |
6716 | GFP_KERNEL); | |
789b5e03 ON |
6717 | |
6718 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6719 | free_scratch_buffer(conf, percpu); | |
6720 | return -ENOMEM; | |
6721 | } | |
6722 | ||
6723 | return 0; | |
6724 | } | |
6725 | ||
29c6d1bb | 6726 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6727 | { |
29c6d1bb SAS |
6728 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6729 | ||
6730 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6731 | return 0; | |
6732 | } | |
36d1c647 | 6733 | |
29c6d1bb SAS |
6734 | static void raid5_free_percpu(struct r5conf *conf) |
6735 | { | |
36d1c647 DW |
6736 | if (!conf->percpu) |
6737 | return; | |
6738 | ||
29c6d1bb | 6739 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6740 | free_percpu(conf->percpu); |
6741 | } | |
6742 | ||
d1688a6d | 6743 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6744 | { |
d7bd398e SL |
6745 | int i; |
6746 | ||
ff875738 AP |
6747 | log_exit(conf); |
6748 | ||
30c89465 | 6749 | if (conf->shrinker.nr_deferred) |
edbe83ab | 6750 | unregister_shrinker(&conf->shrinker); |
5c7e81c3 | 6751 | |
851c30c9 | 6752 | free_thread_groups(conf); |
95fc17aa | 6753 | shrink_stripes(conf); |
36d1c647 | 6754 | raid5_free_percpu(conf); |
d7bd398e SL |
6755 | for (i = 0; i < conf->pool_size; i++) |
6756 | if (conf->disks[i].extra_page) | |
6757 | put_page(conf->disks[i].extra_page); | |
95fc17aa | 6758 | kfree(conf->disks); |
dd7a8f5d N |
6759 | if (conf->bio_split) |
6760 | bioset_free(conf->bio_split); | |
95fc17aa | 6761 | kfree(conf->stripe_hashtbl); |
aaf9f12e | 6762 | kfree(conf->pending_data); |
95fc17aa DW |
6763 | kfree(conf); |
6764 | } | |
6765 | ||
29c6d1bb | 6766 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6767 | { |
29c6d1bb | 6768 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6769 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6770 | ||
29c6d1bb | 6771 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6772 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6773 | __func__, cpu); | |
29c6d1bb | 6774 | return -ENOMEM; |
36d1c647 | 6775 | } |
29c6d1bb | 6776 | return 0; |
36d1c647 | 6777 | } |
36d1c647 | 6778 | |
d1688a6d | 6779 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6780 | { |
789b5e03 | 6781 | int err = 0; |
36d1c647 | 6782 | |
789b5e03 ON |
6783 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6784 | if (!conf->percpu) | |
36d1c647 | 6785 | return -ENOMEM; |
789b5e03 | 6786 | |
29c6d1bb | 6787 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6788 | if (!err) { |
6789 | conf->scribble_disks = max(conf->raid_disks, | |
6790 | conf->previous_raid_disks); | |
6791 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6792 | conf->prev_chunk_sectors); | |
6793 | } | |
36d1c647 DW |
6794 | return err; |
6795 | } | |
6796 | ||
edbe83ab N |
6797 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6798 | struct shrink_control *sc) | |
6799 | { | |
6800 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6801 | unsigned long ret = SHRINK_STOP; |
6802 | ||
6803 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6804 | ret= 0; | |
49895bcc N |
6805 | while (ret < sc->nr_to_scan && |
6806 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6807 | if (drop_one_stripe(conf) == 0) { |
6808 | ret = SHRINK_STOP; | |
6809 | break; | |
6810 | } | |
6811 | ret++; | |
6812 | } | |
6813 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6814 | } |
6815 | return ret; | |
6816 | } | |
6817 | ||
6818 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6819 | struct shrink_control *sc) | |
6820 | { | |
6821 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6822 | ||
6823 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6824 | /* unlikely, but not impossible */ | |
6825 | return 0; | |
6826 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6827 | } | |
6828 | ||
d1688a6d | 6829 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6830 | { |
d1688a6d | 6831 | struct r5conf *conf; |
5e5e3e78 | 6832 | int raid_disk, memory, max_disks; |
3cb03002 | 6833 | struct md_rdev *rdev; |
1da177e4 | 6834 | struct disk_info *disk; |
0232605d | 6835 | char pers_name[6]; |
566c09c5 | 6836 | int i; |
60aaf933 | 6837 | int group_cnt, worker_cnt_per_group; |
6838 | struct r5worker_group *new_group; | |
1da177e4 | 6839 | |
91adb564 N |
6840 | if (mddev->new_level != 5 |
6841 | && mddev->new_level != 4 | |
6842 | && mddev->new_level != 6) { | |
cc6167b4 N |
6843 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6844 | mdname(mddev), mddev->new_level); | |
91adb564 | 6845 | return ERR_PTR(-EIO); |
1da177e4 | 6846 | } |
91adb564 N |
6847 | if ((mddev->new_level == 5 |
6848 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6849 | (mddev->new_level == 6 | |
6850 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6851 | pr_warn("md/raid:%s: layout %d not supported\n", |
6852 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6853 | return ERR_PTR(-EIO); |
99c0fb5f | 6854 | } |
91adb564 | 6855 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6856 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6857 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6858 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6859 | } |
6860 | ||
664e7c41 AN |
6861 | if (!mddev->new_chunk_sectors || |
6862 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6863 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6864 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6865 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6866 | return ERR_PTR(-EINVAL); |
f6705578 N |
6867 | } |
6868 | ||
d1688a6d | 6869 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6870 | if (conf == NULL) |
1da177e4 | 6871 | goto abort; |
aaf9f12e SL |
6872 | INIT_LIST_HEAD(&conf->free_list); |
6873 | INIT_LIST_HEAD(&conf->pending_list); | |
6874 | conf->pending_data = kzalloc(sizeof(struct r5pending_data) * | |
6875 | PENDING_IO_MAX, GFP_KERNEL); | |
6876 | if (!conf->pending_data) | |
6877 | goto abort; | |
6878 | for (i = 0; i < PENDING_IO_MAX; i++) | |
6879 | list_add(&conf->pending_data[i].sibling, &conf->free_list); | |
851c30c9 | 6880 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6881 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6882 | &new_group)) { | |
6883 | conf->group_cnt = group_cnt; | |
6884 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6885 | conf->worker_groups = new_group; | |
6886 | } else | |
851c30c9 | 6887 | goto abort; |
f5efd45a | 6888 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6889 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6890 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6891 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6892 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6893 | init_waitqueue_head(&conf->wait_for_overlap); |
6894 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 6895 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
6896 | INIT_LIST_HEAD(&conf->hold_list); |
6897 | INIT_LIST_HEAD(&conf->delayed_list); | |
6898 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 6899 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6900 | atomic_set(&conf->active_stripes, 0); |
6901 | atomic_set(&conf->preread_active_stripes, 0); | |
6902 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
6903 | spin_lock_init(&conf->pending_bios_lock); |
6904 | conf->batch_bio_dispatch = true; | |
6905 | rdev_for_each(rdev, mddev) { | |
6906 | if (test_bit(Journal, &rdev->flags)) | |
6907 | continue; | |
6908 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
6909 | conf->batch_bio_dispatch = false; | |
6910 | break; | |
6911 | } | |
6912 | } | |
6913 | ||
f5efd45a | 6914 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 6915 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6916 | |
6917 | conf->raid_disks = mddev->raid_disks; | |
6918 | if (mddev->reshape_position == MaxSector) | |
6919 | conf->previous_raid_disks = mddev->raid_disks; | |
6920 | else | |
f6705578 | 6921 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6922 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6923 | |
5e5e3e78 | 6924 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc | 6925 | GFP_KERNEL); |
d7bd398e | 6926 | |
b55e6bfc N |
6927 | if (!conf->disks) |
6928 | goto abort; | |
9ffae0cf | 6929 | |
d7bd398e SL |
6930 | for (i = 0; i < max_disks; i++) { |
6931 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
6932 | if (!conf->disks[i].extra_page) | |
6933 | goto abort; | |
6934 | } | |
6935 | ||
011067b0 | 6936 | conf->bio_split = bioset_create(BIO_POOL_SIZE, 0, 0); |
dd7a8f5d N |
6937 | if (!conf->bio_split) |
6938 | goto abort; | |
1da177e4 LT |
6939 | conf->mddev = mddev; |
6940 | ||
fccddba0 | 6941 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6942 | goto abort; |
1da177e4 | 6943 | |
566c09c5 SL |
6944 | /* We init hash_locks[0] separately to that it can be used |
6945 | * as the reference lock in the spin_lock_nest_lock() call | |
6946 | * in lock_all_device_hash_locks_irq in order to convince | |
6947 | * lockdep that we know what we are doing. | |
6948 | */ | |
6949 | spin_lock_init(conf->hash_locks); | |
6950 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6951 | spin_lock_init(conf->hash_locks + i); | |
6952 | ||
6953 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6954 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6955 | ||
6956 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6957 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6958 | ||
1e6d690b SL |
6959 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
6960 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
6961 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
6962 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
6963 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
6964 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 6965 | |
36d1c647 | 6966 | conf->level = mddev->new_level; |
46d5b785 | 6967 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6968 | if (raid5_alloc_percpu(conf) != 0) |
6969 | goto abort; | |
6970 | ||
0c55e022 | 6971 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6972 | |
dafb20fa | 6973 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6974 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6975 | if (raid_disk >= max_disks |
f2076e7d | 6976 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
6977 | continue; |
6978 | disk = conf->disks + raid_disk; | |
6979 | ||
17045f52 N |
6980 | if (test_bit(Replacement, &rdev->flags)) { |
6981 | if (disk->replacement) | |
6982 | goto abort; | |
6983 | disk->replacement = rdev; | |
6984 | } else { | |
6985 | if (disk->rdev) | |
6986 | goto abort; | |
6987 | disk->rdev = rdev; | |
6988 | } | |
1da177e4 | 6989 | |
b2d444d7 | 6990 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6991 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
6992 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
6993 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6994 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6995 | /* Cannot rely on bitmap to complete recovery */ |
6996 | conf->fullsync = 1; | |
1da177e4 LT |
6997 | } |
6998 | ||
91adb564 | 6999 | conf->level = mddev->new_level; |
584acdd4 | 7000 | if (conf->level == 6) { |
16a53ecc | 7001 | conf->max_degraded = 2; |
584acdd4 MS |
7002 | if (raid6_call.xor_syndrome) |
7003 | conf->rmw_level = PARITY_ENABLE_RMW; | |
7004 | else | |
7005 | conf->rmw_level = PARITY_DISABLE_RMW; | |
7006 | } else { | |
16a53ecc | 7007 | conf->max_degraded = 1; |
584acdd4 MS |
7008 | conf->rmw_level = PARITY_ENABLE_RMW; |
7009 | } | |
91adb564 | 7010 | conf->algorithm = mddev->new_layout; |
fef9c61f | 7011 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 7012 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 7013 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 7014 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
7015 | } else { |
7016 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
7017 | conf->prev_algo = conf->algorithm; | |
e183eaed | 7018 | } |
1da177e4 | 7019 | |
edbe83ab | 7020 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
7021 | if (mddev->reshape_position != MaxSector) { |
7022 | int stripes = max_t(int, | |
7023 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
7024 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
7025 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
7026 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 7027 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
7028 | mdname(mddev), conf->min_nr_stripes); |
7029 | } | |
edbe83ab | 7030 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 7031 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 7032 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 7033 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
7034 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
7035 | mdname(mddev), memory); | |
91adb564 N |
7036 | goto abort; |
7037 | } else | |
cc6167b4 | 7038 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
7039 | /* |
7040 | * Losing a stripe head costs more than the time to refill it, | |
7041 | * it reduces the queue depth and so can hurt throughput. | |
7042 | * So set it rather large, scaled by number of devices. | |
7043 | */ | |
7044 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
7045 | conf->shrinker.scan_objects = raid5_cache_scan; | |
7046 | conf->shrinker.count_objects = raid5_cache_count; | |
7047 | conf->shrinker.batch = 128; | |
7048 | conf->shrinker.flags = 0; | |
6a0f53ff | 7049 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
7050 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
7051 | mdname(mddev)); | |
6a0f53ff CY |
7052 | goto abort; |
7053 | } | |
1da177e4 | 7054 | |
0232605d N |
7055 | sprintf(pers_name, "raid%d", mddev->new_level); |
7056 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 7057 | if (!conf->thread) { |
cc6167b4 N |
7058 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
7059 | mdname(mddev)); | |
16a53ecc N |
7060 | goto abort; |
7061 | } | |
91adb564 N |
7062 | |
7063 | return conf; | |
7064 | ||
7065 | abort: | |
7066 | if (conf) { | |
95fc17aa | 7067 | free_conf(conf); |
91adb564 N |
7068 | return ERR_PTR(-EIO); |
7069 | } else | |
7070 | return ERR_PTR(-ENOMEM); | |
7071 | } | |
7072 | ||
c148ffdc N |
7073 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
7074 | { | |
7075 | switch (algo) { | |
7076 | case ALGORITHM_PARITY_0: | |
7077 | if (raid_disk < max_degraded) | |
7078 | return 1; | |
7079 | break; | |
7080 | case ALGORITHM_PARITY_N: | |
7081 | if (raid_disk >= raid_disks - max_degraded) | |
7082 | return 1; | |
7083 | break; | |
7084 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 7085 | if (raid_disk == 0 || |
c148ffdc N |
7086 | raid_disk == raid_disks - 1) |
7087 | return 1; | |
7088 | break; | |
7089 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7090 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7091 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7092 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7093 | if (raid_disk == raid_disks - 1) | |
7094 | return 1; | |
7095 | } | |
7096 | return 0; | |
7097 | } | |
7098 | ||
849674e4 | 7099 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7100 | { |
d1688a6d | 7101 | struct r5conf *conf; |
9f7c2220 | 7102 | int working_disks = 0; |
c148ffdc | 7103 | int dirty_parity_disks = 0; |
3cb03002 | 7104 | struct md_rdev *rdev; |
713cf5a6 | 7105 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7106 | sector_t reshape_offset = 0; |
17045f52 | 7107 | int i; |
b5254dd5 N |
7108 | long long min_offset_diff = 0; |
7109 | int first = 1; | |
91adb564 | 7110 | |
a415c0f1 N |
7111 | if (mddev_init_writes_pending(mddev) < 0) |
7112 | return -ENOMEM; | |
7113 | ||
8c6ac868 | 7114 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7115 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7116 | mdname(mddev)); | |
b5254dd5 N |
7117 | |
7118 | rdev_for_each(rdev, mddev) { | |
7119 | long long diff; | |
713cf5a6 | 7120 | |
f2076e7d | 7121 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7122 | journal_dev = rdev; |
f2076e7d SL |
7123 | continue; |
7124 | } | |
b5254dd5 N |
7125 | if (rdev->raid_disk < 0) |
7126 | continue; | |
7127 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7128 | if (first) { | |
7129 | min_offset_diff = diff; | |
7130 | first = 0; | |
7131 | } else if (mddev->reshape_backwards && | |
7132 | diff < min_offset_diff) | |
7133 | min_offset_diff = diff; | |
7134 | else if (!mddev->reshape_backwards && | |
7135 | diff > min_offset_diff) | |
7136 | min_offset_diff = diff; | |
7137 | } | |
7138 | ||
230b55fa N |
7139 | if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) && |
7140 | (mddev->bitmap_info.offset || mddev->bitmap_info.file)) { | |
7141 | pr_notice("md/raid:%s: array cannot have both journal and bitmap\n", | |
7142 | mdname(mddev)); | |
7143 | return -EINVAL; | |
7144 | } | |
7145 | ||
91adb564 N |
7146 | if (mddev->reshape_position != MaxSector) { |
7147 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7148 | * Difficulties arise if the stripe we would write to |
7149 | * next is at or after the stripe we would read from next. | |
7150 | * For a reshape that changes the number of devices, this | |
7151 | * is only possible for a very short time, and mdadm makes | |
7152 | * sure that time appears to have past before assembling | |
7153 | * the array. So we fail if that time hasn't passed. | |
7154 | * For a reshape that keeps the number of devices the same | |
7155 | * mdadm must be monitoring the reshape can keeping the | |
7156 | * critical areas read-only and backed up. It will start | |
7157 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7158 | */ |
7159 | sector_t here_new, here_old; | |
7160 | int old_disks; | |
18b00334 | 7161 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7162 | int chunk_sectors; |
7163 | int new_data_disks; | |
91adb564 | 7164 | |
713cf5a6 | 7165 | if (journal_dev) { |
cc6167b4 N |
7166 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7167 | mdname(mddev)); | |
713cf5a6 SL |
7168 | return -EINVAL; |
7169 | } | |
7170 | ||
88ce4930 | 7171 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7172 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7173 | mdname(mddev)); | |
91adb564 N |
7174 | return -EINVAL; |
7175 | } | |
91adb564 N |
7176 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7177 | /* reshape_position must be on a new-stripe boundary, and one | |
7178 | * further up in new geometry must map after here in old | |
7179 | * geometry. | |
05256d98 N |
7180 | * If the chunk sizes are different, then as we perform reshape |
7181 | * in units of the largest of the two, reshape_position needs | |
7182 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7183 | */ |
7184 | here_new = mddev->reshape_position; | |
05256d98 N |
7185 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7186 | new_data_disks = mddev->raid_disks - max_degraded; | |
7187 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7188 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7189 | mdname(mddev)); | |
91adb564 N |
7190 | return -EINVAL; |
7191 | } | |
05256d98 | 7192 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7193 | /* here_new is the stripe we will write to */ |
7194 | here_old = mddev->reshape_position; | |
05256d98 | 7195 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7196 | /* here_old is the first stripe that we might need to read |
7197 | * from */ | |
67ac6011 N |
7198 | if (mddev->delta_disks == 0) { |
7199 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7200 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7201 | * and taking constant backups. |
7202 | * mdadm always starts a situation like this in | |
7203 | * readonly mode so it can take control before | |
7204 | * allowing any writes. So just check for that. | |
7205 | */ | |
b5254dd5 N |
7206 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7207 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7208 | /* not really in-place - so OK */; | |
7209 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7210 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7211 | mdname(mddev)); | |
67ac6011 N |
7212 | return -EINVAL; |
7213 | } | |
2c810cdd | 7214 | } else if (mddev->reshape_backwards |
05256d98 N |
7215 | ? (here_new * chunk_sectors + min_offset_diff <= |
7216 | here_old * chunk_sectors) | |
7217 | : (here_new * chunk_sectors >= | |
7218 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7219 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7220 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7221 | mdname(mddev)); | |
91adb564 N |
7222 | return -EINVAL; |
7223 | } | |
cc6167b4 | 7224 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7225 | /* OK, we should be able to continue; */ |
7226 | } else { | |
7227 | BUG_ON(mddev->level != mddev->new_level); | |
7228 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7229 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7230 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7231 | } |
91adb564 | 7232 | |
3418d036 AP |
7233 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && |
7234 | test_bit(MD_HAS_PPL, &mddev->flags)) { | |
7235 | pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n", | |
7236 | mdname(mddev)); | |
7237 | clear_bit(MD_HAS_PPL, &mddev->flags); | |
ddc08823 | 7238 | clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags); |
3418d036 AP |
7239 | } |
7240 | ||
245f46c2 N |
7241 | if (mddev->private == NULL) |
7242 | conf = setup_conf(mddev); | |
7243 | else | |
7244 | conf = mddev->private; | |
7245 | ||
91adb564 N |
7246 | if (IS_ERR(conf)) |
7247 | return PTR_ERR(conf); | |
7248 | ||
486b0f7b SL |
7249 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7250 | if (!journal_dev) { | |
cc6167b4 N |
7251 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7252 | mdname(mddev)); | |
486b0f7b SL |
7253 | mddev->ro = 1; |
7254 | set_disk_ro(mddev->gendisk, 1); | |
7255 | } else if (mddev->recovery_cp == MaxSector) | |
7256 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7257 | } |
7258 | ||
b5254dd5 | 7259 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7260 | mddev->thread = conf->thread; |
7261 | conf->thread = NULL; | |
7262 | mddev->private = conf; | |
7263 | ||
17045f52 N |
7264 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7265 | i++) { | |
7266 | rdev = conf->disks[i].rdev; | |
7267 | if (!rdev && conf->disks[i].replacement) { | |
7268 | /* The replacement is all we have yet */ | |
7269 | rdev = conf->disks[i].replacement; | |
7270 | conf->disks[i].replacement = NULL; | |
7271 | clear_bit(Replacement, &rdev->flags); | |
7272 | conf->disks[i].rdev = rdev; | |
7273 | } | |
7274 | if (!rdev) | |
c148ffdc | 7275 | continue; |
17045f52 N |
7276 | if (conf->disks[i].replacement && |
7277 | conf->reshape_progress != MaxSector) { | |
7278 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7279 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7280 | goto abort; |
7281 | } | |
2f115882 | 7282 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7283 | working_disks++; |
2f115882 N |
7284 | continue; |
7285 | } | |
c148ffdc N |
7286 | /* This disc is not fully in-sync. However if it |
7287 | * just stored parity (beyond the recovery_offset), | |
7288 | * when we don't need to be concerned about the | |
7289 | * array being dirty. | |
7290 | * When reshape goes 'backwards', we never have | |
7291 | * partially completed devices, so we only need | |
7292 | * to worry about reshape going forwards. | |
7293 | */ | |
7294 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7295 | if (mddev->major_version == 0 && | |
7296 | mddev->minor_version > 90) | |
7297 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7298 | |
c148ffdc N |
7299 | if (rdev->recovery_offset < reshape_offset) { |
7300 | /* We need to check old and new layout */ | |
7301 | if (!only_parity(rdev->raid_disk, | |
7302 | conf->algorithm, | |
7303 | conf->raid_disks, | |
7304 | conf->max_degraded)) | |
7305 | continue; | |
7306 | } | |
7307 | if (!only_parity(rdev->raid_disk, | |
7308 | conf->prev_algo, | |
7309 | conf->previous_raid_disks, | |
7310 | conf->max_degraded)) | |
7311 | continue; | |
7312 | dirty_parity_disks++; | |
7313 | } | |
91adb564 | 7314 | |
17045f52 N |
7315 | /* |
7316 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7317 | */ | |
2e38a37f | 7318 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7319 | |
674806d6 | 7320 | if (has_failed(conf)) { |
cc6167b4 | 7321 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7322 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7323 | goto abort; |
7324 | } | |
7325 | ||
91adb564 | 7326 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7327 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7328 | mddev->resync_max_sectors = mddev->dev_sectors; |
7329 | ||
c148ffdc | 7330 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7331 | mddev->recovery_cp != MaxSector) { |
4536bf9b AP |
7332 | if (test_bit(MD_HAS_PPL, &mddev->flags)) |
7333 | pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n", | |
7334 | mdname(mddev)); | |
7335 | else if (mddev->ok_start_degraded) | |
cc6167b4 N |
7336 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7337 | mdname(mddev)); | |
6ff8d8ec | 7338 | else { |
cc6167b4 N |
7339 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7340 | mdname(mddev)); | |
6ff8d8ec N |
7341 | goto abort; |
7342 | } | |
1da177e4 LT |
7343 | } |
7344 | ||
cc6167b4 N |
7345 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7346 | mdname(mddev), conf->level, | |
7347 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7348 | mddev->new_layout); | |
1da177e4 LT |
7349 | |
7350 | print_raid5_conf(conf); | |
7351 | ||
fef9c61f | 7352 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7353 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7354 | atomic_set(&conf->reshape_stripes, 0); |
7355 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7356 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7357 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7358 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7359 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7360 | "reshape"); |
f6705578 N |
7361 | } |
7362 | ||
1da177e4 | 7363 | /* Ok, everything is just fine now */ |
a64c876f N |
7364 | if (mddev->to_remove == &raid5_attrs_group) |
7365 | mddev->to_remove = NULL; | |
00bcb4ac N |
7366 | else if (mddev->kobj.sd && |
7367 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7368 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7369 | mdname(mddev)); | |
4a5add49 | 7370 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7371 | |
4a5add49 | 7372 | if (mddev->queue) { |
9f7c2220 | 7373 | int chunk_size; |
4a5add49 N |
7374 | /* read-ahead size must cover two whole stripes, which |
7375 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7376 | * number of raid devices | |
7377 | */ | |
7378 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7379 | int stripe = data_disks * | |
7380 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7381 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7382 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7383 | |
9f7c2220 N |
7384 | chunk_size = mddev->chunk_sectors << 9; |
7385 | blk_queue_io_min(mddev->queue, chunk_size); | |
7386 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7387 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7388 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7389 | /* |
7390 | * We can only discard a whole stripe. It doesn't make sense to | |
7391 | * discard data disk but write parity disk | |
7392 | */ | |
7393 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7394 | /* Round up to power of 2, as discard handling |
7395 | * currently assumes that */ | |
7396 | while ((stripe-1) & stripe) | |
7397 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7398 | mddev->queue->limits.discard_alignment = stripe; |
7399 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 | 7400 | |
5026d7a9 | 7401 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 7402 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
5026d7a9 | 7403 | |
05616be5 | 7404 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7405 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7406 | rdev->data_offset << 9); | |
05616be5 N |
7407 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7408 | rdev->new_data_offset << 9); | |
7409 | } | |
620125f2 | 7410 | |
48920ff2 CH |
7411 | /* |
7412 | * zeroing is required, otherwise data | |
7413 | * could be lost. Consider a scenario: discard a stripe | |
7414 | * (the stripe could be inconsistent if | |
7415 | * discard_zeroes_data is 0); write one disk of the | |
7416 | * stripe (the stripe could be inconsistent again | |
7417 | * depending on which disks are used to calculate | |
7418 | * parity); the disk is broken; The stripe data of this | |
7419 | * disk is lost. | |
7420 | * | |
7421 | * We only allow DISCARD if the sysadmin has confirmed that | |
7422 | * only safe devices are in use by setting a module parameter. | |
7423 | * A better idea might be to turn DISCARD into WRITE_ZEROES | |
7424 | * requests, as that is required to be safe. | |
7425 | */ | |
7426 | if (devices_handle_discard_safely && | |
e7597e69 JS |
7427 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7428 | mddev->queue->limits.discard_granularity >= stripe) | |
620125f2 SL |
7429 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
7430 | mddev->queue); | |
7431 | else | |
7432 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
7433 | mddev->queue); | |
1dffdddd SL |
7434 | |
7435 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7436 | } |
23032a0e | 7437 | |
845b9e22 | 7438 | if (log_init(conf, journal_dev, raid5_has_ppl(conf))) |
ff875738 | 7439 | goto abort; |
5c7e81c3 | 7440 | |
1da177e4 LT |
7441 | return 0; |
7442 | abort: | |
01f96c0a | 7443 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7444 | print_raid5_conf(conf); |
7445 | free_conf(conf); | |
1da177e4 | 7446 | mddev->private = NULL; |
cc6167b4 | 7447 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7448 | return -EIO; |
7449 | } | |
7450 | ||
afa0f557 | 7451 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7452 | { |
afa0f557 | 7453 | struct r5conf *conf = priv; |
1da177e4 | 7454 | |
95fc17aa | 7455 | free_conf(conf); |
a64c876f | 7456 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7457 | } |
7458 | ||
849674e4 | 7459 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7460 | { |
d1688a6d | 7461 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7462 | int i; |
7463 | ||
9d8f0363 | 7464 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7465 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7466 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7467 | rcu_read_lock(); |
7468 | for (i = 0; i < conf->raid_disks; i++) { | |
7469 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7470 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7471 | } | |
7472 | rcu_read_unlock(); | |
1da177e4 | 7473 | seq_printf (seq, "]"); |
1da177e4 LT |
7474 | } |
7475 | ||
d1688a6d | 7476 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7477 | { |
7478 | int i; | |
7479 | struct disk_info *tmp; | |
7480 | ||
cc6167b4 | 7481 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7482 | if (!conf) { |
cc6167b4 | 7483 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7484 | return; |
7485 | } | |
cc6167b4 | 7486 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7487 | conf->raid_disks, |
7488 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7489 | |
7490 | for (i = 0; i < conf->raid_disks; i++) { | |
7491 | char b[BDEVNAME_SIZE]; | |
7492 | tmp = conf->disks + i; | |
7493 | if (tmp->rdev) | |
cc6167b4 | 7494 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7495 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7496 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7497 | } |
7498 | } | |
7499 | ||
fd01b88c | 7500 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7501 | { |
7502 | int i; | |
d1688a6d | 7503 | struct r5conf *conf = mddev->private; |
1da177e4 | 7504 | struct disk_info *tmp; |
6b965620 N |
7505 | int count = 0; |
7506 | unsigned long flags; | |
1da177e4 LT |
7507 | |
7508 | for (i = 0; i < conf->raid_disks; i++) { | |
7509 | tmp = conf->disks + i; | |
dd054fce N |
7510 | if (tmp->replacement |
7511 | && tmp->replacement->recovery_offset == MaxSector | |
7512 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7513 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7514 | /* Replacement has just become active. */ | |
7515 | if (!tmp->rdev | |
7516 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7517 | count++; | |
7518 | if (tmp->rdev) { | |
7519 | /* Replaced device not technically faulty, | |
7520 | * but we need to be sure it gets removed | |
7521 | * and never re-added. | |
7522 | */ | |
7523 | set_bit(Faulty, &tmp->rdev->flags); | |
7524 | sysfs_notify_dirent_safe( | |
7525 | tmp->rdev->sysfs_state); | |
7526 | } | |
7527 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7528 | } else if (tmp->rdev | |
70fffd0b | 7529 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7530 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7531 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7532 | count++; |
43c73ca4 | 7533 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7534 | } |
7535 | } | |
6b965620 | 7536 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7537 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7538 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7539 | print_raid5_conf(conf); |
6b965620 | 7540 | return count; |
1da177e4 LT |
7541 | } |
7542 | ||
b8321b68 | 7543 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7544 | { |
d1688a6d | 7545 | struct r5conf *conf = mddev->private; |
1da177e4 | 7546 | int err = 0; |
b8321b68 | 7547 | int number = rdev->raid_disk; |
657e3e4d | 7548 | struct md_rdev **rdevp; |
1da177e4 LT |
7549 | struct disk_info *p = conf->disks + number; |
7550 | ||
7551 | print_raid5_conf(conf); | |
f6b6ec5c | 7552 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
c2bb6242 | 7553 | /* |
f6b6ec5c SL |
7554 | * we can't wait pending write here, as this is called in |
7555 | * raid5d, wait will deadlock. | |
84dd97a6 N |
7556 | * neilb: there is no locking about new writes here, |
7557 | * so this cannot be safe. | |
c2bb6242 | 7558 | */ |
70d466f7 SL |
7559 | if (atomic_read(&conf->active_stripes) || |
7560 | atomic_read(&conf->r5c_cached_full_stripes) || | |
7561 | atomic_read(&conf->r5c_cached_partial_stripes)) { | |
f6b6ec5c | 7562 | return -EBUSY; |
84dd97a6 | 7563 | } |
ff875738 | 7564 | log_exit(conf); |
f6b6ec5c | 7565 | return 0; |
c2bb6242 | 7566 | } |
657e3e4d N |
7567 | if (rdev == p->rdev) |
7568 | rdevp = &p->rdev; | |
7569 | else if (rdev == p->replacement) | |
7570 | rdevp = &p->replacement; | |
7571 | else | |
7572 | return 0; | |
7573 | ||
7574 | if (number >= conf->raid_disks && | |
7575 | conf->reshape_progress == MaxSector) | |
7576 | clear_bit(In_sync, &rdev->flags); | |
7577 | ||
7578 | if (test_bit(In_sync, &rdev->flags) || | |
7579 | atomic_read(&rdev->nr_pending)) { | |
7580 | err = -EBUSY; | |
7581 | goto abort; | |
7582 | } | |
7583 | /* Only remove non-faulty devices if recovery | |
7584 | * isn't possible. | |
7585 | */ | |
7586 | if (!test_bit(Faulty, &rdev->flags) && | |
7587 | mddev->recovery_disabled != conf->recovery_disabled && | |
7588 | !has_failed(conf) && | |
dd054fce | 7589 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7590 | number < conf->raid_disks) { |
7591 | err = -EBUSY; | |
7592 | goto abort; | |
7593 | } | |
7594 | *rdevp = NULL; | |
d787be40 N |
7595 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7596 | synchronize_rcu(); | |
7597 | if (atomic_read(&rdev->nr_pending)) { | |
7598 | /* lost the race, try later */ | |
7599 | err = -EBUSY; | |
7600 | *rdevp = rdev; | |
7601 | } | |
7602 | } | |
6358c239 AP |
7603 | if (!err) { |
7604 | err = log_modify(conf, rdev, false); | |
7605 | if (err) | |
7606 | goto abort; | |
7607 | } | |
d787be40 | 7608 | if (p->replacement) { |
dd054fce N |
7609 | /* We must have just cleared 'rdev' */ |
7610 | p->rdev = p->replacement; | |
7611 | clear_bit(Replacement, &p->replacement->flags); | |
7612 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7613 | * but will never see neither - if they are careful | |
7614 | */ | |
7615 | p->replacement = NULL; | |
6358c239 AP |
7616 | |
7617 | if (!err) | |
7618 | err = log_modify(conf, p->rdev, true); | |
e5bc9c3c GJ |
7619 | } |
7620 | ||
7621 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7622 | abort: |
7623 | ||
7624 | print_raid5_conf(conf); | |
7625 | return err; | |
7626 | } | |
7627 | ||
fd01b88c | 7628 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7629 | { |
d1688a6d | 7630 | struct r5conf *conf = mddev->private; |
199050ea | 7631 | int err = -EEXIST; |
1da177e4 LT |
7632 | int disk; |
7633 | struct disk_info *p; | |
6c2fce2e NB |
7634 | int first = 0; |
7635 | int last = conf->raid_disks - 1; | |
1da177e4 | 7636 | |
f6b6ec5c | 7637 | if (test_bit(Journal, &rdev->flags)) { |
f6b6ec5c SL |
7638 | if (conf->log) |
7639 | return -EBUSY; | |
7640 | ||
7641 | rdev->raid_disk = 0; | |
7642 | /* | |
7643 | * The array is in readonly mode if journal is missing, so no | |
7644 | * write requests running. We should be safe | |
7645 | */ | |
845b9e22 | 7646 | log_init(conf, rdev, false); |
f6b6ec5c SL |
7647 | return 0; |
7648 | } | |
7f0da59b N |
7649 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7650 | return -EBUSY; | |
7651 | ||
dc10c643 | 7652 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7653 | /* no point adding a device */ |
199050ea | 7654 | return -EINVAL; |
1da177e4 | 7655 | |
6c2fce2e NB |
7656 | if (rdev->raid_disk >= 0) |
7657 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7658 | |
7659 | /* | |
16a53ecc N |
7660 | * find the disk ... but prefer rdev->saved_raid_disk |
7661 | * if possible. | |
1da177e4 | 7662 | */ |
16a53ecc | 7663 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7664 | rdev->saved_raid_disk >= first && |
16a53ecc | 7665 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7666 | first = rdev->saved_raid_disk; |
7667 | ||
7668 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7669 | p = conf->disks + disk; |
7670 | if (p->rdev == NULL) { | |
b2d444d7 | 7671 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7672 | rdev->raid_disk = disk; |
72626685 N |
7673 | if (rdev->saved_raid_disk != disk) |
7674 | conf->fullsync = 1; | |
d6065f7b | 7675 | rcu_assign_pointer(p->rdev, rdev); |
6358c239 AP |
7676 | |
7677 | err = log_modify(conf, rdev, true); | |
7678 | ||
5cfb22a1 | 7679 | goto out; |
1da177e4 | 7680 | } |
5cfb22a1 N |
7681 | } |
7682 | for (disk = first; disk <= last; disk++) { | |
7683 | p = conf->disks + disk; | |
7bfec5f3 N |
7684 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7685 | p->replacement == NULL) { | |
7686 | clear_bit(In_sync, &rdev->flags); | |
7687 | set_bit(Replacement, &rdev->flags); | |
7688 | rdev->raid_disk = disk; | |
7689 | err = 0; | |
7690 | conf->fullsync = 1; | |
7691 | rcu_assign_pointer(p->replacement, rdev); | |
7692 | break; | |
7693 | } | |
7694 | } | |
5cfb22a1 | 7695 | out: |
1da177e4 | 7696 | print_raid5_conf(conf); |
199050ea | 7697 | return err; |
1da177e4 LT |
7698 | } |
7699 | ||
fd01b88c | 7700 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7701 | { |
7702 | /* no resync is happening, and there is enough space | |
7703 | * on all devices, so we can resize. | |
7704 | * We need to make sure resync covers any new space. | |
7705 | * If the array is shrinking we should possibly wait until | |
7706 | * any io in the removed space completes, but it hardly seems | |
7707 | * worth it. | |
7708 | */ | |
a4a6125a | 7709 | sector_t newsize; |
3cb5edf4 N |
7710 | struct r5conf *conf = mddev->private; |
7711 | ||
3418d036 | 7712 | if (conf->log || raid5_has_ppl(conf)) |
713cf5a6 | 7713 | return -EINVAL; |
3cb5edf4 | 7714 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7715 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7716 | if (mddev->external_size && | |
7717 | mddev->array_sectors > newsize) | |
b522adcd | 7718 | return -EINVAL; |
a4a6125a N |
7719 | if (mddev->bitmap) { |
7720 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7721 | if (ret) | |
7722 | return ret; | |
7723 | } | |
7724 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7725 | if (sectors > mddev->dev_sectors && |
7726 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7727 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7728 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7729 | } | |
58c0fed4 | 7730 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7731 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7732 | return 0; |
7733 | } | |
7734 | ||
fd01b88c | 7735 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7736 | { |
7737 | /* Can only proceed if there are plenty of stripe_heads. | |
7738 | * We need a minimum of one full stripe,, and for sensible progress | |
7739 | * it is best to have about 4 times that. | |
7740 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7741 | * allow for chunk sizes up to 256K, which is probably OK. | |
7742 | * If the chunk size is greater, user-space should request more | |
7743 | * stripe_heads first. | |
7744 | */ | |
d1688a6d | 7745 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7746 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7747 | > conf->min_nr_stripes || |
01ee22b4 | 7748 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7749 | > conf->min_nr_stripes) { |
cc6167b4 N |
7750 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7751 | mdname(mddev), | |
7752 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
7753 | / STRIPE_SIZE)*4); | |
01ee22b4 N |
7754 | return 0; |
7755 | } | |
7756 | return 1; | |
7757 | } | |
7758 | ||
fd01b88c | 7759 | static int check_reshape(struct mddev *mddev) |
29269553 | 7760 | { |
d1688a6d | 7761 | struct r5conf *conf = mddev->private; |
29269553 | 7762 | |
3418d036 | 7763 | if (conf->log || raid5_has_ppl(conf)) |
713cf5a6 | 7764 | return -EINVAL; |
88ce4930 N |
7765 | if (mddev->delta_disks == 0 && |
7766 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7767 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7768 | return 0; /* nothing to do */ |
674806d6 | 7769 | if (has_failed(conf)) |
ec32a2bd | 7770 | return -EINVAL; |
fdcfbbb6 | 7771 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7772 | /* We might be able to shrink, but the devices must |
7773 | * be made bigger first. | |
7774 | * For raid6, 4 is the minimum size. | |
7775 | * Otherwise 2 is the minimum | |
7776 | */ | |
7777 | int min = 2; | |
7778 | if (mddev->level == 6) | |
7779 | min = 4; | |
7780 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7781 | return -EINVAL; | |
7782 | } | |
29269553 | 7783 | |
01ee22b4 | 7784 | if (!check_stripe_cache(mddev)) |
29269553 | 7785 | return -ENOSPC; |
29269553 | 7786 | |
738a2738 N |
7787 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7788 | mddev->delta_disks > 0) | |
7789 | if (resize_chunks(conf, | |
7790 | conf->previous_raid_disks | |
7791 | + max(0, mddev->delta_disks), | |
7792 | max(mddev->new_chunk_sectors, | |
7793 | mddev->chunk_sectors) | |
7794 | ) < 0) | |
7795 | return -ENOMEM; | |
845b9e22 AP |
7796 | |
7797 | if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size) | |
7798 | return 0; /* never bother to shrink */ | |
e56108d6 N |
7799 | return resize_stripes(conf, (conf->previous_raid_disks |
7800 | + mddev->delta_disks)); | |
63c70c4f N |
7801 | } |
7802 | ||
fd01b88c | 7803 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7804 | { |
d1688a6d | 7805 | struct r5conf *conf = mddev->private; |
3cb03002 | 7806 | struct md_rdev *rdev; |
63c70c4f | 7807 | int spares = 0; |
c04be0aa | 7808 | unsigned long flags; |
63c70c4f | 7809 | |
f416885e | 7810 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7811 | return -EBUSY; |
7812 | ||
01ee22b4 N |
7813 | if (!check_stripe_cache(mddev)) |
7814 | return -ENOSPC; | |
7815 | ||
30b67645 N |
7816 | if (has_failed(conf)) |
7817 | return -EINVAL; | |
7818 | ||
c6563a8c | 7819 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7820 | if (!test_bit(In_sync, &rdev->flags) |
7821 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7822 | spares++; |
c6563a8c | 7823 | } |
63c70c4f | 7824 | |
f416885e | 7825 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7826 | /* Not enough devices even to make a degraded array |
7827 | * of that size | |
7828 | */ | |
7829 | return -EINVAL; | |
7830 | ||
ec32a2bd N |
7831 | /* Refuse to reduce size of the array. Any reductions in |
7832 | * array size must be through explicit setting of array_size | |
7833 | * attribute. | |
7834 | */ | |
7835 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7836 | < mddev->array_sectors) { | |
cc6167b4 N |
7837 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7838 | mdname(mddev)); | |
ec32a2bd N |
7839 | return -EINVAL; |
7840 | } | |
7841 | ||
f6705578 | 7842 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7843 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7844 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7845 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7846 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7847 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7848 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7849 | conf->prev_algo = conf->algorithm; |
7850 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7851 | conf->generation++; |
7852 | /* Code that selects data_offset needs to see the generation update | |
7853 | * if reshape_progress has been set - so a memory barrier needed. | |
7854 | */ | |
7855 | smp_mb(); | |
2c810cdd | 7856 | if (mddev->reshape_backwards) |
fef9c61f N |
7857 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7858 | else | |
7859 | conf->reshape_progress = 0; | |
7860 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7861 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7862 | spin_unlock_irq(&conf->device_lock); |
7863 | ||
4d77e3ba N |
7864 | /* Now make sure any requests that proceeded on the assumption |
7865 | * the reshape wasn't running - like Discard or Read - have | |
7866 | * completed. | |
7867 | */ | |
7868 | mddev_suspend(mddev); | |
7869 | mddev_resume(mddev); | |
7870 | ||
29269553 N |
7871 | /* Add some new drives, as many as will fit. |
7872 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7873 | * Don't add devices if we are reducing the number of |
7874 | * devices in the array. This is because it is not possible | |
7875 | * to correctly record the "partially reconstructed" state of | |
7876 | * such devices during the reshape and confusion could result. | |
29269553 | 7877 | */ |
87a8dec9 | 7878 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7879 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7880 | if (rdev->raid_disk < 0 && |
7881 | !test_bit(Faulty, &rdev->flags)) { | |
7882 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7883 | if (rdev->raid_disk |
9d4c7d87 | 7884 | >= conf->previous_raid_disks) |
87a8dec9 | 7885 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7886 | else |
87a8dec9 | 7887 | rdev->recovery_offset = 0; |
36fad858 NK |
7888 | |
7889 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7890 | /* Failure here is OK */; |
50da0840 | 7891 | } |
87a8dec9 N |
7892 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7893 | && !test_bit(Faulty, &rdev->flags)) { | |
7894 | /* This is a spare that was manually added */ | |
7895 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7896 | } |
29269553 | 7897 | |
87a8dec9 N |
7898 | /* When a reshape changes the number of devices, |
7899 | * ->degraded is measured against the larger of the | |
7900 | * pre and post number of devices. | |
7901 | */ | |
ec32a2bd | 7902 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7903 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
7904 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7905 | } | |
63c70c4f | 7906 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7907 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 7908 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 7909 | |
29269553 N |
7910 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7911 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7912 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7913 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7914 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7915 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7916 | "reshape"); |
29269553 N |
7917 | if (!mddev->sync_thread) { |
7918 | mddev->recovery = 0; | |
7919 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7920 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7921 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7922 | mddev->new_chunk_sectors = |
7923 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7924 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7925 | rdev_for_each(rdev, mddev) |
7926 | rdev->new_data_offset = rdev->data_offset; | |
7927 | smp_wmb(); | |
ba8805b9 | 7928 | conf->generation --; |
fef9c61f | 7929 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7930 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7931 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7932 | spin_unlock_irq(&conf->device_lock); |
7933 | return -EAGAIN; | |
7934 | } | |
c8f517c4 | 7935 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7936 | md_wakeup_thread(mddev->sync_thread); |
7937 | md_new_event(mddev); | |
7938 | return 0; | |
7939 | } | |
29269553 | 7940 | |
ec32a2bd N |
7941 | /* This is called from the reshape thread and should make any |
7942 | * changes needed in 'conf' | |
7943 | */ | |
d1688a6d | 7944 | static void end_reshape(struct r5conf *conf) |
29269553 | 7945 | { |
29269553 | 7946 | |
f6705578 | 7947 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
f6705578 | 7948 | |
f6705578 | 7949 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7950 | conf->previous_raid_disks = conf->raid_disks; |
b5d27718 | 7951 | md_finish_reshape(conf->mddev); |
05616be5 | 7952 | smp_wmb(); |
fef9c61f | 7953 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7954 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 7955 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7956 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7957 | |
7958 | /* read-ahead size must cover two whole stripes, which is | |
7959 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7960 | */ | |
4a5add49 | 7961 | if (conf->mddev->queue) { |
cea9c228 | 7962 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7963 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7964 | / PAGE_SIZE); |
dc3b17cc JK |
7965 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7966 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 7967 | } |
29269553 | 7968 | } |
29269553 N |
7969 | } |
7970 | ||
ec32a2bd N |
7971 | /* This is called from the raid5d thread with mddev_lock held. |
7972 | * It makes config changes to the device. | |
7973 | */ | |
fd01b88c | 7974 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7975 | { |
d1688a6d | 7976 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7977 | |
7978 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7979 | ||
ec32a2bd N |
7980 | if (mddev->delta_disks > 0) { |
7981 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
fe67d19a HM |
7982 | if (mddev->queue) { |
7983 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
7984 | revalidate_disk(mddev->gendisk); | |
7985 | } | |
ec32a2bd N |
7986 | } else { |
7987 | int d; | |
908f4fbd | 7988 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 7989 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 7990 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
7991 | for (d = conf->raid_disks ; |
7992 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7993 | d++) { |
3cb03002 | 7994 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7995 | if (rdev) |
7996 | clear_bit(In_sync, &rdev->flags); | |
7997 | rdev = conf->disks[d].replacement; | |
7998 | if (rdev) | |
7999 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 8000 | } |
cea9c228 | 8001 | } |
88ce4930 | 8002 | mddev->layout = conf->algorithm; |
09c9e5fa | 8003 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
8004 | mddev->reshape_position = MaxSector; |
8005 | mddev->delta_disks = 0; | |
2c810cdd | 8006 | mddev->reshape_backwards = 0; |
cea9c228 N |
8007 | } |
8008 | } | |
8009 | ||
b03e0ccb | 8010 | static void raid5_quiesce(struct mddev *mddev, int quiesce) |
72626685 | 8011 | { |
d1688a6d | 8012 | struct r5conf *conf = mddev->private; |
72626685 | 8013 | |
b03e0ccb N |
8014 | if (quiesce) { |
8015 | /* stop all writes */ | |
566c09c5 | 8016 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
8017 | /* '2' tells resync/reshape to pause so that all |
8018 | * active stripes can drain | |
8019 | */ | |
a39f7afd | 8020 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 8021 | conf->quiesce = 2; |
b1b46486 | 8022 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
8023 | atomic_read(&conf->active_stripes) == 0 && |
8024 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
8025 | unlock_all_device_hash_locks_irq(conf), |
8026 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 8027 | conf->quiesce = 1; |
566c09c5 | 8028 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
8029 | /* allow reshape to continue */ |
8030 | wake_up(&conf->wait_for_overlap); | |
b03e0ccb N |
8031 | } else { |
8032 | /* re-enable writes */ | |
566c09c5 | 8033 | lock_all_device_hash_locks_irq(conf); |
72626685 | 8034 | conf->quiesce = 0; |
b1b46486 | 8035 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 8036 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 8037 | unlock_all_device_hash_locks_irq(conf); |
72626685 | 8038 | } |
b03e0ccb | 8039 | r5l_quiesce(conf->log, quiesce); |
72626685 | 8040 | } |
b15c2e57 | 8041 | |
fd01b88c | 8042 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 8043 | { |
e373ab10 | 8044 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 8045 | sector_t sectors; |
54071b38 | 8046 | |
f1b29bca | 8047 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 8048 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
8049 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
8050 | mdname(mddev)); | |
f1b29bca DW |
8051 | return ERR_PTR(-EINVAL); |
8052 | } | |
8053 | ||
e373ab10 N |
8054 | sectors = raid0_conf->strip_zone[0].zone_end; |
8055 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 8056 | mddev->dev_sectors = sectors; |
f1b29bca | 8057 | mddev->new_level = level; |
54071b38 TM |
8058 | mddev->new_layout = ALGORITHM_PARITY_N; |
8059 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
8060 | mddev->raid_disks += 1; | |
8061 | mddev->delta_disks = 1; | |
8062 | /* make sure it will be not marked as dirty */ | |
8063 | mddev->recovery_cp = MaxSector; | |
8064 | ||
8065 | return setup_conf(mddev); | |
8066 | } | |
8067 | ||
fd01b88c | 8068 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
8069 | { |
8070 | int chunksect; | |
6995f0b2 | 8071 | void *ret; |
d562b0c4 N |
8072 | |
8073 | if (mddev->raid_disks != 2 || | |
8074 | mddev->degraded > 1) | |
8075 | return ERR_PTR(-EINVAL); | |
8076 | ||
8077 | /* Should check if there are write-behind devices? */ | |
8078 | ||
8079 | chunksect = 64*2; /* 64K by default */ | |
8080 | ||
8081 | /* The array must be an exact multiple of chunksize */ | |
8082 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8083 | chunksect >>= 1; | |
8084 | ||
8085 | if ((chunksect<<9) < STRIPE_SIZE) | |
8086 | /* array size does not allow a suitable chunk size */ | |
8087 | return ERR_PTR(-EINVAL); | |
8088 | ||
8089 | mddev->new_level = 5; | |
8090 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8091 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8092 | |
6995f0b2 | 8093 | ret = setup_conf(mddev); |
32cd7cbb | 8094 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8095 | mddev_clear_unsupported_flags(mddev, |
8096 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8097 | return ret; |
d562b0c4 N |
8098 | } |
8099 | ||
fd01b88c | 8100 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8101 | { |
8102 | int new_layout; | |
8103 | ||
8104 | switch (mddev->layout) { | |
8105 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8106 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8107 | break; | |
8108 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8109 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8110 | break; | |
8111 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8112 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8113 | break; | |
8114 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8115 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8116 | break; | |
8117 | case ALGORITHM_PARITY_0_6: | |
8118 | new_layout = ALGORITHM_PARITY_0; | |
8119 | break; | |
8120 | case ALGORITHM_PARITY_N: | |
8121 | new_layout = ALGORITHM_PARITY_N; | |
8122 | break; | |
8123 | default: | |
8124 | return ERR_PTR(-EINVAL); | |
8125 | } | |
8126 | mddev->new_level = 5; | |
8127 | mddev->new_layout = new_layout; | |
8128 | mddev->delta_disks = -1; | |
8129 | mddev->raid_disks -= 1; | |
8130 | return setup_conf(mddev); | |
8131 | } | |
8132 | ||
fd01b88c | 8133 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8134 | { |
88ce4930 N |
8135 | /* For a 2-drive array, the layout and chunk size can be changed |
8136 | * immediately as not restriping is needed. | |
8137 | * For larger arrays we record the new value - after validation | |
8138 | * to be used by a reshape pass. | |
b3546035 | 8139 | */ |
d1688a6d | 8140 | struct r5conf *conf = mddev->private; |
597a711b | 8141 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8142 | |
597a711b | 8143 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8144 | return -EINVAL; |
8145 | if (new_chunk > 0) { | |
0ba459d2 | 8146 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8147 | return -EINVAL; |
597a711b | 8148 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8149 | return -EINVAL; |
597a711b | 8150 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8151 | /* not factor of array size */ |
8152 | return -EINVAL; | |
8153 | } | |
8154 | ||
8155 | /* They look valid */ | |
8156 | ||
88ce4930 | 8157 | if (mddev->raid_disks == 2) { |
597a711b N |
8158 | /* can make the change immediately */ |
8159 | if (mddev->new_layout >= 0) { | |
8160 | conf->algorithm = mddev->new_layout; | |
8161 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8162 | } |
8163 | if (new_chunk > 0) { | |
597a711b N |
8164 | conf->chunk_sectors = new_chunk ; |
8165 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8166 | } |
2953079c | 8167 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8168 | md_wakeup_thread(mddev->thread); |
b3546035 | 8169 | } |
50ac168a | 8170 | return check_reshape(mddev); |
88ce4930 N |
8171 | } |
8172 | ||
fd01b88c | 8173 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8174 | { |
597a711b | 8175 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8176 | |
597a711b | 8177 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8178 | return -EINVAL; |
b3546035 | 8179 | if (new_chunk > 0) { |
0ba459d2 | 8180 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8181 | return -EINVAL; |
597a711b | 8182 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8183 | return -EINVAL; |
597a711b | 8184 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8185 | /* not factor of array size */ |
8186 | return -EINVAL; | |
b3546035 | 8187 | } |
88ce4930 N |
8188 | |
8189 | /* They look valid */ | |
50ac168a | 8190 | return check_reshape(mddev); |
b3546035 N |
8191 | } |
8192 | ||
fd01b88c | 8193 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8194 | { |
8195 | /* raid5 can take over: | |
f1b29bca | 8196 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8197 | * raid1 - if there are two drives. We need to know the chunk size |
8198 | * raid4 - trivial - just use a raid4 layout. | |
8199 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8200 | */ |
f1b29bca DW |
8201 | if (mddev->level == 0) |
8202 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8203 | if (mddev->level == 1) |
8204 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8205 | if (mddev->level == 4) { |
8206 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8207 | mddev->new_level = 5; | |
8208 | return setup_conf(mddev); | |
8209 | } | |
fc9739c6 N |
8210 | if (mddev->level == 6) |
8211 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8212 | |
8213 | return ERR_PTR(-EINVAL); | |
8214 | } | |
8215 | ||
fd01b88c | 8216 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8217 | { |
f1b29bca DW |
8218 | /* raid4 can take over: |
8219 | * raid0 - if there is only one strip zone | |
8220 | * raid5 - if layout is right | |
a78d38a1 | 8221 | */ |
f1b29bca DW |
8222 | if (mddev->level == 0) |
8223 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8224 | if (mddev->level == 5 && |
8225 | mddev->layout == ALGORITHM_PARITY_N) { | |
8226 | mddev->new_layout = 0; | |
8227 | mddev->new_level = 4; | |
8228 | return setup_conf(mddev); | |
8229 | } | |
8230 | return ERR_PTR(-EINVAL); | |
8231 | } | |
d562b0c4 | 8232 | |
84fc4b56 | 8233 | static struct md_personality raid5_personality; |
245f46c2 | 8234 | |
fd01b88c | 8235 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8236 | { |
8237 | /* Currently can only take over a raid5. We map the | |
8238 | * personality to an equivalent raid6 personality | |
8239 | * with the Q block at the end. | |
8240 | */ | |
8241 | int new_layout; | |
8242 | ||
8243 | if (mddev->pers != &raid5_personality) | |
8244 | return ERR_PTR(-EINVAL); | |
8245 | if (mddev->degraded > 1) | |
8246 | return ERR_PTR(-EINVAL); | |
8247 | if (mddev->raid_disks > 253) | |
8248 | return ERR_PTR(-EINVAL); | |
8249 | if (mddev->raid_disks < 3) | |
8250 | return ERR_PTR(-EINVAL); | |
8251 | ||
8252 | switch (mddev->layout) { | |
8253 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8254 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8255 | break; | |
8256 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8257 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8258 | break; | |
8259 | case ALGORITHM_LEFT_SYMMETRIC: | |
8260 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8261 | break; | |
8262 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8263 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8264 | break; | |
8265 | case ALGORITHM_PARITY_0: | |
8266 | new_layout = ALGORITHM_PARITY_0_6; | |
8267 | break; | |
8268 | case ALGORITHM_PARITY_N: | |
8269 | new_layout = ALGORITHM_PARITY_N; | |
8270 | break; | |
8271 | default: | |
8272 | return ERR_PTR(-EINVAL); | |
8273 | } | |
8274 | mddev->new_level = 6; | |
8275 | mddev->new_layout = new_layout; | |
8276 | mddev->delta_disks = 1; | |
8277 | mddev->raid_disks += 1; | |
8278 | return setup_conf(mddev); | |
8279 | } | |
8280 | ||
ba903a3e AP |
8281 | static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf) |
8282 | { | |
8283 | struct r5conf *conf; | |
8284 | int err; | |
8285 | ||
8286 | err = mddev_lock(mddev); | |
8287 | if (err) | |
8288 | return err; | |
8289 | conf = mddev->private; | |
8290 | if (!conf) { | |
8291 | mddev_unlock(mddev); | |
8292 | return -ENODEV; | |
8293 | } | |
8294 | ||
845b9e22 | 8295 | if (strncmp(buf, "ppl", 3) == 0) { |
0bb0c105 | 8296 | /* ppl only works with RAID 5 */ |
845b9e22 AP |
8297 | if (!raid5_has_ppl(conf) && conf->level == 5) { |
8298 | err = log_init(conf, NULL, true); | |
8299 | if (!err) { | |
8300 | err = resize_stripes(conf, conf->pool_size); | |
8301 | if (err) | |
8302 | log_exit(conf); | |
8303 | } | |
0bb0c105 SL |
8304 | } else |
8305 | err = -EINVAL; | |
8306 | } else if (strncmp(buf, "resync", 6) == 0) { | |
8307 | if (raid5_has_ppl(conf)) { | |
8308 | mddev_suspend(mddev); | |
8309 | log_exit(conf); | |
0bb0c105 | 8310 | mddev_resume(mddev); |
845b9e22 | 8311 | err = resize_stripes(conf, conf->pool_size); |
0bb0c105 SL |
8312 | } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) && |
8313 | r5l_log_disk_error(conf)) { | |
8314 | bool journal_dev_exists = false; | |
8315 | struct md_rdev *rdev; | |
8316 | ||
8317 | rdev_for_each(rdev, mddev) | |
8318 | if (test_bit(Journal, &rdev->flags)) { | |
8319 | journal_dev_exists = true; | |
8320 | break; | |
8321 | } | |
8322 | ||
8323 | if (!journal_dev_exists) { | |
8324 | mddev_suspend(mddev); | |
8325 | clear_bit(MD_HAS_JOURNAL, &mddev->flags); | |
8326 | mddev_resume(mddev); | |
8327 | } else /* need remove journal device first */ | |
8328 | err = -EBUSY; | |
8329 | } else | |
8330 | err = -EINVAL; | |
ba903a3e AP |
8331 | } else { |
8332 | err = -EINVAL; | |
8333 | } | |
8334 | ||
8335 | if (!err) | |
8336 | md_update_sb(mddev, 1); | |
8337 | ||
8338 | mddev_unlock(mddev); | |
8339 | ||
8340 | return err; | |
8341 | } | |
8342 | ||
84fc4b56 | 8343 | static struct md_personality raid6_personality = |
16a53ecc N |
8344 | { |
8345 | .name = "raid6", | |
8346 | .level = 6, | |
8347 | .owner = THIS_MODULE, | |
849674e4 SL |
8348 | .make_request = raid5_make_request, |
8349 | .run = raid5_run, | |
afa0f557 | 8350 | .free = raid5_free, |
849674e4 SL |
8351 | .status = raid5_status, |
8352 | .error_handler = raid5_error, | |
16a53ecc N |
8353 | .hot_add_disk = raid5_add_disk, |
8354 | .hot_remove_disk= raid5_remove_disk, | |
8355 | .spare_active = raid5_spare_active, | |
849674e4 | 8356 | .sync_request = raid5_sync_request, |
16a53ecc | 8357 | .resize = raid5_resize, |
80c3a6ce | 8358 | .size = raid5_size, |
50ac168a | 8359 | .check_reshape = raid6_check_reshape, |
f416885e | 8360 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8361 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8362 | .quiesce = raid5_quiesce, |
245f46c2 | 8363 | .takeover = raid6_takeover, |
5c675f83 | 8364 | .congested = raid5_congested, |
0bb0c105 | 8365 | .change_consistency_policy = raid5_change_consistency_policy, |
16a53ecc | 8366 | }; |
84fc4b56 | 8367 | static struct md_personality raid5_personality = |
1da177e4 LT |
8368 | { |
8369 | .name = "raid5", | |
2604b703 | 8370 | .level = 5, |
1da177e4 | 8371 | .owner = THIS_MODULE, |
849674e4 SL |
8372 | .make_request = raid5_make_request, |
8373 | .run = raid5_run, | |
afa0f557 | 8374 | .free = raid5_free, |
849674e4 SL |
8375 | .status = raid5_status, |
8376 | .error_handler = raid5_error, | |
1da177e4 LT |
8377 | .hot_add_disk = raid5_add_disk, |
8378 | .hot_remove_disk= raid5_remove_disk, | |
8379 | .spare_active = raid5_spare_active, | |
849674e4 | 8380 | .sync_request = raid5_sync_request, |
1da177e4 | 8381 | .resize = raid5_resize, |
80c3a6ce | 8382 | .size = raid5_size, |
63c70c4f N |
8383 | .check_reshape = raid5_check_reshape, |
8384 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8385 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8386 | .quiesce = raid5_quiesce, |
d562b0c4 | 8387 | .takeover = raid5_takeover, |
5c675f83 | 8388 | .congested = raid5_congested, |
ba903a3e | 8389 | .change_consistency_policy = raid5_change_consistency_policy, |
1da177e4 LT |
8390 | }; |
8391 | ||
84fc4b56 | 8392 | static struct md_personality raid4_personality = |
1da177e4 | 8393 | { |
2604b703 N |
8394 | .name = "raid4", |
8395 | .level = 4, | |
8396 | .owner = THIS_MODULE, | |
849674e4 SL |
8397 | .make_request = raid5_make_request, |
8398 | .run = raid5_run, | |
afa0f557 | 8399 | .free = raid5_free, |
849674e4 SL |
8400 | .status = raid5_status, |
8401 | .error_handler = raid5_error, | |
2604b703 N |
8402 | .hot_add_disk = raid5_add_disk, |
8403 | .hot_remove_disk= raid5_remove_disk, | |
8404 | .spare_active = raid5_spare_active, | |
849674e4 | 8405 | .sync_request = raid5_sync_request, |
2604b703 | 8406 | .resize = raid5_resize, |
80c3a6ce | 8407 | .size = raid5_size, |
3d37890b N |
8408 | .check_reshape = raid5_check_reshape, |
8409 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8410 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8411 | .quiesce = raid5_quiesce, |
a78d38a1 | 8412 | .takeover = raid4_takeover, |
5c675f83 | 8413 | .congested = raid5_congested, |
0bb0c105 | 8414 | .change_consistency_policy = raid5_change_consistency_policy, |
2604b703 N |
8415 | }; |
8416 | ||
8417 | static int __init raid5_init(void) | |
8418 | { | |
29c6d1bb SAS |
8419 | int ret; |
8420 | ||
851c30c9 SL |
8421 | raid5_wq = alloc_workqueue("raid5wq", |
8422 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8423 | if (!raid5_wq) | |
8424 | return -ENOMEM; | |
29c6d1bb SAS |
8425 | |
8426 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8427 | "md/raid5:prepare", | |
8428 | raid456_cpu_up_prepare, | |
8429 | raid456_cpu_dead); | |
8430 | if (ret) { | |
8431 | destroy_workqueue(raid5_wq); | |
8432 | return ret; | |
8433 | } | |
16a53ecc | 8434 | register_md_personality(&raid6_personality); |
2604b703 N |
8435 | register_md_personality(&raid5_personality); |
8436 | register_md_personality(&raid4_personality); | |
8437 | return 0; | |
1da177e4 LT |
8438 | } |
8439 | ||
2604b703 | 8440 | static void raid5_exit(void) |
1da177e4 | 8441 | { |
16a53ecc | 8442 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8443 | unregister_md_personality(&raid5_personality); |
8444 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8445 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8446 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8447 | } |
8448 | ||
8449 | module_init(raid5_init); | |
8450 | module_exit(raid5_exit); | |
8451 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8452 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8453 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8454 | MODULE_ALIAS("md-raid5"); |
8455 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8456 | MODULE_ALIAS("md-level-5"); |
8457 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8458 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8459 | MODULE_ALIAS("md-raid6"); | |
8460 | MODULE_ALIAS("md-level-6"); | |
8461 | ||
8462 | /* This used to be two separate modules, they were: */ | |
8463 | MODULE_ALIAS("raid5"); | |
8464 | MODULE_ALIAS("raid6"); |