]>
Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * raid5.c : Multiple Devices driver for Linux | |
4 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
5 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 6 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 7 | * |
16a53ecc N |
8 | * RAID-4/5/6 management functions. |
9 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
10 | * by donating a test server! | |
1da177e4 LT |
11 | */ |
12 | ||
ae3c20cc N |
13 | /* |
14 | * BITMAP UNPLUGGING: | |
15 | * | |
16 | * The sequencing for updating the bitmap reliably is a little | |
17 | * subtle (and I got it wrong the first time) so it deserves some | |
18 | * explanation. | |
19 | * | |
20 | * We group bitmap updates into batches. Each batch has a number. | |
21 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
22 | * conf->seq_write is the number of the last batch successfully written. |
23 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
24 | * new additions. |
25 | * When we discover that we will need to write to any block in a stripe | |
26 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 27 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
28 | * When we are ready to do a write, if that batch hasn't been written yet, |
29 | * we plug the array and queue the stripe for later. | |
30 | * When an unplug happens, we increment bm_flush, thus closing the current | |
31 | * batch. | |
32 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
33 | * to the bitmap, and advance bm_write to where bm_flush was. | |
34 | * This may occasionally write a bit out twice, but is sure never to | |
35 | * miss any bits. | |
36 | */ | |
1da177e4 | 37 | |
bff61975 | 38 | #include <linux/blkdev.h> |
f6705578 | 39 | #include <linux/kthread.h> |
f701d589 | 40 | #include <linux/raid/pq.h> |
91c00924 | 41 | #include <linux/async_tx.h> |
056075c7 | 42 | #include <linux/module.h> |
07a3b417 | 43 | #include <linux/async.h> |
bff61975 | 44 | #include <linux/seq_file.h> |
36d1c647 | 45 | #include <linux/cpu.h> |
5a0e3ad6 | 46 | #include <linux/slab.h> |
8bda470e | 47 | #include <linux/ratelimit.h> |
851c30c9 | 48 | #include <linux/nodemask.h> |
3f07c014 | 49 | |
a9add5d9 | 50 | #include <trace/events/block.h> |
aaf9f12e | 51 | #include <linux/list_sort.h> |
a9add5d9 | 52 | |
43b2e5d8 | 53 | #include "md.h" |
bff61975 | 54 | #include "raid5.h" |
54071b38 | 55 | #include "raid0.h" |
935fe098 | 56 | #include "md-bitmap.h" |
ff875738 | 57 | #include "raid5-log.h" |
72626685 | 58 | |
394ed8e4 SL |
59 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
60 | ||
851c30c9 SL |
61 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
62 | #define ANY_GROUP NUMA_NO_NODE | |
63 | ||
8e0e99ba N |
64 | static bool devices_handle_discard_safely = false; |
65 | module_param(devices_handle_discard_safely, bool, 0644); | |
66 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
67 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 68 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 69 | |
d1688a6d | 70 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 | 71 | { |
c911c46c | 72 | int hash = (sect >> RAID5_STRIPE_SHIFT(conf)) & HASH_MASK; |
db298e19 N |
73 | return &conf->stripe_hashtbl[hash]; |
74 | } | |
1da177e4 | 75 | |
c911c46c | 76 | static inline int stripe_hash_locks_hash(struct r5conf *conf, sector_t sect) |
566c09c5 | 77 | { |
c911c46c | 78 | return (sect >> RAID5_STRIPE_SHIFT(conf)) & STRIPE_HASH_LOCKS_MASK; |
566c09c5 SL |
79 | } |
80 | ||
81 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
82 | { | |
83 | spin_lock_irq(conf->hash_locks + hash); | |
84 | spin_lock(&conf->device_lock); | |
85 | } | |
86 | ||
87 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
88 | { | |
89 | spin_unlock(&conf->device_lock); | |
90 | spin_unlock_irq(conf->hash_locks + hash); | |
91 | } | |
92 | ||
93 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
94 | { | |
95 | int i; | |
3d05f3ae | 96 | spin_lock_irq(conf->hash_locks); |
566c09c5 SL |
97 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) |
98 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
99 | spin_lock(&conf->device_lock); | |
100 | } | |
101 | ||
102 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
103 | { | |
104 | int i; | |
105 | spin_unlock(&conf->device_lock); | |
3d05f3ae JC |
106 | for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--) |
107 | spin_unlock(conf->hash_locks + i); | |
108 | spin_unlock_irq(conf->hash_locks); | |
566c09c5 SL |
109 | } |
110 | ||
d0dabf7e N |
111 | /* Find first data disk in a raid6 stripe */ |
112 | static inline int raid6_d0(struct stripe_head *sh) | |
113 | { | |
67cc2b81 N |
114 | if (sh->ddf_layout) |
115 | /* ddf always start from first device */ | |
116 | return 0; | |
117 | /* md starts just after Q block */ | |
d0dabf7e N |
118 | if (sh->qd_idx == sh->disks - 1) |
119 | return 0; | |
120 | else | |
121 | return sh->qd_idx + 1; | |
122 | } | |
16a53ecc N |
123 | static inline int raid6_next_disk(int disk, int raid_disks) |
124 | { | |
125 | disk++; | |
126 | return (disk < raid_disks) ? disk : 0; | |
127 | } | |
a4456856 | 128 | |
d0dabf7e N |
129 | /* When walking through the disks in a raid5, starting at raid6_d0, |
130 | * We need to map each disk to a 'slot', where the data disks are slot | |
131 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
132 | * is raid_disks-1. This help does that mapping. | |
133 | */ | |
67cc2b81 N |
134 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
135 | int *count, int syndrome_disks) | |
d0dabf7e | 136 | { |
6629542e | 137 | int slot = *count; |
67cc2b81 | 138 | |
e4424fee | 139 | if (sh->ddf_layout) |
6629542e | 140 | (*count)++; |
d0dabf7e | 141 | if (idx == sh->pd_idx) |
67cc2b81 | 142 | return syndrome_disks; |
d0dabf7e | 143 | if (idx == sh->qd_idx) |
67cc2b81 | 144 | return syndrome_disks + 1; |
e4424fee | 145 | if (!sh->ddf_layout) |
6629542e | 146 | (*count)++; |
d0dabf7e N |
147 | return slot; |
148 | } | |
149 | ||
d1688a6d | 150 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 151 | |
600aa109 DW |
152 | static int stripe_operations_active(struct stripe_head *sh) |
153 | { | |
154 | return sh->check_state || sh->reconstruct_state || | |
155 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
156 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
157 | } | |
158 | ||
535ae4eb SL |
159 | static bool stripe_is_lowprio(struct stripe_head *sh) |
160 | { | |
161 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
162 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
163 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
164 | } | |
165 | ||
851c30c9 SL |
166 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
167 | { | |
168 | struct r5conf *conf = sh->raid_conf; | |
169 | struct r5worker_group *group; | |
bfc90cb0 | 170 | int thread_cnt; |
851c30c9 SL |
171 | int i, cpu = sh->cpu; |
172 | ||
173 | if (!cpu_online(cpu)) { | |
174 | cpu = cpumask_any(cpu_online_mask); | |
175 | sh->cpu = cpu; | |
176 | } | |
177 | ||
178 | if (list_empty(&sh->lru)) { | |
179 | struct r5worker_group *group; | |
180 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
181 | if (stripe_is_lowprio(sh)) |
182 | list_add_tail(&sh->lru, &group->loprio_list); | |
183 | else | |
184 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
185 | group->stripes_cnt++; |
186 | sh->group = group; | |
851c30c9 SL |
187 | } |
188 | ||
189 | if (conf->worker_cnt_per_group == 0) { | |
190 | md_wakeup_thread(conf->mddev->thread); | |
191 | return; | |
192 | } | |
193 | ||
194 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
195 | ||
bfc90cb0 SL |
196 | group->workers[0].working = true; |
197 | /* at least one worker should run to avoid race */ | |
198 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
199 | ||
200 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
201 | /* wakeup more workers */ | |
202 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
203 | if (group->workers[i].working == false) { | |
204 | group->workers[i].working = true; | |
205 | queue_work_on(sh->cpu, raid5_wq, | |
206 | &group->workers[i].work); | |
207 | thread_cnt--; | |
208 | } | |
209 | } | |
851c30c9 SL |
210 | } |
211 | ||
566c09c5 SL |
212 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
213 | struct list_head *temp_inactive_list) | |
1da177e4 | 214 | { |
1e6d690b SL |
215 | int i; |
216 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
217 | ||
4eb788df SL |
218 | BUG_ON(!list_empty(&sh->lru)); |
219 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
220 | |
221 | if (r5c_is_writeback(conf->log)) | |
222 | for (i = sh->disks; i--; ) | |
223 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
224 | injournal++; | |
a39f7afd | 225 | /* |
5ddf0440 SL |
226 | * In the following cases, the stripe cannot be released to cached |
227 | * lists. Therefore, we make the stripe write out and set | |
228 | * STRIPE_HANDLE: | |
229 | * 1. when quiesce in r5c write back; | |
230 | * 2. when resync is requested fot the stripe. | |
a39f7afd | 231 | */ |
5ddf0440 SL |
232 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) || |
233 | (conf->quiesce && r5c_is_writeback(conf->log) && | |
234 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) { | |
a39f7afd SL |
235 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) |
236 | r5c_make_stripe_write_out(sh); | |
237 | set_bit(STRIPE_HANDLE, &sh->state); | |
238 | } | |
1e6d690b | 239 | |
4eb788df SL |
240 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
241 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 242 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 243 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 244 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
245 | sh->bm_seq - conf->seq_write > 0) |
246 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
247 | else { | |
248 | clear_bit(STRIPE_DELAYED, &sh->state); | |
249 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 250 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
251 | if (stripe_is_lowprio(sh)) |
252 | list_add_tail(&sh->lru, | |
253 | &conf->loprio_list); | |
254 | else | |
255 | list_add_tail(&sh->lru, | |
256 | &conf->handle_list); | |
851c30c9 SL |
257 | } else { |
258 | raid5_wakeup_stripe_thread(sh); | |
259 | return; | |
260 | } | |
4eb788df SL |
261 | } |
262 | md_wakeup_thread(conf->mddev->thread); | |
263 | } else { | |
264 | BUG_ON(stripe_operations_active(sh)); | |
265 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
266 | if (atomic_dec_return(&conf->preread_active_stripes) | |
267 | < IO_THRESHOLD) | |
268 | md_wakeup_thread(conf->mddev->thread); | |
269 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
270 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
271 | if (!r5c_is_writeback(conf->log)) | |
272 | list_add_tail(&sh->lru, temp_inactive_list); | |
273 | else { | |
274 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
275 | if (injournal == 0) | |
276 | list_add_tail(&sh->lru, temp_inactive_list); | |
277 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
278 | /* full stripe */ | |
279 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
280 | atomic_inc(&conf->r5c_cached_full_stripes); | |
281 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
282 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
283 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 284 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
285 | } else |
286 | /* | |
287 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
288 | * r5c_try_caching_write(). No need to | |
289 | * set it again. | |
290 | */ | |
1e6d690b | 291 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
292 | } |
293 | } | |
1da177e4 LT |
294 | } |
295 | } | |
d0dabf7e | 296 | |
566c09c5 SL |
297 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
298 | struct list_head *temp_inactive_list) | |
4eb788df SL |
299 | { |
300 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
301 | do_release_stripe(conf, sh, temp_inactive_list); |
302 | } | |
303 | ||
304 | /* | |
305 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
306 | * | |
307 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
308 | * given time. Adding stripes only takes device lock, while deleting stripes | |
309 | * only takes hash lock. | |
310 | */ | |
311 | static void release_inactive_stripe_list(struct r5conf *conf, | |
312 | struct list_head *temp_inactive_list, | |
313 | int hash) | |
314 | { | |
315 | int size; | |
6ab2a4b8 | 316 | bool do_wakeup = false; |
566c09c5 SL |
317 | unsigned long flags; |
318 | ||
319 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
320 | size = NR_STRIPE_HASH_LOCKS; | |
321 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
322 | } else | |
323 | size = 1; | |
324 | while (size) { | |
325 | struct list_head *list = &temp_inactive_list[size - 1]; | |
326 | ||
327 | /* | |
6d036f7d | 328 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
329 | * remove stripes from the list |
330 | */ | |
331 | if (!list_empty_careful(list)) { | |
332 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
333 | if (list_empty(conf->inactive_list + hash) && |
334 | !list_empty(list)) | |
335 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 336 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 337 | do_wakeup = true; |
566c09c5 SL |
338 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
339 | } | |
340 | size--; | |
341 | hash--; | |
342 | } | |
343 | ||
344 | if (do_wakeup) { | |
6ab2a4b8 | 345 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
346 | if (atomic_read(&conf->active_stripes) == 0) |
347 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
348 | if (conf->retry_read_aligned) |
349 | md_wakeup_thread(conf->mddev->thread); | |
350 | } | |
4eb788df SL |
351 | } |
352 | ||
773ca82f | 353 | /* should hold conf->device_lock already */ |
566c09c5 SL |
354 | static int release_stripe_list(struct r5conf *conf, |
355 | struct list_head *temp_inactive_list) | |
773ca82f | 356 | { |
eae8263f | 357 | struct stripe_head *sh, *t; |
773ca82f SL |
358 | int count = 0; |
359 | struct llist_node *head; | |
360 | ||
361 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 362 | head = llist_reverse_order(head); |
eae8263f | 363 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
364 | int hash; |
365 | ||
773ca82f SL |
366 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
367 | smp_mb(); | |
368 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
369 | /* | |
370 | * Don't worry the bit is set here, because if the bit is set | |
371 | * again, the count is always > 1. This is true for | |
372 | * STRIPE_ON_UNPLUG_LIST bit too. | |
373 | */ | |
566c09c5 SL |
374 | hash = sh->hash_lock_index; |
375 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
376 | count++; |
377 | } | |
378 | ||
379 | return count; | |
380 | } | |
381 | ||
6d036f7d | 382 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 383 | { |
d1688a6d | 384 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 385 | unsigned long flags; |
566c09c5 SL |
386 | struct list_head list; |
387 | int hash; | |
773ca82f | 388 | bool wakeup; |
16a53ecc | 389 | |
cf170f3f ES |
390 | /* Avoid release_list until the last reference. |
391 | */ | |
392 | if (atomic_add_unless(&sh->count, -1, 1)) | |
393 | return; | |
394 | ||
ad4068de | 395 | if (unlikely(!conf->mddev->thread) || |
396 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
397 | goto slow_path; |
398 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
399 | if (wakeup) | |
400 | md_wakeup_thread(conf->mddev->thread); | |
401 | return; | |
402 | slow_path: | |
773ca82f | 403 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
685dbcaa | 404 | if (atomic_dec_and_lock_irqsave(&sh->count, &conf->device_lock, flags)) { |
566c09c5 SL |
405 | INIT_LIST_HEAD(&list); |
406 | hash = sh->hash_lock_index; | |
407 | do_release_stripe(conf, sh, &list); | |
08edaaa6 | 408 | spin_unlock_irqrestore(&conf->device_lock, flags); |
566c09c5 | 409 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df | 410 | } |
1da177e4 LT |
411 | } |
412 | ||
fccddba0 | 413 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 414 | { |
45b4233c DW |
415 | pr_debug("remove_hash(), stripe %llu\n", |
416 | (unsigned long long)sh->sector); | |
1da177e4 | 417 | |
fccddba0 | 418 | hlist_del_init(&sh->hash); |
1da177e4 LT |
419 | } |
420 | ||
d1688a6d | 421 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 422 | { |
fccddba0 | 423 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 424 | |
45b4233c DW |
425 | pr_debug("insert_hash(), stripe %llu\n", |
426 | (unsigned long long)sh->sector); | |
1da177e4 | 427 | |
fccddba0 | 428 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
429 | } |
430 | ||
1da177e4 | 431 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 432 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
433 | { |
434 | struct stripe_head *sh = NULL; | |
435 | struct list_head *first; | |
436 | ||
566c09c5 | 437 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 438 | goto out; |
566c09c5 | 439 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
440 | sh = list_entry(first, struct stripe_head, lru); |
441 | list_del_init(first); | |
442 | remove_hash(sh); | |
443 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 444 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
445 | if (list_empty(conf->inactive_list + hash)) |
446 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
447 | out: |
448 | return sh; | |
449 | } | |
450 | ||
e4e11e38 | 451 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
452 | { |
453 | struct page *p; | |
454 | int i; | |
e4e11e38 | 455 | int num = sh->raid_conf->pool_size; |
1da177e4 | 456 | |
e4e11e38 | 457 | for (i = 0; i < num ; i++) { |
d592a996 | 458 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
459 | p = sh->dev[i].page; |
460 | if (!p) | |
461 | continue; | |
462 | sh->dev[i].page = NULL; | |
2d1f3b5d | 463 | put_page(p); |
1da177e4 LT |
464 | } |
465 | } | |
466 | ||
a9683a79 | 467 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
468 | { |
469 | int i; | |
e4e11e38 | 470 | int num = sh->raid_conf->pool_size; |
1da177e4 | 471 | |
e4e11e38 | 472 | for (i = 0; i < num; i++) { |
1da177e4 LT |
473 | struct page *page; |
474 | ||
a9683a79 | 475 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
476 | return 1; |
477 | } | |
478 | sh->dev[i].page = page; | |
d592a996 | 479 | sh->dev[i].orig_page = page; |
1da177e4 | 480 | } |
3418d036 | 481 | |
1da177e4 LT |
482 | return 0; |
483 | } | |
484 | ||
d1688a6d | 485 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 486 | struct stripe_head *sh); |
1da177e4 | 487 | |
b5663ba4 | 488 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 489 | { |
d1688a6d | 490 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 491 | int i, seq; |
1da177e4 | 492 | |
78bafebd ES |
493 | BUG_ON(atomic_read(&sh->count) != 0); |
494 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 495 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 496 | BUG_ON(sh->batch_head); |
d84e0f10 | 497 | |
45b4233c | 498 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 499 | (unsigned long long)sector); |
566c09c5 SL |
500 | retry: |
501 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 502 | sh->generation = conf->generation - previous; |
b5663ba4 | 503 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 504 | sh->sector = sector; |
911d4ee8 | 505 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
506 | sh->state = 0; |
507 | ||
7ecaa1e6 | 508 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
509 | struct r5dev *dev = &sh->dev[i]; |
510 | ||
d84e0f10 | 511 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 512 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 513 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 514 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 515 | dev->read, dev->towrite, dev->written, |
1da177e4 | 516 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 517 | WARN_ON(1); |
1da177e4 LT |
518 | } |
519 | dev->flags = 0; | |
27a4ff8f | 520 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 | 521 | } |
566c09c5 SL |
522 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
523 | goto retry; | |
7a87f434 | 524 | sh->overwrite_disks = 0; |
1da177e4 | 525 | insert_hash(conf, sh); |
851c30c9 | 526 | sh->cpu = smp_processor_id(); |
da41ba65 | 527 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
528 | } |
529 | ||
d1688a6d | 530 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 531 | short generation) |
1da177e4 LT |
532 | { |
533 | struct stripe_head *sh; | |
534 | ||
45b4233c | 535 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 536 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 537 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 538 | return sh; |
45b4233c | 539 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
540 | return NULL; |
541 | } | |
542 | ||
674806d6 N |
543 | /* |
544 | * Need to check if array has failed when deciding whether to: | |
545 | * - start an array | |
546 | * - remove non-faulty devices | |
547 | * - add a spare | |
548 | * - allow a reshape | |
549 | * This determination is simple when no reshape is happening. | |
550 | * However if there is a reshape, we need to carefully check | |
551 | * both the before and after sections. | |
552 | * This is because some failed devices may only affect one | |
553 | * of the two sections, and some non-in_sync devices may | |
554 | * be insync in the section most affected by failed devices. | |
555 | */ | |
2e38a37f | 556 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 557 | { |
908f4fbd | 558 | int degraded, degraded2; |
674806d6 | 559 | int i; |
674806d6 N |
560 | |
561 | rcu_read_lock(); | |
562 | degraded = 0; | |
563 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 564 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
565 | if (rdev && test_bit(Faulty, &rdev->flags)) |
566 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
567 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
568 | degraded++; | |
569 | else if (test_bit(In_sync, &rdev->flags)) | |
570 | ; | |
571 | else | |
572 | /* not in-sync or faulty. | |
573 | * If the reshape increases the number of devices, | |
574 | * this is being recovered by the reshape, so | |
575 | * this 'previous' section is not in_sync. | |
576 | * If the number of devices is being reduced however, | |
577 | * the device can only be part of the array if | |
578 | * we are reverting a reshape, so this section will | |
579 | * be in-sync. | |
580 | */ | |
581 | if (conf->raid_disks >= conf->previous_raid_disks) | |
582 | degraded++; | |
583 | } | |
584 | rcu_read_unlock(); | |
908f4fbd N |
585 | if (conf->raid_disks == conf->previous_raid_disks) |
586 | return degraded; | |
674806d6 | 587 | rcu_read_lock(); |
908f4fbd | 588 | degraded2 = 0; |
674806d6 | 589 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 590 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
591 | if (rdev && test_bit(Faulty, &rdev->flags)) |
592 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 593 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 594 | degraded2++; |
674806d6 N |
595 | else if (test_bit(In_sync, &rdev->flags)) |
596 | ; | |
597 | else | |
598 | /* not in-sync or faulty. | |
599 | * If reshape increases the number of devices, this | |
600 | * section has already been recovered, else it | |
601 | * almost certainly hasn't. | |
602 | */ | |
603 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 604 | degraded2++; |
674806d6 N |
605 | } |
606 | rcu_read_unlock(); | |
908f4fbd N |
607 | if (degraded2 > degraded) |
608 | return degraded2; | |
609 | return degraded; | |
610 | } | |
611 | ||
612 | static int has_failed(struct r5conf *conf) | |
613 | { | |
614 | int degraded; | |
615 | ||
616 | if (conf->mddev->reshape_position == MaxSector) | |
617 | return conf->mddev->degraded > conf->max_degraded; | |
618 | ||
2e38a37f | 619 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
620 | if (degraded > conf->max_degraded) |
621 | return 1; | |
622 | return 0; | |
623 | } | |
624 | ||
6d036f7d SL |
625 | struct stripe_head * |
626 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
627 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
628 | { |
629 | struct stripe_head *sh; | |
c911c46c | 630 | int hash = stripe_hash_locks_hash(conf, sector); |
ff00d3b4 | 631 | int inc_empty_inactive_list_flag; |
1da177e4 | 632 | |
45b4233c | 633 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 634 | |
566c09c5 | 635 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
636 | |
637 | do { | |
b1b46486 | 638 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 639 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 640 | *(conf->hash_locks + hash)); |
86b42c71 | 641 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 642 | if (!sh) { |
edbe83ab | 643 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 644 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
645 | if (!sh && !test_bit(R5_DID_ALLOC, |
646 | &conf->cache_state)) | |
edbe83ab N |
647 | set_bit(R5_ALLOC_MORE, |
648 | &conf->cache_state); | |
649 | } | |
1da177e4 LT |
650 | if (noblock && sh == NULL) |
651 | break; | |
a39f7afd SL |
652 | |
653 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 654 | if (!sh) { |
5423399a N |
655 | set_bit(R5_INACTIVE_BLOCKED, |
656 | &conf->cache_state); | |
a39f7afd | 657 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
658 | wait_event_lock_irq( |
659 | conf->wait_for_stripe, | |
566c09c5 SL |
660 | !list_empty(conf->inactive_list + hash) && |
661 | (atomic_read(&conf->active_stripes) | |
662 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
663 | || !test_bit(R5_INACTIVE_BLOCKED, |
664 | &conf->cache_state)), | |
6ab2a4b8 | 665 | *(conf->hash_locks + hash)); |
5423399a N |
666 | clear_bit(R5_INACTIVE_BLOCKED, |
667 | &conf->cache_state); | |
7da9d450 | 668 | } else { |
b5663ba4 | 669 | init_stripe(sh, sector, previous); |
7da9d450 N |
670 | atomic_inc(&sh->count); |
671 | } | |
e240c183 | 672 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 673 | spin_lock(&conf->device_lock); |
e240c183 | 674 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
675 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
676 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
677 | BUG_ON(list_empty(&sh->lru) && |
678 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
679 | inc_empty_inactive_list_flag = 0; |
680 | if (!list_empty(conf->inactive_list + hash)) | |
681 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 682 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
683 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
684 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
685 | if (sh->group) { |
686 | sh->group->stripes_cnt--; | |
687 | sh->group = NULL; | |
688 | } | |
1da177e4 | 689 | } |
7da9d450 | 690 | atomic_inc(&sh->count); |
6d183de4 | 691 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
692 | } |
693 | } while (sh == NULL); | |
694 | ||
566c09c5 | 695 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
696 | return sh; |
697 | } | |
698 | ||
7a87f434 | 699 | static bool is_full_stripe_write(struct stripe_head *sh) |
700 | { | |
701 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
702 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
703 | } | |
704 | ||
59fc630b | 705 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
368ecade CH |
706 | __acquires(&sh1->stripe_lock) |
707 | __acquires(&sh2->stripe_lock) | |
59fc630b | 708 | { |
59fc630b | 709 | if (sh1 > sh2) { |
3d05f3ae | 710 | spin_lock_irq(&sh2->stripe_lock); |
59fc630b | 711 | spin_lock_nested(&sh1->stripe_lock, 1); |
712 | } else { | |
3d05f3ae | 713 | spin_lock_irq(&sh1->stripe_lock); |
59fc630b | 714 | spin_lock_nested(&sh2->stripe_lock, 1); |
715 | } | |
716 | } | |
717 | ||
718 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
368ecade CH |
719 | __releases(&sh1->stripe_lock) |
720 | __releases(&sh2->stripe_lock) | |
59fc630b | 721 | { |
722 | spin_unlock(&sh1->stripe_lock); | |
3d05f3ae | 723 | spin_unlock_irq(&sh2->stripe_lock); |
59fc630b | 724 | } |
725 | ||
726 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
727 | static bool stripe_can_batch(struct stripe_head *sh) | |
728 | { | |
9c3e333d SL |
729 | struct r5conf *conf = sh->raid_conf; |
730 | ||
e254de6b | 731 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
9c3e333d | 732 | return false; |
59fc630b | 733 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 734 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 735 | is_full_stripe_write(sh); |
736 | } | |
737 | ||
738 | /* we only do back search */ | |
739 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
740 | { | |
741 | struct stripe_head *head; | |
742 | sector_t head_sector, tmp_sec; | |
743 | int hash; | |
744 | int dd_idx; | |
ff00d3b4 | 745 | int inc_empty_inactive_list_flag; |
59fc630b | 746 | |
59fc630b | 747 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
748 | tmp_sec = sh->sector; | |
749 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
750 | return; | |
c911c46c | 751 | head_sector = sh->sector - RAID5_STRIPE_SECTORS(conf); |
59fc630b | 752 | |
c911c46c | 753 | hash = stripe_hash_locks_hash(conf, head_sector); |
59fc630b | 754 | spin_lock_irq(conf->hash_locks + hash); |
755 | head = __find_stripe(conf, head_sector, conf->generation); | |
756 | if (head && !atomic_inc_not_zero(&head->count)) { | |
757 | spin_lock(&conf->device_lock); | |
758 | if (!atomic_read(&head->count)) { | |
759 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
760 | atomic_inc(&conf->active_stripes); | |
761 | BUG_ON(list_empty(&head->lru) && | |
762 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
763 | inc_empty_inactive_list_flag = 0; |
764 | if (!list_empty(conf->inactive_list + hash)) | |
765 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 766 | list_del_init(&head->lru); |
ff00d3b4 ZL |
767 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
768 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 769 | if (head->group) { |
770 | head->group->stripes_cnt--; | |
771 | head->group = NULL; | |
772 | } | |
773 | } | |
774 | atomic_inc(&head->count); | |
775 | spin_unlock(&conf->device_lock); | |
776 | } | |
777 | spin_unlock_irq(conf->hash_locks + hash); | |
778 | ||
779 | if (!head) | |
780 | return; | |
781 | if (!stripe_can_batch(head)) | |
782 | goto out; | |
783 | ||
784 | lock_two_stripes(head, sh); | |
785 | /* clear_batch_ready clear the flag */ | |
786 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
787 | goto unlock_out; | |
788 | ||
789 | if (sh->batch_head) | |
790 | goto unlock_out; | |
791 | ||
792 | dd_idx = 0; | |
793 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
794 | dd_idx++; | |
1eff9d32 | 795 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 796 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 797 | goto unlock_out; |
798 | ||
799 | if (head->batch_head) { | |
800 | spin_lock(&head->batch_head->batch_lock); | |
801 | /* This batch list is already running */ | |
802 | if (!stripe_can_batch(head)) { | |
803 | spin_unlock(&head->batch_head->batch_lock); | |
804 | goto unlock_out; | |
805 | } | |
3664847d SL |
806 | /* |
807 | * We must assign batch_head of this stripe within the | |
808 | * batch_lock, otherwise clear_batch_ready of batch head | |
809 | * stripe could clear BATCH_READY bit of this stripe and | |
810 | * this stripe->batch_head doesn't get assigned, which | |
811 | * could confuse clear_batch_ready for this stripe | |
812 | */ | |
813 | sh->batch_head = head->batch_head; | |
59fc630b | 814 | |
815 | /* | |
816 | * at this point, head's BATCH_READY could be cleared, but we | |
817 | * can still add the stripe to batch list | |
818 | */ | |
819 | list_add(&sh->batch_list, &head->batch_list); | |
820 | spin_unlock(&head->batch_head->batch_lock); | |
59fc630b | 821 | } else { |
822 | head->batch_head = head; | |
823 | sh->batch_head = head->batch_head; | |
824 | spin_lock(&head->batch_lock); | |
825 | list_add_tail(&sh->batch_list, &head->batch_list); | |
826 | spin_unlock(&head->batch_lock); | |
827 | } | |
828 | ||
829 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
830 | if (atomic_dec_return(&conf->preread_active_stripes) | |
831 | < IO_THRESHOLD) | |
832 | md_wakeup_thread(conf->mddev->thread); | |
833 | ||
2b6b2457 N |
834 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
835 | int seq = sh->bm_seq; | |
836 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
837 | sh->batch_head->bm_seq > seq) | |
838 | seq = sh->batch_head->bm_seq; | |
839 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
840 | sh->batch_head->bm_seq = seq; | |
841 | } | |
842 | ||
59fc630b | 843 | atomic_inc(&sh->count); |
844 | unlock_out: | |
845 | unlock_two_stripes(head, sh); | |
846 | out: | |
6d036f7d | 847 | raid5_release_stripe(head); |
59fc630b | 848 | } |
849 | ||
05616be5 N |
850 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
851 | * in this stripe_head. | |
852 | */ | |
853 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
854 | { | |
855 | sector_t progress = conf->reshape_progress; | |
856 | /* Need a memory barrier to make sure we see the value | |
857 | * of conf->generation, or ->data_offset that was set before | |
858 | * reshape_progress was updated. | |
859 | */ | |
860 | smp_rmb(); | |
861 | if (progress == MaxSector) | |
862 | return 0; | |
863 | if (sh->generation == conf->generation - 1) | |
864 | return 0; | |
865 | /* We are in a reshape, and this is a new-generation stripe, | |
866 | * so use new_data_offset. | |
867 | */ | |
868 | return 1; | |
869 | } | |
870 | ||
aaf9f12e | 871 | static void dispatch_bio_list(struct bio_list *tmp) |
765d704d | 872 | { |
765d704d SL |
873 | struct bio *bio; |
874 | ||
aaf9f12e | 875 | while ((bio = bio_list_pop(tmp))) |
ed00aabd | 876 | submit_bio_noacct(bio); |
aaf9f12e SL |
877 | } |
878 | ||
879 | static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b) | |
880 | { | |
881 | const struct r5pending_data *da = list_entry(a, | |
882 | struct r5pending_data, sibling); | |
883 | const struct r5pending_data *db = list_entry(b, | |
884 | struct r5pending_data, sibling); | |
885 | if (da->sector > db->sector) | |
886 | return 1; | |
887 | if (da->sector < db->sector) | |
888 | return -1; | |
889 | return 0; | |
890 | } | |
891 | ||
892 | static void dispatch_defer_bios(struct r5conf *conf, int target, | |
893 | struct bio_list *list) | |
894 | { | |
895 | struct r5pending_data *data; | |
896 | struct list_head *first, *next = NULL; | |
897 | int cnt = 0; | |
898 | ||
899 | if (conf->pending_data_cnt == 0) | |
900 | return; | |
901 | ||
902 | list_sort(NULL, &conf->pending_list, cmp_stripe); | |
903 | ||
904 | first = conf->pending_list.next; | |
905 | ||
906 | /* temporarily move the head */ | |
907 | if (conf->next_pending_data) | |
908 | list_move_tail(&conf->pending_list, | |
909 | &conf->next_pending_data->sibling); | |
910 | ||
911 | while (!list_empty(&conf->pending_list)) { | |
912 | data = list_first_entry(&conf->pending_list, | |
913 | struct r5pending_data, sibling); | |
914 | if (&data->sibling == first) | |
915 | first = data->sibling.next; | |
916 | next = data->sibling.next; | |
917 | ||
918 | bio_list_merge(list, &data->bios); | |
919 | list_move(&data->sibling, &conf->free_list); | |
920 | cnt++; | |
921 | if (cnt >= target) | |
922 | break; | |
923 | } | |
924 | conf->pending_data_cnt -= cnt; | |
925 | BUG_ON(conf->pending_data_cnt < 0 || cnt < target); | |
926 | ||
927 | if (next != &conf->pending_list) | |
928 | conf->next_pending_data = list_entry(next, | |
929 | struct r5pending_data, sibling); | |
930 | else | |
931 | conf->next_pending_data = NULL; | |
932 | /* list isn't empty */ | |
933 | if (first != &conf->pending_list) | |
934 | list_move_tail(&conf->pending_list, first); | |
935 | } | |
936 | ||
937 | static void flush_deferred_bios(struct r5conf *conf) | |
938 | { | |
939 | struct bio_list tmp = BIO_EMPTY_LIST; | |
940 | ||
941 | if (conf->pending_data_cnt == 0) | |
765d704d SL |
942 | return; |
943 | ||
765d704d | 944 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
945 | dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp); |
946 | BUG_ON(conf->pending_data_cnt != 0); | |
765d704d SL |
947 | spin_unlock(&conf->pending_bios_lock); |
948 | ||
aaf9f12e | 949 | dispatch_bio_list(&tmp); |
765d704d SL |
950 | } |
951 | ||
aaf9f12e SL |
952 | static void defer_issue_bios(struct r5conf *conf, sector_t sector, |
953 | struct bio_list *bios) | |
765d704d | 954 | { |
aaf9f12e SL |
955 | struct bio_list tmp = BIO_EMPTY_LIST; |
956 | struct r5pending_data *ent; | |
957 | ||
765d704d | 958 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
959 | ent = list_first_entry(&conf->free_list, struct r5pending_data, |
960 | sibling); | |
961 | list_move_tail(&ent->sibling, &conf->pending_list); | |
962 | ent->sector = sector; | |
963 | bio_list_init(&ent->bios); | |
964 | bio_list_merge(&ent->bios, bios); | |
965 | conf->pending_data_cnt++; | |
966 | if (conf->pending_data_cnt >= PENDING_IO_MAX) | |
967 | dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp); | |
968 | ||
765d704d | 969 | spin_unlock(&conf->pending_bios_lock); |
aaf9f12e SL |
970 | |
971 | dispatch_bio_list(&tmp); | |
765d704d SL |
972 | } |
973 | ||
6712ecf8 | 974 | static void |
4246a0b6 | 975 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 976 | static void |
4246a0b6 | 977 | raid5_end_write_request(struct bio *bi); |
91c00924 | 978 | |
c4e5ac0a | 979 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 980 | { |
d1688a6d | 981 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 982 | int i, disks = sh->disks; |
59fc630b | 983 | struct stripe_head *head_sh = sh; |
aaf9f12e SL |
984 | struct bio_list pending_bios = BIO_EMPTY_LIST; |
985 | bool should_defer; | |
91c00924 DW |
986 | |
987 | might_sleep(); | |
988 | ||
ff875738 AP |
989 | if (log_stripe(sh, s) == 0) |
990 | return; | |
1e6d690b | 991 | |
aaf9f12e | 992 | should_defer = conf->batch_bio_dispatch && conf->group_cnt; |
1e6d690b | 993 | |
91c00924 | 994 | for (i = disks; i--; ) { |
796a5cf0 | 995 | int op, op_flags = 0; |
9a3e1101 | 996 | int replace_only = 0; |
977df362 N |
997 | struct bio *bi, *rbi; |
998 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 999 | |
1000 | sh = head_sh; | |
e9c7469b | 1001 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 1002 | op = REQ_OP_WRITE; |
e9c7469b | 1003 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 1004 | op_flags = REQ_FUA; |
9e444768 | 1005 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 1006 | op = REQ_OP_DISCARD; |
e9c7469b | 1007 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 1008 | op = REQ_OP_READ; |
9a3e1101 N |
1009 | else if (test_and_clear_bit(R5_WantReplace, |
1010 | &sh->dev[i].flags)) { | |
796a5cf0 | 1011 | op = REQ_OP_WRITE; |
9a3e1101 N |
1012 | replace_only = 1; |
1013 | } else | |
91c00924 | 1014 | continue; |
bc0934f0 | 1015 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 1016 | op_flags |= REQ_SYNC; |
91c00924 | 1017 | |
59fc630b | 1018 | again: |
91c00924 | 1019 | bi = &sh->dev[i].req; |
977df362 | 1020 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 1021 | |
91c00924 | 1022 | rcu_read_lock(); |
9a3e1101 | 1023 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
1024 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
1025 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1026 | if (!rdev) { | |
1027 | rdev = rrdev; | |
1028 | rrdev = NULL; | |
1029 | } | |
796a5cf0 | 1030 | if (op_is_write(op)) { |
9a3e1101 N |
1031 | if (replace_only) |
1032 | rdev = NULL; | |
dd054fce N |
1033 | if (rdev == rrdev) |
1034 | /* We raced and saw duplicates */ | |
1035 | rrdev = NULL; | |
9a3e1101 | 1036 | } else { |
59fc630b | 1037 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
1038 | rdev = rrdev; |
1039 | rrdev = NULL; | |
1040 | } | |
977df362 | 1041 | |
91c00924 DW |
1042 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1043 | rdev = NULL; | |
1044 | if (rdev) | |
1045 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
1046 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
1047 | rrdev = NULL; | |
1048 | if (rrdev) | |
1049 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1050 | rcu_read_unlock(); |
1051 | ||
73e92e51 | 1052 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1053 | * need to check for writes. We never accept write errors |
1054 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1055 | */ |
796a5cf0 | 1056 | while (op_is_write(op) && rdev && |
73e92e51 N |
1057 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1058 | sector_t first_bad; | |
1059 | int bad_sectors; | |
c911c46c | 1060 | int bad = is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), |
73e92e51 N |
1061 | &first_bad, &bad_sectors); |
1062 | if (!bad) | |
1063 | break; | |
1064 | ||
1065 | if (bad < 0) { | |
1066 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1067 | if (!conf->mddev->external && | |
2953079c | 1068 | conf->mddev->sb_flags) { |
73e92e51 N |
1069 | /* It is very unlikely, but we might |
1070 | * still need to write out the | |
1071 | * bad block log - better give it | |
1072 | * a chance*/ | |
1073 | md_check_recovery(conf->mddev); | |
1074 | } | |
1850753d | 1075 | /* |
1076 | * Because md_wait_for_blocked_rdev | |
1077 | * will dec nr_pending, we must | |
1078 | * increment it first. | |
1079 | */ | |
1080 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1081 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1082 | } else { | |
1083 | /* Acknowledged bad block - skip the write */ | |
1084 | rdev_dec_pending(rdev, conf->mddev); | |
1085 | rdev = NULL; | |
1086 | } | |
1087 | } | |
1088 | ||
91c00924 | 1089 | if (rdev) { |
9a3e1101 N |
1090 | if (s->syncing || s->expanding || s->expanded |
1091 | || s->replacing) | |
c911c46c | 1092 | md_sync_acct(rdev->bdev, RAID5_STRIPE_SECTORS(conf)); |
91c00924 | 1093 | |
2b7497f0 DW |
1094 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1095 | ||
74d46992 | 1096 | bio_set_dev(bi, rdev->bdev); |
796a5cf0 MC |
1097 | bio_set_op_attrs(bi, op, op_flags); |
1098 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1099 | ? raid5_end_write_request |
1100 | : raid5_end_read_request; | |
1101 | bi->bi_private = sh; | |
1102 | ||
6296b960 | 1103 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1104 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1105 | bi->bi_opf, i); |
91c00924 | 1106 | atomic_inc(&sh->count); |
59fc630b | 1107 | if (sh != head_sh) |
1108 | atomic_inc(&head_sh->count); | |
05616be5 | 1109 | if (use_new_offset(conf, sh)) |
4f024f37 | 1110 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1111 | + rdev->new_data_offset); |
1112 | else | |
4f024f37 | 1113 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1114 | + rdev->data_offset); |
59fc630b | 1115 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1116 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1117 | |
d592a996 SL |
1118 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1119 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1120 | |
1121 | if (!op_is_write(op) && | |
1122 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1123 | /* | |
1124 | * issuing read for a page in journal, this | |
1125 | * must be preparing for prexor in rmw; read | |
1126 | * the data into orig_page | |
1127 | */ | |
1128 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1129 | else | |
1130 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1131 | bi->bi_vcnt = 1; |
c911c46c | 1132 | bi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf); |
91c00924 | 1133 | bi->bi_io_vec[0].bv_offset = 0; |
c911c46c | 1134 | bi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf); |
2cd259a7 MD |
1135 | bi->bi_write_hint = sh->dev[i].write_hint; |
1136 | if (!rrdev) | |
f1934892 | 1137 | sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET; |
37c61ff3 SL |
1138 | /* |
1139 | * If this is discard request, set bi_vcnt 0. We don't | |
1140 | * want to confuse SCSI because SCSI will replace payload | |
1141 | */ | |
796a5cf0 | 1142 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1143 | bi->bi_vcnt = 0; |
977df362 N |
1144 | if (rrdev) |
1145 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1146 | |
1147 | if (conf->mddev->gendisk) | |
74d46992 | 1148 | trace_block_bio_remap(bi->bi_disk->queue, |
e3620a3a JB |
1149 | bi, disk_devt(conf->mddev->gendisk), |
1150 | sh->dev[i].sector); | |
aaf9f12e SL |
1151 | if (should_defer && op_is_write(op)) |
1152 | bio_list_add(&pending_bios, bi); | |
1153 | else | |
ed00aabd | 1154 | submit_bio_noacct(bi); |
977df362 N |
1155 | } |
1156 | if (rrdev) { | |
9a3e1101 N |
1157 | if (s->syncing || s->expanding || s->expanded |
1158 | || s->replacing) | |
c911c46c | 1159 | md_sync_acct(rrdev->bdev, RAID5_STRIPE_SECTORS(conf)); |
977df362 N |
1160 | |
1161 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1162 | ||
74d46992 | 1163 | bio_set_dev(rbi, rrdev->bdev); |
796a5cf0 MC |
1164 | bio_set_op_attrs(rbi, op, op_flags); |
1165 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1166 | rbi->bi_end_io = raid5_end_write_request; |
1167 | rbi->bi_private = sh; | |
1168 | ||
6296b960 | 1169 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1170 | "replacement disc %d\n", |
1171 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1172 | rbi->bi_opf, i); |
977df362 | 1173 | atomic_inc(&sh->count); |
59fc630b | 1174 | if (sh != head_sh) |
1175 | atomic_inc(&head_sh->count); | |
05616be5 | 1176 | if (use_new_offset(conf, sh)) |
4f024f37 | 1177 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1178 | + rrdev->new_data_offset); |
1179 | else | |
4f024f37 | 1180 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1181 | + rrdev->data_offset); |
d592a996 SL |
1182 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1183 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1184 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1185 | rbi->bi_vcnt = 1; |
c911c46c | 1186 | rbi->bi_io_vec[0].bv_len = RAID5_STRIPE_SIZE(conf); |
977df362 | 1187 | rbi->bi_io_vec[0].bv_offset = 0; |
c911c46c | 1188 | rbi->bi_iter.bi_size = RAID5_STRIPE_SIZE(conf); |
2cd259a7 | 1189 | rbi->bi_write_hint = sh->dev[i].write_hint; |
f1934892 | 1190 | sh->dev[i].write_hint = RWH_WRITE_LIFE_NOT_SET; |
37c61ff3 SL |
1191 | /* |
1192 | * If this is discard request, set bi_vcnt 0. We don't | |
1193 | * want to confuse SCSI because SCSI will replace payload | |
1194 | */ | |
796a5cf0 | 1195 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1196 | rbi->bi_vcnt = 0; |
e3620a3a | 1197 | if (conf->mddev->gendisk) |
74d46992 | 1198 | trace_block_bio_remap(rbi->bi_disk->queue, |
e3620a3a JB |
1199 | rbi, disk_devt(conf->mddev->gendisk), |
1200 | sh->dev[i].sector); | |
aaf9f12e SL |
1201 | if (should_defer && op_is_write(op)) |
1202 | bio_list_add(&pending_bios, rbi); | |
1203 | else | |
ed00aabd | 1204 | submit_bio_noacct(rbi); |
977df362 N |
1205 | } |
1206 | if (!rdev && !rrdev) { | |
796a5cf0 | 1207 | if (op_is_write(op)) |
91c00924 | 1208 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1209 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1210 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1211 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1212 | set_bit(STRIPE_HANDLE, &sh->state); | |
1213 | } | |
59fc630b | 1214 | |
1215 | if (!head_sh->batch_head) | |
1216 | continue; | |
1217 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1218 | batch_list); | |
1219 | if (sh != head_sh) | |
1220 | goto again; | |
91c00924 | 1221 | } |
aaf9f12e SL |
1222 | |
1223 | if (should_defer && !bio_list_empty(&pending_bios)) | |
1224 | defer_issue_bios(conf, head_sh->sector, &pending_bios); | |
91c00924 DW |
1225 | } |
1226 | ||
1227 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1228 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1229 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1230 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1231 | { |
7988613b KO |
1232 | struct bio_vec bvl; |
1233 | struct bvec_iter iter; | |
91c00924 | 1234 | struct page *bio_page; |
91c00924 | 1235 | int page_offset; |
a08abd8c | 1236 | struct async_submit_ctl submit; |
0403e382 | 1237 | enum async_tx_flags flags = 0; |
c911c46c | 1238 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1239 | |
4f024f37 KO |
1240 | if (bio->bi_iter.bi_sector >= sector) |
1241 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1242 | else |
4f024f37 | 1243 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1244 | |
0403e382 DW |
1245 | if (frombio) |
1246 | flags |= ASYNC_TX_FENCE; | |
1247 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1248 | ||
7988613b KO |
1249 | bio_for_each_segment(bvl, bio, iter) { |
1250 | int len = bvl.bv_len; | |
91c00924 DW |
1251 | int clen; |
1252 | int b_offset = 0; | |
1253 | ||
1254 | if (page_offset < 0) { | |
1255 | b_offset = -page_offset; | |
1256 | page_offset += b_offset; | |
1257 | len -= b_offset; | |
1258 | } | |
1259 | ||
c911c46c YY |
1260 | if (len > 0 && page_offset + len > RAID5_STRIPE_SIZE(conf)) |
1261 | clen = RAID5_STRIPE_SIZE(conf) - page_offset; | |
91c00924 DW |
1262 | else |
1263 | clen = len; | |
1264 | ||
1265 | if (clen > 0) { | |
7988613b KO |
1266 | b_offset += bvl.bv_offset; |
1267 | bio_page = bvl.bv_page; | |
d592a996 | 1268 | if (frombio) { |
c911c46c | 1269 | if (conf->skip_copy && |
d592a996 | 1270 | b_offset == 0 && page_offset == 0 && |
c911c46c | 1271 | clen == RAID5_STRIPE_SIZE(conf) && |
1e6d690b | 1272 | !no_skipcopy) |
d592a996 SL |
1273 | *page = bio_page; |
1274 | else | |
1275 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1276 | b_offset, clen, &submit); |
d592a996 SL |
1277 | } else |
1278 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1279 | page_offset, clen, &submit); |
91c00924 | 1280 | } |
a08abd8c DW |
1281 | /* chain the operations */ |
1282 | submit.depend_tx = tx; | |
1283 | ||
91c00924 DW |
1284 | if (clen < len) /* hit end of page */ |
1285 | break; | |
1286 | page_offset += len; | |
1287 | } | |
1288 | ||
1289 | return tx; | |
1290 | } | |
1291 | ||
1292 | static void ops_complete_biofill(void *stripe_head_ref) | |
1293 | { | |
1294 | struct stripe_head *sh = stripe_head_ref; | |
e4d84909 | 1295 | int i; |
c911c46c | 1296 | struct r5conf *conf = sh->raid_conf; |
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 < |
c911c46c YY |
1317 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
1318 | rbi2 = r5_next_bio(conf, 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 | 1334 | int i; |
c911c46c | 1335 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1336 | |
59fc630b | 1337 | BUG_ON(sh->batch_head); |
e46b272b | 1338 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1339 | (unsigned long long)sh->sector); |
1340 | ||
1341 | for (i = sh->disks; i--; ) { | |
1342 | struct r5dev *dev = &sh->dev[i]; | |
1343 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1344 | struct bio *rbi; | |
b17459c0 | 1345 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1346 | dev->read = rbi = dev->toread; |
1347 | dev->toread = NULL; | |
b17459c0 | 1348 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1349 | while (rbi && rbi->bi_iter.bi_sector < |
c911c46c | 1350 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
d592a996 | 1351 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1352 | dev->sector, tx, sh, 0); |
c911c46c | 1353 | rbi = r5_next_bio(conf, rbi, dev->sector); |
91c00924 DW |
1354 | } |
1355 | } | |
1356 | } | |
1357 | ||
1358 | atomic_inc(&sh->count); | |
a08abd8c DW |
1359 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1360 | async_trigger_callback(&submit); | |
91c00924 DW |
1361 | } |
1362 | ||
4e7d2c0a | 1363 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1364 | { |
4e7d2c0a | 1365 | struct r5dev *tgt; |
91c00924 | 1366 | |
4e7d2c0a DW |
1367 | if (target < 0) |
1368 | return; | |
91c00924 | 1369 | |
4e7d2c0a | 1370 | tgt = &sh->dev[target]; |
91c00924 DW |
1371 | set_bit(R5_UPTODATE, &tgt->flags); |
1372 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1373 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1374 | } |
1375 | ||
ac6b53b6 | 1376 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1377 | { |
1378 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1379 | |
e46b272b | 1380 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1381 | (unsigned long long)sh->sector); |
1382 | ||
ac6b53b6 | 1383 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1384 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1385 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1386 | |
ecc65c9b DW |
1387 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1388 | if (sh->check_state == check_state_compute_run) | |
1389 | sh->check_state = check_state_compute_result; | |
91c00924 | 1390 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1391 | raid5_release_stripe(sh); |
91c00924 DW |
1392 | } |
1393 | ||
d6f38f31 | 1394 | /* return a pointer to the address conversion region of the scribble buffer */ |
b330e6a4 | 1395 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) |
d6f38f31 | 1396 | { |
b330e6a4 | 1397 | return percpu->scribble + i * percpu->scribble_obj_size; |
46d5b785 | 1398 | } |
1399 | ||
1400 | /* return a pointer to the address conversion region of the scribble buffer */ | |
b330e6a4 KO |
1401 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, |
1402 | struct raid5_percpu *percpu, int i) | |
46d5b785 | 1403 | { |
b330e6a4 | 1404 | return (void *) (to_addr_page(percpu, i) + sh->disks + 2); |
d6f38f31 DW |
1405 | } |
1406 | ||
1407 | static struct dma_async_tx_descriptor * | |
1408 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1409 | { |
91c00924 | 1410 | int disks = sh->disks; |
46d5b785 | 1411 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1412 | int target = sh->ops.target; |
1413 | struct r5dev *tgt = &sh->dev[target]; | |
1414 | struct page *xor_dest = tgt->page; | |
1415 | int count = 0; | |
1416 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1417 | struct async_submit_ctl submit; |
91c00924 DW |
1418 | int i; |
1419 | ||
59fc630b | 1420 | BUG_ON(sh->batch_head); |
1421 | ||
91c00924 | 1422 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1423 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1424 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1425 | ||
1426 | for (i = disks; i--; ) | |
1427 | if (i != target) | |
1428 | xor_srcs[count++] = sh->dev[i].page; | |
1429 | ||
1430 | atomic_inc(&sh->count); | |
1431 | ||
0403e382 | 1432 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1433 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1434 | if (unlikely(count == 1)) |
c911c46c YY |
1435 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, |
1436 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
91c00924 | 1437 | else |
c911c46c YY |
1438 | tx = async_xor(xor_dest, xor_srcs, 0, count, |
1439 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
91c00924 | 1440 | |
91c00924 DW |
1441 | return tx; |
1442 | } | |
1443 | ||
ac6b53b6 DW |
1444 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1445 | * @srcs - (struct page *) array of size sh->disks | |
1446 | * @sh - stripe_head to parse | |
1447 | * | |
1448 | * Populates srcs in proper layout order for the stripe and returns the | |
1449 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1450 | * destination buffer is recorded in srcs[count] and the Q destination | |
1451 | * is recorded in srcs[count+1]]. | |
1452 | */ | |
584acdd4 MS |
1453 | static int set_syndrome_sources(struct page **srcs, |
1454 | struct stripe_head *sh, | |
1455 | int srctype) | |
ac6b53b6 DW |
1456 | { |
1457 | int disks = sh->disks; | |
1458 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1459 | int d0_idx = raid6_d0(sh); | |
1460 | int count; | |
1461 | int i; | |
1462 | ||
1463 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1464 | srcs[i] = NULL; |
ac6b53b6 DW |
1465 | |
1466 | count = 0; | |
1467 | i = d0_idx; | |
1468 | do { | |
1469 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1470 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1471 | |
584acdd4 MS |
1472 | if (i == sh->qd_idx || i == sh->pd_idx || |
1473 | (srctype == SYNDROME_SRC_ALL) || | |
1474 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1475 | (test_bit(R5_Wantdrain, &dev->flags) || |
1476 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1477 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1478 | (dev->written || |
1479 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1480 | if (test_bit(R5_InJournal, &dev->flags)) |
1481 | srcs[slot] = sh->dev[i].orig_page; | |
1482 | else | |
1483 | srcs[slot] = sh->dev[i].page; | |
1484 | } | |
ac6b53b6 DW |
1485 | i = raid6_next_disk(i, disks); |
1486 | } while (i != d0_idx); | |
ac6b53b6 | 1487 | |
e4424fee | 1488 | return syndrome_disks; |
ac6b53b6 DW |
1489 | } |
1490 | ||
1491 | static struct dma_async_tx_descriptor * | |
1492 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1493 | { | |
1494 | int disks = sh->disks; | |
46d5b785 | 1495 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1496 | int target; |
1497 | int qd_idx = sh->qd_idx; | |
1498 | struct dma_async_tx_descriptor *tx; | |
1499 | struct async_submit_ctl submit; | |
1500 | struct r5dev *tgt; | |
1501 | struct page *dest; | |
1502 | int i; | |
1503 | int count; | |
1504 | ||
59fc630b | 1505 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1506 | if (sh->ops.target < 0) |
1507 | target = sh->ops.target2; | |
1508 | else if (sh->ops.target2 < 0) | |
1509 | target = sh->ops.target; | |
91c00924 | 1510 | else |
ac6b53b6 DW |
1511 | /* we should only have one valid target */ |
1512 | BUG(); | |
1513 | BUG_ON(target < 0); | |
1514 | pr_debug("%s: stripe %llu block: %d\n", | |
1515 | __func__, (unsigned long long)sh->sector, target); | |
1516 | ||
1517 | tgt = &sh->dev[target]; | |
1518 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1519 | dest = tgt->page; | |
1520 | ||
1521 | atomic_inc(&sh->count); | |
1522 | ||
1523 | if (target == qd_idx) { | |
584acdd4 | 1524 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1525 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1526 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1527 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1528 | ops_complete_compute, sh, | |
46d5b785 | 1529 | to_addr_conv(sh, percpu, 0)); |
c911c46c YY |
1530 | tx = async_gen_syndrome(blocks, 0, count+2, |
1531 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
ac6b53b6 DW |
1532 | } else { |
1533 | /* Compute any data- or p-drive using XOR */ | |
1534 | count = 0; | |
1535 | for (i = disks; i-- ; ) { | |
1536 | if (i == target || i == qd_idx) | |
1537 | continue; | |
1538 | blocks[count++] = sh->dev[i].page; | |
1539 | } | |
1540 | ||
0403e382 DW |
1541 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1542 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1543 | to_addr_conv(sh, percpu, 0)); |
c911c46c YY |
1544 | tx = async_xor(dest, blocks, 0, count, |
1545 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
ac6b53b6 | 1546 | } |
91c00924 | 1547 | |
91c00924 DW |
1548 | return tx; |
1549 | } | |
1550 | ||
ac6b53b6 DW |
1551 | static struct dma_async_tx_descriptor * |
1552 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1553 | { | |
1554 | int i, count, disks = sh->disks; | |
1555 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1556 | int d0_idx = raid6_d0(sh); | |
1557 | int faila = -1, failb = -1; | |
1558 | int target = sh->ops.target; | |
1559 | int target2 = sh->ops.target2; | |
1560 | struct r5dev *tgt = &sh->dev[target]; | |
1561 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1562 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1563 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1564 | struct async_submit_ctl submit; |
1565 | ||
59fc630b | 1566 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1567 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1568 | __func__, (unsigned long long)sh->sector, target, target2); | |
1569 | BUG_ON(target < 0 || target2 < 0); | |
1570 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1571 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1572 | ||
6c910a78 | 1573 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1574 | * slot number conversion for 'faila' and 'failb' |
1575 | */ | |
1576 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1577 | blocks[i] = NULL; |
ac6b53b6 DW |
1578 | count = 0; |
1579 | i = d0_idx; | |
1580 | do { | |
1581 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1582 | ||
1583 | blocks[slot] = sh->dev[i].page; | |
1584 | ||
1585 | if (i == target) | |
1586 | faila = slot; | |
1587 | if (i == target2) | |
1588 | failb = slot; | |
1589 | i = raid6_next_disk(i, disks); | |
1590 | } while (i != d0_idx); | |
ac6b53b6 DW |
1591 | |
1592 | BUG_ON(faila == failb); | |
1593 | if (failb < faila) | |
1594 | swap(faila, failb); | |
1595 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1596 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1597 | ||
1598 | atomic_inc(&sh->count); | |
1599 | ||
1600 | if (failb == syndrome_disks+1) { | |
1601 | /* Q disk is one of the missing disks */ | |
1602 | if (faila == syndrome_disks) { | |
1603 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1604 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1605 | ops_complete_compute, sh, | |
46d5b785 | 1606 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1607 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
c911c46c YY |
1608 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1609 | &submit); | |
ac6b53b6 DW |
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)); |
c911c46c YY |
1632 | tx = async_xor(dest, blocks, 0, count, |
1633 | RAID5_STRIPE_SIZE(sh->raid_conf), | |
ac6b53b6 DW |
1634 | &submit); |
1635 | ||
584acdd4 | 1636 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1637 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1638 | ops_complete_compute, sh, | |
46d5b785 | 1639 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 | 1640 | return async_gen_syndrome(blocks, 0, count+2, |
c911c46c YY |
1641 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1642 | &submit); | |
ac6b53b6 | 1643 | } |
ac6b53b6 | 1644 | } else { |
6c910a78 DW |
1645 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1646 | ops_complete_compute, sh, | |
46d5b785 | 1647 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1648 | if (failb == syndrome_disks) { |
1649 | /* We're missing D+P. */ | |
1650 | return async_raid6_datap_recov(syndrome_disks+2, | |
c911c46c YY |
1651 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1652 | faila, | |
1653 | blocks, &submit); | |
6c910a78 DW |
1654 | } else { |
1655 | /* We're missing D+D. */ | |
1656 | return async_raid6_2data_recov(syndrome_disks+2, | |
c911c46c YY |
1657 | RAID5_STRIPE_SIZE(sh->raid_conf), |
1658 | faila, failb, | |
1659 | blocks, &submit); | |
6c910a78 | 1660 | } |
ac6b53b6 DW |
1661 | } |
1662 | } | |
1663 | ||
91c00924 DW |
1664 | static void ops_complete_prexor(void *stripe_head_ref) |
1665 | { | |
1666 | struct stripe_head *sh = stripe_head_ref; | |
1667 | ||
e46b272b | 1668 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1669 | (unsigned long long)sh->sector); |
1e6d690b SL |
1670 | |
1671 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1672 | /* | |
1673 | * raid5-cache write back uses orig_page during prexor. | |
1674 | * After prexor, it is time to free orig_page | |
1675 | */ | |
1676 | r5c_release_extra_page(sh); | |
91c00924 DW |
1677 | } |
1678 | ||
1679 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1680 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1681 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1682 | { |
91c00924 | 1683 | int disks = sh->disks; |
46d5b785 | 1684 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1685 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1686 | struct async_submit_ctl submit; |
91c00924 DW |
1687 | |
1688 | /* existing parity data subtracted */ | |
1689 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1690 | ||
59fc630b | 1691 | BUG_ON(sh->batch_head); |
e46b272b | 1692 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1693 | (unsigned long long)sh->sector); |
1694 | ||
1695 | for (i = disks; i--; ) { | |
1696 | struct r5dev *dev = &sh->dev[i]; | |
1697 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1698 | if (test_bit(R5_InJournal, &dev->flags)) |
1699 | xor_srcs[count++] = dev->orig_page; | |
1700 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1701 | xor_srcs[count++] = dev->page; |
1702 | } | |
1703 | ||
0403e382 | 1704 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1705 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
c911c46c YY |
1706 | tx = async_xor(xor_dest, xor_srcs, 0, count, |
1707 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
91c00924 DW |
1708 | |
1709 | return tx; | |
1710 | } | |
1711 | ||
584acdd4 MS |
1712 | static struct dma_async_tx_descriptor * |
1713 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1714 | struct dma_async_tx_descriptor *tx) | |
1715 | { | |
1716 | struct page **blocks = to_addr_page(percpu, 0); | |
1717 | int count; | |
1718 | struct async_submit_ctl submit; | |
1719 | ||
1720 | pr_debug("%s: stripe %llu\n", __func__, | |
1721 | (unsigned long long)sh->sector); | |
1722 | ||
1723 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1724 | ||
1725 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1726 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
c911c46c YY |
1727 | tx = async_gen_syndrome(blocks, 0, count+2, |
1728 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
584acdd4 MS |
1729 | |
1730 | return tx; | |
1731 | } | |
1732 | ||
91c00924 | 1733 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1734 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1735 | { |
1e6d690b | 1736 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1737 | int disks = sh->disks; |
d8ee0728 | 1738 | int i; |
59fc630b | 1739 | struct stripe_head *head_sh = sh; |
91c00924 | 1740 | |
e46b272b | 1741 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1742 | (unsigned long long)sh->sector); |
1743 | ||
1744 | for (i = disks; i--; ) { | |
59fc630b | 1745 | struct r5dev *dev; |
91c00924 | 1746 | struct bio *chosen; |
91c00924 | 1747 | |
59fc630b | 1748 | sh = head_sh; |
1749 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1750 | struct bio *wbi; |
1751 | ||
59fc630b | 1752 | again: |
1753 | dev = &sh->dev[i]; | |
1e6d690b SL |
1754 | /* |
1755 | * clear R5_InJournal, so when rewriting a page in | |
1756 | * journal, it is not skipped by r5l_log_stripe() | |
1757 | */ | |
1758 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1759 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1760 | chosen = dev->towrite; |
1761 | dev->towrite = NULL; | |
7a87f434 | 1762 | sh->overwrite_disks = 0; |
91c00924 DW |
1763 | BUG_ON(dev->written); |
1764 | wbi = dev->written = chosen; | |
b17459c0 | 1765 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1766 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1767 | |
4f024f37 | 1768 | while (wbi && wbi->bi_iter.bi_sector < |
c911c46c | 1769 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
1eff9d32 | 1770 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1771 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1772 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1773 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1774 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1775 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1776 | else { |
1777 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1778 | dev->sector, tx, sh, |
1779 | r5c_is_writeback(conf->log)); | |
1780 | if (dev->page != dev->orig_page && | |
1781 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1782 | set_bit(R5_SkipCopy, &dev->flags); |
1783 | clear_bit(R5_UPTODATE, &dev->flags); | |
1784 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1785 | } | |
1786 | } | |
c911c46c | 1787 | wbi = r5_next_bio(conf, wbi, dev->sector); |
91c00924 | 1788 | } |
59fc630b | 1789 | |
1790 | if (head_sh->batch_head) { | |
1791 | sh = list_first_entry(&sh->batch_list, | |
1792 | struct stripe_head, | |
1793 | batch_list); | |
1794 | if (sh == head_sh) | |
1795 | continue; | |
1796 | goto again; | |
1797 | } | |
91c00924 DW |
1798 | } |
1799 | } | |
1800 | ||
1801 | return tx; | |
1802 | } | |
1803 | ||
ac6b53b6 | 1804 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1805 | { |
1806 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1807 | int disks = sh->disks; |
1808 | int pd_idx = sh->pd_idx; | |
1809 | int qd_idx = sh->qd_idx; | |
1810 | int i; | |
9e444768 | 1811 | bool fua = false, sync = false, discard = false; |
91c00924 | 1812 | |
e46b272b | 1813 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1814 | (unsigned long long)sh->sector); |
1815 | ||
bc0934f0 | 1816 | for (i = disks; i--; ) { |
e9c7469b | 1817 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1818 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1819 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1820 | } |
e9c7469b | 1821 | |
91c00924 DW |
1822 | for (i = disks; i--; ) { |
1823 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1824 | |
e9c7469b | 1825 | if (dev->written || i == pd_idx || i == qd_idx) { |
235b6003 | 1826 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) { |
9e444768 | 1827 | set_bit(R5_UPTODATE, &dev->flags); |
235b6003 N |
1828 | if (test_bit(STRIPE_EXPAND_READY, &sh->state)) |
1829 | set_bit(R5_Expanded, &dev->flags); | |
1830 | } | |
e9c7469b TH |
1831 | if (fua) |
1832 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1833 | if (sync) |
1834 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1835 | } |
91c00924 DW |
1836 | } |
1837 | ||
d8ee0728 DW |
1838 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1839 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1840 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1841 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1842 | else { | |
1843 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1844 | sh->reconstruct_state = reconstruct_state_result; | |
1845 | } | |
91c00924 DW |
1846 | |
1847 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1848 | raid5_release_stripe(sh); |
91c00924 DW |
1849 | } |
1850 | ||
1851 | static void | |
ac6b53b6 DW |
1852 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1853 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1854 | { |
91c00924 | 1855 | int disks = sh->disks; |
59fc630b | 1856 | struct page **xor_srcs; |
a08abd8c | 1857 | struct async_submit_ctl submit; |
59fc630b | 1858 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1859 | struct page *xor_dest; |
d8ee0728 | 1860 | int prexor = 0; |
91c00924 | 1861 | unsigned long flags; |
59fc630b | 1862 | int j = 0; |
1863 | struct stripe_head *head_sh = sh; | |
1864 | int last_stripe; | |
91c00924 | 1865 | |
e46b272b | 1866 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1867 | (unsigned long long)sh->sector); |
1868 | ||
620125f2 SL |
1869 | for (i = 0; i < sh->disks; i++) { |
1870 | if (pd_idx == i) | |
1871 | continue; | |
1872 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1873 | break; | |
1874 | } | |
1875 | if (i >= sh->disks) { | |
1876 | atomic_inc(&sh->count); | |
620125f2 SL |
1877 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1878 | ops_complete_reconstruct(sh); | |
1879 | return; | |
1880 | } | |
59fc630b | 1881 | again: |
1882 | count = 0; | |
1883 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1884 | /* check if prexor is active which means only process blocks |
1885 | * that are part of a read-modify-write (written) | |
1886 | */ | |
59fc630b | 1887 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1888 | prexor = 1; |
91c00924 DW |
1889 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1890 | for (i = disks; i--; ) { | |
1891 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1892 | if (head_sh->dev[i].written || |
1893 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1894 | xor_srcs[count++] = dev->page; |
1895 | } | |
1896 | } else { | |
1897 | xor_dest = sh->dev[pd_idx].page; | |
1898 | for (i = disks; i--; ) { | |
1899 | struct r5dev *dev = &sh->dev[i]; | |
1900 | if (i != pd_idx) | |
1901 | xor_srcs[count++] = dev->page; | |
1902 | } | |
1903 | } | |
1904 | ||
91c00924 DW |
1905 | /* 1/ if we prexor'd then the dest is reused as a source |
1906 | * 2/ if we did not prexor then we are redoing the parity | |
1907 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1908 | * for the synchronous xor case | |
1909 | */ | |
59fc630b | 1910 | last_stripe = !head_sh->batch_head || |
1911 | list_first_entry(&sh->batch_list, | |
1912 | struct stripe_head, batch_list) == head_sh; | |
1913 | if (last_stripe) { | |
1914 | flags = ASYNC_TX_ACK | | |
1915 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1916 | ||
1917 | atomic_inc(&head_sh->count); | |
1918 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1919 | to_addr_conv(sh, percpu, j)); | |
1920 | } else { | |
1921 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1922 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1923 | to_addr_conv(sh, percpu, j)); | |
1924 | } | |
91c00924 | 1925 | |
a08abd8c | 1926 | if (unlikely(count == 1)) |
c911c46c YY |
1927 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, |
1928 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
a08abd8c | 1929 | else |
c911c46c YY |
1930 | tx = async_xor(xor_dest, xor_srcs, 0, count, |
1931 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
59fc630b | 1932 | if (!last_stripe) { |
1933 | j++; | |
1934 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1935 | batch_list); | |
1936 | goto again; | |
1937 | } | |
91c00924 DW |
1938 | } |
1939 | ||
ac6b53b6 DW |
1940 | static void |
1941 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1942 | struct dma_async_tx_descriptor *tx) | |
1943 | { | |
1944 | struct async_submit_ctl submit; | |
59fc630b | 1945 | struct page **blocks; |
1946 | int count, i, j = 0; | |
1947 | struct stripe_head *head_sh = sh; | |
1948 | int last_stripe; | |
584acdd4 MS |
1949 | int synflags; |
1950 | unsigned long txflags; | |
ac6b53b6 DW |
1951 | |
1952 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1953 | ||
620125f2 SL |
1954 | for (i = 0; i < sh->disks; i++) { |
1955 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1956 | continue; | |
1957 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1958 | break; | |
1959 | } | |
1960 | if (i >= sh->disks) { | |
1961 | atomic_inc(&sh->count); | |
620125f2 SL |
1962 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1963 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1964 | ops_complete_reconstruct(sh); | |
1965 | return; | |
1966 | } | |
1967 | ||
59fc630b | 1968 | again: |
1969 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1970 | |
1971 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1972 | synflags = SYNDROME_SRC_WRITTEN; | |
1973 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1974 | } else { | |
1975 | synflags = SYNDROME_SRC_ALL; | |
1976 | txflags = ASYNC_TX_ACK; | |
1977 | } | |
1978 | ||
1979 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1980 | last_stripe = !head_sh->batch_head || |
1981 | list_first_entry(&sh->batch_list, | |
1982 | struct stripe_head, batch_list) == head_sh; | |
1983 | ||
1984 | if (last_stripe) { | |
1985 | atomic_inc(&head_sh->count); | |
584acdd4 | 1986 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1987 | head_sh, to_addr_conv(sh, percpu, j)); |
1988 | } else | |
1989 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1990 | to_addr_conv(sh, percpu, j)); | |
c911c46c YY |
1991 | tx = async_gen_syndrome(blocks, 0, count+2, |
1992 | RAID5_STRIPE_SIZE(sh->raid_conf), &submit); | |
59fc630b | 1993 | if (!last_stripe) { |
1994 | j++; | |
1995 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1996 | batch_list); | |
1997 | goto again; | |
1998 | } | |
91c00924 DW |
1999 | } |
2000 | ||
2001 | static void ops_complete_check(void *stripe_head_ref) | |
2002 | { | |
2003 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 2004 | |
e46b272b | 2005 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2006 | (unsigned long long)sh->sector); |
2007 | ||
ecc65c9b | 2008 | sh->check_state = check_state_check_result; |
91c00924 | 2009 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2010 | raid5_release_stripe(sh); |
91c00924 DW |
2011 | } |
2012 | ||
ac6b53b6 | 2013 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 2014 | { |
91c00924 | 2015 | int disks = sh->disks; |
ac6b53b6 DW |
2016 | int pd_idx = sh->pd_idx; |
2017 | int qd_idx = sh->qd_idx; | |
2018 | struct page *xor_dest; | |
46d5b785 | 2019 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 2020 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 2021 | struct async_submit_ctl submit; |
ac6b53b6 DW |
2022 | int count; |
2023 | int i; | |
91c00924 | 2024 | |
e46b272b | 2025 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2026 | (unsigned long long)sh->sector); |
2027 | ||
59fc630b | 2028 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
2029 | count = 0; |
2030 | xor_dest = sh->dev[pd_idx].page; | |
2031 | xor_srcs[count++] = xor_dest; | |
91c00924 | 2032 | for (i = disks; i--; ) { |
ac6b53b6 DW |
2033 | if (i == pd_idx || i == qd_idx) |
2034 | continue; | |
2035 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
2036 | } |
2037 | ||
d6f38f31 | 2038 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 2039 | to_addr_conv(sh, percpu, 0)); |
c911c46c YY |
2040 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, |
2041 | RAID5_STRIPE_SIZE(sh->raid_conf), | |
a08abd8c | 2042 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 2043 | |
91c00924 | 2044 | atomic_inc(&sh->count); |
a08abd8c DW |
2045 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
2046 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
2047 | } |
2048 | ||
ac6b53b6 DW |
2049 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
2050 | { | |
46d5b785 | 2051 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
2052 | struct async_submit_ctl submit; |
2053 | int count; | |
2054 | ||
2055 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
2056 | (unsigned long long)sh->sector, checkp); | |
2057 | ||
59fc630b | 2058 | BUG_ON(sh->batch_head); |
584acdd4 | 2059 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
2060 | if (!checkp) |
2061 | srcs[count] = NULL; | |
91c00924 | 2062 | |
91c00924 | 2063 | atomic_inc(&sh->count); |
ac6b53b6 | 2064 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 2065 | sh, to_addr_conv(sh, percpu, 0)); |
c911c46c YY |
2066 | async_syndrome_val(srcs, 0, count+2, |
2067 | RAID5_STRIPE_SIZE(sh->raid_conf), | |
ac6b53b6 | 2068 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); |
91c00924 DW |
2069 | } |
2070 | ||
51acbcec | 2071 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
2072 | { |
2073 | int overlap_clear = 0, i, disks = sh->disks; | |
2074 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 2075 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2076 | int level = conf->level; |
d6f38f31 DW |
2077 | struct raid5_percpu *percpu; |
2078 | unsigned long cpu; | |
91c00924 | 2079 | |
d6f38f31 DW |
2080 | cpu = get_cpu(); |
2081 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2082 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2083 | ops_run_biofill(sh); |
2084 | overlap_clear++; | |
2085 | } | |
2086 | ||
7b3a871e | 2087 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2088 | if (level < 6) |
2089 | tx = ops_run_compute5(sh, percpu); | |
2090 | else { | |
2091 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2092 | tx = ops_run_compute6_1(sh, percpu); | |
2093 | else | |
2094 | tx = ops_run_compute6_2(sh, percpu); | |
2095 | } | |
2096 | /* terminate the chain if reconstruct is not set to be run */ | |
2097 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2098 | async_tx_ack(tx); |
2099 | } | |
91c00924 | 2100 | |
584acdd4 MS |
2101 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2102 | if (level < 6) | |
2103 | tx = ops_run_prexor5(sh, percpu, tx); | |
2104 | else | |
2105 | tx = ops_run_prexor6(sh, percpu, tx); | |
2106 | } | |
91c00924 | 2107 | |
ae1713e2 AP |
2108 | if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request)) |
2109 | tx = ops_run_partial_parity(sh, percpu, tx); | |
2110 | ||
600aa109 | 2111 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2112 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2113 | overlap_clear++; |
2114 | } | |
2115 | ||
ac6b53b6 DW |
2116 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2117 | if (level < 6) | |
2118 | ops_run_reconstruct5(sh, percpu, tx); | |
2119 | else | |
2120 | ops_run_reconstruct6(sh, percpu, tx); | |
2121 | } | |
91c00924 | 2122 | |
ac6b53b6 DW |
2123 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2124 | if (sh->check_state == check_state_run) | |
2125 | ops_run_check_p(sh, percpu); | |
2126 | else if (sh->check_state == check_state_run_q) | |
2127 | ops_run_check_pq(sh, percpu, 0); | |
2128 | else if (sh->check_state == check_state_run_pq) | |
2129 | ops_run_check_pq(sh, percpu, 1); | |
2130 | else | |
2131 | BUG(); | |
2132 | } | |
91c00924 | 2133 | |
59fc630b | 2134 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2135 | for (i = disks; i--; ) { |
2136 | struct r5dev *dev = &sh->dev[i]; | |
2137 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2138 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2139 | } | |
d6f38f31 | 2140 | put_cpu(); |
91c00924 DW |
2141 | } |
2142 | ||
845b9e22 AP |
2143 | static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh) |
2144 | { | |
2145 | if (sh->ppl_page) | |
2146 | __free_page(sh->ppl_page); | |
2147 | kmem_cache_free(sc, sh); | |
2148 | } | |
2149 | ||
5f9d1fde | 2150 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
845b9e22 | 2151 | int disks, struct r5conf *conf) |
f18c1a35 N |
2152 | { |
2153 | struct stripe_head *sh; | |
5f9d1fde | 2154 | int i; |
f18c1a35 N |
2155 | |
2156 | sh = kmem_cache_zalloc(sc, gfp); | |
2157 | if (sh) { | |
2158 | spin_lock_init(&sh->stripe_lock); | |
2159 | spin_lock_init(&sh->batch_lock); | |
2160 | INIT_LIST_HEAD(&sh->batch_list); | |
2161 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2162 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2163 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2164 | atomic_set(&sh->count, 1); |
845b9e22 | 2165 | sh->raid_conf = conf; |
a39f7afd | 2166 | sh->log_start = MaxSector; |
5f9d1fde SL |
2167 | for (i = 0; i < disks; i++) { |
2168 | struct r5dev *dev = &sh->dev[i]; | |
2169 | ||
3a83f467 ML |
2170 | bio_init(&dev->req, &dev->vec, 1); |
2171 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2172 | } |
845b9e22 AP |
2173 | |
2174 | if (raid5_has_ppl(conf)) { | |
2175 | sh->ppl_page = alloc_page(gfp); | |
2176 | if (!sh->ppl_page) { | |
2177 | free_stripe(sc, sh); | |
2178 | sh = NULL; | |
2179 | } | |
2180 | } | |
f18c1a35 N |
2181 | } |
2182 | return sh; | |
2183 | } | |
486f0644 | 2184 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2185 | { |
2186 | struct stripe_head *sh; | |
f18c1a35 | 2187 | |
845b9e22 | 2188 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf); |
3f294f4f N |
2189 | if (!sh) |
2190 | return 0; | |
6ce32846 | 2191 | |
a9683a79 | 2192 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2193 | shrink_buffers(sh); |
845b9e22 | 2194 | free_stripe(conf->slab_cache, sh); |
3f294f4f N |
2195 | return 0; |
2196 | } | |
486f0644 N |
2197 | sh->hash_lock_index = |
2198 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2199 | /* we just created an active stripe so... */ |
3f294f4f | 2200 | atomic_inc(&conf->active_stripes); |
59fc630b | 2201 | |
6d036f7d | 2202 | raid5_release_stripe(sh); |
486f0644 | 2203 | conf->max_nr_stripes++; |
3f294f4f N |
2204 | return 1; |
2205 | } | |
2206 | ||
d1688a6d | 2207 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2208 | { |
e18b890b | 2209 | struct kmem_cache *sc; |
53b8d89d | 2210 | size_t namelen = sizeof(conf->cache_name[0]); |
5e5e3e78 | 2211 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2212 | |
f4be6b43 | 2213 | if (conf->mddev->gendisk) |
53b8d89d | 2214 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 N |
2215 | "raid%d-%s", conf->level, mdname(conf->mddev)); |
2216 | else | |
53b8d89d | 2217 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 | 2218 | "raid%d-%p", conf->level, conf->mddev); |
53b8d89d | 2219 | snprintf(conf->cache_name[1], namelen, "%.27s-alt", conf->cache_name[0]); |
f4be6b43 | 2220 | |
ad01c9e3 N |
2221 | conf->active_name = 0; |
2222 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2223 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2224 | 0, 0, NULL); |
1da177e4 LT |
2225 | if (!sc) |
2226 | return 1; | |
2227 | conf->slab_cache = sc; | |
ad01c9e3 | 2228 | conf->pool_size = devs; |
486f0644 N |
2229 | while (num--) |
2230 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2231 | return 1; |
486f0644 | 2232 | |
1da177e4 LT |
2233 | return 0; |
2234 | } | |
29269553 | 2235 | |
d6f38f31 | 2236 | /** |
7f8a30e5 CL |
2237 | * scribble_alloc - allocate percpu scribble buffer for required size |
2238 | * of the scribble region | |
2aada5b1 DLM |
2239 | * @percpu: from for_each_present_cpu() of the caller |
2240 | * @num: total number of disks in the array | |
2241 | * @cnt: scribble objs count for required size of the scribble region | |
d6f38f31 | 2242 | * |
7f8a30e5 | 2243 | * The scribble buffer size must be enough to contain: |
d6f38f31 DW |
2244 | * 1/ a struct page pointer for each device in the array +2 |
2245 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2246 | * (dma_map_page()) or page (page_address()) address. | |
2247 | * | |
2248 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2249 | * calculate over all devices (not just the data blocks), using zeros in place | |
2250 | * of the P and Q blocks. | |
2251 | */ | |
b330e6a4 | 2252 | static int scribble_alloc(struct raid5_percpu *percpu, |
ba54d4d4 | 2253 | int num, int cnt) |
d6f38f31 | 2254 | { |
b330e6a4 KO |
2255 | size_t obj_size = |
2256 | sizeof(struct page *) * (num+2) + | |
2257 | sizeof(addr_conv_t) * (num+2); | |
2258 | void *scribble; | |
d6f38f31 | 2259 | |
ba54d4d4 CL |
2260 | /* |
2261 | * If here is in raid array suspend context, it is in memalloc noio | |
2262 | * context as well, there is no potential recursive memory reclaim | |
2263 | * I/Os with the GFP_KERNEL flag. | |
2264 | */ | |
2265 | scribble = kvmalloc_array(cnt, obj_size, GFP_KERNEL); | |
b330e6a4 KO |
2266 | if (!scribble) |
2267 | return -ENOMEM; | |
2268 | ||
2269 | kvfree(percpu->scribble); | |
2270 | ||
2271 | percpu->scribble = scribble; | |
2272 | percpu->scribble_obj_size = obj_size; | |
2273 | return 0; | |
d6f38f31 DW |
2274 | } |
2275 | ||
738a2738 N |
2276 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2277 | { | |
2278 | unsigned long cpu; | |
2279 | int err = 0; | |
2280 | ||
27a353c0 SL |
2281 | /* |
2282 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2283 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2284 | * should equal to new_disks and new_sectors | |
2285 | */ | |
2286 | if (conf->scribble_disks >= new_disks && | |
2287 | conf->scribble_sectors >= new_sectors) | |
2288 | return 0; | |
738a2738 N |
2289 | mddev_suspend(conf->mddev); |
2290 | get_online_cpus(); | |
b330e6a4 | 2291 | |
738a2738 N |
2292 | for_each_present_cpu(cpu) { |
2293 | struct raid5_percpu *percpu; | |
738a2738 N |
2294 | |
2295 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
b330e6a4 | 2296 | err = scribble_alloc(percpu, new_disks, |
c911c46c | 2297 | new_sectors / RAID5_STRIPE_SECTORS(conf)); |
b330e6a4 | 2298 | if (err) |
738a2738 | 2299 | break; |
738a2738 | 2300 | } |
b330e6a4 | 2301 | |
738a2738 N |
2302 | put_online_cpus(); |
2303 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2304 | if (!err) { |
2305 | conf->scribble_disks = new_disks; | |
2306 | conf->scribble_sectors = new_sectors; | |
2307 | } | |
738a2738 N |
2308 | return err; |
2309 | } | |
2310 | ||
d1688a6d | 2311 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2312 | { |
2313 | /* Make all the stripes able to hold 'newsize' devices. | |
2314 | * New slots in each stripe get 'page' set to a new page. | |
2315 | * | |
2316 | * This happens in stages: | |
2317 | * 1/ create a new kmem_cache and allocate the required number of | |
2318 | * stripe_heads. | |
83f0d77a | 2319 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2320 | * to the new stripe_heads. This will have the side effect of |
2321 | * freezing the array as once all stripe_heads have been collected, | |
2322 | * no IO will be possible. Old stripe heads are freed once their | |
2323 | * pages have been transferred over, and the old kmem_cache is | |
2324 | * freed when all stripes are done. | |
2325 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
3560741e | 2326 | * we simple return a failure status - no need to clean anything up. |
ad01c9e3 N |
2327 | * 4/ allocate new pages for the new slots in the new stripe_heads. |
2328 | * If this fails, we don't bother trying the shrink the | |
2329 | * stripe_heads down again, we just leave them as they are. | |
2330 | * As each stripe_head is processed the new one is released into | |
2331 | * active service. | |
2332 | * | |
2333 | * Once step2 is started, we cannot afford to wait for a write, | |
2334 | * so we use GFP_NOIO allocations. | |
2335 | */ | |
2336 | struct stripe_head *osh, *nsh; | |
2337 | LIST_HEAD(newstripes); | |
2338 | struct disk_info *ndisks; | |
2214c260 | 2339 | int err = 0; |
e18b890b | 2340 | struct kmem_cache *sc; |
ad01c9e3 | 2341 | int i; |
566c09c5 | 2342 | int hash, cnt; |
ad01c9e3 | 2343 | |
2214c260 | 2344 | md_allow_write(conf->mddev); |
2a2275d6 | 2345 | |
ad01c9e3 N |
2346 | /* Step 1 */ |
2347 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2348 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2349 | 0, 0, NULL); |
ad01c9e3 N |
2350 | if (!sc) |
2351 | return -ENOMEM; | |
2352 | ||
2d5b569b N |
2353 | /* Need to ensure auto-resizing doesn't interfere */ |
2354 | mutex_lock(&conf->cache_size_mutex); | |
2355 | ||
ad01c9e3 | 2356 | for (i = conf->max_nr_stripes; i; i--) { |
845b9e22 | 2357 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf); |
ad01c9e3 N |
2358 | if (!nsh) |
2359 | break; | |
2360 | ||
ad01c9e3 N |
2361 | list_add(&nsh->lru, &newstripes); |
2362 | } | |
2363 | if (i) { | |
2364 | /* didn't get enough, give up */ | |
2365 | while (!list_empty(&newstripes)) { | |
2366 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2367 | list_del(&nsh->lru); | |
845b9e22 | 2368 | free_stripe(sc, nsh); |
ad01c9e3 N |
2369 | } |
2370 | kmem_cache_destroy(sc); | |
2d5b569b | 2371 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2372 | return -ENOMEM; |
2373 | } | |
2374 | /* Step 2 - Must use GFP_NOIO now. | |
2375 | * OK, we have enough stripes, start collecting inactive | |
2376 | * stripes and copying them over | |
2377 | */ | |
566c09c5 SL |
2378 | hash = 0; |
2379 | cnt = 0; | |
ad01c9e3 | 2380 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2381 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2382 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2383 | !list_empty(conf->inactive_list + hash), |
2384 | unlock_device_hash_lock(conf, hash), | |
2385 | lock_device_hash_lock(conf, hash)); | |
2386 | osh = get_free_stripe(conf, hash); | |
2387 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2388 | |
d592a996 | 2389 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2390 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2391 | nsh->dev[i].orig_page = osh->dev[i].page; |
2392 | } | |
566c09c5 | 2393 | nsh->hash_lock_index = hash; |
845b9e22 | 2394 | free_stripe(conf->slab_cache, osh); |
566c09c5 SL |
2395 | cnt++; |
2396 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2397 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2398 | hash++; | |
2399 | cnt = 0; | |
2400 | } | |
ad01c9e3 N |
2401 | } |
2402 | kmem_cache_destroy(conf->slab_cache); | |
2403 | ||
2404 | /* Step 3. | |
2405 | * At this point, we are holding all the stripes so the array | |
2406 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2407 | * conf->disks and the scribble region |
ad01c9e3 | 2408 | */ |
6396bb22 | 2409 | ndisks = kcalloc(newsize, sizeof(struct disk_info), GFP_NOIO); |
ad01c9e3 | 2410 | if (ndisks) { |
d7bd398e | 2411 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2412 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2413 | |
2414 | for (i = conf->pool_size; i < newsize; i++) { | |
2415 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2416 | if (!ndisks[i].extra_page) | |
2417 | err = -ENOMEM; | |
2418 | } | |
2419 | ||
2420 | if (err) { | |
2421 | for (i = conf->pool_size; i < newsize; i++) | |
2422 | if (ndisks[i].extra_page) | |
2423 | put_page(ndisks[i].extra_page); | |
2424 | kfree(ndisks); | |
2425 | } else { | |
2426 | kfree(conf->disks); | |
2427 | conf->disks = ndisks; | |
2428 | } | |
ad01c9e3 N |
2429 | } else |
2430 | err = -ENOMEM; | |
2431 | ||
2d5b569b | 2432 | mutex_unlock(&conf->cache_size_mutex); |
583da48e DY |
2433 | |
2434 | conf->slab_cache = sc; | |
2435 | conf->active_name = 1-conf->active_name; | |
2436 | ||
ad01c9e3 N |
2437 | /* Step 4, return new stripes to service */ |
2438 | while(!list_empty(&newstripes)) { | |
2439 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2440 | list_del_init(&nsh->lru); | |
d6f38f31 | 2441 | |
ad01c9e3 N |
2442 | for (i=conf->raid_disks; i < newsize; i++) |
2443 | if (nsh->dev[i].page == NULL) { | |
2444 | struct page *p = alloc_page(GFP_NOIO); | |
2445 | nsh->dev[i].page = p; | |
d592a996 | 2446 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2447 | if (!p) |
2448 | err = -ENOMEM; | |
2449 | } | |
6d036f7d | 2450 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2451 | } |
2452 | /* critical section pass, GFP_NOIO no longer needed */ | |
2453 | ||
6e9eac2d N |
2454 | if (!err) |
2455 | conf->pool_size = newsize; | |
ad01c9e3 N |
2456 | return err; |
2457 | } | |
1da177e4 | 2458 | |
486f0644 | 2459 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2460 | { |
2461 | struct stripe_head *sh; | |
49895bcc | 2462 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2463 | |
566c09c5 SL |
2464 | spin_lock_irq(conf->hash_locks + hash); |
2465 | sh = get_free_stripe(conf, hash); | |
2466 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2467 | if (!sh) |
2468 | return 0; | |
78bafebd | 2469 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2470 | shrink_buffers(sh); |
845b9e22 | 2471 | free_stripe(conf->slab_cache, sh); |
3f294f4f | 2472 | atomic_dec(&conf->active_stripes); |
486f0644 | 2473 | conf->max_nr_stripes--; |
3f294f4f N |
2474 | return 1; |
2475 | } | |
2476 | ||
d1688a6d | 2477 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2478 | { |
486f0644 N |
2479 | while (conf->max_nr_stripes && |
2480 | drop_one_stripe(conf)) | |
2481 | ; | |
3f294f4f | 2482 | |
644df1a8 | 2483 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2484 | conf->slab_cache = NULL; |
2485 | } | |
2486 | ||
4246a0b6 | 2487 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2488 | { |
99c0fb5f | 2489 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2490 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2491 | int disks = sh->disks, i; |
d6950432 | 2492 | char b[BDEVNAME_SIZE]; |
dd054fce | 2493 | struct md_rdev *rdev = NULL; |
05616be5 | 2494 | sector_t s; |
1da177e4 LT |
2495 | |
2496 | for (i=0 ; i<disks; i++) | |
2497 | if (bi == &sh->dev[i].req) | |
2498 | break; | |
2499 | ||
4246a0b6 | 2500 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2501 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2502 | bi->bi_status); |
1da177e4 | 2503 | if (i == disks) { |
5f9d1fde | 2504 | bio_reset(bi); |
1da177e4 | 2505 | BUG(); |
6712ecf8 | 2506 | return; |
1da177e4 | 2507 | } |
14a75d3e | 2508 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2509 | /* If replacement finished while this request was outstanding, |
2510 | * 'replacement' might be NULL already. | |
2511 | * In that case it moved down to 'rdev'. | |
2512 | * rdev is not removed until all requests are finished. | |
2513 | */ | |
14a75d3e | 2514 | rdev = conf->disks[i].replacement; |
dd054fce | 2515 | if (!rdev) |
14a75d3e | 2516 | rdev = conf->disks[i].rdev; |
1da177e4 | 2517 | |
05616be5 N |
2518 | if (use_new_offset(conf, sh)) |
2519 | s = sh->sector + rdev->new_data_offset; | |
2520 | else | |
2521 | s = sh->sector + rdev->data_offset; | |
4e4cbee9 | 2522 | if (!bi->bi_status) { |
1da177e4 | 2523 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2524 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2525 | /* Note that this cannot happen on a |
2526 | * replacement device. We just fail those on | |
2527 | * any error | |
2528 | */ | |
cc6167b4 N |
2529 | pr_info_ratelimited( |
2530 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
c911c46c | 2531 | mdname(conf->mddev), RAID5_STRIPE_SECTORS(conf), |
05616be5 | 2532 | (unsigned long long)s, |
8bda470e | 2533 | bdevname(rdev->bdev, b)); |
c911c46c | 2534 | atomic_add(RAID5_STRIPE_SECTORS(conf), &rdev->corrected_errors); |
4e5314b5 N |
2535 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2536 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2537 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2538 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2539 | ||
86aa1397 SL |
2540 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2541 | /* | |
2542 | * end read for a page in journal, this | |
2543 | * must be preparing for prexor in rmw | |
2544 | */ | |
2545 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2546 | ||
14a75d3e N |
2547 | if (atomic_read(&rdev->read_errors)) |
2548 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2549 | } else { |
14a75d3e | 2550 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2551 | int retry = 0; |
2e8ac303 | 2552 | int set_bad = 0; |
d6950432 | 2553 | |
1da177e4 | 2554 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
b76b4715 NC |
2555 | if (!(bi->bi_status == BLK_STS_PROTECTION)) |
2556 | atomic_inc(&rdev->read_errors); | |
14a75d3e | 2557 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2558 | pr_warn_ratelimited( |
2559 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2560 | mdname(conf->mddev), |
05616be5 | 2561 | (unsigned long long)s, |
14a75d3e | 2562 | bdn); |
2e8ac303 | 2563 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2564 | set_bad = 1; | |
cc6167b4 N |
2565 | pr_warn_ratelimited( |
2566 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2567 | mdname(conf->mddev), |
05616be5 | 2568 | (unsigned long long)s, |
8bda470e | 2569 | bdn); |
2e8ac303 | 2570 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2571 | /* Oh, no!!! */ |
2e8ac303 | 2572 | set_bad = 1; |
cc6167b4 N |
2573 | pr_warn_ratelimited( |
2574 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2575 | mdname(conf->mddev), |
05616be5 | 2576 | (unsigned long long)s, |
8bda470e | 2577 | bdn); |
2e8ac303 | 2578 | } else if (atomic_read(&rdev->read_errors) |
0009fad0 NC |
2579 | > conf->max_nr_stripes) { |
2580 | if (!test_bit(Faulty, &rdev->flags)) { | |
2581 | pr_warn("md/raid:%s: %d read_errors > %d stripes\n", | |
2582 | mdname(conf->mddev), | |
2583 | atomic_read(&rdev->read_errors), | |
2584 | conf->max_nr_stripes); | |
2585 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", | |
2586 | mdname(conf->mddev), bdn); | |
2587 | } | |
2588 | } else | |
ba22dcbf | 2589 | retry = 1; |
edfa1f65 BY |
2590 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2591 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2592 | retry = 1; | |
ba22dcbf | 2593 | if (retry) |
143f6e73 XN |
2594 | if (sh->qd_idx >= 0 && sh->pd_idx == i) |
2595 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2596 | else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { | |
3f9e7c14 | 2597 | set_bit(R5_ReadError, &sh->dev[i].flags); |
2598 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2599 | } else | |
2600 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2601 | else { |
4e5314b5 N |
2602 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2603 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2604 | if (!(set_bad |
2605 | && test_bit(In_sync, &rdev->flags) | |
2606 | && rdev_set_badblocks( | |
c911c46c | 2607 | rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), 0))) |
2e8ac303 | 2608 | md_error(conf->mddev, rdev); |
ba22dcbf | 2609 | } |
1da177e4 | 2610 | } |
14a75d3e | 2611 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2612 | bio_reset(bi); |
1da177e4 LT |
2613 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2614 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2615 | raid5_release_stripe(sh); |
1da177e4 LT |
2616 | } |
2617 | ||
4246a0b6 | 2618 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2619 | { |
99c0fb5f | 2620 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2621 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2622 | int disks = sh->disks, i; |
3f649ab7 | 2623 | struct md_rdev *rdev; |
b84db560 N |
2624 | sector_t first_bad; |
2625 | int bad_sectors; | |
977df362 | 2626 | int replacement = 0; |
1da177e4 | 2627 | |
977df362 N |
2628 | for (i = 0 ; i < disks; i++) { |
2629 | if (bi == &sh->dev[i].req) { | |
2630 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2631 | break; |
977df362 N |
2632 | } |
2633 | if (bi == &sh->dev[i].rreq) { | |
2634 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2635 | if (rdev) |
2636 | replacement = 1; | |
2637 | else | |
2638 | /* rdev was removed and 'replacement' | |
2639 | * replaced it. rdev is not removed | |
2640 | * until all requests are finished. | |
2641 | */ | |
2642 | rdev = conf->disks[i].rdev; | |
977df362 N |
2643 | break; |
2644 | } | |
2645 | } | |
4246a0b6 | 2646 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2647 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2648 | bi->bi_status); |
1da177e4 | 2649 | if (i == disks) { |
5f9d1fde | 2650 | bio_reset(bi); |
1da177e4 | 2651 | BUG(); |
6712ecf8 | 2652 | return; |
1da177e4 LT |
2653 | } |
2654 | ||
977df362 | 2655 | if (replacement) { |
4e4cbee9 | 2656 | if (bi->bi_status) |
977df362 N |
2657 | md_error(conf->mddev, rdev); |
2658 | else if (is_badblock(rdev, sh->sector, | |
c911c46c | 2659 | RAID5_STRIPE_SECTORS(conf), |
977df362 N |
2660 | &first_bad, &bad_sectors)) |
2661 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2662 | } else { | |
4e4cbee9 | 2663 | if (bi->bi_status) { |
9f97e4b1 | 2664 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2665 | set_bit(WriteErrorSeen, &rdev->flags); |
2666 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2667 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2668 | set_bit(MD_RECOVERY_NEEDED, | |
2669 | &rdev->mddev->recovery); | |
977df362 | 2670 | } else if (is_badblock(rdev, sh->sector, |
c911c46c | 2671 | RAID5_STRIPE_SECTORS(conf), |
c0b32972 | 2672 | &first_bad, &bad_sectors)) { |
977df362 | 2673 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2674 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2675 | /* That was a successful write so make | |
2676 | * sure it looks like we already did | |
2677 | * a re-write. | |
2678 | */ | |
2679 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2680 | } | |
977df362 N |
2681 | } |
2682 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2683 | |
4e4cbee9 | 2684 | if (sh->batch_head && bi->bi_status && !replacement) |
72ac7330 | 2685 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2686 | ||
c9445555 | 2687 | bio_reset(bi); |
977df362 N |
2688 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2689 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2690 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2691 | raid5_release_stripe(sh); |
59fc630b | 2692 | |
2693 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2694 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2695 | } |
2696 | ||
849674e4 | 2697 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2698 | { |
2699 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2700 | struct r5conf *conf = mddev->private; |
908f4fbd | 2701 | unsigned long flags; |
0c55e022 | 2702 | pr_debug("raid456: error called\n"); |
1da177e4 | 2703 | |
908f4fbd | 2704 | spin_lock_irqsave(&conf->device_lock, flags); |
fb73b357 MT |
2705 | |
2706 | if (test_bit(In_sync, &rdev->flags) && | |
2707 | mddev->degraded == conf->max_degraded) { | |
2708 | /* | |
2709 | * Don't allow to achieve failed state | |
2710 | * Don't try to recover this device | |
2711 | */ | |
2712 | conf->recovery_disabled = mddev->recovery_disabled; | |
2713 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2714 | return; | |
2715 | } | |
2716 | ||
aff69d89 | 2717 | set_bit(Faulty, &rdev->flags); |
908f4fbd | 2718 | clear_bit(In_sync, &rdev->flags); |
2e38a37f | 2719 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2720 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2721 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2722 | ||
de393cde | 2723 | set_bit(Blocked, &rdev->flags); |
2953079c SL |
2724 | set_mask_bits(&mddev->sb_flags, 0, |
2725 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2726 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2727 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2728 | mdname(mddev), | |
2729 | bdevname(rdev->bdev, b), | |
2730 | mdname(mddev), | |
2731 | conf->raid_disks - mddev->degraded); | |
70d466f7 | 2732 | r5c_update_on_rdev_error(mddev, rdev); |
16a53ecc | 2733 | } |
1da177e4 LT |
2734 | |
2735 | /* | |
2736 | * Input: a 'big' sector number, | |
2737 | * Output: index of the data and parity disk, and the sector # in them. | |
2738 | */ | |
6d036f7d SL |
2739 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2740 | int previous, int *dd_idx, | |
2741 | struct stripe_head *sh) | |
1da177e4 | 2742 | { |
6e3b96ed | 2743 | sector_t stripe, stripe2; |
35f2a591 | 2744 | sector_t chunk_number; |
1da177e4 | 2745 | unsigned int chunk_offset; |
911d4ee8 | 2746 | int pd_idx, qd_idx; |
67cc2b81 | 2747 | int ddf_layout = 0; |
1da177e4 | 2748 | sector_t new_sector; |
e183eaed N |
2749 | int algorithm = previous ? conf->prev_algo |
2750 | : conf->algorithm; | |
09c9e5fa AN |
2751 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2752 | : conf->chunk_sectors; | |
112bf897 N |
2753 | int raid_disks = previous ? conf->previous_raid_disks |
2754 | : conf->raid_disks; | |
2755 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2756 | |
2757 | /* First compute the information on this sector */ | |
2758 | ||
2759 | /* | |
2760 | * Compute the chunk number and the sector offset inside the chunk | |
2761 | */ | |
2762 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2763 | chunk_number = r_sector; | |
1da177e4 LT |
2764 | |
2765 | /* | |
2766 | * Compute the stripe number | |
2767 | */ | |
35f2a591 N |
2768 | stripe = chunk_number; |
2769 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2770 | stripe2 = stripe; |
1da177e4 LT |
2771 | /* |
2772 | * Select the parity disk based on the user selected algorithm. | |
2773 | */ | |
84789554 | 2774 | pd_idx = qd_idx = -1; |
16a53ecc N |
2775 | switch(conf->level) { |
2776 | case 4: | |
911d4ee8 | 2777 | pd_idx = data_disks; |
16a53ecc N |
2778 | break; |
2779 | case 5: | |
e183eaed | 2780 | switch (algorithm) { |
1da177e4 | 2781 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2782 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2783 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2784 | (*dd_idx)++; |
2785 | break; | |
2786 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2787 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2788 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2789 | (*dd_idx)++; |
2790 | break; | |
2791 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2792 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2793 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2794 | break; |
2795 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2796 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2797 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2798 | break; |
99c0fb5f N |
2799 | case ALGORITHM_PARITY_0: |
2800 | pd_idx = 0; | |
2801 | (*dd_idx)++; | |
2802 | break; | |
2803 | case ALGORITHM_PARITY_N: | |
2804 | pd_idx = data_disks; | |
2805 | break; | |
1da177e4 | 2806 | default: |
99c0fb5f | 2807 | BUG(); |
16a53ecc N |
2808 | } |
2809 | break; | |
2810 | case 6: | |
2811 | ||
e183eaed | 2812 | switch (algorithm) { |
16a53ecc | 2813 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2814 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2815 | qd_idx = pd_idx + 1; |
2816 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2817 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2818 | qd_idx = 0; |
2819 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2820 | (*dd_idx) += 2; /* D D P Q D */ |
2821 | break; | |
2822 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2823 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2824 | qd_idx = pd_idx + 1; |
2825 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2826 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2827 | qd_idx = 0; |
2828 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2829 | (*dd_idx) += 2; /* D D P Q D */ |
2830 | break; | |
2831 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2832 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2833 | qd_idx = (pd_idx + 1) % raid_disks; |
2834 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2835 | break; |
2836 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2837 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2838 | qd_idx = (pd_idx + 1) % raid_disks; |
2839 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2840 | break; |
99c0fb5f N |
2841 | |
2842 | case ALGORITHM_PARITY_0: | |
2843 | pd_idx = 0; | |
2844 | qd_idx = 1; | |
2845 | (*dd_idx) += 2; | |
2846 | break; | |
2847 | case ALGORITHM_PARITY_N: | |
2848 | pd_idx = data_disks; | |
2849 | qd_idx = data_disks + 1; | |
2850 | break; | |
2851 | ||
2852 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2853 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2854 | * of blocks for computing Q is different. | |
2855 | */ | |
6e3b96ed | 2856 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2857 | qd_idx = pd_idx + 1; |
2858 | if (pd_idx == raid_disks-1) { | |
2859 | (*dd_idx)++; /* Q D D D P */ | |
2860 | qd_idx = 0; | |
2861 | } else if (*dd_idx >= pd_idx) | |
2862 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2863 | ddf_layout = 1; |
99c0fb5f N |
2864 | break; |
2865 | ||
2866 | case ALGORITHM_ROTATING_N_RESTART: | |
2867 | /* Same a left_asymmetric, by first stripe is | |
2868 | * D D D P Q rather than | |
2869 | * Q D D D P | |
2870 | */ | |
6e3b96ed N |
2871 | stripe2 += 1; |
2872 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2873 | qd_idx = pd_idx + 1; |
2874 | if (pd_idx == raid_disks-1) { | |
2875 | (*dd_idx)++; /* Q D D D P */ | |
2876 | qd_idx = 0; | |
2877 | } else if (*dd_idx >= pd_idx) | |
2878 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2879 | ddf_layout = 1; |
99c0fb5f N |
2880 | break; |
2881 | ||
2882 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2883 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2884 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2885 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2886 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2887 | ddf_layout = 1; |
99c0fb5f N |
2888 | break; |
2889 | ||
2890 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2891 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2892 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2893 | if (*dd_idx >= pd_idx) |
2894 | (*dd_idx)++; | |
2895 | qd_idx = raid_disks - 1; | |
2896 | break; | |
2897 | ||
2898 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2899 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2900 | if (*dd_idx >= pd_idx) |
2901 | (*dd_idx)++; | |
2902 | qd_idx = raid_disks - 1; | |
2903 | break; | |
2904 | ||
2905 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2906 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2907 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2908 | qd_idx = raid_disks - 1; | |
2909 | break; | |
2910 | ||
2911 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2912 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2913 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2914 | qd_idx = raid_disks - 1; | |
2915 | break; | |
2916 | ||
2917 | case ALGORITHM_PARITY_0_6: | |
2918 | pd_idx = 0; | |
2919 | (*dd_idx)++; | |
2920 | qd_idx = raid_disks - 1; | |
2921 | break; | |
2922 | ||
16a53ecc | 2923 | default: |
99c0fb5f | 2924 | BUG(); |
16a53ecc N |
2925 | } |
2926 | break; | |
1da177e4 LT |
2927 | } |
2928 | ||
911d4ee8 N |
2929 | if (sh) { |
2930 | sh->pd_idx = pd_idx; | |
2931 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2932 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2933 | } |
1da177e4 LT |
2934 | /* |
2935 | * Finally, compute the new sector number | |
2936 | */ | |
2937 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2938 | return new_sector; | |
2939 | } | |
2940 | ||
6d036f7d | 2941 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2942 | { |
d1688a6d | 2943 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2944 | int raid_disks = sh->disks; |
2945 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2946 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2947 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2948 | : conf->chunk_sectors; | |
e183eaed N |
2949 | int algorithm = previous ? conf->prev_algo |
2950 | : conf->algorithm; | |
1da177e4 LT |
2951 | sector_t stripe; |
2952 | int chunk_offset; | |
35f2a591 N |
2953 | sector_t chunk_number; |
2954 | int dummy1, dd_idx = i; | |
1da177e4 | 2955 | sector_t r_sector; |
911d4ee8 | 2956 | struct stripe_head sh2; |
1da177e4 LT |
2957 | |
2958 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2959 | stripe = new_sector; | |
1da177e4 | 2960 | |
16a53ecc N |
2961 | if (i == sh->pd_idx) |
2962 | return 0; | |
2963 | switch(conf->level) { | |
2964 | case 4: break; | |
2965 | case 5: | |
e183eaed | 2966 | switch (algorithm) { |
1da177e4 LT |
2967 | case ALGORITHM_LEFT_ASYMMETRIC: |
2968 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2969 | if (i > sh->pd_idx) | |
2970 | i--; | |
2971 | break; | |
2972 | case ALGORITHM_LEFT_SYMMETRIC: | |
2973 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2974 | if (i < sh->pd_idx) | |
2975 | i += raid_disks; | |
2976 | i -= (sh->pd_idx + 1); | |
2977 | break; | |
99c0fb5f N |
2978 | case ALGORITHM_PARITY_0: |
2979 | i -= 1; | |
2980 | break; | |
2981 | case ALGORITHM_PARITY_N: | |
2982 | break; | |
1da177e4 | 2983 | default: |
99c0fb5f | 2984 | BUG(); |
16a53ecc N |
2985 | } |
2986 | break; | |
2987 | case 6: | |
d0dabf7e | 2988 | if (i == sh->qd_idx) |
16a53ecc | 2989 | return 0; /* It is the Q disk */ |
e183eaed | 2990 | switch (algorithm) { |
16a53ecc N |
2991 | case ALGORITHM_LEFT_ASYMMETRIC: |
2992 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2993 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2994 | case ALGORITHM_ROTATING_N_RESTART: | |
2995 | if (sh->pd_idx == raid_disks-1) | |
2996 | i--; /* Q D D D P */ | |
16a53ecc N |
2997 | else if (i > sh->pd_idx) |
2998 | i -= 2; /* D D P Q D */ | |
2999 | break; | |
3000 | case ALGORITHM_LEFT_SYMMETRIC: | |
3001 | case ALGORITHM_RIGHT_SYMMETRIC: | |
3002 | if (sh->pd_idx == raid_disks-1) | |
3003 | i--; /* Q D D D P */ | |
3004 | else { | |
3005 | /* D D P Q D */ | |
3006 | if (i < sh->pd_idx) | |
3007 | i += raid_disks; | |
3008 | i -= (sh->pd_idx + 2); | |
3009 | } | |
3010 | break; | |
99c0fb5f N |
3011 | case ALGORITHM_PARITY_0: |
3012 | i -= 2; | |
3013 | break; | |
3014 | case ALGORITHM_PARITY_N: | |
3015 | break; | |
3016 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 3017 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
3018 | if (sh->pd_idx == 0) |
3019 | i--; /* P D D D Q */ | |
e4424fee N |
3020 | else { |
3021 | /* D D Q P D */ | |
3022 | if (i < sh->pd_idx) | |
3023 | i += raid_disks; | |
3024 | i -= (sh->pd_idx + 1); | |
3025 | } | |
99c0fb5f N |
3026 | break; |
3027 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
3028 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
3029 | if (i > sh->pd_idx) | |
3030 | i--; | |
3031 | break; | |
3032 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
3033 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
3034 | if (i < sh->pd_idx) | |
3035 | i += data_disks + 1; | |
3036 | i -= (sh->pd_idx + 1); | |
3037 | break; | |
3038 | case ALGORITHM_PARITY_0_6: | |
3039 | i -= 1; | |
3040 | break; | |
16a53ecc | 3041 | default: |
99c0fb5f | 3042 | BUG(); |
16a53ecc N |
3043 | } |
3044 | break; | |
1da177e4 LT |
3045 | } |
3046 | ||
3047 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 3048 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 3049 | |
112bf897 | 3050 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 3051 | previous, &dummy1, &sh2); |
911d4ee8 N |
3052 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
3053 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
3054 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
3055 | mdname(conf->mddev)); | |
1da177e4 LT |
3056 | return 0; |
3057 | } | |
3058 | return r_sector; | |
3059 | } | |
3060 | ||
07e83364 SL |
3061 | /* |
3062 | * There are cases where we want handle_stripe_dirtying() and | |
3063 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
3064 | * | |
3065 | * This function checks whether we want to delay the towrite. Specifically, | |
3066 | * we delay the towrite when: | |
3067 | * | |
3068 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
3069 | * stripe has data in journal (for other devices). | |
3070 | * | |
3071 | * In this case, when reading data for the non-overwrite dev, it is | |
3072 | * necessary to handle complex rmw of write back cache (prexor with | |
3073 | * orig_page, and xor with page). To keep read path simple, we would | |
3074 | * like to flush data in journal to RAID disks first, so complex rmw | |
3075 | * is handled in the write patch (handle_stripe_dirtying). | |
3076 | * | |
39b99586 SL |
3077 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
3078 | * | |
3079 | * It is important to be able to flush all stripes in raid5-cache. | |
3080 | * Therefore, we need reserve some space on the journal device for | |
3081 | * these flushes. If flush operation includes pending writes to the | |
3082 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
3083 | * for the flush out. If we exclude these pending writes from flush | |
3084 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
3085 | * Therefore, excluding pending writes in these cases enables more | |
3086 | * efficient use of the journal device. | |
3087 | * | |
3088 | * Note: To make sure the stripe makes progress, we only delay | |
3089 | * towrite for stripes with data already in journal (injournal > 0). | |
3090 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
3091 | * no_space_stripes list. | |
3092 | * | |
70d466f7 SL |
3093 | * 3. during journal failure |
3094 | * In journal failure, we try to flush all cached data to raid disks | |
3095 | * based on data in stripe cache. The array is read-only to upper | |
3096 | * layers, so we would skip all pending writes. | |
3097 | * | |
07e83364 | 3098 | */ |
39b99586 SL |
3099 | static inline bool delay_towrite(struct r5conf *conf, |
3100 | struct r5dev *dev, | |
3101 | struct stripe_head_state *s) | |
07e83364 | 3102 | { |
39b99586 SL |
3103 | /* case 1 above */ |
3104 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
3105 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
3106 | return true; | |
3107 | /* case 2 above */ | |
3108 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
3109 | s->injournal > 0) | |
3110 | return true; | |
70d466f7 SL |
3111 | /* case 3 above */ |
3112 | if (s->log_failed && s->injournal) | |
3113 | return true; | |
39b99586 | 3114 | return false; |
07e83364 SL |
3115 | } |
3116 | ||
600aa109 | 3117 | static void |
c0f7bddb | 3118 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3119 | int rcw, int expand) |
e33129d8 | 3120 | { |
584acdd4 | 3121 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3122 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3123 | int level = conf->level; |
e33129d8 DW |
3124 | |
3125 | if (rcw) { | |
1e6d690b SL |
3126 | /* |
3127 | * In some cases, handle_stripe_dirtying initially decided to | |
3128 | * run rmw and allocates extra page for prexor. However, rcw is | |
3129 | * cheaper later on. We need to free the extra page now, | |
3130 | * because we won't be able to do that in ops_complete_prexor(). | |
3131 | */ | |
3132 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3133 | |
3134 | for (i = disks; i--; ) { | |
3135 | struct r5dev *dev = &sh->dev[i]; | |
3136 | ||
39b99586 | 3137 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3138 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3139 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3140 | if (!expand) |
3141 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3142 | s->locked++; |
1e6d690b SL |
3143 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3144 | set_bit(R5_LOCKED, &dev->flags); | |
3145 | s->locked++; | |
e33129d8 DW |
3146 | } |
3147 | } | |
ce7d363a N |
3148 | /* if we are not expanding this is a proper write request, and |
3149 | * there will be bios with new data to be drained into the | |
3150 | * stripe cache | |
3151 | */ | |
3152 | if (!expand) { | |
3153 | if (!s->locked) | |
3154 | /* False alarm, nothing to do */ | |
3155 | return; | |
3156 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3157 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3158 | } else | |
3159 | sh->reconstruct_state = reconstruct_state_run; | |
3160 | ||
3161 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3162 | ||
c0f7bddb | 3163 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3164 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3165 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3166 | } else { |
3167 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3168 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3169 | BUG_ON(level == 6 && |
3170 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3171 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3172 | |
e33129d8 DW |
3173 | for (i = disks; i--; ) { |
3174 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3175 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3176 | continue; |
3177 | ||
e33129d8 DW |
3178 | if (dev->towrite && |
3179 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3180 | test_bit(R5_Wantcompute, &dev->flags))) { |
3181 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3182 | set_bit(R5_LOCKED, &dev->flags); |
3183 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3184 | s->locked++; |
1e6d690b SL |
3185 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3186 | set_bit(R5_LOCKED, &dev->flags); | |
3187 | s->locked++; | |
e33129d8 DW |
3188 | } |
3189 | } | |
ce7d363a N |
3190 | if (!s->locked) |
3191 | /* False alarm - nothing to do */ | |
3192 | return; | |
3193 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3194 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3195 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3196 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3197 | } |
3198 | ||
c0f7bddb | 3199 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3200 | * are in flight |
3201 | */ | |
3202 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3203 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3204 | s->locked++; |
e33129d8 | 3205 | |
c0f7bddb YT |
3206 | if (level == 6) { |
3207 | int qd_idx = sh->qd_idx; | |
3208 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3209 | ||
3210 | set_bit(R5_LOCKED, &dev->flags); | |
3211 | clear_bit(R5_UPTODATE, &dev->flags); | |
3212 | s->locked++; | |
3213 | } | |
3214 | ||
845b9e22 | 3215 | if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page && |
3418d036 AP |
3216 | test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) && |
3217 | !test_bit(STRIPE_FULL_WRITE, &sh->state) && | |
3218 | test_bit(R5_Insync, &sh->dev[pd_idx].flags)) | |
3219 | set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request); | |
3220 | ||
600aa109 | 3221 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3222 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3223 | s->locked, s->ops_request); |
e33129d8 | 3224 | } |
16a53ecc | 3225 | |
1da177e4 LT |
3226 | /* |
3227 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3228 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3229 | * The bi_next chain must be in order. |
3230 | */ | |
da41ba65 | 3231 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3232 | int forwrite, int previous) | |
1da177e4 LT |
3233 | { |
3234 | struct bio **bip; | |
d1688a6d | 3235 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3236 | int firstwrite=0; |
1da177e4 | 3237 | |
cbe47ec5 | 3238 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3239 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3240 | (unsigned long long)sh->sector); |
3241 | ||
b17459c0 | 3242 | spin_lock_irq(&sh->stripe_lock); |
2cd259a7 | 3243 | sh->dev[dd_idx].write_hint = bi->bi_write_hint; |
59fc630b | 3244 | /* Don't allow new IO added to stripes in batch list */ |
3245 | if (sh->batch_head) | |
3246 | goto overlap; | |
72626685 | 3247 | if (forwrite) { |
1da177e4 | 3248 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3249 | if (*bip == NULL) |
72626685 N |
3250 | firstwrite = 1; |
3251 | } else | |
1da177e4 | 3252 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3253 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3254 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3255 | goto overlap; |
3256 | bip = & (*bip)->bi_next; | |
3257 | } | |
4f024f37 | 3258 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3259 | goto overlap; |
3260 | ||
3418d036 AP |
3261 | if (forwrite && raid5_has_ppl(conf)) { |
3262 | /* | |
3263 | * With PPL only writes to consecutive data chunks within a | |
3264 | * stripe are allowed because for a single stripe_head we can | |
3265 | * only have one PPL entry at a time, which describes one data | |
3266 | * range. Not really an overlap, but wait_for_overlap can be | |
3267 | * used to handle this. | |
3268 | */ | |
3269 | sector_t sector; | |
3270 | sector_t first = 0; | |
3271 | sector_t last = 0; | |
3272 | int count = 0; | |
3273 | int i; | |
3274 | ||
3275 | for (i = 0; i < sh->disks; i++) { | |
3276 | if (i != sh->pd_idx && | |
3277 | (i == dd_idx || sh->dev[i].towrite)) { | |
3278 | sector = sh->dev[i].sector; | |
3279 | if (count == 0 || sector < first) | |
3280 | first = sector; | |
3281 | if (sector > last) | |
3282 | last = sector; | |
3283 | count++; | |
3284 | } | |
3285 | } | |
3286 | ||
3287 | if (first + conf->chunk_sectors * (count - 1) != last) | |
3288 | goto overlap; | |
3289 | } | |
3290 | ||
da41ba65 | 3291 | if (!forwrite || previous) |
3292 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3293 | ||
78bafebd | 3294 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3295 | if (*bip) |
3296 | bi->bi_next = *bip; | |
3297 | *bip = bi; | |
016c76ac | 3298 | bio_inc_remaining(bi); |
49728050 | 3299 | md_write_inc(conf->mddev, bi); |
72626685 | 3300 | |
1da177e4 LT |
3301 | if (forwrite) { |
3302 | /* check if page is covered */ | |
3303 | sector_t sector = sh->dev[dd_idx].sector; | |
3304 | for (bi=sh->dev[dd_idx].towrite; | |
c911c46c | 3305 | sector < sh->dev[dd_idx].sector + RAID5_STRIPE_SECTORS(conf) && |
4f024f37 | 3306 | bi && bi->bi_iter.bi_sector <= sector; |
c911c46c | 3307 | bi = r5_next_bio(conf, bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3308 | if (bio_end_sector(bi) >= sector) |
3309 | sector = bio_end_sector(bi); | |
1da177e4 | 3310 | } |
c911c46c | 3311 | if (sector >= sh->dev[dd_idx].sector + RAID5_STRIPE_SECTORS(conf)) |
7a87f434 | 3312 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3313 | sh->overwrite_disks++; | |
1da177e4 | 3314 | } |
cbe47ec5 N |
3315 | |
3316 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3317 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3318 | (unsigned long long)sh->sector, dd_idx); |
3319 | ||
3320 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3321 | /* Cannot hold spinlock over bitmap_startwrite, |
3322 | * but must ensure this isn't added to a batch until | |
3323 | * we have added to the bitmap and set bm_seq. | |
3324 | * So set STRIPE_BITMAP_PENDING to prevent | |
3325 | * batching. | |
3326 | * If multiple add_stripe_bio() calls race here they | |
3327 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3328 | * to complete "bitmap_startwrite" gets to set | |
3329 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3330 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3331 | * any more. | |
3332 | */ | |
3333 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3334 | spin_unlock_irq(&sh->stripe_lock); | |
e64e4018 | 3335 | md_bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3336 | RAID5_STRIPE_SECTORS(conf), 0); |
d0852df5 N |
3337 | spin_lock_irq(&sh->stripe_lock); |
3338 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3339 | if (!sh->batch_head) { | |
3340 | sh->bm_seq = conf->seq_flush+1; | |
3341 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3342 | } | |
cbe47ec5 | 3343 | } |
d0852df5 | 3344 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3345 | |
3346 | if (stripe_can_batch(sh)) | |
3347 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3348 | return 1; |
3349 | ||
3350 | overlap: | |
3351 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3352 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3353 | return 0; |
3354 | } | |
3355 | ||
d1688a6d | 3356 | static void end_reshape(struct r5conf *conf); |
29269553 | 3357 | |
d1688a6d | 3358 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3359 | struct stripe_head *sh) |
ccfcc3c1 | 3360 | { |
784052ec | 3361 | int sectors_per_chunk = |
09c9e5fa | 3362 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3363 | int dd_idx; |
2d2063ce | 3364 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3365 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3366 | |
112bf897 N |
3367 | raid5_compute_sector(conf, |
3368 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3369 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3370 | previous, |
911d4ee8 | 3371 | &dd_idx, sh); |
ccfcc3c1 N |
3372 | } |
3373 | ||
a4456856 | 3374 | static void |
d1688a6d | 3375 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
bd83d0a2 | 3376 | struct stripe_head_state *s, int disks) |
a4456856 DW |
3377 | { |
3378 | int i; | |
59fc630b | 3379 | BUG_ON(sh->batch_head); |
a4456856 DW |
3380 | for (i = disks; i--; ) { |
3381 | struct bio *bi; | |
3382 | int bitmap_end = 0; | |
3383 | ||
3384 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3385 | struct md_rdev *rdev; |
a4456856 DW |
3386 | rcu_read_lock(); |
3387 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3388 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3389 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3390 | atomic_inc(&rdev->nr_pending); |
3391 | else | |
3392 | rdev = NULL; | |
a4456856 | 3393 | rcu_read_unlock(); |
7f0da59b N |
3394 | if (rdev) { |
3395 | if (!rdev_set_badblocks( | |
3396 | rdev, | |
3397 | sh->sector, | |
c911c46c | 3398 | RAID5_STRIPE_SECTORS(conf), 0)) |
7f0da59b N |
3399 | md_error(conf->mddev, rdev); |
3400 | rdev_dec_pending(rdev, conf->mddev); | |
3401 | } | |
a4456856 | 3402 | } |
b17459c0 | 3403 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3404 | /* fail all writes first */ |
3405 | bi = sh->dev[i].towrite; | |
3406 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3407 | sh->overwrite_disks = 0; |
b17459c0 | 3408 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3409 | if (bi) |
a4456856 | 3410 | bitmap_end = 1; |
a4456856 | 3411 | |
ff875738 | 3412 | log_stripe_write_finished(sh); |
0576b1c6 | 3413 | |
a4456856 DW |
3414 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3415 | wake_up(&conf->wait_for_overlap); | |
3416 | ||
4f024f37 | 3417 | while (bi && bi->bi_iter.bi_sector < |
c911c46c YY |
3418 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
3419 | struct bio *nextbi = r5_next_bio(conf, bi, sh->dev[i].sector); | |
4246a0b6 | 3420 | |
49728050 | 3421 | md_write_end(conf->mddev); |
6308d8e3 | 3422 | bio_io_error(bi); |
a4456856 DW |
3423 | bi = nextbi; |
3424 | } | |
7eaf7e8e | 3425 | if (bitmap_end) |
e64e4018 | 3426 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3427 | RAID5_STRIPE_SECTORS(conf), 0, 0); |
7eaf7e8e | 3428 | bitmap_end = 0; |
a4456856 DW |
3429 | /* and fail all 'written' */ |
3430 | bi = sh->dev[i].written; | |
3431 | sh->dev[i].written = NULL; | |
d592a996 SL |
3432 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3433 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3434 | sh->dev[i].page = sh->dev[i].orig_page; | |
3435 | } | |
3436 | ||
a4456856 | 3437 | if (bi) bitmap_end = 1; |
4f024f37 | 3438 | while (bi && bi->bi_iter.bi_sector < |
c911c46c YY |
3439 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
3440 | struct bio *bi2 = r5_next_bio(conf, bi, sh->dev[i].sector); | |
4246a0b6 | 3441 | |
49728050 | 3442 | md_write_end(conf->mddev); |
6308d8e3 | 3443 | bio_io_error(bi); |
a4456856 DW |
3444 | bi = bi2; |
3445 | } | |
3446 | ||
b5e98d65 DW |
3447 | /* fail any reads if this device is non-operational and |
3448 | * the data has not reached the cache yet. | |
3449 | */ | |
3450 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3451 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3452 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3453 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3454 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3455 | bi = sh->dev[i].toread; |
3456 | sh->dev[i].toread = NULL; | |
143c4d05 | 3457 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3458 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3459 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3460 | if (bi) |
3461 | s->to_read--; | |
4f024f37 | 3462 | while (bi && bi->bi_iter.bi_sector < |
c911c46c | 3463 | sh->dev[i].sector + RAID5_STRIPE_SECTORS(conf)) { |
a4456856 | 3464 | struct bio *nextbi = |
c911c46c | 3465 | r5_next_bio(conf, bi, sh->dev[i].sector); |
4246a0b6 | 3466 | |
6308d8e3 | 3467 | bio_io_error(bi); |
a4456856 DW |
3468 | bi = nextbi; |
3469 | } | |
3470 | } | |
a4456856 | 3471 | if (bitmap_end) |
e64e4018 | 3472 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3473 | RAID5_STRIPE_SECTORS(conf), 0, 0); |
8cfa7b0f N |
3474 | /* If we were in the middle of a write the parity block might |
3475 | * still be locked - so just clear all R5_LOCKED flags | |
3476 | */ | |
3477 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3478 | } |
ebda780b SL |
3479 | s->to_write = 0; |
3480 | s->written = 0; | |
a4456856 | 3481 | |
8b3e6cdc DW |
3482 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3483 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3484 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3485 | } |
3486 | ||
7f0da59b | 3487 | static void |
d1688a6d | 3488 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3489 | struct stripe_head_state *s) |
3490 | { | |
3491 | int abort = 0; | |
3492 | int i; | |
3493 | ||
59fc630b | 3494 | BUG_ON(sh->batch_head); |
7f0da59b | 3495 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3496 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3497 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3498 | s->syncing = 0; |
9a3e1101 | 3499 | s->replacing = 0; |
7f0da59b | 3500 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3501 | * Don't even need to abort as that is handled elsewhere |
3502 | * if needed, and not always wanted e.g. if there is a known | |
3503 | * bad block here. | |
9a3e1101 | 3504 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3505 | * non-sync devices, or abort the recovery |
3506 | */ | |
18b9837e N |
3507 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3508 | /* During recovery devices cannot be removed, so | |
3509 | * locking and refcounting of rdevs is not needed | |
3510 | */ | |
e50d3992 | 3511 | rcu_read_lock(); |
18b9837e | 3512 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3513 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3514 | if (rdev |
3515 | && !test_bit(Faulty, &rdev->flags) | |
3516 | && !test_bit(In_sync, &rdev->flags) | |
3517 | && !rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 3518 | RAID5_STRIPE_SECTORS(conf), 0)) |
18b9837e | 3519 | abort = 1; |
e50d3992 | 3520 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3521 | if (rdev |
3522 | && !test_bit(Faulty, &rdev->flags) | |
3523 | && !test_bit(In_sync, &rdev->flags) | |
3524 | && !rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 3525 | RAID5_STRIPE_SECTORS(conf), 0)) |
18b9837e N |
3526 | abort = 1; |
3527 | } | |
e50d3992 | 3528 | rcu_read_unlock(); |
18b9837e N |
3529 | if (abort) |
3530 | conf->recovery_disabled = | |
3531 | conf->mddev->recovery_disabled; | |
7f0da59b | 3532 | } |
c911c46c | 3533 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), !abort); |
7f0da59b N |
3534 | } |
3535 | ||
9a3e1101 N |
3536 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3537 | { | |
3538 | struct md_rdev *rdev; | |
3539 | int rv = 0; | |
3f232d6a N |
3540 | |
3541 | rcu_read_lock(); | |
3542 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3543 | if (rdev |
3544 | && !test_bit(Faulty, &rdev->flags) | |
3545 | && !test_bit(In_sync, &rdev->flags) | |
3546 | && (rdev->recovery_offset <= sh->sector | |
3547 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3548 | rv = 1; | |
3f232d6a | 3549 | rcu_read_unlock(); |
9a3e1101 N |
3550 | return rv; |
3551 | } | |
3552 | ||
2c58f06e N |
3553 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3554 | int disk_idx, int disks) | |
a4456856 | 3555 | { |
5599becc | 3556 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3557 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3558 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3559 | int i; |
45a4d8fd | 3560 | bool force_rcw = (sh->raid_conf->rmw_level == PARITY_DISABLE_RMW); |
5599becc | 3561 | |
a79cfe12 N |
3562 | |
3563 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3564 | test_bit(R5_UPTODATE, &dev->flags)) | |
3565 | /* No point reading this as we already have it or have | |
3566 | * decided to get it. | |
3567 | */ | |
3568 | return 0; | |
3569 | ||
3570 | if (dev->toread || | |
3571 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3572 | /* We need this block to directly satisfy a request */ | |
3573 | return 1; | |
3574 | ||
3575 | if (s->syncing || s->expanding || | |
3576 | (s->replacing && want_replace(sh, disk_idx))) | |
3577 | /* When syncing, or expanding we read everything. | |
3578 | * When replacing, we need the replaced block. | |
3579 | */ | |
3580 | return 1; | |
3581 | ||
3582 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3583 | (s->failed >= 2 && fdev[1]->toread)) | |
3584 | /* If we want to read from a failed device, then | |
3585 | * we need to actually read every other device. | |
3586 | */ | |
3587 | return 1; | |
3588 | ||
a9d56950 N |
3589 | /* Sometimes neither read-modify-write nor reconstruct-write |
3590 | * cycles can work. In those cases we read every block we | |
3591 | * can. Then the parity-update is certain to have enough to | |
3592 | * work with. | |
3593 | * This can only be a problem when we need to write something, | |
3594 | * and some device has failed. If either of those tests | |
3595 | * fail we need look no further. | |
3596 | */ | |
3597 | if (!s->failed || !s->to_write) | |
3598 | return 0; | |
3599 | ||
3600 | if (test_bit(R5_Insync, &dev->flags) && | |
3601 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3602 | /* Pre-reads at not permitted until after short delay | |
3603 | * to gather multiple requests. However if this | |
3560741e | 3604 | * device is no Insync, the block could only be computed |
a9d56950 N |
3605 | * and there is no need to delay that. |
3606 | */ | |
3607 | return 0; | |
ea664c82 | 3608 | |
36707bb2 | 3609 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3610 | if (fdev[i]->towrite && |
3611 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3612 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3613 | /* If we have a partial write to a failed | |
3614 | * device, then we will need to reconstruct | |
3615 | * the content of that device, so all other | |
3616 | * devices must be read. | |
3617 | */ | |
3618 | return 1; | |
45a4d8fd CP |
3619 | |
3620 | if (s->failed >= 2 && | |
3621 | (fdev[i]->towrite || | |
3622 | s->failed_num[i] == sh->pd_idx || | |
3623 | s->failed_num[i] == sh->qd_idx) && | |
3624 | !test_bit(R5_UPTODATE, &fdev[i]->flags)) | |
3625 | /* In max degraded raid6, If the failed disk is P, Q, | |
3626 | * or we want to read the failed disk, we need to do | |
3627 | * reconstruct-write. | |
3628 | */ | |
3629 | force_rcw = true; | |
ea664c82 N |
3630 | } |
3631 | ||
45a4d8fd CP |
3632 | /* If we are forced to do a reconstruct-write, because parity |
3633 | * cannot be trusted and we are currently recovering it, there | |
3634 | * is extra need to be careful. | |
ea664c82 N |
3635 | * If one of the devices that we would need to read, because |
3636 | * it is not being overwritten (and maybe not written at all) | |
3637 | * is missing/faulty, then we need to read everything we can. | |
3638 | */ | |
45a4d8fd | 3639 | if (!force_rcw && |
ea664c82 N |
3640 | sh->sector < sh->raid_conf->mddev->recovery_cp) |
3641 | /* reconstruct-write isn't being forced */ | |
3642 | return 0; | |
36707bb2 | 3643 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3644 | if (s->failed_num[i] != sh->pd_idx && |
3645 | s->failed_num[i] != sh->qd_idx && | |
3646 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3647 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3648 | return 1; | |
3649 | } | |
3650 | ||
2c58f06e N |
3651 | return 0; |
3652 | } | |
3653 | ||
ba02684d SL |
3654 | /* fetch_block - checks the given member device to see if its data needs |
3655 | * to be read or computed to satisfy a request. | |
3656 | * | |
3657 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3658 | * 0 to tell the loop in handle_stripe_fill to continue | |
3659 | */ | |
2c58f06e N |
3660 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3661 | int disk_idx, int disks) | |
3662 | { | |
3663 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3664 | ||
3665 | /* is the data in this block needed, and can we get it? */ | |
3666 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3667 | /* we would like to get this block, possibly by computing it, |
3668 | * otherwise read it if the backing disk is insync | |
3669 | */ | |
3670 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3671 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3672 | BUG_ON(sh->batch_head); |
7471fb77 N |
3673 | |
3674 | /* | |
3675 | * In the raid6 case if the only non-uptodate disk is P | |
3676 | * then we already trusted P to compute the other failed | |
3677 | * drives. It is safe to compute rather than re-read P. | |
3678 | * In other cases we only compute blocks from failed | |
3679 | * devices, otherwise check/repair might fail to detect | |
3680 | * a real inconsistency. | |
3681 | */ | |
3682 | ||
5599becc | 3683 | if ((s->uptodate == disks - 1) && |
7471fb77 | 3684 | ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) || |
f2b3b44d | 3685 | (s->failed && (disk_idx == s->failed_num[0] || |
7471fb77 | 3686 | disk_idx == s->failed_num[1])))) { |
5599becc YT |
3687 | /* have disk failed, and we're requested to fetch it; |
3688 | * do compute it | |
a4456856 | 3689 | */ |
5599becc YT |
3690 | pr_debug("Computing stripe %llu block %d\n", |
3691 | (unsigned long long)sh->sector, disk_idx); | |
3692 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3693 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3694 | set_bit(R5_Wantcompute, &dev->flags); | |
3695 | sh->ops.target = disk_idx; | |
3696 | sh->ops.target2 = -1; /* no 2nd target */ | |
3697 | s->req_compute = 1; | |
93b3dbce N |
3698 | /* Careful: from this point on 'uptodate' is in the eye |
3699 | * of raid_run_ops which services 'compute' operations | |
3700 | * before writes. R5_Wantcompute flags a block that will | |
3701 | * be R5_UPTODATE by the time it is needed for a | |
3702 | * subsequent operation. | |
3703 | */ | |
5599becc YT |
3704 | s->uptodate++; |
3705 | return 1; | |
3706 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3707 | /* Computing 2-failure is *very* expensive; only | |
3708 | * do it if failed >= 2 | |
3709 | */ | |
3710 | int other; | |
3711 | for (other = disks; other--; ) { | |
3712 | if (other == disk_idx) | |
3713 | continue; | |
3714 | if (!test_bit(R5_UPTODATE, | |
3715 | &sh->dev[other].flags)) | |
3716 | break; | |
a4456856 | 3717 | } |
5599becc YT |
3718 | BUG_ON(other < 0); |
3719 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3720 | (unsigned long long)sh->sector, | |
3721 | disk_idx, other); | |
3722 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3723 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3724 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3725 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3726 | sh->ops.target = disk_idx; | |
3727 | sh->ops.target2 = other; | |
3728 | s->uptodate += 2; | |
3729 | s->req_compute = 1; | |
3730 | return 1; | |
3731 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3732 | set_bit(R5_LOCKED, &dev->flags); | |
3733 | set_bit(R5_Wantread, &dev->flags); | |
3734 | s->locked++; | |
3735 | pr_debug("Reading block %d (sync=%d)\n", | |
3736 | disk_idx, s->syncing); | |
a4456856 DW |
3737 | } |
3738 | } | |
5599becc YT |
3739 | |
3740 | return 0; | |
3741 | } | |
3742 | ||
2aada5b1 | 3743 | /* |
93b3dbce | 3744 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3745 | */ |
93b3dbce N |
3746 | static void handle_stripe_fill(struct stripe_head *sh, |
3747 | struct stripe_head_state *s, | |
3748 | int disks) | |
5599becc YT |
3749 | { |
3750 | int i; | |
3751 | ||
3752 | /* look for blocks to read/compute, skip this if a compute | |
3753 | * is already in flight, or if the stripe contents are in the | |
3754 | * midst of changing due to a write | |
3755 | */ | |
3756 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3757 | !sh->reconstruct_state) { |
3758 | ||
3759 | /* | |
3760 | * For degraded stripe with data in journal, do not handle | |
3761 | * read requests yet, instead, flush the stripe to raid | |
3762 | * disks first, this avoids handling complex rmw of write | |
3763 | * back cache (prexor with orig_page, and then xor with | |
3764 | * page) in the read path | |
3765 | */ | |
3766 | if (s->injournal && s->failed) { | |
3767 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3768 | r5c_make_stripe_write_out(sh); | |
3769 | goto out; | |
3770 | } | |
3771 | ||
5599becc | 3772 | for (i = disks; i--; ) |
93b3dbce | 3773 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3774 | break; |
07e83364 SL |
3775 | } |
3776 | out: | |
a4456856 DW |
3777 | set_bit(STRIPE_HANDLE, &sh->state); |
3778 | } | |
3779 | ||
787b76fa N |
3780 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3781 | unsigned long handle_flags); | |
1fe797e6 | 3782 | /* handle_stripe_clean_event |
a4456856 DW |
3783 | * any written block on an uptodate or failed drive can be returned. |
3784 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3785 | * never LOCKED, so we don't need to test 'failed' directly. | |
3786 | */ | |
d1688a6d | 3787 | static void handle_stripe_clean_event(struct r5conf *conf, |
bd83d0a2 | 3788 | struct stripe_head *sh, int disks) |
a4456856 DW |
3789 | { |
3790 | int i; | |
3791 | struct r5dev *dev; | |
f8dfcffd | 3792 | int discard_pending = 0; |
59fc630b | 3793 | struct stripe_head *head_sh = sh; |
3794 | bool do_endio = false; | |
a4456856 DW |
3795 | |
3796 | for (i = disks; i--; ) | |
3797 | if (sh->dev[i].written) { | |
3798 | dev = &sh->dev[i]; | |
3799 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3800 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3801 | test_bit(R5_Discard, &dev->flags) || |
3802 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3803 | /* We can return any write requests */ |
3804 | struct bio *wbi, *wbi2; | |
45b4233c | 3805 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3806 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3807 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3808 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3809 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3810 | } |
59fc630b | 3811 | do_endio = true; |
3812 | ||
3813 | returnbi: | |
3814 | dev->page = dev->orig_page; | |
a4456856 DW |
3815 | wbi = dev->written; |
3816 | dev->written = NULL; | |
4f024f37 | 3817 | while (wbi && wbi->bi_iter.bi_sector < |
c911c46c YY |
3818 | dev->sector + RAID5_STRIPE_SECTORS(conf)) { |
3819 | wbi2 = r5_next_bio(conf, wbi, dev->sector); | |
49728050 | 3820 | md_write_end(conf->mddev); |
016c76ac | 3821 | bio_endio(wbi); |
a4456856 DW |
3822 | wbi = wbi2; |
3823 | } | |
e64e4018 | 3824 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
c911c46c | 3825 | RAID5_STRIPE_SECTORS(conf), |
e64e4018 AS |
3826 | !test_bit(STRIPE_DEGRADED, &sh->state), |
3827 | 0); | |
59fc630b | 3828 | if (head_sh->batch_head) { |
3829 | sh = list_first_entry(&sh->batch_list, | |
3830 | struct stripe_head, | |
3831 | batch_list); | |
3832 | if (sh != head_sh) { | |
3833 | dev = &sh->dev[i]; | |
3834 | goto returnbi; | |
3835 | } | |
3836 | } | |
3837 | sh = head_sh; | |
3838 | dev = &sh->dev[i]; | |
f8dfcffd N |
3839 | } else if (test_bit(R5_Discard, &dev->flags)) |
3840 | discard_pending = 1; | |
3841 | } | |
f6bed0ef | 3842 | |
ff875738 | 3843 | log_stripe_write_finished(sh); |
0576b1c6 | 3844 | |
f8dfcffd N |
3845 | if (!discard_pending && |
3846 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3847 | int hash; |
f8dfcffd N |
3848 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3849 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3850 | if (sh->qd_idx >= 0) { | |
3851 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3852 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3853 | } | |
3854 | /* now that discard is done we can proceed with any sync */ | |
3855 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3856 | /* |
3857 | * SCSI discard will change some bio fields and the stripe has | |
3858 | * no updated data, so remove it from hash list and the stripe | |
3859 | * will be reinitialized | |
3860 | */ | |
59fc630b | 3861 | unhash: |
b8a9d66d RG |
3862 | hash = sh->hash_lock_index; |
3863 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3864 | remove_hash(sh); |
b8a9d66d | 3865 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3866 | if (head_sh->batch_head) { |
3867 | sh = list_first_entry(&sh->batch_list, | |
3868 | struct stripe_head, batch_list); | |
3869 | if (sh != head_sh) | |
3870 | goto unhash; | |
3871 | } | |
59fc630b | 3872 | sh = head_sh; |
3873 | ||
f8dfcffd N |
3874 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3875 | set_bit(STRIPE_HANDLE, &sh->state); | |
3876 | ||
3877 | } | |
8b3e6cdc DW |
3878 | |
3879 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3880 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3881 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3882 | |
787b76fa N |
3883 | if (head_sh->batch_head && do_endio) |
3884 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3885 | } |
3886 | ||
86aa1397 SL |
3887 | /* |
3888 | * For RMW in write back cache, we need extra page in prexor to store the | |
3889 | * old data. This page is stored in dev->orig_page. | |
3890 | * | |
3891 | * This function checks whether we have data for prexor. The exact logic | |
3892 | * is: | |
3893 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3894 | */ | |
3895 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3896 | { | |
3897 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3898 | (!test_bit(R5_InJournal, &dev->flags) || | |
3899 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3900 | } | |
3901 | ||
d7bd398e SL |
3902 | static int handle_stripe_dirtying(struct r5conf *conf, |
3903 | struct stripe_head *sh, | |
3904 | struct stripe_head_state *s, | |
3905 | int disks) | |
a4456856 DW |
3906 | { |
3907 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3908 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3909 | ||
584acdd4 | 3910 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3911 | * If yes, then the array is dirty (after unclean shutdown or |
3912 | * initial creation), so parity in some stripes might be inconsistent. | |
3913 | * In this case, we need to always do reconstruct-write, to ensure | |
3914 | * that in case of drive failure or read-error correction, we | |
3915 | * generate correct data from the parity. | |
3916 | */ | |
584acdd4 | 3917 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3918 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3919 | s->failed == 0)) { | |
a7854487 | 3920 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3921 | * look like rcw is cheaper |
3922 | */ | |
3923 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3924 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3925 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3926 | (unsigned long long)sh->sector); |
c8ac1803 | 3927 | } else for (i = disks; i--; ) { |
a4456856 DW |
3928 | /* would I have to read this buffer for read_modify_write */ |
3929 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3930 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3931 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3932 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3933 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3934 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3935 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3936 | if (test_bit(R5_Insync, &dev->flags)) |
3937 | rmw++; | |
3938 | else | |
3939 | rmw += 2*disks; /* cannot read it */ | |
3940 | } | |
3941 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3942 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3943 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3944 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3945 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3946 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3947 | if (test_bit(R5_Insync, &dev->flags)) |
3948 | rcw++; | |
a4456856 DW |
3949 | else |
3950 | rcw += 2*disks; | |
3951 | } | |
3952 | } | |
1e6d690b | 3953 | |
39b99586 SL |
3954 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3955 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3956 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3957 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3958 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3959 | if (conf->mddev->queue) |
3960 | blk_add_trace_msg(conf->mddev->queue, | |
3961 | "raid5 rmw %llu %d", | |
3962 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3963 | for (i = disks; i--; ) { |
3964 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3965 | if (test_bit(R5_InJournal, &dev->flags) && |
3966 | dev->page == dev->orig_page && | |
3967 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3968 | /* alloc page for prexor */ | |
d7bd398e SL |
3969 | struct page *p = alloc_page(GFP_NOIO); |
3970 | ||
3971 | if (p) { | |
3972 | dev->orig_page = p; | |
3973 | continue; | |
3974 | } | |
3975 | ||
3976 | /* | |
3977 | * alloc_page() failed, try use | |
3978 | * disk_info->extra_page | |
3979 | */ | |
3980 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3981 | &conf->cache_state)) { | |
3982 | r5c_use_extra_page(sh); | |
3983 | break; | |
3984 | } | |
1e6d690b | 3985 | |
d7bd398e SL |
3986 | /* extra_page in use, add to delayed_list */ |
3987 | set_bit(STRIPE_DELAYED, &sh->state); | |
3988 | s->waiting_extra_page = 1; | |
3989 | return -EAGAIN; | |
1e6d690b | 3990 | } |
d7bd398e | 3991 | } |
1e6d690b | 3992 | |
d7bd398e SL |
3993 | for (i = disks; i--; ) { |
3994 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3995 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3996 | i == sh->pd_idx || i == sh->qd_idx || |
3997 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3998 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3999 | !(uptodate_for_rmw(dev) || |
1e6d690b | 4000 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 4001 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
4002 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
4003 | &sh->state)) { | |
4004 | pr_debug("Read_old block %d for r-m-w\n", | |
4005 | i); | |
a4456856 DW |
4006 | set_bit(R5_LOCKED, &dev->flags); |
4007 | set_bit(R5_Wantread, &dev->flags); | |
4008 | s->locked++; | |
e3914d59 | 4009 | } else |
a4456856 | 4010 | set_bit(STRIPE_DELAYED, &sh->state); |
a4456856 DW |
4011 | } |
4012 | } | |
a9add5d9 | 4013 | } |
41257580 | 4014 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 4015 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 4016 | int qread =0; |
c8ac1803 | 4017 | rcw = 0; |
a4456856 DW |
4018 | for (i = disks; i--; ) { |
4019 | struct r5dev *dev = &sh->dev[i]; | |
4020 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 4021 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 4022 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 4023 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
4024 | test_bit(R5_Wantcompute, &dev->flags))) { |
4025 | rcw++; | |
67f45548 N |
4026 | if (test_bit(R5_Insync, &dev->flags) && |
4027 | test_bit(STRIPE_PREREAD_ACTIVE, | |
4028 | &sh->state)) { | |
45b4233c | 4029 | pr_debug("Read_old block " |
a4456856 DW |
4030 | "%d for Reconstruct\n", i); |
4031 | set_bit(R5_LOCKED, &dev->flags); | |
4032 | set_bit(R5_Wantread, &dev->flags); | |
4033 | s->locked++; | |
a9add5d9 | 4034 | qread++; |
e3914d59 | 4035 | } else |
a4456856 | 4036 | set_bit(STRIPE_DELAYED, &sh->state); |
a4456856 DW |
4037 | } |
4038 | } | |
e3620a3a | 4039 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
4040 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
4041 | (unsigned long long)sh->sector, | |
4042 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 4043 | } |
b1b02fe9 N |
4044 | |
4045 | if (rcw > disks && rmw > disks && | |
4046 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4047 | set_bit(STRIPE_DELAYED, &sh->state); | |
4048 | ||
a4456856 DW |
4049 | /* now if nothing is locked, and if we have enough data, |
4050 | * we can start a write request | |
4051 | */ | |
f38e1219 DW |
4052 | /* since handle_stripe can be called at any time we need to handle the |
4053 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
4054 | * subsequent call wants to start a write request. raid_run_ops only |
4055 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
4056 | * simultaneously. If this is not the case then new writes need to be |
4057 | * held off until the compute completes. | |
4058 | */ | |
976ea8d4 DW |
4059 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
4060 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 4061 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 4062 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 4063 | return 0; |
a4456856 DW |
4064 | } |
4065 | ||
d1688a6d | 4066 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
4067 | struct stripe_head_state *s, int disks) |
4068 | { | |
ecc65c9b | 4069 | struct r5dev *dev = NULL; |
bd2ab670 | 4070 | |
59fc630b | 4071 | BUG_ON(sh->batch_head); |
a4456856 | 4072 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 4073 | |
ecc65c9b DW |
4074 | switch (sh->check_state) { |
4075 | case check_state_idle: | |
4076 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 4077 | if (s->failed == 0) { |
bd2ab670 | 4078 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
4079 | sh->check_state = check_state_run; |
4080 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 4081 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 4082 | s->uptodate--; |
ecc65c9b | 4083 | break; |
bd2ab670 | 4084 | } |
f2b3b44d | 4085 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
4086 | /* fall through */ |
4087 | case check_state_compute_result: | |
4088 | sh->check_state = check_state_idle; | |
4089 | if (!dev) | |
4090 | dev = &sh->dev[sh->pd_idx]; | |
4091 | ||
4092 | /* check that a write has not made the stripe insync */ | |
4093 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4094 | break; | |
c8894419 | 4095 | |
a4456856 | 4096 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
4097 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
4098 | BUG_ON(s->uptodate != disks); | |
4099 | ||
4100 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 4101 | s->locked++; |
a4456856 | 4102 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 4103 | |
a4456856 | 4104 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 4105 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
4106 | break; |
4107 | case check_state_run: | |
4108 | break; /* we will be called again upon completion */ | |
4109 | case check_state_check_result: | |
4110 | sh->check_state = check_state_idle; | |
4111 | ||
4112 | /* if a failure occurred during the check operation, leave | |
4113 | * STRIPE_INSYNC not set and let the stripe be handled again | |
4114 | */ | |
4115 | if (s->failed) | |
4116 | break; | |
4117 | ||
4118 | /* handle a successful check operation, if parity is correct | |
4119 | * we are done. Otherwise update the mismatch count and repair | |
4120 | * parity if !MD_RECOVERY_CHECK | |
4121 | */ | |
ad283ea4 | 4122 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
4123 | /* parity is correct (on disc, |
4124 | * not in buffer any more) | |
4125 | */ | |
4126 | set_bit(STRIPE_INSYNC, &sh->state); | |
4127 | else { | |
c911c46c | 4128 | atomic64_add(RAID5_STRIPE_SECTORS(conf), &conf->mddev->resync_mismatches); |
e1539036 | 4129 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
ecc65c9b DW |
4130 | /* don't try to repair!! */ |
4131 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4132 | pr_warn_ratelimited("%s: mismatch sector in range " |
4133 | "%llu-%llu\n", mdname(conf->mddev), | |
4134 | (unsigned long long) sh->sector, | |
4135 | (unsigned long long) sh->sector + | |
c911c46c | 4136 | RAID5_STRIPE_SECTORS(conf)); |
e1539036 | 4137 | } else { |
ecc65c9b | 4138 | sh->check_state = check_state_compute_run; |
976ea8d4 | 4139 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
4140 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
4141 | set_bit(R5_Wantcompute, | |
4142 | &sh->dev[sh->pd_idx].flags); | |
4143 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 4144 | sh->ops.target2 = -1; |
ecc65c9b DW |
4145 | s->uptodate++; |
4146 | } | |
4147 | } | |
4148 | break; | |
4149 | case check_state_compute_run: | |
4150 | break; | |
4151 | default: | |
cc6167b4 | 4152 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
4153 | __func__, sh->check_state, |
4154 | (unsigned long long) sh->sector); | |
4155 | BUG(); | |
a4456856 DW |
4156 | } |
4157 | } | |
4158 | ||
d1688a6d | 4159 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4160 | struct stripe_head_state *s, |
f2b3b44d | 4161 | int disks) |
a4456856 | 4162 | { |
a4456856 | 4163 | int pd_idx = sh->pd_idx; |
34e04e87 | 4164 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4165 | struct r5dev *dev; |
a4456856 | 4166 | |
59fc630b | 4167 | BUG_ON(sh->batch_head); |
a4456856 DW |
4168 | set_bit(STRIPE_HANDLE, &sh->state); |
4169 | ||
4170 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4171 | |
a4456856 DW |
4172 | /* Want to check and possibly repair P and Q. |
4173 | * However there could be one 'failed' device, in which | |
4174 | * case we can only check one of them, possibly using the | |
4175 | * other to generate missing data | |
4176 | */ | |
4177 | ||
d82dfee0 DW |
4178 | switch (sh->check_state) { |
4179 | case check_state_idle: | |
4180 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4181 | if (s->failed == s->q_failed) { |
d82dfee0 | 4182 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4183 | * makes sense to check P (If anything else were failed, |
4184 | * we would have used P to recreate it). | |
4185 | */ | |
d82dfee0 | 4186 | sh->check_state = check_state_run; |
a4456856 | 4187 | } |
f2b3b44d | 4188 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4189 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4190 | * anything, so it makes sense to check it |
4191 | */ | |
d82dfee0 DW |
4192 | if (sh->check_state == check_state_run) |
4193 | sh->check_state = check_state_run_pq; | |
4194 | else | |
4195 | sh->check_state = check_state_run_q; | |
a4456856 | 4196 | } |
a4456856 | 4197 | |
d82dfee0 DW |
4198 | /* discard potentially stale zero_sum_result */ |
4199 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4200 | |
d82dfee0 DW |
4201 | if (sh->check_state == check_state_run) { |
4202 | /* async_xor_zero_sum destroys the contents of P */ | |
4203 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4204 | s->uptodate--; | |
a4456856 | 4205 | } |
d82dfee0 DW |
4206 | if (sh->check_state >= check_state_run && |
4207 | sh->check_state <= check_state_run_pq) { | |
4208 | /* async_syndrome_zero_sum preserves P and Q, so | |
4209 | * no need to mark them !uptodate here | |
4210 | */ | |
4211 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4212 | break; | |
a4456856 DW |
4213 | } |
4214 | ||
d82dfee0 DW |
4215 | /* we have 2-disk failure */ |
4216 | BUG_ON(s->failed != 2); | |
4217 | /* fall through */ | |
4218 | case check_state_compute_result: | |
4219 | sh->check_state = check_state_idle; | |
a4456856 | 4220 | |
d82dfee0 DW |
4221 | /* check that a write has not made the stripe insync */ |
4222 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4223 | break; | |
a4456856 DW |
4224 | |
4225 | /* now write out any block on a failed drive, | |
d82dfee0 | 4226 | * or P or Q if they were recomputed |
a4456856 | 4227 | */ |
b2176a1d | 4228 | dev = NULL; |
a4456856 | 4229 | if (s->failed == 2) { |
f2b3b44d | 4230 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4231 | s->locked++; |
4232 | set_bit(R5_LOCKED, &dev->flags); | |
4233 | set_bit(R5_Wantwrite, &dev->flags); | |
4234 | } | |
4235 | if (s->failed >= 1) { | |
f2b3b44d | 4236 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4237 | s->locked++; |
4238 | set_bit(R5_LOCKED, &dev->flags); | |
4239 | set_bit(R5_Wantwrite, &dev->flags); | |
4240 | } | |
d82dfee0 | 4241 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4242 | dev = &sh->dev[pd_idx]; |
4243 | s->locked++; | |
4244 | set_bit(R5_LOCKED, &dev->flags); | |
4245 | set_bit(R5_Wantwrite, &dev->flags); | |
4246 | } | |
d82dfee0 | 4247 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4248 | dev = &sh->dev[qd_idx]; |
4249 | s->locked++; | |
4250 | set_bit(R5_LOCKED, &dev->flags); | |
4251 | set_bit(R5_Wantwrite, &dev->flags); | |
4252 | } | |
b2176a1d NC |
4253 | if (WARN_ONCE(dev && !test_bit(R5_UPTODATE, &dev->flags), |
4254 | "%s: disk%td not up to date\n", | |
4255 | mdname(conf->mddev), | |
4256 | dev - (struct r5dev *) &sh->dev)) { | |
4257 | clear_bit(R5_LOCKED, &dev->flags); | |
4258 | clear_bit(R5_Wantwrite, &dev->flags); | |
4259 | s->locked--; | |
4260 | } | |
a4456856 DW |
4261 | clear_bit(STRIPE_DEGRADED, &sh->state); |
4262 | ||
4263 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4264 | break; |
4265 | case check_state_run: | |
4266 | case check_state_run_q: | |
4267 | case check_state_run_pq: | |
4268 | break; /* we will be called again upon completion */ | |
4269 | case check_state_check_result: | |
4270 | sh->check_state = check_state_idle; | |
4271 | ||
4272 | /* handle a successful check operation, if parity is correct | |
4273 | * we are done. Otherwise update the mismatch count and repair | |
4274 | * parity if !MD_RECOVERY_CHECK | |
4275 | */ | |
4276 | if (sh->ops.zero_sum_result == 0) { | |
a25d8c32 SL |
4277 | /* both parities are correct */ |
4278 | if (!s->failed) | |
4279 | set_bit(STRIPE_INSYNC, &sh->state); | |
4280 | else { | |
4281 | /* in contrast to the raid5 case we can validate | |
4282 | * parity, but still have a failure to write | |
4283 | * back | |
4284 | */ | |
4285 | sh->check_state = check_state_compute_result; | |
4286 | /* Returning at this point means that we may go | |
4287 | * off and bring p and/or q uptodate again so | |
4288 | * we make sure to check zero_sum_result again | |
4289 | * to verify if p or q need writeback | |
4290 | */ | |
4291 | } | |
d82dfee0 | 4292 | } else { |
c911c46c | 4293 | atomic64_add(RAID5_STRIPE_SECTORS(conf), &conf->mddev->resync_mismatches); |
e1539036 | 4294 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
d82dfee0 DW |
4295 | /* don't try to repair!! */ |
4296 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4297 | pr_warn_ratelimited("%s: mismatch sector in range " |
4298 | "%llu-%llu\n", mdname(conf->mddev), | |
4299 | (unsigned long long) sh->sector, | |
4300 | (unsigned long long) sh->sector + | |
c911c46c | 4301 | RAID5_STRIPE_SECTORS(conf)); |
e1539036 | 4302 | } else { |
d82dfee0 DW |
4303 | int *target = &sh->ops.target; |
4304 | ||
4305 | sh->ops.target = -1; | |
4306 | sh->ops.target2 = -1; | |
4307 | sh->check_state = check_state_compute_run; | |
4308 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4309 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4310 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4311 | set_bit(R5_Wantcompute, | |
4312 | &sh->dev[pd_idx].flags); | |
4313 | *target = pd_idx; | |
4314 | target = &sh->ops.target2; | |
4315 | s->uptodate++; | |
4316 | } | |
4317 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4318 | set_bit(R5_Wantcompute, | |
4319 | &sh->dev[qd_idx].flags); | |
4320 | *target = qd_idx; | |
4321 | s->uptodate++; | |
4322 | } | |
4323 | } | |
4324 | } | |
4325 | break; | |
4326 | case check_state_compute_run: | |
4327 | break; | |
4328 | default: | |
cc6167b4 N |
4329 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4330 | __func__, sh->check_state, | |
4331 | (unsigned long long) sh->sector); | |
d82dfee0 | 4332 | BUG(); |
a4456856 DW |
4333 | } |
4334 | } | |
4335 | ||
d1688a6d | 4336 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4337 | { |
4338 | int i; | |
4339 | ||
4340 | /* We have read all the blocks in this stripe and now we need to | |
4341 | * copy some of them into a target stripe for expand. | |
4342 | */ | |
f0a50d37 | 4343 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4344 | BUG_ON(sh->batch_head); |
a4456856 DW |
4345 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4346 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4347 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4348 | int dd_idx, j; |
a4456856 | 4349 | struct stripe_head *sh2; |
a08abd8c | 4350 | struct async_submit_ctl submit; |
a4456856 | 4351 | |
6d036f7d | 4352 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4353 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4354 | &dd_idx, NULL); | |
6d036f7d | 4355 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4356 | if (sh2 == NULL) |
4357 | /* so far only the early blocks of this stripe | |
4358 | * have been requested. When later blocks | |
4359 | * get requested, we will try again | |
4360 | */ | |
4361 | continue; | |
4362 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4363 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4364 | /* must have already done this block */ | |
6d036f7d | 4365 | raid5_release_stripe(sh2); |
a4456856 DW |
4366 | continue; |
4367 | } | |
f0a50d37 DW |
4368 | |
4369 | /* place all the copies on one channel */ | |
a08abd8c | 4370 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4371 | tx = async_memcpy(sh2->dev[dd_idx].page, |
c911c46c | 4372 | sh->dev[i].page, 0, 0, RAID5_STRIPE_SIZE(conf), |
a08abd8c | 4373 | &submit); |
f0a50d37 | 4374 | |
a4456856 DW |
4375 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4376 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4377 | for (j = 0; j < conf->raid_disks; j++) | |
4378 | if (j != sh2->pd_idx && | |
86c374ba | 4379 | j != sh2->qd_idx && |
a4456856 DW |
4380 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4381 | break; | |
4382 | if (j == conf->raid_disks) { | |
4383 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4384 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4385 | } | |
6d036f7d | 4386 | raid5_release_stripe(sh2); |
f0a50d37 | 4387 | |
a4456856 | 4388 | } |
a2e08551 | 4389 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4390 | async_tx_quiesce(&tx); |
a4456856 | 4391 | } |
1da177e4 LT |
4392 | |
4393 | /* | |
4394 | * handle_stripe - do things to a stripe. | |
4395 | * | |
9a3e1101 N |
4396 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4397 | * state of various bits to see what needs to be done. | |
1da177e4 | 4398 | * Possible results: |
9a3e1101 N |
4399 | * return some read requests which now have data |
4400 | * return some write requests which are safely on storage | |
1da177e4 LT |
4401 | * schedule a read on some buffers |
4402 | * schedule a write of some buffers | |
4403 | * return confirmation of parity correctness | |
4404 | * | |
1da177e4 | 4405 | */ |
a4456856 | 4406 | |
acfe726b | 4407 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4408 | { |
d1688a6d | 4409 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4410 | int disks = sh->disks; |
474af965 N |
4411 | struct r5dev *dev; |
4412 | int i; | |
9a3e1101 | 4413 | int do_recovery = 0; |
1da177e4 | 4414 | |
acfe726b N |
4415 | memset(s, 0, sizeof(*s)); |
4416 | ||
dabc4ec6 | 4417 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4418 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4419 | s->failed_num[0] = -1; |
4420 | s->failed_num[1] = -1; | |
6e74a9cf | 4421 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4422 | |
acfe726b | 4423 | /* Now to look around and see what can be done */ |
1da177e4 | 4424 | rcu_read_lock(); |
16a53ecc | 4425 | for (i=disks; i--; ) { |
3cb03002 | 4426 | struct md_rdev *rdev; |
31c176ec N |
4427 | sector_t first_bad; |
4428 | int bad_sectors; | |
4429 | int is_bad = 0; | |
acfe726b | 4430 | |
16a53ecc | 4431 | dev = &sh->dev[i]; |
1da177e4 | 4432 | |
45b4233c | 4433 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4434 | i, dev->flags, |
4435 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4436 | /* maybe we can reply to a read |
4437 | * | |
4438 | * new wantfill requests are only permitted while | |
4439 | * ops_complete_biofill is guaranteed to be inactive | |
4440 | */ | |
4441 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4442 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4443 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4444 | |
16a53ecc | 4445 | /* now count some things */ |
cc94015a N |
4446 | if (test_bit(R5_LOCKED, &dev->flags)) |
4447 | s->locked++; | |
4448 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4449 | s->uptodate++; | |
2d6e4ecc | 4450 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4451 | s->compute++; |
4452 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4453 | } |
1da177e4 | 4454 | |
acfe726b | 4455 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4456 | s->to_fill++; |
acfe726b | 4457 | else if (dev->toread) |
cc94015a | 4458 | s->to_read++; |
16a53ecc | 4459 | if (dev->towrite) { |
cc94015a | 4460 | s->to_write++; |
16a53ecc | 4461 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4462 | s->non_overwrite++; |
16a53ecc | 4463 | } |
a4456856 | 4464 | if (dev->written) |
cc94015a | 4465 | s->written++; |
14a75d3e N |
4466 | /* Prefer to use the replacement for reads, but only |
4467 | * if it is recovered enough and has no bad blocks. | |
4468 | */ | |
4469 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4470 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
c911c46c YY |
4471 | rdev->recovery_offset >= sh->sector + RAID5_STRIPE_SECTORS(conf) && |
4472 | !is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), | |
14a75d3e N |
4473 | &first_bad, &bad_sectors)) |
4474 | set_bit(R5_ReadRepl, &dev->flags); | |
4475 | else { | |
e6030cb0 | 4476 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4477 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4478 | else |
4479 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4480 | rdev = rcu_dereference(conf->disks[i].rdev); |
4481 | clear_bit(R5_ReadRepl, &dev->flags); | |
4482 | } | |
9283d8c5 N |
4483 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4484 | rdev = NULL; | |
31c176ec | 4485 | if (rdev) { |
c911c46c | 4486 | is_bad = is_badblock(rdev, sh->sector, RAID5_STRIPE_SECTORS(conf), |
31c176ec N |
4487 | &first_bad, &bad_sectors); |
4488 | if (s->blocked_rdev == NULL | |
4489 | && (test_bit(Blocked, &rdev->flags) | |
4490 | || is_bad < 0)) { | |
4491 | if (is_bad < 0) | |
4492 | set_bit(BlockedBadBlocks, | |
4493 | &rdev->flags); | |
4494 | s->blocked_rdev = rdev; | |
4495 | atomic_inc(&rdev->nr_pending); | |
4496 | } | |
6bfe0b49 | 4497 | } |
415e72d0 N |
4498 | clear_bit(R5_Insync, &dev->flags); |
4499 | if (!rdev) | |
4500 | /* Not in-sync */; | |
31c176ec N |
4501 | else if (is_bad) { |
4502 | /* also not in-sync */ | |
18b9837e N |
4503 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4504 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4505 | /* treat as in-sync, but with a read error |
4506 | * which we can now try to correct | |
4507 | */ | |
4508 | set_bit(R5_Insync, &dev->flags); | |
4509 | set_bit(R5_ReadError, &dev->flags); | |
4510 | } | |
4511 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4512 | set_bit(R5_Insync, &dev->flags); |
c911c46c | 4513 | else if (sh->sector + RAID5_STRIPE_SECTORS(conf) <= rdev->recovery_offset) |
415e72d0 | 4514 | /* in sync if before recovery_offset */ |
30d7a483 N |
4515 | set_bit(R5_Insync, &dev->flags); |
4516 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4517 | test_bit(R5_Expanded, &dev->flags)) | |
4518 | /* If we've reshaped into here, we assume it is Insync. | |
4519 | * We will shortly update recovery_offset to make | |
4520 | * it official. | |
4521 | */ | |
4522 | set_bit(R5_Insync, &dev->flags); | |
4523 | ||
1cc03eb9 | 4524 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4525 | /* This flag does not apply to '.replacement' |
4526 | * only to .rdev, so make sure to check that*/ | |
4527 | struct md_rdev *rdev2 = rcu_dereference( | |
4528 | conf->disks[i].rdev); | |
4529 | if (rdev2 == rdev) | |
4530 | clear_bit(R5_Insync, &dev->flags); | |
4531 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4532 | s->handle_bad_blocks = 1; |
14a75d3e | 4533 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4534 | } else |
4535 | clear_bit(R5_WriteError, &dev->flags); | |
4536 | } | |
1cc03eb9 | 4537 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4538 | /* This flag does not apply to '.replacement' |
4539 | * only to .rdev, so make sure to check that*/ | |
4540 | struct md_rdev *rdev2 = rcu_dereference( | |
4541 | conf->disks[i].rdev); | |
4542 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4543 | s->handle_bad_blocks = 1; |
14a75d3e | 4544 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4545 | } else |
4546 | clear_bit(R5_MadeGood, &dev->flags); | |
4547 | } | |
977df362 N |
4548 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4549 | struct md_rdev *rdev2 = rcu_dereference( | |
4550 | conf->disks[i].replacement); | |
4551 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4552 | s->handle_bad_blocks = 1; | |
4553 | atomic_inc(&rdev2->nr_pending); | |
4554 | } else | |
4555 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4556 | } | |
415e72d0 | 4557 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4558 | /* The ReadError flag will just be confusing now */ |
4559 | clear_bit(R5_ReadError, &dev->flags); | |
4560 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4561 | } |
415e72d0 N |
4562 | if (test_bit(R5_ReadError, &dev->flags)) |
4563 | clear_bit(R5_Insync, &dev->flags); | |
4564 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4565 | if (s->failed < 2) |
4566 | s->failed_num[s->failed] = i; | |
4567 | s->failed++; | |
9a3e1101 N |
4568 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4569 | do_recovery = 1; | |
d63e2fc8 BC |
4570 | else if (!rdev) { |
4571 | rdev = rcu_dereference( | |
4572 | conf->disks[i].replacement); | |
4573 | if (rdev && !test_bit(Faulty, &rdev->flags)) | |
4574 | do_recovery = 1; | |
4575 | } | |
415e72d0 | 4576 | } |
2ded3703 SL |
4577 | |
4578 | if (test_bit(R5_InJournal, &dev->flags)) | |
4579 | s->injournal++; | |
1e6d690b SL |
4580 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4581 | s->just_cached++; | |
1da177e4 | 4582 | } |
9a3e1101 N |
4583 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4584 | /* If there is a failed device being replaced, | |
4585 | * we must be recovering. | |
4586 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4587 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4588 | * else we can only be replacing |
4589 | * sync and recovery both need to read all devices, and so | |
4590 | * use the same flag. | |
4591 | */ | |
4592 | if (do_recovery || | |
c6d2e084 | 4593 | sh->sector >= conf->mddev->recovery_cp || |
4594 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4595 | s->syncing = 1; |
4596 | else | |
4597 | s->replacing = 1; | |
4598 | } | |
1da177e4 | 4599 | rcu_read_unlock(); |
cc94015a N |
4600 | } |
4601 | ||
cb9902db GJ |
4602 | /* |
4603 | * Return '1' if this is a member of batch, or '0' if it is a lone stripe or | |
4604 | * a head which can now be handled. | |
4605 | */ | |
59fc630b | 4606 | static int clear_batch_ready(struct stripe_head *sh) |
4607 | { | |
4608 | struct stripe_head *tmp; | |
4609 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4610 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4611 | spin_lock(&sh->stripe_lock); |
4612 | if (!sh->batch_head) { | |
4613 | spin_unlock(&sh->stripe_lock); | |
4614 | return 0; | |
4615 | } | |
4616 | ||
4617 | /* | |
4618 | * this stripe could be added to a batch list before we check | |
4619 | * BATCH_READY, skips it | |
4620 | */ | |
4621 | if (sh->batch_head != sh) { | |
4622 | spin_unlock(&sh->stripe_lock); | |
4623 | return 1; | |
4624 | } | |
4625 | spin_lock(&sh->batch_lock); | |
4626 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4627 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4628 | spin_unlock(&sh->batch_lock); | |
4629 | spin_unlock(&sh->stripe_lock); | |
4630 | ||
4631 | /* | |
4632 | * BATCH_READY is cleared, no new stripes can be added. | |
4633 | * batch_list can be accessed without lock | |
4634 | */ | |
4635 | return 0; | |
4636 | } | |
4637 | ||
3960ce79 N |
4638 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4639 | unsigned long handle_flags) | |
72ac7330 | 4640 | { |
4e3d62ff | 4641 | struct stripe_head *sh, *next; |
72ac7330 | 4642 | int i; |
fb642b92 | 4643 | int do_wakeup = 0; |
72ac7330 | 4644 | |
bb27051f N |
4645 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4646 | ||
72ac7330 | 4647 | list_del_init(&sh->batch_list); |
4648 | ||
fb3229d5 | 4649 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4650 | (1 << STRIPE_SYNCING) | |
4651 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4652 | (1 << STRIPE_DELAYED) | |
4653 | (1 << STRIPE_BIT_DELAY) | | |
4654 | (1 << STRIPE_FULL_WRITE) | | |
4655 | (1 << STRIPE_BIOFILL_RUN) | | |
4656 | (1 << STRIPE_COMPUTE_RUN) | | |
1b956f7a N |
4657 | (1 << STRIPE_DISCARD) | |
4658 | (1 << STRIPE_BATCH_READY) | | |
4659 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4660 | (1 << STRIPE_BITMAP_PENDING)), |
4661 | "stripe state: %lx\n", sh->state); | |
4662 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4663 | (1 << STRIPE_REPLACED)), | |
4664 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4665 | |
4666 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4667 | (1 << STRIPE_PREREAD_ACTIVE) | |
184a09eb DY |
4668 | (1 << STRIPE_DEGRADED) | |
4669 | (1 << STRIPE_ON_UNPLUG_LIST)), | |
1b956f7a N |
4670 | head_sh->state & (1 << STRIPE_INSYNC)); |
4671 | ||
72ac7330 | 4672 | sh->check_state = head_sh->check_state; |
4673 | sh->reconstruct_state = head_sh->reconstruct_state; | |
448ec638 AC |
4674 | spin_lock_irq(&sh->stripe_lock); |
4675 | sh->batch_head = NULL; | |
4676 | spin_unlock_irq(&sh->stripe_lock); | |
fb642b92 N |
4677 | for (i = 0; i < sh->disks; i++) { |
4678 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4679 | do_wakeup = 1; | |
72ac7330 | 4680 | sh->dev[i].flags = head_sh->dev[i].flags & |
4681 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4682 | } |
3960ce79 N |
4683 | if (handle_flags == 0 || |
4684 | sh->state & handle_flags) | |
4685 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4686 | raid5_release_stripe(sh); |
72ac7330 | 4687 | } |
fb642b92 N |
4688 | spin_lock_irq(&head_sh->stripe_lock); |
4689 | head_sh->batch_head = NULL; | |
4690 | spin_unlock_irq(&head_sh->stripe_lock); | |
4691 | for (i = 0; i < head_sh->disks; i++) | |
4692 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4693 | do_wakeup = 1; | |
3960ce79 N |
4694 | if (head_sh->state & handle_flags) |
4695 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4696 | |
4697 | if (do_wakeup) | |
4698 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4699 | } |
4700 | ||
cc94015a N |
4701 | static void handle_stripe(struct stripe_head *sh) |
4702 | { | |
4703 | struct stripe_head_state s; | |
d1688a6d | 4704 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4705 | int i; |
84789554 N |
4706 | int prexor; |
4707 | int disks = sh->disks; | |
474af965 | 4708 | struct r5dev *pdev, *qdev; |
cc94015a N |
4709 | |
4710 | clear_bit(STRIPE_HANDLE, &sh->state); | |
a377a472 GJ |
4711 | |
4712 | /* | |
4713 | * handle_stripe should not continue handle the batched stripe, only | |
4714 | * the head of batch list or lone stripe can continue. Otherwise we | |
4715 | * could see break_stripe_batch_list warns about the STRIPE_ACTIVE | |
4716 | * is set for the batched stripe. | |
4717 | */ | |
4718 | if (clear_batch_ready(sh)) | |
4719 | return; | |
4720 | ||
257a4b42 | 4721 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4722 | /* already being handled, ensure it gets handled |
4723 | * again when current action finishes */ | |
4724 | set_bit(STRIPE_HANDLE, &sh->state); | |
4725 | return; | |
4726 | } | |
4727 | ||
4e3d62ff | 4728 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4729 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4730 | |
dabc4ec6 | 4731 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd | 4732 | spin_lock(&sh->stripe_lock); |
5ddf0440 SL |
4733 | /* |
4734 | * Cannot process 'sync' concurrently with 'discard'. | |
4735 | * Flush data in r5cache before 'sync'. | |
4736 | */ | |
4737 | if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) && | |
4738 | !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) && | |
4739 | !test_bit(STRIPE_DISCARD, &sh->state) && | |
f8dfcffd N |
4740 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
4741 | set_bit(STRIPE_SYNCING, &sh->state); | |
4742 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4743 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4744 | } |
4745 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4746 | } |
4747 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4748 | ||
4749 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4750 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4751 | (unsigned long long)sh->sector, sh->state, | |
4752 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4753 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4754 | |
acfe726b | 4755 | analyse_stripe(sh, &s); |
c5a31000 | 4756 | |
b70abcb2 SL |
4757 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4758 | goto finish; | |
4759 | ||
16d997b7 N |
4760 | if (s.handle_bad_blocks || |
4761 | test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { | |
bc2607f3 N |
4762 | set_bit(STRIPE_HANDLE, &sh->state); |
4763 | goto finish; | |
4764 | } | |
4765 | ||
474af965 N |
4766 | if (unlikely(s.blocked_rdev)) { |
4767 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4768 | s.replacing || s.to_write || s.written) { |
474af965 N |
4769 | set_bit(STRIPE_HANDLE, &sh->state); |
4770 | goto finish; | |
4771 | } | |
4772 | /* There is nothing for the blocked_rdev to block */ | |
4773 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4774 | s.blocked_rdev = NULL; | |
4775 | } | |
4776 | ||
4777 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4778 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4779 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4780 | } | |
4781 | ||
4782 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4783 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4784 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4785 | s.failed_num[0], s.failed_num[1]); | |
70d466f7 SL |
4786 | /* |
4787 | * check if the array has lost more than max_degraded devices and, | |
474af965 | 4788 | * if so, some requests might need to be failed. |
70d466f7 SL |
4789 | * |
4790 | * When journal device failed (log_failed), we will only process | |
4791 | * the stripe if there is data need write to raid disks | |
474af965 | 4792 | */ |
70d466f7 SL |
4793 | if (s.failed > conf->max_degraded || |
4794 | (s.log_failed && s.injournal == 0)) { | |
9a3f530f N |
4795 | sh->check_state = 0; |
4796 | sh->reconstruct_state = 0; | |
626f2092 | 4797 | break_stripe_batch_list(sh, 0); |
9a3f530f | 4798 | if (s.to_read+s.to_write+s.written) |
bd83d0a2 | 4799 | handle_failed_stripe(conf, sh, &s, disks); |
9a3e1101 | 4800 | if (s.syncing + s.replacing) |
9a3f530f N |
4801 | handle_failed_sync(conf, sh, &s); |
4802 | } | |
474af965 | 4803 | |
84789554 N |
4804 | /* Now we check to see if any write operations have recently |
4805 | * completed | |
4806 | */ | |
4807 | prexor = 0; | |
4808 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4809 | prexor = 1; | |
4810 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4811 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4812 | sh->reconstruct_state = reconstruct_state_idle; | |
4813 | ||
4814 | /* All the 'written' buffers and the parity block are ready to | |
4815 | * be written back to disk | |
4816 | */ | |
9e444768 SL |
4817 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4818 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4819 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4820 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4821 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4822 | for (i = disks; i--; ) { |
4823 | struct r5dev *dev = &sh->dev[i]; | |
4824 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4825 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4826 | dev->written || test_bit(R5_InJournal, |
4827 | &dev->flags))) { | |
84789554 N |
4828 | pr_debug("Writing block %d\n", i); |
4829 | set_bit(R5_Wantwrite, &dev->flags); | |
4830 | if (prexor) | |
4831 | continue; | |
9c4bdf69 N |
4832 | if (s.failed > 1) |
4833 | continue; | |
84789554 N |
4834 | if (!test_bit(R5_Insync, &dev->flags) || |
4835 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4836 | s.failed == 0)) | |
4837 | set_bit(STRIPE_INSYNC, &sh->state); | |
4838 | } | |
4839 | } | |
4840 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4841 | s.dec_preread_active = 1; | |
4842 | } | |
4843 | ||
ef5b7c69 N |
4844 | /* |
4845 | * might be able to return some write requests if the parity blocks | |
4846 | * are safe, or on a failed drive | |
4847 | */ | |
4848 | pdev = &sh->dev[sh->pd_idx]; | |
4849 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4850 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4851 | qdev = &sh->dev[sh->qd_idx]; | |
4852 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4853 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4854 | || conf->level < 6; | |
4855 | ||
4856 | if (s.written && | |
4857 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4858 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4859 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4860 | test_bit(R5_Discard, &pdev->flags))))) && | |
4861 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4862 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4863 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4864 | test_bit(R5_Discard, &qdev->flags)))))) | |
bd83d0a2 | 4865 | handle_stripe_clean_event(conf, sh, disks); |
ef5b7c69 | 4866 | |
1e6d690b | 4867 | if (s.just_cached) |
bd83d0a2 | 4868 | r5c_handle_cached_data_endio(conf, sh, disks); |
ff875738 | 4869 | log_stripe_write_finished(sh); |
1e6d690b | 4870 | |
ef5b7c69 N |
4871 | /* Now we might consider reading some blocks, either to check/generate |
4872 | * parity, or to satisfy requests | |
4873 | * or to load a block that is being partially written. | |
4874 | */ | |
4875 | if (s.to_read || s.non_overwrite | |
a1c6ae3d | 4876 | || (s.to_write && s.failed) |
ef5b7c69 N |
4877 | || (s.syncing && (s.uptodate + s.compute < disks)) |
4878 | || s.replacing | |
4879 | || s.expanding) | |
4880 | handle_stripe_fill(sh, &s, disks); | |
4881 | ||
2ded3703 SL |
4882 | /* |
4883 | * When the stripe finishes full journal write cycle (write to journal | |
4884 | * and raid disk), this is the clean up procedure so it is ready for | |
4885 | * next operation. | |
4886 | */ | |
4887 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4888 | ||
4889 | /* | |
4890 | * Now to consider new write requests, cache write back and what else, | |
4891 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4892 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4893 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4894 | * block. | |
2ded3703 | 4895 | * 3/ A r5c cache log write is in flight. |
84789554 | 4896 | */ |
2ded3703 SL |
4897 | |
4898 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4899 | if (!r5c_is_writeback(conf->log)) { | |
4900 | if (s.to_write) | |
4901 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4902 | } else { /* write back cache */ | |
4903 | int ret = 0; | |
4904 | ||
4905 | /* First, try handle writes in caching phase */ | |
4906 | if (s.to_write) | |
4907 | ret = r5c_try_caching_write(conf, sh, &s, | |
4908 | disks); | |
4909 | /* | |
4910 | * If caching phase failed: ret == -EAGAIN | |
4911 | * OR | |
4912 | * stripe under reclaim: !caching && injournal | |
4913 | * | |
4914 | * fall back to handle_stripe_dirtying() | |
4915 | */ | |
4916 | if (ret == -EAGAIN || | |
4917 | /* stripe under reclaim: !caching && injournal */ | |
4918 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4919 | s.injournal > 0)) { |
4920 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4921 | disks); | |
4922 | if (ret == -EAGAIN) | |
4923 | goto finish; | |
4924 | } | |
2ded3703 SL |
4925 | } |
4926 | } | |
84789554 N |
4927 | |
4928 | /* maybe we need to check and possibly fix the parity for this stripe | |
4929 | * Any reads will already have been scheduled, so we just see if enough | |
4930 | * data is available. The parity check is held off while parity | |
4931 | * dependent operations are in flight. | |
4932 | */ | |
4933 | if (sh->check_state || | |
4934 | (s.syncing && s.locked == 0 && | |
4935 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4936 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4937 | if (conf->level == 6) | |
4938 | handle_parity_checks6(conf, sh, &s, disks); | |
4939 | else | |
4940 | handle_parity_checks5(conf, sh, &s, disks); | |
4941 | } | |
c5a31000 | 4942 | |
f94c0b66 N |
4943 | if ((s.replacing || s.syncing) && s.locked == 0 |
4944 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4945 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4946 | /* Write out to replacement devices where possible */ |
4947 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4948 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4949 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4950 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4951 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4952 | s.locked++; | |
4953 | } | |
f94c0b66 N |
4954 | if (s.replacing) |
4955 | set_bit(STRIPE_INSYNC, &sh->state); | |
4956 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4957 | } |
4958 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4959 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4960 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c911c46c | 4961 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), 1); |
c5a31000 | 4962 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
4963 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4964 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4965 | } |
4966 | ||
4967 | /* If the failed drives are just a ReadError, then we might need | |
4968 | * to progress the repair/check process | |
4969 | */ | |
4970 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4971 | for (i = 0; i < s.failed; i++) { | |
4972 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4973 | if (test_bit(R5_ReadError, &dev->flags) | |
4974 | && !test_bit(R5_LOCKED, &dev->flags) | |
4975 | && test_bit(R5_UPTODATE, &dev->flags) | |
4976 | ) { | |
4977 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4978 | set_bit(R5_Wantwrite, &dev->flags); | |
4979 | set_bit(R5_ReWrite, &dev->flags); | |
3a31cf3d | 4980 | } else |
c5a31000 N |
4981 | /* let's read it back */ |
4982 | set_bit(R5_Wantread, &dev->flags); | |
3a31cf3d GJ |
4983 | set_bit(R5_LOCKED, &dev->flags); |
4984 | s.locked++; | |
c5a31000 N |
4985 | } |
4986 | } | |
4987 | ||
3687c061 N |
4988 | /* Finish reconstruct operations initiated by the expansion process */ |
4989 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4990 | struct stripe_head *sh_src | |
6d036f7d | 4991 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4992 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4993 | /* sh cannot be written until sh_src has been read. | |
4994 | * so arrange for sh to be delayed a little | |
4995 | */ | |
4996 | set_bit(STRIPE_DELAYED, &sh->state); | |
4997 | set_bit(STRIPE_HANDLE, &sh->state); | |
4998 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4999 | &sh_src->state)) | |
5000 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 5001 | raid5_release_stripe(sh_src); |
3687c061 N |
5002 | goto finish; |
5003 | } | |
5004 | if (sh_src) | |
6d036f7d | 5005 | raid5_release_stripe(sh_src); |
3687c061 N |
5006 | |
5007 | sh->reconstruct_state = reconstruct_state_idle; | |
5008 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
5009 | for (i = conf->raid_disks; i--; ) { | |
5010 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
5011 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
5012 | s.locked++; | |
5013 | } | |
5014 | } | |
f416885e | 5015 | |
3687c061 N |
5016 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
5017 | !sh->reconstruct_state) { | |
5018 | /* Need to write out all blocks after computing parity */ | |
5019 | sh->disks = conf->raid_disks; | |
5020 | stripe_set_idx(sh->sector, conf, 0, sh); | |
5021 | schedule_reconstruction(sh, &s, 1, 1); | |
5022 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
5023 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
5024 | atomic_dec(&conf->reshape_stripes); | |
5025 | wake_up(&conf->wait_for_overlap); | |
c911c46c | 5026 | md_done_sync(conf->mddev, RAID5_STRIPE_SECTORS(conf), 1); |
3687c061 N |
5027 | } |
5028 | ||
5029 | if (s.expanding && s.locked == 0 && | |
5030 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
5031 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 5032 | |
3687c061 | 5033 | finish: |
6bfe0b49 | 5034 | /* wait for this device to become unblocked */ |
5f066c63 N |
5035 | if (unlikely(s.blocked_rdev)) { |
5036 | if (conf->mddev->external) | |
5037 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
5038 | conf->mddev); | |
5039 | else | |
5040 | /* Internal metadata will immediately | |
5041 | * be written by raid5d, so we don't | |
5042 | * need to wait here. | |
5043 | */ | |
5044 | rdev_dec_pending(s.blocked_rdev, | |
5045 | conf->mddev); | |
5046 | } | |
6bfe0b49 | 5047 | |
bc2607f3 N |
5048 | if (s.handle_bad_blocks) |
5049 | for (i = disks; i--; ) { | |
3cb03002 | 5050 | struct md_rdev *rdev; |
bc2607f3 N |
5051 | struct r5dev *dev = &sh->dev[i]; |
5052 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
5053 | /* We own a safe reference to the rdev */ | |
5054 | rdev = conf->disks[i].rdev; | |
5055 | if (!rdev_set_badblocks(rdev, sh->sector, | |
c911c46c | 5056 | RAID5_STRIPE_SECTORS(conf), 0)) |
bc2607f3 N |
5057 | md_error(conf->mddev, rdev); |
5058 | rdev_dec_pending(rdev, conf->mddev); | |
5059 | } | |
b84db560 N |
5060 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
5061 | rdev = conf->disks[i].rdev; | |
5062 | rdev_clear_badblocks(rdev, sh->sector, | |
c911c46c | 5063 | RAID5_STRIPE_SECTORS(conf), 0); |
b84db560 N |
5064 | rdev_dec_pending(rdev, conf->mddev); |
5065 | } | |
977df362 N |
5066 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
5067 | rdev = conf->disks[i].replacement; | |
dd054fce N |
5068 | if (!rdev) |
5069 | /* rdev have been moved down */ | |
5070 | rdev = conf->disks[i].rdev; | |
977df362 | 5071 | rdev_clear_badblocks(rdev, sh->sector, |
c911c46c | 5072 | RAID5_STRIPE_SECTORS(conf), 0); |
977df362 N |
5073 | rdev_dec_pending(rdev, conf->mddev); |
5074 | } | |
bc2607f3 N |
5075 | } |
5076 | ||
6c0069c0 YT |
5077 | if (s.ops_request) |
5078 | raid_run_ops(sh, s.ops_request); | |
5079 | ||
f0e43bcd | 5080 | ops_run_io(sh, &s); |
16a53ecc | 5081 | |
c5709ef6 | 5082 | if (s.dec_preread_active) { |
729a1866 | 5083 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 5084 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
5085 | * have actually been submitted. |
5086 | */ | |
5087 | atomic_dec(&conf->preread_active_stripes); | |
5088 | if (atomic_read(&conf->preread_active_stripes) < | |
5089 | IO_THRESHOLD) | |
5090 | md_wakeup_thread(conf->mddev->thread); | |
5091 | } | |
5092 | ||
257a4b42 | 5093 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
5094 | } |
5095 | ||
d1688a6d | 5096 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
5097 | { |
5098 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
5099 | while (!list_empty(&conf->delayed_list)) { | |
5100 | struct list_head *l = conf->delayed_list.next; | |
5101 | struct stripe_head *sh; | |
5102 | sh = list_entry(l, struct stripe_head, lru); | |
5103 | list_del_init(l); | |
5104 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5105 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5106 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 5107 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 5108 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 5109 | } |
482c0834 | 5110 | } |
16a53ecc N |
5111 | } |
5112 | ||
566c09c5 SL |
5113 | static void activate_bit_delay(struct r5conf *conf, |
5114 | struct list_head *temp_inactive_list) | |
16a53ecc N |
5115 | { |
5116 | /* device_lock is held */ | |
5117 | struct list_head head; | |
5118 | list_add(&head, &conf->bitmap_list); | |
5119 | list_del_init(&conf->bitmap_list); | |
5120 | while (!list_empty(&head)) { | |
5121 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 5122 | int hash; |
16a53ecc N |
5123 | list_del_init(&sh->lru); |
5124 | atomic_inc(&sh->count); | |
566c09c5 SL |
5125 | hash = sh->hash_lock_index; |
5126 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
5127 | } |
5128 | } | |
5129 | ||
fd01b88c | 5130 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 5131 | { |
3cb5edf4 | 5132 | struct r5conf *conf = mddev->private; |
10433d04 | 5133 | sector_t sector = bio->bi_iter.bi_sector; |
3cb5edf4 | 5134 | unsigned int chunk_sectors; |
aa8b57aa | 5135 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 5136 | |
10433d04 CH |
5137 | WARN_ON_ONCE(bio->bi_partno); |
5138 | ||
3cb5edf4 | 5139 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
5140 | return chunk_sectors >= |
5141 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
5142 | } | |
5143 | ||
46031f9a RBJ |
5144 | /* |
5145 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
5146 | * later sampled by raid5d. | |
5147 | */ | |
d1688a6d | 5148 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
5149 | { |
5150 | unsigned long flags; | |
5151 | ||
5152 | spin_lock_irqsave(&conf->device_lock, flags); | |
5153 | ||
5154 | bi->bi_next = conf->retry_read_aligned_list; | |
5155 | conf->retry_read_aligned_list = bi; | |
5156 | ||
5157 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5158 | md_wakeup_thread(conf->mddev->thread); | |
5159 | } | |
5160 | ||
0472a42b N |
5161 | static struct bio *remove_bio_from_retry(struct r5conf *conf, |
5162 | unsigned int *offset) | |
46031f9a RBJ |
5163 | { |
5164 | struct bio *bi; | |
5165 | ||
5166 | bi = conf->retry_read_aligned; | |
5167 | if (bi) { | |
0472a42b | 5168 | *offset = conf->retry_read_offset; |
46031f9a RBJ |
5169 | conf->retry_read_aligned = NULL; |
5170 | return bi; | |
5171 | } | |
5172 | bi = conf->retry_read_aligned_list; | |
5173 | if(bi) { | |
387bb173 | 5174 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5175 | bi->bi_next = NULL; |
0472a42b | 5176 | *offset = 0; |
46031f9a RBJ |
5177 | } |
5178 | ||
5179 | return bi; | |
5180 | } | |
5181 | ||
f679623f RBJ |
5182 | /* |
5183 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5184 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5185 | * first). | |
5186 | * If the read failed.. | |
5187 | */ | |
4246a0b6 | 5188 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5189 | { |
5190 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5191 | struct mddev *mddev; |
d1688a6d | 5192 | struct r5conf *conf; |
3cb03002 | 5193 | struct md_rdev *rdev; |
4e4cbee9 | 5194 | blk_status_t error = bi->bi_status; |
46031f9a | 5195 | |
f679623f | 5196 | bio_put(bi); |
46031f9a | 5197 | |
46031f9a RBJ |
5198 | rdev = (void*)raid_bi->bi_next; |
5199 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5200 | mddev = rdev->mddev; |
5201 | conf = mddev->private; | |
46031f9a RBJ |
5202 | |
5203 | rdev_dec_pending(rdev, conf->mddev); | |
5204 | ||
9b81c842 | 5205 | if (!error) { |
4246a0b6 | 5206 | bio_endio(raid_bi); |
46031f9a | 5207 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5208 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5209 | return; |
46031f9a RBJ |
5210 | } |
5211 | ||
45b4233c | 5212 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5213 | |
5214 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5215 | } |
5216 | ||
7ef6b12a | 5217 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5218 | { |
d1688a6d | 5219 | struct r5conf *conf = mddev->private; |
8553fe7e | 5220 | int dd_idx; |
f679623f | 5221 | struct bio* align_bi; |
3cb03002 | 5222 | struct md_rdev *rdev; |
671488cc | 5223 | sector_t end_sector; |
f679623f RBJ |
5224 | |
5225 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5226 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5227 | return 0; |
5228 | } | |
5229 | /* | |
d7a10308 | 5230 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5231 | */ |
afeee514 | 5232 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, &mddev->bio_set); |
f679623f RBJ |
5233 | if (!align_bi) |
5234 | return 0; | |
5235 | /* | |
5236 | * set bi_end_io to a new function, and set bi_private to the | |
5237 | * original bio. | |
5238 | */ | |
5239 | align_bi->bi_end_io = raid5_align_endio; | |
5240 | align_bi->bi_private = raid_bio; | |
5241 | /* | |
5242 | * compute position | |
5243 | */ | |
4f024f37 KO |
5244 | align_bi->bi_iter.bi_sector = |
5245 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5246 | 0, &dd_idx, NULL); | |
f679623f | 5247 | |
f73a1c7d | 5248 | end_sector = bio_end_sector(align_bi); |
f679623f | 5249 | rcu_read_lock(); |
671488cc N |
5250 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5251 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5252 | rdev->recovery_offset < end_sector) { | |
5253 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5254 | if (rdev && | |
5255 | (test_bit(Faulty, &rdev->flags) || | |
5256 | !(test_bit(In_sync, &rdev->flags) || | |
5257 | rdev->recovery_offset >= end_sector))) | |
5258 | rdev = NULL; | |
5259 | } | |
03b047f4 SL |
5260 | |
5261 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5262 | rcu_read_unlock(); | |
5263 | bio_put(align_bi); | |
5264 | return 0; | |
5265 | } | |
5266 | ||
671488cc | 5267 | if (rdev) { |
31c176ec N |
5268 | sector_t first_bad; |
5269 | int bad_sectors; | |
5270 | ||
f679623f RBJ |
5271 | atomic_inc(&rdev->nr_pending); |
5272 | rcu_read_unlock(); | |
46031f9a | 5273 | raid_bio->bi_next = (void*)rdev; |
74d46992 | 5274 | bio_set_dev(align_bi, rdev->bdev); |
46031f9a | 5275 | |
7140aafc | 5276 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5277 | bio_sectors(align_bi), |
31c176ec | 5278 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5279 | bio_put(align_bi); |
5280 | rdev_dec_pending(rdev, mddev); | |
5281 | return 0; | |
5282 | } | |
5283 | ||
6c0544e2 | 5284 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5285 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5286 | |
46031f9a | 5287 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5288 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5289 | conf->quiesce == 0, |
eed8c02e | 5290 | conf->device_lock); |
46031f9a RBJ |
5291 | atomic_inc(&conf->active_aligned_reads); |
5292 | spin_unlock_irq(&conf->device_lock); | |
5293 | ||
e3620a3a | 5294 | if (mddev->gendisk) |
74d46992 | 5295 | trace_block_bio_remap(align_bi->bi_disk->queue, |
e3620a3a | 5296 | align_bi, disk_devt(mddev->gendisk), |
4f024f37 | 5297 | raid_bio->bi_iter.bi_sector); |
ed00aabd | 5298 | submit_bio_noacct(align_bi); |
f679623f RBJ |
5299 | return 1; |
5300 | } else { | |
5301 | rcu_read_unlock(); | |
46031f9a | 5302 | bio_put(align_bi); |
f679623f RBJ |
5303 | return 0; |
5304 | } | |
5305 | } | |
5306 | ||
7ef6b12a ML |
5307 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5308 | { | |
5309 | struct bio *split; | |
dd7a8f5d N |
5310 | sector_t sector = raid_bio->bi_iter.bi_sector; |
5311 | unsigned chunk_sects = mddev->chunk_sectors; | |
5312 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
7ef6b12a | 5313 | |
dd7a8f5d N |
5314 | if (sectors < bio_sectors(raid_bio)) { |
5315 | struct r5conf *conf = mddev->private; | |
afeee514 | 5316 | split = bio_split(raid_bio, sectors, GFP_NOIO, &conf->bio_split); |
dd7a8f5d | 5317 | bio_chain(split, raid_bio); |
ed00aabd | 5318 | submit_bio_noacct(raid_bio); |
dd7a8f5d N |
5319 | raid_bio = split; |
5320 | } | |
7ef6b12a | 5321 | |
dd7a8f5d N |
5322 | if (!raid5_read_one_chunk(mddev, raid_bio)) |
5323 | return raid_bio; | |
7ef6b12a ML |
5324 | |
5325 | return NULL; | |
5326 | } | |
5327 | ||
8b3e6cdc DW |
5328 | /* __get_priority_stripe - get the next stripe to process |
5329 | * | |
5330 | * Full stripe writes are allowed to pass preread active stripes up until | |
5331 | * the bypass_threshold is exceeded. In general the bypass_count | |
5332 | * increments when the handle_list is handled before the hold_list; however, it | |
5333 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5334 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5335 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5336 | * handle_list. | |
5337 | */ | |
851c30c9 | 5338 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5339 | { |
535ae4eb | 5340 | struct stripe_head *sh, *tmp; |
851c30c9 | 5341 | struct list_head *handle_list = NULL; |
535ae4eb | 5342 | struct r5worker_group *wg; |
70d466f7 SL |
5343 | bool second_try = !r5c_is_writeback(conf->log) && |
5344 | !r5l_log_disk_error(conf); | |
5345 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) || | |
5346 | r5l_log_disk_error(conf); | |
851c30c9 | 5347 | |
535ae4eb SL |
5348 | again: |
5349 | wg = NULL; | |
5350 | sh = NULL; | |
851c30c9 | 5351 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5352 | handle_list = try_loprio ? &conf->loprio_list : |
5353 | &conf->handle_list; | |
851c30c9 | 5354 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5355 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5356 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5357 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5358 | } else { |
5359 | int i; | |
5360 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5361 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5362 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5363 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5364 | if (!list_empty(handle_list)) |
5365 | break; | |
5366 | } | |
5367 | } | |
8b3e6cdc DW |
5368 | |
5369 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5370 | __func__, | |
851c30c9 | 5371 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5372 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5373 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5374 | ||
851c30c9 SL |
5375 | if (!list_empty(handle_list)) { |
5376 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5377 | |
5378 | if (list_empty(&conf->hold_list)) | |
5379 | conf->bypass_count = 0; | |
5380 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5381 | if (conf->hold_list.next == conf->last_hold) | |
5382 | conf->bypass_count++; | |
5383 | else { | |
5384 | conf->last_hold = conf->hold_list.next; | |
5385 | conf->bypass_count -= conf->bypass_threshold; | |
5386 | if (conf->bypass_count < 0) | |
5387 | conf->bypass_count = 0; | |
5388 | } | |
5389 | } | |
5390 | } else if (!list_empty(&conf->hold_list) && | |
5391 | ((conf->bypass_threshold && | |
5392 | conf->bypass_count > conf->bypass_threshold) || | |
5393 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5394 | |
5395 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5396 | if (conf->worker_cnt_per_group == 0 || | |
5397 | group == ANY_GROUP || | |
5398 | !cpu_online(tmp->cpu) || | |
5399 | cpu_to_group(tmp->cpu) == group) { | |
5400 | sh = tmp; | |
5401 | break; | |
5402 | } | |
5403 | } | |
5404 | ||
5405 | if (sh) { | |
5406 | conf->bypass_count -= conf->bypass_threshold; | |
5407 | if (conf->bypass_count < 0) | |
5408 | conf->bypass_count = 0; | |
5409 | } | |
bfc90cb0 | 5410 | wg = NULL; |
851c30c9 SL |
5411 | } |
5412 | ||
535ae4eb SL |
5413 | if (!sh) { |
5414 | if (second_try) | |
5415 | return NULL; | |
5416 | second_try = true; | |
5417 | try_loprio = !try_loprio; | |
5418 | goto again; | |
5419 | } | |
8b3e6cdc | 5420 | |
bfc90cb0 SL |
5421 | if (wg) { |
5422 | wg->stripes_cnt--; | |
5423 | sh->group = NULL; | |
5424 | } | |
8b3e6cdc | 5425 | list_del_init(&sh->lru); |
c7a6d35e | 5426 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5427 | return sh; |
5428 | } | |
f679623f | 5429 | |
8811b596 SL |
5430 | struct raid5_plug_cb { |
5431 | struct blk_plug_cb cb; | |
5432 | struct list_head list; | |
566c09c5 | 5433 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5434 | }; |
5435 | ||
5436 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5437 | { | |
5438 | struct raid5_plug_cb *cb = container_of( | |
5439 | blk_cb, struct raid5_plug_cb, cb); | |
5440 | struct stripe_head *sh; | |
5441 | struct mddev *mddev = cb->cb.data; | |
5442 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5443 | int cnt = 0; |
566c09c5 | 5444 | int hash; |
8811b596 SL |
5445 | |
5446 | if (cb->list.next && !list_empty(&cb->list)) { | |
5447 | spin_lock_irq(&conf->device_lock); | |
5448 | while (!list_empty(&cb->list)) { | |
5449 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5450 | list_del_init(&sh->lru); | |
5451 | /* | |
5452 | * avoid race release_stripe_plug() sees | |
5453 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5454 | * is still in our list | |
5455 | */ | |
4e857c58 | 5456 | smp_mb__before_atomic(); |
8811b596 | 5457 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5458 | /* |
5459 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5460 | * case, the count is always > 1 here | |
5461 | */ | |
566c09c5 SL |
5462 | hash = sh->hash_lock_index; |
5463 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5464 | cnt++; |
8811b596 SL |
5465 | } |
5466 | spin_unlock_irq(&conf->device_lock); | |
5467 | } | |
566c09c5 SL |
5468 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5469 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5470 | if (mddev->queue) |
5471 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5472 | kfree(cb); |
5473 | } | |
5474 | ||
5475 | static void release_stripe_plug(struct mddev *mddev, | |
5476 | struct stripe_head *sh) | |
5477 | { | |
5478 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5479 | raid5_unplug, mddev, | |
5480 | sizeof(struct raid5_plug_cb)); | |
5481 | struct raid5_plug_cb *cb; | |
5482 | ||
5483 | if (!blk_cb) { | |
6d036f7d | 5484 | raid5_release_stripe(sh); |
8811b596 SL |
5485 | return; |
5486 | } | |
5487 | ||
5488 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5489 | ||
566c09c5 SL |
5490 | if (cb->list.next == NULL) { |
5491 | int i; | |
8811b596 | 5492 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5493 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5494 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5495 | } | |
8811b596 SL |
5496 | |
5497 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5498 | list_add_tail(&sh->lru, &cb->list); | |
5499 | else | |
6d036f7d | 5500 | raid5_release_stripe(sh); |
8811b596 SL |
5501 | } |
5502 | ||
620125f2 SL |
5503 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5504 | { | |
5505 | struct r5conf *conf = mddev->private; | |
5506 | sector_t logical_sector, last_sector; | |
5507 | struct stripe_head *sh; | |
620125f2 SL |
5508 | int stripe_sectors; |
5509 | ||
5510 | if (mddev->reshape_position != MaxSector) | |
5511 | /* Skip discard while reshape is happening */ | |
5512 | return; | |
5513 | ||
c911c46c | 5514 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
b0f01ecf | 5515 | last_sector = bio_end_sector(bi); |
620125f2 SL |
5516 | |
5517 | bi->bi_next = NULL; | |
620125f2 SL |
5518 | |
5519 | stripe_sectors = conf->chunk_sectors * | |
5520 | (conf->raid_disks - conf->max_degraded); | |
5521 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5522 | stripe_sectors); | |
5523 | sector_div(last_sector, stripe_sectors); | |
5524 | ||
5525 | logical_sector *= conf->chunk_sectors; | |
5526 | last_sector *= conf->chunk_sectors; | |
5527 | ||
5528 | for (; logical_sector < last_sector; | |
c911c46c | 5529 | logical_sector += RAID5_STRIPE_SECTORS(conf)) { |
620125f2 SL |
5530 | DEFINE_WAIT(w); |
5531 | int d; | |
5532 | again: | |
6d036f7d | 5533 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5534 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5535 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5536 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5537 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5538 | raid5_release_stripe(sh); |
f8dfcffd N |
5539 | schedule(); |
5540 | goto again; | |
5541 | } | |
5542 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5543 | spin_lock_irq(&sh->stripe_lock); |
5544 | for (d = 0; d < conf->raid_disks; d++) { | |
5545 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5546 | continue; | |
5547 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5548 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5549 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5550 | raid5_release_stripe(sh); |
620125f2 SL |
5551 | schedule(); |
5552 | goto again; | |
5553 | } | |
5554 | } | |
f8dfcffd | 5555 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5556 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5557 | sh->overwrite_disks = 0; |
620125f2 SL |
5558 | for (d = 0; d < conf->raid_disks; d++) { |
5559 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5560 | continue; | |
5561 | sh->dev[d].towrite = bi; | |
5562 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
016c76ac | 5563 | bio_inc_remaining(bi); |
49728050 | 5564 | md_write_inc(mddev, bi); |
7a87f434 | 5565 | sh->overwrite_disks++; |
620125f2 SL |
5566 | } |
5567 | spin_unlock_irq(&sh->stripe_lock); | |
5568 | if (conf->mddev->bitmap) { | |
5569 | for (d = 0; | |
5570 | d < conf->raid_disks - conf->max_degraded; | |
5571 | d++) | |
e64e4018 AS |
5572 | md_bitmap_startwrite(mddev->bitmap, |
5573 | sh->sector, | |
c911c46c | 5574 | RAID5_STRIPE_SECTORS(conf), |
e64e4018 | 5575 | 0); |
620125f2 SL |
5576 | sh->bm_seq = conf->seq_flush + 1; |
5577 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5578 | } | |
5579 | ||
5580 | set_bit(STRIPE_HANDLE, &sh->state); | |
5581 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5582 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5583 | atomic_inc(&conf->preread_active_stripes); | |
5584 | release_stripe_plug(mddev, sh); | |
5585 | } | |
5586 | ||
016c76ac | 5587 | bio_endio(bi); |
620125f2 SL |
5588 | } |
5589 | ||
cc27b0c7 | 5590 | static bool raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5591 | { |
d1688a6d | 5592 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5593 | int dd_idx; |
1da177e4 LT |
5594 | sector_t new_sector; |
5595 | sector_t logical_sector, last_sector; | |
5596 | struct stripe_head *sh; | |
a362357b | 5597 | const int rw = bio_data_dir(bi); |
27c0f68f SL |
5598 | DEFINE_WAIT(w); |
5599 | bool do_prepare; | |
3bddb7f8 | 5600 | bool do_flush = false; |
1da177e4 | 5601 | |
1eff9d32 | 5602 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
1532d9e8 | 5603 | int ret = log_handle_flush_request(conf, bi); |
828cbe98 SL |
5604 | |
5605 | if (ret == 0) | |
cc27b0c7 | 5606 | return true; |
828cbe98 | 5607 | if (ret == -ENODEV) { |
775d7831 DJ |
5608 | if (md_flush_request(mddev, bi)) |
5609 | return true; | |
828cbe98 SL |
5610 | } |
5611 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5612 | /* |
5613 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5614 | * we need to flush journal device | |
5615 | */ | |
5616 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5617 | } |
5618 | ||
cc27b0c7 N |
5619 | if (!md_write_start(mddev, bi)) |
5620 | return false; | |
9ffc8f7c EM |
5621 | /* |
5622 | * If array is degraded, better not do chunk aligned read because | |
5623 | * later we might have to read it again in order to reconstruct | |
5624 | * data on failed drives. | |
5625 | */ | |
5626 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5627 | mddev->reshape_position == MaxSector) { |
5628 | bi = chunk_aligned_read(mddev, bi); | |
5629 | if (!bi) | |
cc27b0c7 | 5630 | return true; |
7ef6b12a | 5631 | } |
52488615 | 5632 | |
796a5cf0 | 5633 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 | 5634 | make_discard_request(mddev, bi); |
cc27b0c7 N |
5635 | md_write_end(mddev); |
5636 | return true; | |
620125f2 SL |
5637 | } |
5638 | ||
c911c46c | 5639 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
f73a1c7d | 5640 | last_sector = bio_end_sector(bi); |
1da177e4 | 5641 | bi->bi_next = NULL; |
06d91a5f | 5642 | |
27c0f68f | 5643 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
c911c46c | 5644 | for (; logical_sector < last_sector; logical_sector += RAID5_STRIPE_SECTORS(conf)) { |
b5663ba4 | 5645 | int previous; |
c46501b2 | 5646 | int seq; |
b578d55f | 5647 | |
27c0f68f | 5648 | do_prepare = false; |
7ecaa1e6 | 5649 | retry: |
c46501b2 | 5650 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5651 | previous = 0; |
27c0f68f SL |
5652 | if (do_prepare) |
5653 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5654 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5655 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5656 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5657 | * 64bit on a 32bit platform, and so it might be |
5658 | * possible to see a half-updated value | |
aeb878b0 | 5659 | * Of course reshape_progress could change after |
df8e7f76 N |
5660 | * the lock is dropped, so once we get a reference |
5661 | * to the stripe that we think it is, we will have | |
5662 | * to check again. | |
5663 | */ | |
7ecaa1e6 | 5664 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5665 | if (mddev->reshape_backwards |
fef9c61f N |
5666 | ? logical_sector < conf->reshape_progress |
5667 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5668 | previous = 1; |
5669 | } else { | |
2c810cdd | 5670 | if (mddev->reshape_backwards |
fef9c61f N |
5671 | ? logical_sector < conf->reshape_safe |
5672 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5673 | spin_unlock_irq(&conf->device_lock); |
5674 | schedule(); | |
27c0f68f | 5675 | do_prepare = true; |
b578d55f N |
5676 | goto retry; |
5677 | } | |
5678 | } | |
7ecaa1e6 N |
5679 | spin_unlock_irq(&conf->device_lock); |
5680 | } | |
16a53ecc | 5681 | |
112bf897 N |
5682 | new_sector = raid5_compute_sector(conf, logical_sector, |
5683 | previous, | |
911d4ee8 | 5684 | &dd_idx, NULL); |
849674e4 | 5685 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5686 | (unsigned long long)new_sector, |
1da177e4 LT |
5687 | (unsigned long long)logical_sector); |
5688 | ||
6d036f7d | 5689 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5690 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5691 | if (sh) { |
b0f9ec04 | 5692 | if (unlikely(previous)) { |
7ecaa1e6 | 5693 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5694 | * stripe, so we must do the range check again. |
5695 | * Expansion could still move past after this | |
5696 | * test, but as we are holding a reference to | |
5697 | * 'sh', we know that if that happens, | |
5698 | * STRIPE_EXPANDING will get set and the expansion | |
5699 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5700 | */ |
5701 | int must_retry = 0; | |
5702 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5703 | if (mddev->reshape_backwards |
b0f9ec04 N |
5704 | ? logical_sector >= conf->reshape_progress |
5705 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5706 | /* mismatch, need to try again */ |
5707 | must_retry = 1; | |
5708 | spin_unlock_irq(&conf->device_lock); | |
5709 | if (must_retry) { | |
6d036f7d | 5710 | raid5_release_stripe(sh); |
7a3ab908 | 5711 | schedule(); |
27c0f68f | 5712 | do_prepare = true; |
7ecaa1e6 N |
5713 | goto retry; |
5714 | } | |
5715 | } | |
c46501b2 N |
5716 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5717 | /* Might have got the wrong stripe_head | |
5718 | * by accident | |
5719 | */ | |
6d036f7d | 5720 | raid5_release_stripe(sh); |
c46501b2 N |
5721 | goto retry; |
5722 | } | |
e62e58a5 | 5723 | |
7ecaa1e6 | 5724 | if (test_bit(STRIPE_EXPANDING, &sh->state) || |
da41ba65 | 5725 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5726 | /* Stripe is busy expanding or |
5727 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5728 | * and wait a while |
5729 | */ | |
482c0834 | 5730 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5731 | raid5_release_stripe(sh); |
1da177e4 | 5732 | schedule(); |
27c0f68f | 5733 | do_prepare = true; |
1da177e4 LT |
5734 | goto retry; |
5735 | } | |
3bddb7f8 SL |
5736 | if (do_flush) { |
5737 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5738 | /* we only need flush for one stripe */ | |
5739 | do_flush = false; | |
5740 | } | |
5741 | ||
1684e975 | 5742 | set_bit(STRIPE_HANDLE, &sh->state); |
6ed3003c | 5743 | clear_bit(STRIPE_DELAYED, &sh->state); |
59fc630b | 5744 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5745 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5746 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5747 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5748 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5749 | } else { |
5750 | /* cannot get stripe for read-ahead, just give-up */ | |
4e4cbee9 | 5751 | bi->bi_status = BLK_STS_IOERR; |
1da177e4 LT |
5752 | break; |
5753 | } | |
1da177e4 | 5754 | } |
27c0f68f | 5755 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5756 | |
49728050 N |
5757 | if (rw == WRITE) |
5758 | md_write_end(mddev); | |
016c76ac | 5759 | bio_endio(bi); |
cc27b0c7 | 5760 | return true; |
1da177e4 LT |
5761 | } |
5762 | ||
fd01b88c | 5763 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5764 | |
fd01b88c | 5765 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5766 | { |
52c03291 N |
5767 | /* reshaping is quite different to recovery/resync so it is |
5768 | * handled quite separately ... here. | |
5769 | * | |
5770 | * On each call to sync_request, we gather one chunk worth of | |
5771 | * destination stripes and flag them as expanding. | |
5772 | * Then we find all the source stripes and request reads. | |
5773 | * As the reads complete, handle_stripe will copy the data | |
5774 | * into the destination stripe and release that stripe. | |
5775 | */ | |
d1688a6d | 5776 | struct r5conf *conf = mddev->private; |
1da177e4 | 5777 | struct stripe_head *sh; |
db0505d3 | 5778 | struct md_rdev *rdev; |
ccfcc3c1 | 5779 | sector_t first_sector, last_sector; |
f416885e N |
5780 | int raid_disks = conf->previous_raid_disks; |
5781 | int data_disks = raid_disks - conf->max_degraded; | |
5782 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5783 | int i; |
5784 | int dd_idx; | |
c8f517c4 | 5785 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5786 | sector_t stripe_addr; |
7a661381 | 5787 | int reshape_sectors; |
ab69ae12 | 5788 | struct list_head stripes; |
92140480 | 5789 | sector_t retn; |
52c03291 | 5790 | |
fef9c61f N |
5791 | if (sector_nr == 0) { |
5792 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5793 | if (mddev->reshape_backwards && |
fef9c61f N |
5794 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5795 | sector_nr = raid5_size(mddev, 0, 0) | |
5796 | - conf->reshape_progress; | |
6cbd8148 N |
5797 | } else if (mddev->reshape_backwards && |
5798 | conf->reshape_progress == MaxSector) { | |
5799 | /* shouldn't happen, but just in case, finish up.*/ | |
5800 | sector_nr = MaxSector; | |
2c810cdd | 5801 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5802 | conf->reshape_progress > 0) |
5803 | sector_nr = conf->reshape_progress; | |
f416885e | 5804 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5805 | if (sector_nr) { |
8dee7211 | 5806 | mddev->curr_resync_completed = sector_nr; |
e1a86dbb | 5807 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
fef9c61f | 5808 | *skipped = 1; |
92140480 N |
5809 | retn = sector_nr; |
5810 | goto finish; | |
fef9c61f | 5811 | } |
52c03291 N |
5812 | } |
5813 | ||
7a661381 N |
5814 | /* We need to process a full chunk at a time. |
5815 | * If old and new chunk sizes differ, we need to process the | |
5816 | * largest of these | |
5817 | */ | |
3cb5edf4 N |
5818 | |
5819 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5820 | |
b5254dd5 N |
5821 | /* We update the metadata at least every 10 seconds, or when |
5822 | * the data about to be copied would over-write the source of | |
5823 | * the data at the front of the range. i.e. one new_stripe | |
5824 | * along from reshape_progress new_maps to after where | |
5825 | * reshape_safe old_maps to | |
52c03291 | 5826 | */ |
fef9c61f | 5827 | writepos = conf->reshape_progress; |
f416885e | 5828 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5829 | readpos = conf->reshape_progress; |
5830 | sector_div(readpos, data_disks); | |
fef9c61f | 5831 | safepos = conf->reshape_safe; |
f416885e | 5832 | sector_div(safepos, data_disks); |
2c810cdd | 5833 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5834 | BUG_ON(writepos < reshape_sectors); |
5835 | writepos -= reshape_sectors; | |
c8f517c4 | 5836 | readpos += reshape_sectors; |
7a661381 | 5837 | safepos += reshape_sectors; |
fef9c61f | 5838 | } else { |
7a661381 | 5839 | writepos += reshape_sectors; |
c74c0d76 N |
5840 | /* readpos and safepos are worst-case calculations. |
5841 | * A negative number is overly pessimistic, and causes | |
5842 | * obvious problems for unsigned storage. So clip to 0. | |
5843 | */ | |
ed37d83e N |
5844 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5845 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5846 | } |
52c03291 | 5847 | |
b5254dd5 N |
5848 | /* Having calculated the 'writepos' possibly use it |
5849 | * to set 'stripe_addr' which is where we will write to. | |
5850 | */ | |
5851 | if (mddev->reshape_backwards) { | |
5852 | BUG_ON(conf->reshape_progress == 0); | |
5853 | stripe_addr = writepos; | |
5854 | BUG_ON((mddev->dev_sectors & | |
5855 | ~((sector_t)reshape_sectors - 1)) | |
5856 | - reshape_sectors - stripe_addr | |
5857 | != sector_nr); | |
5858 | } else { | |
5859 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5860 | stripe_addr = sector_nr; | |
5861 | } | |
5862 | ||
c8f517c4 N |
5863 | /* 'writepos' is the most advanced device address we might write. |
5864 | * 'readpos' is the least advanced device address we might read. | |
5865 | * 'safepos' is the least address recorded in the metadata as having | |
5866 | * been reshaped. | |
b5254dd5 N |
5867 | * If there is a min_offset_diff, these are adjusted either by |
5868 | * increasing the safepos/readpos if diff is negative, or | |
5869 | * increasing writepos if diff is positive. | |
5870 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5871 | * ensure safety in the face of a crash - that must be done by userspace |
5872 | * making a backup of the data. So in that case there is no particular | |
5873 | * rush to update metadata. | |
5874 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5875 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5876 | * we can be safe in the event of a crash. | |
5877 | * So we insist on updating metadata if safepos is behind writepos and | |
5878 | * readpos is beyond writepos. | |
5879 | * In any case, update the metadata every 10 seconds. | |
5880 | * Maybe that number should be configurable, but I'm not sure it is | |
5881 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5882 | */ | |
b5254dd5 N |
5883 | if (conf->min_offset_diff < 0) { |
5884 | safepos += -conf->min_offset_diff; | |
5885 | readpos += -conf->min_offset_diff; | |
5886 | } else | |
5887 | writepos += conf->min_offset_diff; | |
5888 | ||
2c810cdd | 5889 | if ((mddev->reshape_backwards |
c8f517c4 N |
5890 | ? (safepos > writepos && readpos < writepos) |
5891 | : (safepos < writepos && readpos > writepos)) || | |
5892 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5893 | /* Cannot proceed until we've updated the superblock... */ |
5894 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5895 | atomic_read(&conf->reshape_stripes)==0 |
5896 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5897 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5898 | return 0; | |
fef9c61f | 5899 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5900 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
5901 | if (!mddev->reshape_backwards) |
5902 | /* Can update recovery_offset */ | |
5903 | rdev_for_each(rdev, mddev) | |
5904 | if (rdev->raid_disk >= 0 && | |
5905 | !test_bit(Journal, &rdev->flags) && | |
5906 | !test_bit(In_sync, &rdev->flags) && | |
5907 | rdev->recovery_offset < sector_nr) | |
5908 | rdev->recovery_offset = sector_nr; | |
5909 | ||
c8f517c4 | 5910 | conf->reshape_checkpoint = jiffies; |
2953079c | 5911 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5912 | md_wakeup_thread(mddev->thread); |
2953079c | 5913 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5914 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5915 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5916 | return 0; | |
52c03291 | 5917 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5918 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5919 | spin_unlock_irq(&conf->device_lock); |
5920 | wake_up(&conf->wait_for_overlap); | |
e1a86dbb | 5921 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
52c03291 N |
5922 | } |
5923 | ||
ab69ae12 | 5924 | INIT_LIST_HEAD(&stripes); |
c911c46c | 5925 | for (i = 0; i < reshape_sectors; i += RAID5_STRIPE_SECTORS(conf)) { |
52c03291 | 5926 | int j; |
a9f326eb | 5927 | int skipped_disk = 0; |
6d036f7d | 5928 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5929 | set_bit(STRIPE_EXPANDING, &sh->state); |
5930 | atomic_inc(&conf->reshape_stripes); | |
5931 | /* If any of this stripe is beyond the end of the old | |
5932 | * array, then we need to zero those blocks | |
5933 | */ | |
5934 | for (j=sh->disks; j--;) { | |
5935 | sector_t s; | |
5936 | if (j == sh->pd_idx) | |
5937 | continue; | |
f416885e | 5938 | if (conf->level == 6 && |
d0dabf7e | 5939 | j == sh->qd_idx) |
f416885e | 5940 | continue; |
6d036f7d | 5941 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5942 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5943 | skipped_disk = 1; |
52c03291 N |
5944 | continue; |
5945 | } | |
c911c46c | 5946 | memset(page_address(sh->dev[j].page), 0, RAID5_STRIPE_SIZE(conf)); |
52c03291 N |
5947 | set_bit(R5_Expanded, &sh->dev[j].flags); |
5948 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5949 | } | |
a9f326eb | 5950 | if (!skipped_disk) { |
52c03291 N |
5951 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5952 | set_bit(STRIPE_HANDLE, &sh->state); | |
5953 | } | |
ab69ae12 | 5954 | list_add(&sh->lru, &stripes); |
52c03291 N |
5955 | } |
5956 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5957 | if (mddev->reshape_backwards) |
7a661381 | 5958 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5959 | else |
7a661381 | 5960 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5961 | spin_unlock_irq(&conf->device_lock); |
5962 | /* Ok, those stripe are ready. We can start scheduling | |
5963 | * reads on the source stripes. | |
5964 | * The source stripes are determined by mapping the first and last | |
5965 | * block on the destination stripes. | |
5966 | */ | |
52c03291 | 5967 | first_sector = |
ec32a2bd | 5968 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5969 | 1, &dd_idx, NULL); |
52c03291 | 5970 | last_sector = |
0e6e0271 | 5971 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5972 | * new_data_disks - 1), |
911d4ee8 | 5973 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5974 | if (last_sector >= mddev->dev_sectors) |
5975 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5976 | while (first_sector <= last_sector) { |
6d036f7d | 5977 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5978 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5979 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5980 | raid5_release_stripe(sh); |
c911c46c | 5981 | first_sector += RAID5_STRIPE_SECTORS(conf); |
52c03291 | 5982 | } |
ab69ae12 N |
5983 | /* Now that the sources are clearly marked, we can release |
5984 | * the destination stripes | |
5985 | */ | |
5986 | while (!list_empty(&stripes)) { | |
5987 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5988 | list_del_init(&sh->lru); | |
6d036f7d | 5989 | raid5_release_stripe(sh); |
ab69ae12 | 5990 | } |
c6207277 N |
5991 | /* If this takes us to the resync_max point where we have to pause, |
5992 | * then we need to write out the superblock. | |
5993 | */ | |
7a661381 | 5994 | sector_nr += reshape_sectors; |
92140480 N |
5995 | retn = reshape_sectors; |
5996 | finish: | |
c5e19d90 N |
5997 | if (mddev->curr_resync_completed > mddev->resync_max || |
5998 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5999 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
6000 | /* Cannot proceed until we've updated the superblock... */ |
6001 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
6002 | atomic_read(&conf->reshape_stripes) == 0 |
6003 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
6004 | if (atomic_read(&conf->reshape_stripes) != 0) | |
6005 | goto ret; | |
fef9c61f | 6006 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 6007 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
6008 | if (!mddev->reshape_backwards) |
6009 | /* Can update recovery_offset */ | |
6010 | rdev_for_each(rdev, mddev) | |
6011 | if (rdev->raid_disk >= 0 && | |
6012 | !test_bit(Journal, &rdev->flags) && | |
6013 | !test_bit(In_sync, &rdev->flags) && | |
6014 | rdev->recovery_offset < sector_nr) | |
6015 | rdev->recovery_offset = sector_nr; | |
c8f517c4 | 6016 | conf->reshape_checkpoint = jiffies; |
2953079c | 6017 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
6018 | md_wakeup_thread(mddev->thread); |
6019 | wait_event(mddev->sb_wait, | |
2953079c | 6020 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
6021 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
6022 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
6023 | goto ret; | |
c6207277 | 6024 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 6025 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
6026 | spin_unlock_irq(&conf->device_lock); |
6027 | wake_up(&conf->wait_for_overlap); | |
e1a86dbb | 6028 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
c6207277 | 6029 | } |
c91abf5a | 6030 | ret: |
92140480 | 6031 | return retn; |
52c03291 N |
6032 | } |
6033 | ||
849674e4 SL |
6034 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
6035 | int *skipped) | |
52c03291 | 6036 | { |
d1688a6d | 6037 | struct r5conf *conf = mddev->private; |
52c03291 | 6038 | struct stripe_head *sh; |
58c0fed4 | 6039 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 6040 | sector_t sync_blocks; |
16a53ecc N |
6041 | int still_degraded = 0; |
6042 | int i; | |
1da177e4 | 6043 | |
72626685 | 6044 | if (sector_nr >= max_sector) { |
1da177e4 | 6045 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 6046 | |
29269553 N |
6047 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
6048 | end_reshape(conf); | |
6049 | return 0; | |
6050 | } | |
72626685 N |
6051 | |
6052 | if (mddev->curr_resync < max_sector) /* aborted */ | |
e64e4018 AS |
6053 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
6054 | &sync_blocks, 1); | |
16a53ecc | 6055 | else /* completed sync */ |
72626685 | 6056 | conf->fullsync = 0; |
e64e4018 | 6057 | md_bitmap_close_sync(mddev->bitmap); |
72626685 | 6058 | |
1da177e4 LT |
6059 | return 0; |
6060 | } | |
ccfcc3c1 | 6061 | |
64bd660b N |
6062 | /* Allow raid5_quiesce to complete */ |
6063 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
6064 | ||
52c03291 N |
6065 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
6066 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 6067 | |
c6207277 N |
6068 | /* No need to check resync_max as we never do more than one |
6069 | * stripe, and as resync_max will always be on a chunk boundary, | |
6070 | * if the check in md_do_sync didn't fire, there is no chance | |
6071 | * of overstepping resync_max here | |
6072 | */ | |
6073 | ||
16a53ecc | 6074 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
6075 | * to resync, then assert that we are finished, because there is |
6076 | * nothing we can do. | |
6077 | */ | |
3285edf1 | 6078 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 6079 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 6080 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 6081 | *skipped = 1; |
1da177e4 LT |
6082 | return rv; |
6083 | } | |
6f608040 | 6084 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
6085 | !conf->fullsync && | |
e64e4018 | 6086 | !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
c911c46c | 6087 | sync_blocks >= RAID5_STRIPE_SECTORS(conf)) { |
72626685 | 6088 | /* we can skip this block, and probably more */ |
83c3e5e1 | 6089 | do_div(sync_blocks, RAID5_STRIPE_SECTORS(conf)); |
72626685 | 6090 | *skipped = 1; |
c911c46c YY |
6091 | /* keep things rounded to whole stripes */ |
6092 | return sync_blocks * RAID5_STRIPE_SECTORS(conf); | |
72626685 | 6093 | } |
1da177e4 | 6094 | |
e64e4018 | 6095 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 6096 | |
6d036f7d | 6097 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 6098 | if (sh == NULL) { |
6d036f7d | 6099 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 6100 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 6101 | * is trying to get access |
1da177e4 | 6102 | */ |
66c006a5 | 6103 | schedule_timeout_uninterruptible(1); |
1da177e4 | 6104 | } |
16a53ecc | 6105 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
6106 | * Note in case of > 1 drive failures it's possible we're rebuilding |
6107 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 6108 | */ |
16d9cfab EM |
6109 | rcu_read_lock(); |
6110 | for (i = 0; i < conf->raid_disks; i++) { | |
6aa7de05 | 6111 | struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev); |
16d9cfab EM |
6112 | |
6113 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 6114 | still_degraded = 1; |
16d9cfab EM |
6115 | } |
6116 | rcu_read_unlock(); | |
16a53ecc | 6117 | |
e64e4018 | 6118 | md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); |
16a53ecc | 6119 | |
83206d66 | 6120 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 6121 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 6122 | |
6d036f7d | 6123 | raid5_release_stripe(sh); |
1da177e4 | 6124 | |
c911c46c | 6125 | return RAID5_STRIPE_SECTORS(conf); |
1da177e4 LT |
6126 | } |
6127 | ||
0472a42b N |
6128 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio, |
6129 | unsigned int offset) | |
46031f9a RBJ |
6130 | { |
6131 | /* We may not be able to submit a whole bio at once as there | |
6132 | * may not be enough stripe_heads available. | |
6133 | * We cannot pre-allocate enough stripe_heads as we may need | |
6134 | * more than exist in the cache (if we allow ever large chunks). | |
6135 | * So we do one stripe head at a time and record in | |
6136 | * ->bi_hw_segments how many have been done. | |
6137 | * | |
6138 | * We *know* that this entire raid_bio is in one chunk, so | |
6139 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
6140 | */ | |
6141 | struct stripe_head *sh; | |
911d4ee8 | 6142 | int dd_idx; |
46031f9a RBJ |
6143 | sector_t sector, logical_sector, last_sector; |
6144 | int scnt = 0; | |
46031f9a RBJ |
6145 | int handled = 0; |
6146 | ||
4f024f37 | 6147 | logical_sector = raid_bio->bi_iter.bi_sector & |
c911c46c | 6148 | ~((sector_t)RAID5_STRIPE_SECTORS(conf)-1); |
112bf897 | 6149 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6150 | 0, &dd_idx, NULL); |
f73a1c7d | 6151 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6152 | |
6153 | for (; logical_sector < last_sector; | |
c911c46c YY |
6154 | logical_sector += RAID5_STRIPE_SECTORS(conf), |
6155 | sector += RAID5_STRIPE_SECTORS(conf), | |
387bb173 | 6156 | scnt++) { |
46031f9a | 6157 | |
0472a42b | 6158 | if (scnt < offset) |
46031f9a RBJ |
6159 | /* already done this stripe */ |
6160 | continue; | |
6161 | ||
6d036f7d | 6162 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6163 | |
6164 | if (!sh) { | |
6165 | /* failed to get a stripe - must wait */ | |
46031f9a | 6166 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6167 | conf->retry_read_offset = scnt; |
46031f9a RBJ |
6168 | return handled; |
6169 | } | |
6170 | ||
da41ba65 | 6171 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6172 | raid5_release_stripe(sh); |
387bb173 | 6173 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6174 | conf->retry_read_offset = scnt; |
387bb173 NB |
6175 | return handled; |
6176 | } | |
6177 | ||
3f9e7c14 | 6178 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6179 | handle_stripe(sh); |
6d036f7d | 6180 | raid5_release_stripe(sh); |
46031f9a RBJ |
6181 | handled++; |
6182 | } | |
016c76ac N |
6183 | |
6184 | bio_endio(raid_bio); | |
6185 | ||
46031f9a | 6186 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6187 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6188 | return handled; |
6189 | } | |
6190 | ||
bfc90cb0 | 6191 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6192 | struct r5worker *worker, |
6193 | struct list_head *temp_inactive_list) | |
efcd487c CH |
6194 | __releases(&conf->device_lock) |
6195 | __acquires(&conf->device_lock) | |
46a06401 SL |
6196 | { |
6197 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6198 | int i, batch_size = 0, hash; |
6199 | bool release_inactive = false; | |
46a06401 SL |
6200 | |
6201 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6202 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6203 | batch[batch_size++] = sh; |
6204 | ||
566c09c5 SL |
6205 | if (batch_size == 0) { |
6206 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6207 | if (!list_empty(temp_inactive_list + i)) | |
6208 | break; | |
a8c34f91 SL |
6209 | if (i == NR_STRIPE_HASH_LOCKS) { |
6210 | spin_unlock_irq(&conf->device_lock); | |
1532d9e8 | 6211 | log_flush_stripe_to_raid(conf); |
a8c34f91 | 6212 | spin_lock_irq(&conf->device_lock); |
566c09c5 | 6213 | return batch_size; |
a8c34f91 | 6214 | } |
566c09c5 SL |
6215 | release_inactive = true; |
6216 | } | |
46a06401 SL |
6217 | spin_unlock_irq(&conf->device_lock); |
6218 | ||
566c09c5 SL |
6219 | release_inactive_stripe_list(conf, temp_inactive_list, |
6220 | NR_STRIPE_HASH_LOCKS); | |
6221 | ||
a8c34f91 | 6222 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6223 | if (release_inactive) { |
6224 | spin_lock_irq(&conf->device_lock); | |
6225 | return 0; | |
6226 | } | |
6227 | ||
46a06401 SL |
6228 | for (i = 0; i < batch_size; i++) |
6229 | handle_stripe(batch[i]); | |
ff875738 | 6230 | log_write_stripe_run(conf); |
46a06401 SL |
6231 | |
6232 | cond_resched(); | |
6233 | ||
6234 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6235 | for (i = 0; i < batch_size; i++) { |
6236 | hash = batch[i]->hash_lock_index; | |
6237 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6238 | } | |
46a06401 SL |
6239 | return batch_size; |
6240 | } | |
46031f9a | 6241 | |
851c30c9 SL |
6242 | static void raid5_do_work(struct work_struct *work) |
6243 | { | |
6244 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6245 | struct r5worker_group *group = worker->group; | |
6246 | struct r5conf *conf = group->conf; | |
16d997b7 | 6247 | struct mddev *mddev = conf->mddev; |
851c30c9 SL |
6248 | int group_id = group - conf->worker_groups; |
6249 | int handled; | |
6250 | struct blk_plug plug; | |
6251 | ||
6252 | pr_debug("+++ raid5worker active\n"); | |
6253 | ||
6254 | blk_start_plug(&plug); | |
6255 | handled = 0; | |
6256 | spin_lock_irq(&conf->device_lock); | |
6257 | while (1) { | |
6258 | int batch_size, released; | |
6259 | ||
566c09c5 | 6260 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6261 | |
566c09c5 SL |
6262 | batch_size = handle_active_stripes(conf, group_id, worker, |
6263 | worker->temp_inactive_list); | |
bfc90cb0 | 6264 | worker->working = false; |
851c30c9 SL |
6265 | if (!batch_size && !released) |
6266 | break; | |
6267 | handled += batch_size; | |
16d997b7 N |
6268 | wait_event_lock_irq(mddev->sb_wait, |
6269 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags), | |
6270 | conf->device_lock); | |
851c30c9 SL |
6271 | } |
6272 | pr_debug("%d stripes handled\n", handled); | |
6273 | ||
6274 | spin_unlock_irq(&conf->device_lock); | |
7e96d559 | 6275 | |
9c72a18e SL |
6276 | flush_deferred_bios(conf); |
6277 | ||
6278 | r5l_flush_stripe_to_raid(conf->log); | |
6279 | ||
7e96d559 | 6280 | async_tx_issue_pending_all(); |
851c30c9 SL |
6281 | blk_finish_plug(&plug); |
6282 | ||
6283 | pr_debug("--- raid5worker inactive\n"); | |
6284 | } | |
6285 | ||
1da177e4 LT |
6286 | /* |
6287 | * This is our raid5 kernel thread. | |
6288 | * | |
6289 | * We scan the hash table for stripes which can be handled now. | |
6290 | * During the scan, completed stripes are saved for us by the interrupt | |
6291 | * handler, so that they will not have to wait for our next wakeup. | |
6292 | */ | |
4ed8731d | 6293 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6294 | { |
4ed8731d | 6295 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6296 | struct r5conf *conf = mddev->private; |
1da177e4 | 6297 | int handled; |
e1dfa0a2 | 6298 | struct blk_plug plug; |
1da177e4 | 6299 | |
45b4233c | 6300 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6301 | |
6302 | md_check_recovery(mddev); | |
1da177e4 | 6303 | |
e1dfa0a2 | 6304 | blk_start_plug(&plug); |
1da177e4 LT |
6305 | handled = 0; |
6306 | spin_lock_irq(&conf->device_lock); | |
6307 | while (1) { | |
46031f9a | 6308 | struct bio *bio; |
773ca82f | 6309 | int batch_size, released; |
0472a42b | 6310 | unsigned int offset; |
773ca82f | 6311 | |
566c09c5 | 6312 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6313 | if (released) |
6314 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6315 | |
0021b7bc | 6316 | if ( |
7c13edc8 N |
6317 | !list_empty(&conf->bitmap_list)) { |
6318 | /* Now is a good time to flush some bitmap updates */ | |
6319 | conf->seq_flush++; | |
700e432d | 6320 | spin_unlock_irq(&conf->device_lock); |
e64e4018 | 6321 | md_bitmap_unplug(mddev->bitmap); |
700e432d | 6322 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6323 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6324 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6325 | } |
0021b7bc | 6326 | raid5_activate_delayed(conf); |
72626685 | 6327 | |
0472a42b | 6328 | while ((bio = remove_bio_from_retry(conf, &offset))) { |
46031f9a RBJ |
6329 | int ok; |
6330 | spin_unlock_irq(&conf->device_lock); | |
0472a42b | 6331 | ok = retry_aligned_read(conf, bio, offset); |
46031f9a RBJ |
6332 | spin_lock_irq(&conf->device_lock); |
6333 | if (!ok) | |
6334 | break; | |
6335 | handled++; | |
6336 | } | |
6337 | ||
566c09c5 SL |
6338 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6339 | conf->temp_inactive_list); | |
773ca82f | 6340 | if (!batch_size && !released) |
1da177e4 | 6341 | break; |
46a06401 | 6342 | handled += batch_size; |
1da177e4 | 6343 | |
2953079c | 6344 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6345 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6346 | md_check_recovery(mddev); |
46a06401 SL |
6347 | spin_lock_irq(&conf->device_lock); |
6348 | } | |
1da177e4 | 6349 | } |
45b4233c | 6350 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6351 | |
6352 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6353 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6354 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6355 | grow_one_stripe(conf, __GFP_NOWARN); |
6356 | /* Set flag even if allocation failed. This helps | |
6357 | * slow down allocation requests when mem is short | |
6358 | */ | |
6359 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6360 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6361 | } |
1da177e4 | 6362 | |
765d704d SL |
6363 | flush_deferred_bios(conf); |
6364 | ||
0576b1c6 SL |
6365 | r5l_flush_stripe_to_raid(conf->log); |
6366 | ||
c9f21aaf | 6367 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6368 | blk_finish_plug(&plug); |
1da177e4 | 6369 | |
45b4233c | 6370 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6371 | } |
6372 | ||
3f294f4f | 6373 | static ssize_t |
fd01b88c | 6374 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6375 | { |
7b1485ba N |
6376 | struct r5conf *conf; |
6377 | int ret = 0; | |
6378 | spin_lock(&mddev->lock); | |
6379 | conf = mddev->private; | |
96de1e66 | 6380 | if (conf) |
edbe83ab | 6381 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6382 | spin_unlock(&mddev->lock); |
6383 | return ret; | |
3f294f4f N |
6384 | } |
6385 | ||
c41d4ac4 | 6386 | int |
fd01b88c | 6387 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6388 | { |
483cbbed | 6389 | int result = 0; |
d1688a6d | 6390 | struct r5conf *conf = mddev->private; |
b5470dc5 | 6391 | |
c41d4ac4 | 6392 | if (size <= 16 || size > 32768) |
3f294f4f | 6393 | return -EINVAL; |
486f0644 | 6394 | |
edbe83ab | 6395 | conf->min_nr_stripes = size; |
2d5b569b | 6396 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6397 | while (size < conf->max_nr_stripes && |
6398 | drop_one_stripe(conf)) | |
6399 | ; | |
2d5b569b | 6400 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6401 | |
2214c260 | 6402 | md_allow_write(mddev); |
486f0644 | 6403 | |
2d5b569b | 6404 | mutex_lock(&conf->cache_size_mutex); |
486f0644 | 6405 | while (size > conf->max_nr_stripes) |
483cbbed AN |
6406 | if (!grow_one_stripe(conf, GFP_KERNEL)) { |
6407 | conf->min_nr_stripes = conf->max_nr_stripes; | |
6408 | result = -ENOMEM; | |
486f0644 | 6409 | break; |
483cbbed | 6410 | } |
2d5b569b | 6411 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6412 | |
483cbbed | 6413 | return result; |
c41d4ac4 N |
6414 | } |
6415 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6416 | ||
6417 | static ssize_t | |
fd01b88c | 6418 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6419 | { |
6791875e | 6420 | struct r5conf *conf; |
c41d4ac4 N |
6421 | unsigned long new; |
6422 | int err; | |
6423 | ||
6424 | if (len >= PAGE_SIZE) | |
6425 | return -EINVAL; | |
b29bebd6 | 6426 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6427 | return -EINVAL; |
6791875e | 6428 | err = mddev_lock(mddev); |
c41d4ac4 N |
6429 | if (err) |
6430 | return err; | |
6791875e N |
6431 | conf = mddev->private; |
6432 | if (!conf) | |
6433 | err = -ENODEV; | |
6434 | else | |
6435 | err = raid5_set_cache_size(mddev, new); | |
6436 | mddev_unlock(mddev); | |
6437 | ||
6438 | return err ?: len; | |
3f294f4f | 6439 | } |
007583c9 | 6440 | |
96de1e66 N |
6441 | static struct md_sysfs_entry |
6442 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6443 | raid5_show_stripe_cache_size, | |
6444 | raid5_store_stripe_cache_size); | |
3f294f4f | 6445 | |
d06f191f MS |
6446 | static ssize_t |
6447 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6448 | { | |
6449 | struct r5conf *conf = mddev->private; | |
6450 | if (conf) | |
6451 | return sprintf(page, "%d\n", conf->rmw_level); | |
6452 | else | |
6453 | return 0; | |
6454 | } | |
6455 | ||
6456 | static ssize_t | |
6457 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6458 | { | |
6459 | struct r5conf *conf = mddev->private; | |
6460 | unsigned long new; | |
6461 | ||
6462 | if (!conf) | |
6463 | return -ENODEV; | |
6464 | ||
6465 | if (len >= PAGE_SIZE) | |
6466 | return -EINVAL; | |
6467 | ||
6468 | if (kstrtoul(page, 10, &new)) | |
6469 | return -EINVAL; | |
6470 | ||
6471 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6472 | return -EINVAL; | |
6473 | ||
6474 | if (new != PARITY_DISABLE_RMW && | |
6475 | new != PARITY_ENABLE_RMW && | |
6476 | new != PARITY_PREFER_RMW) | |
6477 | return -EINVAL; | |
6478 | ||
6479 | conf->rmw_level = new; | |
6480 | return len; | |
6481 | } | |
6482 | ||
6483 | static struct md_sysfs_entry | |
6484 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6485 | raid5_show_rmw_level, | |
6486 | raid5_store_rmw_level); | |
6487 | ||
3b5408b9 YY |
6488 | static ssize_t |
6489 | raid5_show_stripe_size(struct mddev *mddev, char *page) | |
6490 | { | |
6491 | struct r5conf *conf; | |
6492 | int ret = 0; | |
6493 | ||
6494 | spin_lock(&mddev->lock); | |
6495 | conf = mddev->private; | |
6496 | if (conf) | |
6497 | ret = sprintf(page, "%lu\n", RAID5_STRIPE_SIZE(conf)); | |
6498 | spin_unlock(&mddev->lock); | |
6499 | return ret; | |
6500 | } | |
6501 | ||
6502 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE | |
6503 | static ssize_t | |
6504 | raid5_store_stripe_size(struct mddev *mddev, const char *page, size_t len) | |
6505 | { | |
6506 | struct r5conf *conf; | |
6507 | unsigned long new; | |
6508 | int err; | |
6509 | ||
6510 | if (len >= PAGE_SIZE) | |
6511 | return -EINVAL; | |
6512 | if (kstrtoul(page, 10, &new)) | |
6513 | return -EINVAL; | |
6514 | ||
6515 | /* | |
6516 | * The value should not be bigger than PAGE_SIZE. It requires to | |
6517 | * be multiple of DEFAULT_STRIPE_SIZE. | |
6518 | */ | |
6519 | if (new % DEFAULT_STRIPE_SIZE != 0 || new > PAGE_SIZE || new == 0) | |
6520 | return -EINVAL; | |
6521 | ||
6522 | err = mddev_lock(mddev); | |
6523 | if (err) | |
6524 | return err; | |
6525 | ||
6526 | conf = mddev->private; | |
6527 | if (!conf) { | |
6528 | err = -ENODEV; | |
6529 | goto out_unlock; | |
6530 | } | |
6531 | ||
6532 | if (new == conf->stripe_size) | |
6533 | goto out_unlock; | |
6534 | ||
6535 | pr_debug("md/raid: change stripe_size from %lu to %lu\n", | |
6536 | conf->stripe_size, new); | |
6537 | ||
6538 | mddev_suspend(mddev); | |
6539 | conf->stripe_size = new; | |
6540 | conf->stripe_shift = ilog2(new) - 9; | |
6541 | conf->stripe_sectors = new >> 9; | |
6542 | mddev_resume(mddev); | |
6543 | ||
6544 | out_unlock: | |
6545 | mddev_unlock(mddev); | |
6546 | return err ?: len; | |
6547 | } | |
6548 | ||
6549 | static struct md_sysfs_entry | |
6550 | raid5_stripe_size = __ATTR(stripe_size, 0644, | |
6551 | raid5_show_stripe_size, | |
6552 | raid5_store_stripe_size); | |
6553 | #else | |
6554 | static struct md_sysfs_entry | |
6555 | raid5_stripe_size = __ATTR(stripe_size, 0444, | |
6556 | raid5_show_stripe_size, | |
6557 | NULL); | |
6558 | #endif | |
d06f191f | 6559 | |
8b3e6cdc | 6560 | static ssize_t |
fd01b88c | 6561 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6562 | { |
7b1485ba N |
6563 | struct r5conf *conf; |
6564 | int ret = 0; | |
6565 | spin_lock(&mddev->lock); | |
6566 | conf = mddev->private; | |
8b3e6cdc | 6567 | if (conf) |
7b1485ba N |
6568 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6569 | spin_unlock(&mddev->lock); | |
6570 | return ret; | |
8b3e6cdc DW |
6571 | } |
6572 | ||
6573 | static ssize_t | |
fd01b88c | 6574 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6575 | { |
6791875e | 6576 | struct r5conf *conf; |
4ef197d8 | 6577 | unsigned long new; |
6791875e N |
6578 | int err; |
6579 | ||
8b3e6cdc DW |
6580 | if (len >= PAGE_SIZE) |
6581 | return -EINVAL; | |
b29bebd6 | 6582 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6583 | return -EINVAL; |
6791875e N |
6584 | |
6585 | err = mddev_lock(mddev); | |
6586 | if (err) | |
6587 | return err; | |
6588 | conf = mddev->private; | |
6589 | if (!conf) | |
6590 | err = -ENODEV; | |
edbe83ab | 6591 | else if (new > conf->min_nr_stripes) |
6791875e N |
6592 | err = -EINVAL; |
6593 | else | |
6594 | conf->bypass_threshold = new; | |
6595 | mddev_unlock(mddev); | |
6596 | return err ?: len; | |
8b3e6cdc DW |
6597 | } |
6598 | ||
6599 | static struct md_sysfs_entry | |
6600 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6601 | S_IRUGO | S_IWUSR, | |
6602 | raid5_show_preread_threshold, | |
6603 | raid5_store_preread_threshold); | |
6604 | ||
d592a996 SL |
6605 | static ssize_t |
6606 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6607 | { | |
7b1485ba N |
6608 | struct r5conf *conf; |
6609 | int ret = 0; | |
6610 | spin_lock(&mddev->lock); | |
6611 | conf = mddev->private; | |
d592a996 | 6612 | if (conf) |
7b1485ba N |
6613 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6614 | spin_unlock(&mddev->lock); | |
6615 | return ret; | |
d592a996 SL |
6616 | } |
6617 | ||
6618 | static ssize_t | |
6619 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6620 | { | |
6791875e | 6621 | struct r5conf *conf; |
d592a996 | 6622 | unsigned long new; |
6791875e N |
6623 | int err; |
6624 | ||
d592a996 SL |
6625 | if (len >= PAGE_SIZE) |
6626 | return -EINVAL; | |
d592a996 SL |
6627 | if (kstrtoul(page, 10, &new)) |
6628 | return -EINVAL; | |
6629 | new = !!new; | |
6791875e N |
6630 | |
6631 | err = mddev_lock(mddev); | |
6632 | if (err) | |
6633 | return err; | |
6634 | conf = mddev->private; | |
6635 | if (!conf) | |
6636 | err = -ENODEV; | |
6637 | else if (new != conf->skip_copy) { | |
6638 | mddev_suspend(mddev); | |
6639 | conf->skip_copy = new; | |
6640 | if (new) | |
dc3b17cc | 6641 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6642 | BDI_CAP_STABLE_WRITES; |
6643 | else | |
dc3b17cc | 6644 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6645 | ~BDI_CAP_STABLE_WRITES; |
6646 | mddev_resume(mddev); | |
6647 | } | |
6648 | mddev_unlock(mddev); | |
6649 | return err ?: len; | |
d592a996 SL |
6650 | } |
6651 | ||
6652 | static struct md_sysfs_entry | |
6653 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6654 | raid5_show_skip_copy, | |
6655 | raid5_store_skip_copy); | |
6656 | ||
3f294f4f | 6657 | static ssize_t |
fd01b88c | 6658 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6659 | { |
d1688a6d | 6660 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6661 | if (conf) |
6662 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6663 | else | |
6664 | return 0; | |
3f294f4f N |
6665 | } |
6666 | ||
96de1e66 N |
6667 | static struct md_sysfs_entry |
6668 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6669 | |
b721420e SL |
6670 | static ssize_t |
6671 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6672 | { | |
7b1485ba N |
6673 | struct r5conf *conf; |
6674 | int ret = 0; | |
6675 | spin_lock(&mddev->lock); | |
6676 | conf = mddev->private; | |
b721420e | 6677 | if (conf) |
7b1485ba N |
6678 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6679 | spin_unlock(&mddev->lock); | |
6680 | return ret; | |
b721420e SL |
6681 | } |
6682 | ||
60aaf933 | 6683 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6684 | int *group_cnt, | |
60aaf933 | 6685 | struct r5worker_group **worker_groups); |
b721420e SL |
6686 | static ssize_t |
6687 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6688 | { | |
6791875e | 6689 | struct r5conf *conf; |
7d5d7b50 | 6690 | unsigned int new; |
b721420e | 6691 | int err; |
60aaf933 | 6692 | struct r5worker_group *new_groups, *old_groups; |
d2c9ad41 | 6693 | int group_cnt; |
b721420e SL |
6694 | |
6695 | if (len >= PAGE_SIZE) | |
6696 | return -EINVAL; | |
7d5d7b50 SL |
6697 | if (kstrtouint(page, 10, &new)) |
6698 | return -EINVAL; | |
6699 | /* 8192 should be big enough */ | |
6700 | if (new > 8192) | |
b721420e SL |
6701 | return -EINVAL; |
6702 | ||
6791875e N |
6703 | err = mddev_lock(mddev); |
6704 | if (err) | |
6705 | return err; | |
6706 | conf = mddev->private; | |
6707 | if (!conf) | |
6708 | err = -ENODEV; | |
6709 | else if (new != conf->worker_cnt_per_group) { | |
6710 | mddev_suspend(mddev); | |
b721420e | 6711 | |
6791875e N |
6712 | old_groups = conf->worker_groups; |
6713 | if (old_groups) | |
6714 | flush_workqueue(raid5_wq); | |
d206dcfa | 6715 | |
d2c9ad41 | 6716 | err = alloc_thread_groups(conf, new, &group_cnt, &new_groups); |
6791875e N |
6717 | if (!err) { |
6718 | spin_lock_irq(&conf->device_lock); | |
6719 | conf->group_cnt = group_cnt; | |
d2c9ad41 | 6720 | conf->worker_cnt_per_group = new; |
6791875e N |
6721 | conf->worker_groups = new_groups; |
6722 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6723 | |
6791875e N |
6724 | if (old_groups) |
6725 | kfree(old_groups[0].workers); | |
6726 | kfree(old_groups); | |
6727 | } | |
6728 | mddev_resume(mddev); | |
b721420e | 6729 | } |
6791875e | 6730 | mddev_unlock(mddev); |
b721420e | 6731 | |
6791875e | 6732 | return err ?: len; |
b721420e SL |
6733 | } |
6734 | ||
6735 | static struct md_sysfs_entry | |
6736 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6737 | raid5_show_group_thread_cnt, | |
6738 | raid5_store_group_thread_cnt); | |
6739 | ||
007583c9 | 6740 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6741 | &raid5_stripecache_size.attr, |
6742 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6743 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6744 | &raid5_group_thread_cnt.attr, |
d592a996 | 6745 | &raid5_skip_copy.attr, |
d06f191f | 6746 | &raid5_rmw_level.attr, |
3b5408b9 | 6747 | &raid5_stripe_size.attr, |
2c7da14b | 6748 | &r5c_journal_mode.attr, |
a596d086 | 6749 | &ppl_write_hint.attr, |
3f294f4f N |
6750 | NULL, |
6751 | }; | |
007583c9 N |
6752 | static struct attribute_group raid5_attrs_group = { |
6753 | .name = NULL, | |
6754 | .attrs = raid5_attrs, | |
3f294f4f N |
6755 | }; |
6756 | ||
d2c9ad41 | 6757 | static int alloc_thread_groups(struct r5conf *conf, int cnt, int *group_cnt, |
60aaf933 | 6758 | struct r5worker_group **worker_groups) |
851c30c9 | 6759 | { |
566c09c5 | 6760 | int i, j, k; |
851c30c9 SL |
6761 | ssize_t size; |
6762 | struct r5worker *workers; | |
6763 | ||
851c30c9 | 6764 | if (cnt == 0) { |
60aaf933 | 6765 | *group_cnt = 0; |
6766 | *worker_groups = NULL; | |
851c30c9 SL |
6767 | return 0; |
6768 | } | |
60aaf933 | 6769 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6770 | size = sizeof(struct r5worker) * cnt; |
6396bb22 KC |
6771 | workers = kcalloc(size, *group_cnt, GFP_NOIO); |
6772 | *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group), | |
6773 | GFP_NOIO); | |
60aaf933 | 6774 | if (!*worker_groups || !workers) { |
851c30c9 | 6775 | kfree(workers); |
60aaf933 | 6776 | kfree(*worker_groups); |
851c30c9 SL |
6777 | return -ENOMEM; |
6778 | } | |
6779 | ||
60aaf933 | 6780 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6781 | struct r5worker_group *group; |
6782 | ||
0c775d52 | 6783 | group = &(*worker_groups)[i]; |
851c30c9 | 6784 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6785 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6786 | group->conf = conf; |
6787 | group->workers = workers + i * cnt; | |
6788 | ||
6789 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6790 | struct r5worker *worker = group->workers + j; |
6791 | worker->group = group; | |
6792 | INIT_WORK(&worker->work, raid5_do_work); | |
6793 | ||
6794 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6795 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6796 | } |
6797 | } | |
6798 | ||
6799 | return 0; | |
6800 | } | |
6801 | ||
6802 | static void free_thread_groups(struct r5conf *conf) | |
6803 | { | |
6804 | if (conf->worker_groups) | |
6805 | kfree(conf->worker_groups[0].workers); | |
6806 | kfree(conf->worker_groups); | |
6807 | conf->worker_groups = NULL; | |
6808 | } | |
6809 | ||
80c3a6ce | 6810 | static sector_t |
fd01b88c | 6811 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6812 | { |
d1688a6d | 6813 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6814 | |
6815 | if (!sectors) | |
6816 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6817 | if (!raid_disks) |
7ec05478 | 6818 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6819 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6820 | |
3cb5edf4 N |
6821 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6822 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6823 | return sectors * (raid_disks - conf->max_degraded); |
6824 | } | |
6825 | ||
789b5e03 ON |
6826 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6827 | { | |
6828 | safe_put_page(percpu->spare_page); | |
789b5e03 | 6829 | percpu->spare_page = NULL; |
b330e6a4 | 6830 | kvfree(percpu->scribble); |
789b5e03 ON |
6831 | percpu->scribble = NULL; |
6832 | } | |
6833 | ||
6834 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6835 | { | |
b330e6a4 | 6836 | if (conf->level == 6 && !percpu->spare_page) { |
789b5e03 | 6837 | percpu->spare_page = alloc_page(GFP_KERNEL); |
b330e6a4 KO |
6838 | if (!percpu->spare_page) |
6839 | return -ENOMEM; | |
6840 | } | |
6841 | ||
6842 | if (scribble_alloc(percpu, | |
6843 | max(conf->raid_disks, | |
6844 | conf->previous_raid_disks), | |
6845 | max(conf->chunk_sectors, | |
6846 | conf->prev_chunk_sectors) | |
c911c46c | 6847 | / RAID5_STRIPE_SECTORS(conf))) { |
789b5e03 ON |
6848 | free_scratch_buffer(conf, percpu); |
6849 | return -ENOMEM; | |
6850 | } | |
6851 | ||
6852 | return 0; | |
6853 | } | |
6854 | ||
29c6d1bb | 6855 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6856 | { |
29c6d1bb SAS |
6857 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6858 | ||
6859 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6860 | return 0; | |
6861 | } | |
36d1c647 | 6862 | |
29c6d1bb SAS |
6863 | static void raid5_free_percpu(struct r5conf *conf) |
6864 | { | |
36d1c647 DW |
6865 | if (!conf->percpu) |
6866 | return; | |
6867 | ||
29c6d1bb | 6868 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6869 | free_percpu(conf->percpu); |
6870 | } | |
6871 | ||
d1688a6d | 6872 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6873 | { |
d7bd398e SL |
6874 | int i; |
6875 | ||
ff875738 AP |
6876 | log_exit(conf); |
6877 | ||
565e0450 | 6878 | unregister_shrinker(&conf->shrinker); |
851c30c9 | 6879 | free_thread_groups(conf); |
95fc17aa | 6880 | shrink_stripes(conf); |
36d1c647 | 6881 | raid5_free_percpu(conf); |
d7bd398e SL |
6882 | for (i = 0; i < conf->pool_size; i++) |
6883 | if (conf->disks[i].extra_page) | |
6884 | put_page(conf->disks[i].extra_page); | |
95fc17aa | 6885 | kfree(conf->disks); |
afeee514 | 6886 | bioset_exit(&conf->bio_split); |
95fc17aa | 6887 | kfree(conf->stripe_hashtbl); |
aaf9f12e | 6888 | kfree(conf->pending_data); |
95fc17aa DW |
6889 | kfree(conf); |
6890 | } | |
6891 | ||
29c6d1bb | 6892 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6893 | { |
29c6d1bb | 6894 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6895 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6896 | ||
29c6d1bb | 6897 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6898 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6899 | __func__, cpu); | |
29c6d1bb | 6900 | return -ENOMEM; |
36d1c647 | 6901 | } |
29c6d1bb | 6902 | return 0; |
36d1c647 | 6903 | } |
36d1c647 | 6904 | |
d1688a6d | 6905 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6906 | { |
789b5e03 | 6907 | int err = 0; |
36d1c647 | 6908 | |
789b5e03 ON |
6909 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6910 | if (!conf->percpu) | |
36d1c647 | 6911 | return -ENOMEM; |
789b5e03 | 6912 | |
29c6d1bb | 6913 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6914 | if (!err) { |
6915 | conf->scribble_disks = max(conf->raid_disks, | |
6916 | conf->previous_raid_disks); | |
6917 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6918 | conf->prev_chunk_sectors); | |
6919 | } | |
36d1c647 DW |
6920 | return err; |
6921 | } | |
6922 | ||
edbe83ab N |
6923 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6924 | struct shrink_control *sc) | |
6925 | { | |
6926 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6927 | unsigned long ret = SHRINK_STOP; |
6928 | ||
6929 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6930 | ret= 0; | |
49895bcc N |
6931 | while (ret < sc->nr_to_scan && |
6932 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6933 | if (drop_one_stripe(conf) == 0) { |
6934 | ret = SHRINK_STOP; | |
6935 | break; | |
6936 | } | |
6937 | ret++; | |
6938 | } | |
6939 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6940 | } |
6941 | return ret; | |
6942 | } | |
6943 | ||
6944 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6945 | struct shrink_control *sc) | |
6946 | { | |
6947 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6948 | ||
6949 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6950 | /* unlikely, but not impossible */ | |
6951 | return 0; | |
6952 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6953 | } | |
6954 | ||
d1688a6d | 6955 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6956 | { |
d1688a6d | 6957 | struct r5conf *conf; |
5e5e3e78 | 6958 | int raid_disk, memory, max_disks; |
3cb03002 | 6959 | struct md_rdev *rdev; |
1da177e4 | 6960 | struct disk_info *disk; |
0232605d | 6961 | char pers_name[6]; |
566c09c5 | 6962 | int i; |
d2c9ad41 | 6963 | int group_cnt; |
60aaf933 | 6964 | struct r5worker_group *new_group; |
afeee514 | 6965 | int ret; |
1da177e4 | 6966 | |
91adb564 N |
6967 | if (mddev->new_level != 5 |
6968 | && mddev->new_level != 4 | |
6969 | && mddev->new_level != 6) { | |
cc6167b4 N |
6970 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6971 | mdname(mddev), mddev->new_level); | |
91adb564 | 6972 | return ERR_PTR(-EIO); |
1da177e4 | 6973 | } |
91adb564 N |
6974 | if ((mddev->new_level == 5 |
6975 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6976 | (mddev->new_level == 6 | |
6977 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6978 | pr_warn("md/raid:%s: layout %d not supported\n", |
6979 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6980 | return ERR_PTR(-EIO); |
99c0fb5f | 6981 | } |
91adb564 | 6982 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6983 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6984 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6985 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6986 | } |
6987 | ||
664e7c41 AN |
6988 | if (!mddev->new_chunk_sectors || |
6989 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6990 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6991 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6992 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6993 | return ERR_PTR(-EINVAL); |
f6705578 N |
6994 | } |
6995 | ||
d1688a6d | 6996 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6997 | if (conf == NULL) |
1da177e4 | 6998 | goto abort; |
c911c46c | 6999 | |
e2368582 YY |
7000 | #if PAGE_SIZE != DEFAULT_STRIPE_SIZE |
7001 | conf->stripe_size = DEFAULT_STRIPE_SIZE; | |
7002 | conf->stripe_shift = ilog2(DEFAULT_STRIPE_SIZE) - 9; | |
7003 | conf->stripe_sectors = DEFAULT_STRIPE_SIZE >> 9; | |
7004 | #endif | |
aaf9f12e SL |
7005 | INIT_LIST_HEAD(&conf->free_list); |
7006 | INIT_LIST_HEAD(&conf->pending_list); | |
6396bb22 KC |
7007 | conf->pending_data = kcalloc(PENDING_IO_MAX, |
7008 | sizeof(struct r5pending_data), | |
7009 | GFP_KERNEL); | |
aaf9f12e SL |
7010 | if (!conf->pending_data) |
7011 | goto abort; | |
7012 | for (i = 0; i < PENDING_IO_MAX; i++) | |
7013 | list_add(&conf->pending_data[i].sibling, &conf->free_list); | |
851c30c9 | 7014 | /* Don't enable multi-threading by default*/ |
d2c9ad41 | 7015 | if (!alloc_thread_groups(conf, 0, &group_cnt, &new_group)) { |
60aaf933 | 7016 | conf->group_cnt = group_cnt; |
d2c9ad41 | 7017 | conf->worker_cnt_per_group = 0; |
60aaf933 | 7018 | conf->worker_groups = new_group; |
7019 | } else | |
851c30c9 | 7020 | goto abort; |
f5efd45a | 7021 | spin_lock_init(&conf->device_lock); |
0a87b25f | 7022 | seqcount_spinlock_init(&conf->gen_lock, &conf->device_lock); |
2d5b569b | 7023 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 7024 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 7025 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
7026 | init_waitqueue_head(&conf->wait_for_overlap); |
7027 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 7028 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
7029 | INIT_LIST_HEAD(&conf->hold_list); |
7030 | INIT_LIST_HEAD(&conf->delayed_list); | |
7031 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 7032 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
7033 | atomic_set(&conf->active_stripes, 0); |
7034 | atomic_set(&conf->preread_active_stripes, 0); | |
7035 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
7036 | spin_lock_init(&conf->pending_bios_lock); |
7037 | conf->batch_bio_dispatch = true; | |
7038 | rdev_for_each(rdev, mddev) { | |
7039 | if (test_bit(Journal, &rdev->flags)) | |
7040 | continue; | |
7041 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
7042 | conf->batch_bio_dispatch = false; | |
7043 | break; | |
7044 | } | |
7045 | } | |
7046 | ||
f5efd45a | 7047 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 7048 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
7049 | |
7050 | conf->raid_disks = mddev->raid_disks; | |
7051 | if (mddev->reshape_position == MaxSector) | |
7052 | conf->previous_raid_disks = mddev->raid_disks; | |
7053 | else | |
f6705578 | 7054 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 7055 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 7056 | |
6396bb22 | 7057 | conf->disks = kcalloc(max_disks, sizeof(struct disk_info), |
b55e6bfc | 7058 | GFP_KERNEL); |
d7bd398e | 7059 | |
b55e6bfc N |
7060 | if (!conf->disks) |
7061 | goto abort; | |
9ffae0cf | 7062 | |
d7bd398e SL |
7063 | for (i = 0; i < max_disks; i++) { |
7064 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
7065 | if (!conf->disks[i].extra_page) | |
7066 | goto abort; | |
7067 | } | |
7068 | ||
afeee514 KO |
7069 | ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
7070 | if (ret) | |
dd7a8f5d | 7071 | goto abort; |
1da177e4 LT |
7072 | conf->mddev = mddev; |
7073 | ||
fccddba0 | 7074 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 7075 | goto abort; |
1da177e4 | 7076 | |
566c09c5 SL |
7077 | /* We init hash_locks[0] separately to that it can be used |
7078 | * as the reference lock in the spin_lock_nest_lock() call | |
7079 | * in lock_all_device_hash_locks_irq in order to convince | |
7080 | * lockdep that we know what we are doing. | |
7081 | */ | |
7082 | spin_lock_init(conf->hash_locks); | |
7083 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
7084 | spin_lock_init(conf->hash_locks + i); | |
7085 | ||
7086 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
7087 | INIT_LIST_HEAD(conf->inactive_list + i); | |
7088 | ||
7089 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
7090 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
7091 | ||
1e6d690b SL |
7092 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
7093 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
7094 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
7095 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
7096 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
7097 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 7098 | |
36d1c647 | 7099 | conf->level = mddev->new_level; |
46d5b785 | 7100 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
7101 | if (raid5_alloc_percpu(conf) != 0) |
7102 | goto abort; | |
7103 | ||
0c55e022 | 7104 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 7105 | |
dafb20fa | 7106 | rdev_for_each(rdev, mddev) { |
1da177e4 | 7107 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 7108 | if (raid_disk >= max_disks |
f2076e7d | 7109 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
7110 | continue; |
7111 | disk = conf->disks + raid_disk; | |
7112 | ||
17045f52 N |
7113 | if (test_bit(Replacement, &rdev->flags)) { |
7114 | if (disk->replacement) | |
7115 | goto abort; | |
7116 | disk->replacement = rdev; | |
7117 | } else { | |
7118 | if (disk->rdev) | |
7119 | goto abort; | |
7120 | disk->rdev = rdev; | |
7121 | } | |
1da177e4 | 7122 | |
b2d444d7 | 7123 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 7124 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
7125 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
7126 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 7127 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
7128 | /* Cannot rely on bitmap to complete recovery */ |
7129 | conf->fullsync = 1; | |
1da177e4 LT |
7130 | } |
7131 | ||
91adb564 | 7132 | conf->level = mddev->new_level; |
584acdd4 | 7133 | if (conf->level == 6) { |
16a53ecc | 7134 | conf->max_degraded = 2; |
584acdd4 MS |
7135 | if (raid6_call.xor_syndrome) |
7136 | conf->rmw_level = PARITY_ENABLE_RMW; | |
7137 | else | |
7138 | conf->rmw_level = PARITY_DISABLE_RMW; | |
7139 | } else { | |
16a53ecc | 7140 | conf->max_degraded = 1; |
584acdd4 MS |
7141 | conf->rmw_level = PARITY_ENABLE_RMW; |
7142 | } | |
91adb564 | 7143 | conf->algorithm = mddev->new_layout; |
fef9c61f | 7144 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 7145 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 7146 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 7147 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
7148 | } else { |
7149 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
7150 | conf->prev_algo = conf->algorithm; | |
e183eaed | 7151 | } |
1da177e4 | 7152 | |
edbe83ab | 7153 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
7154 | if (mddev->reshape_position != MaxSector) { |
7155 | int stripes = max_t(int, | |
c911c46c YY |
7156 | ((mddev->chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4, |
7157 | ((mddev->new_chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4); | |
ad5b0f76 SL |
7158 | conf->min_nr_stripes = max(NR_STRIPES, stripes); |
7159 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 7160 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
7161 | mdname(mddev), conf->min_nr_stripes); |
7162 | } | |
edbe83ab | 7163 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 7164 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 7165 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 7166 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
7167 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
7168 | mdname(mddev), memory); | |
91adb564 N |
7169 | goto abort; |
7170 | } else | |
cc6167b4 | 7171 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
7172 | /* |
7173 | * Losing a stripe head costs more than the time to refill it, | |
7174 | * it reduces the queue depth and so can hurt throughput. | |
7175 | * So set it rather large, scaled by number of devices. | |
7176 | */ | |
7177 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
7178 | conf->shrinker.scan_objects = raid5_cache_scan; | |
7179 | conf->shrinker.count_objects = raid5_cache_count; | |
7180 | conf->shrinker.batch = 128; | |
7181 | conf->shrinker.flags = 0; | |
6a0f53ff | 7182 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
7183 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
7184 | mdname(mddev)); | |
6a0f53ff CY |
7185 | goto abort; |
7186 | } | |
1da177e4 | 7187 | |
0232605d N |
7188 | sprintf(pers_name, "raid%d", mddev->new_level); |
7189 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 7190 | if (!conf->thread) { |
cc6167b4 N |
7191 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
7192 | mdname(mddev)); | |
16a53ecc N |
7193 | goto abort; |
7194 | } | |
91adb564 N |
7195 | |
7196 | return conf; | |
7197 | ||
7198 | abort: | |
7199 | if (conf) { | |
95fc17aa | 7200 | free_conf(conf); |
91adb564 N |
7201 | return ERR_PTR(-EIO); |
7202 | } else | |
7203 | return ERR_PTR(-ENOMEM); | |
7204 | } | |
7205 | ||
c148ffdc N |
7206 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
7207 | { | |
7208 | switch (algo) { | |
7209 | case ALGORITHM_PARITY_0: | |
7210 | if (raid_disk < max_degraded) | |
7211 | return 1; | |
7212 | break; | |
7213 | case ALGORITHM_PARITY_N: | |
7214 | if (raid_disk >= raid_disks - max_degraded) | |
7215 | return 1; | |
7216 | break; | |
7217 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 7218 | if (raid_disk == 0 || |
c148ffdc N |
7219 | raid_disk == raid_disks - 1) |
7220 | return 1; | |
7221 | break; | |
7222 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7223 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7224 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7225 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7226 | if (raid_disk == raid_disks - 1) | |
7227 | return 1; | |
7228 | } | |
7229 | return 0; | |
7230 | } | |
7231 | ||
849674e4 | 7232 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7233 | { |
d1688a6d | 7234 | struct r5conf *conf; |
9f7c2220 | 7235 | int working_disks = 0; |
c148ffdc | 7236 | int dirty_parity_disks = 0; |
3cb03002 | 7237 | struct md_rdev *rdev; |
713cf5a6 | 7238 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7239 | sector_t reshape_offset = 0; |
17045f52 | 7240 | int i; |
b5254dd5 N |
7241 | long long min_offset_diff = 0; |
7242 | int first = 1; | |
91adb564 | 7243 | |
a415c0f1 N |
7244 | if (mddev_init_writes_pending(mddev) < 0) |
7245 | return -ENOMEM; | |
7246 | ||
8c6ac868 | 7247 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7248 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7249 | mdname(mddev)); | |
b5254dd5 N |
7250 | |
7251 | rdev_for_each(rdev, mddev) { | |
7252 | long long diff; | |
713cf5a6 | 7253 | |
f2076e7d | 7254 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7255 | journal_dev = rdev; |
f2076e7d SL |
7256 | continue; |
7257 | } | |
b5254dd5 N |
7258 | if (rdev->raid_disk < 0) |
7259 | continue; | |
7260 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7261 | if (first) { | |
7262 | min_offset_diff = diff; | |
7263 | first = 0; | |
7264 | } else if (mddev->reshape_backwards && | |
7265 | diff < min_offset_diff) | |
7266 | min_offset_diff = diff; | |
7267 | else if (!mddev->reshape_backwards && | |
7268 | diff > min_offset_diff) | |
7269 | min_offset_diff = diff; | |
7270 | } | |
7271 | ||
230b55fa N |
7272 | if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) && |
7273 | (mddev->bitmap_info.offset || mddev->bitmap_info.file)) { | |
7274 | pr_notice("md/raid:%s: array cannot have both journal and bitmap\n", | |
7275 | mdname(mddev)); | |
7276 | return -EINVAL; | |
7277 | } | |
7278 | ||
91adb564 N |
7279 | if (mddev->reshape_position != MaxSector) { |
7280 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7281 | * Difficulties arise if the stripe we would write to |
7282 | * next is at or after the stripe we would read from next. | |
7283 | * For a reshape that changes the number of devices, this | |
7284 | * is only possible for a very short time, and mdadm makes | |
7285 | * sure that time appears to have past before assembling | |
7286 | * the array. So we fail if that time hasn't passed. | |
7287 | * For a reshape that keeps the number of devices the same | |
7288 | * mdadm must be monitoring the reshape can keeping the | |
7289 | * critical areas read-only and backed up. It will start | |
7290 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7291 | */ |
7292 | sector_t here_new, here_old; | |
7293 | int old_disks; | |
18b00334 | 7294 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7295 | int chunk_sectors; |
7296 | int new_data_disks; | |
91adb564 | 7297 | |
713cf5a6 | 7298 | if (journal_dev) { |
cc6167b4 N |
7299 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7300 | mdname(mddev)); | |
713cf5a6 SL |
7301 | return -EINVAL; |
7302 | } | |
7303 | ||
88ce4930 | 7304 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7305 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7306 | mdname(mddev)); | |
91adb564 N |
7307 | return -EINVAL; |
7308 | } | |
91adb564 N |
7309 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7310 | /* reshape_position must be on a new-stripe boundary, and one | |
7311 | * further up in new geometry must map after here in old | |
7312 | * geometry. | |
05256d98 N |
7313 | * If the chunk sizes are different, then as we perform reshape |
7314 | * in units of the largest of the two, reshape_position needs | |
7315 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7316 | */ |
7317 | here_new = mddev->reshape_position; | |
05256d98 N |
7318 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7319 | new_data_disks = mddev->raid_disks - max_degraded; | |
7320 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7321 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7322 | mdname(mddev)); | |
91adb564 N |
7323 | return -EINVAL; |
7324 | } | |
05256d98 | 7325 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7326 | /* here_new is the stripe we will write to */ |
7327 | here_old = mddev->reshape_position; | |
05256d98 | 7328 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7329 | /* here_old is the first stripe that we might need to read |
7330 | * from */ | |
67ac6011 N |
7331 | if (mddev->delta_disks == 0) { |
7332 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7333 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7334 | * and taking constant backups. |
7335 | * mdadm always starts a situation like this in | |
7336 | * readonly mode so it can take control before | |
7337 | * allowing any writes. So just check for that. | |
7338 | */ | |
b5254dd5 N |
7339 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7340 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7341 | /* not really in-place - so OK */; | |
7342 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7343 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7344 | mdname(mddev)); | |
67ac6011 N |
7345 | return -EINVAL; |
7346 | } | |
2c810cdd | 7347 | } else if (mddev->reshape_backwards |
05256d98 N |
7348 | ? (here_new * chunk_sectors + min_offset_diff <= |
7349 | here_old * chunk_sectors) | |
7350 | : (here_new * chunk_sectors >= | |
7351 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7352 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7353 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7354 | mdname(mddev)); | |
91adb564 N |
7355 | return -EINVAL; |
7356 | } | |
cc6167b4 | 7357 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7358 | /* OK, we should be able to continue; */ |
7359 | } else { | |
7360 | BUG_ON(mddev->level != mddev->new_level); | |
7361 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7362 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7363 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7364 | } |
91adb564 | 7365 | |
3418d036 AP |
7366 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && |
7367 | test_bit(MD_HAS_PPL, &mddev->flags)) { | |
7368 | pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n", | |
7369 | mdname(mddev)); | |
7370 | clear_bit(MD_HAS_PPL, &mddev->flags); | |
ddc08823 | 7371 | clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags); |
3418d036 AP |
7372 | } |
7373 | ||
245f46c2 N |
7374 | if (mddev->private == NULL) |
7375 | conf = setup_conf(mddev); | |
7376 | else | |
7377 | conf = mddev->private; | |
7378 | ||
91adb564 N |
7379 | if (IS_ERR(conf)) |
7380 | return PTR_ERR(conf); | |
7381 | ||
486b0f7b SL |
7382 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7383 | if (!journal_dev) { | |
cc6167b4 N |
7384 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7385 | mdname(mddev)); | |
486b0f7b SL |
7386 | mddev->ro = 1; |
7387 | set_disk_ro(mddev->gendisk, 1); | |
7388 | } else if (mddev->recovery_cp == MaxSector) | |
7389 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7390 | } |
7391 | ||
b5254dd5 | 7392 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7393 | mddev->thread = conf->thread; |
7394 | conf->thread = NULL; | |
7395 | mddev->private = conf; | |
7396 | ||
17045f52 N |
7397 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7398 | i++) { | |
7399 | rdev = conf->disks[i].rdev; | |
7400 | if (!rdev && conf->disks[i].replacement) { | |
7401 | /* The replacement is all we have yet */ | |
7402 | rdev = conf->disks[i].replacement; | |
7403 | conf->disks[i].replacement = NULL; | |
7404 | clear_bit(Replacement, &rdev->flags); | |
7405 | conf->disks[i].rdev = rdev; | |
7406 | } | |
7407 | if (!rdev) | |
c148ffdc | 7408 | continue; |
17045f52 N |
7409 | if (conf->disks[i].replacement && |
7410 | conf->reshape_progress != MaxSector) { | |
7411 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7412 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7413 | goto abort; |
7414 | } | |
2f115882 | 7415 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7416 | working_disks++; |
2f115882 N |
7417 | continue; |
7418 | } | |
c148ffdc N |
7419 | /* This disc is not fully in-sync. However if it |
7420 | * just stored parity (beyond the recovery_offset), | |
7421 | * when we don't need to be concerned about the | |
7422 | * array being dirty. | |
7423 | * When reshape goes 'backwards', we never have | |
7424 | * partially completed devices, so we only need | |
7425 | * to worry about reshape going forwards. | |
7426 | */ | |
7427 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7428 | if (mddev->major_version == 0 && | |
7429 | mddev->minor_version > 90) | |
7430 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7431 | |
c148ffdc N |
7432 | if (rdev->recovery_offset < reshape_offset) { |
7433 | /* We need to check old and new layout */ | |
7434 | if (!only_parity(rdev->raid_disk, | |
7435 | conf->algorithm, | |
7436 | conf->raid_disks, | |
7437 | conf->max_degraded)) | |
7438 | continue; | |
7439 | } | |
7440 | if (!only_parity(rdev->raid_disk, | |
7441 | conf->prev_algo, | |
7442 | conf->previous_raid_disks, | |
7443 | conf->max_degraded)) | |
7444 | continue; | |
7445 | dirty_parity_disks++; | |
7446 | } | |
91adb564 | 7447 | |
17045f52 N |
7448 | /* |
7449 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7450 | */ | |
2e38a37f | 7451 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7452 | |
674806d6 | 7453 | if (has_failed(conf)) { |
cc6167b4 | 7454 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7455 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7456 | goto abort; |
7457 | } | |
7458 | ||
91adb564 | 7459 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7460 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7461 | mddev->resync_max_sectors = mddev->dev_sectors; |
7462 | ||
c148ffdc | 7463 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7464 | mddev->recovery_cp != MaxSector) { |
4536bf9b AP |
7465 | if (test_bit(MD_HAS_PPL, &mddev->flags)) |
7466 | pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n", | |
7467 | mdname(mddev)); | |
7468 | else if (mddev->ok_start_degraded) | |
cc6167b4 N |
7469 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7470 | mdname(mddev)); | |
6ff8d8ec | 7471 | else { |
cc6167b4 N |
7472 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7473 | mdname(mddev)); | |
6ff8d8ec N |
7474 | goto abort; |
7475 | } | |
1da177e4 LT |
7476 | } |
7477 | ||
cc6167b4 N |
7478 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7479 | mdname(mddev), conf->level, | |
7480 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7481 | mddev->new_layout); | |
1da177e4 LT |
7482 | |
7483 | print_raid5_conf(conf); | |
7484 | ||
fef9c61f | 7485 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7486 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7487 | atomic_set(&conf->reshape_stripes, 0); |
7488 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7489 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7490 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7491 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7492 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7493 | "reshape"); |
e406f12d AP |
7494 | if (!mddev->sync_thread) |
7495 | goto abort; | |
f6705578 N |
7496 | } |
7497 | ||
1da177e4 | 7498 | /* Ok, everything is just fine now */ |
a64c876f N |
7499 | if (mddev->to_remove == &raid5_attrs_group) |
7500 | mddev->to_remove = NULL; | |
00bcb4ac N |
7501 | else if (mddev->kobj.sd && |
7502 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7503 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7504 | mdname(mddev)); | |
4a5add49 | 7505 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7506 | |
4a5add49 | 7507 | if (mddev->queue) { |
9f7c2220 | 7508 | int chunk_size; |
4a5add49 N |
7509 | /* read-ahead size must cover two whole stripes, which |
7510 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7511 | * number of raid devices | |
7512 | */ | |
7513 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7514 | int stripe = data_disks * | |
7515 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7516 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7517 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7518 | |
9f7c2220 N |
7519 | chunk_size = mddev->chunk_sectors << 9; |
7520 | blk_queue_io_min(mddev->queue, chunk_size); | |
7521 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7522 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7523 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7524 | /* |
7525 | * We can only discard a whole stripe. It doesn't make sense to | |
7526 | * discard data disk but write parity disk | |
7527 | */ | |
7528 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7529 | /* Round up to power of 2, as discard handling |
7530 | * currently assumes that */ | |
7531 | while ((stripe-1) & stripe) | |
7532 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7533 | mddev->queue->limits.discard_alignment = stripe; |
7534 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 | 7535 | |
5026d7a9 | 7536 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 7537 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
5026d7a9 | 7538 | |
05616be5 | 7539 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7540 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7541 | rdev->data_offset << 9); | |
05616be5 N |
7542 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7543 | rdev->new_data_offset << 9); | |
7544 | } | |
620125f2 | 7545 | |
48920ff2 CH |
7546 | /* |
7547 | * zeroing is required, otherwise data | |
7548 | * could be lost. Consider a scenario: discard a stripe | |
7549 | * (the stripe could be inconsistent if | |
7550 | * discard_zeroes_data is 0); write one disk of the | |
7551 | * stripe (the stripe could be inconsistent again | |
7552 | * depending on which disks are used to calculate | |
7553 | * parity); the disk is broken; The stripe data of this | |
7554 | * disk is lost. | |
7555 | * | |
7556 | * We only allow DISCARD if the sysadmin has confirmed that | |
7557 | * only safe devices are in use by setting a module parameter. | |
7558 | * A better idea might be to turn DISCARD into WRITE_ZEROES | |
7559 | * requests, as that is required to be safe. | |
7560 | */ | |
7561 | if (devices_handle_discard_safely && | |
e7597e69 JS |
7562 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7563 | mddev->queue->limits.discard_granularity >= stripe) | |
8b904b5b | 7564 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
620125f2 SL |
7565 | mddev->queue); |
7566 | else | |
8b904b5b | 7567 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
620125f2 | 7568 | mddev->queue); |
1dffdddd SL |
7569 | |
7570 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7571 | } |
23032a0e | 7572 | |
845b9e22 | 7573 | if (log_init(conf, journal_dev, raid5_has_ppl(conf))) |
ff875738 | 7574 | goto abort; |
5c7e81c3 | 7575 | |
1da177e4 LT |
7576 | return 0; |
7577 | abort: | |
01f96c0a | 7578 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7579 | print_raid5_conf(conf); |
7580 | free_conf(conf); | |
1da177e4 | 7581 | mddev->private = NULL; |
cc6167b4 | 7582 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7583 | return -EIO; |
7584 | } | |
7585 | ||
afa0f557 | 7586 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7587 | { |
afa0f557 | 7588 | struct r5conf *conf = priv; |
1da177e4 | 7589 | |
95fc17aa | 7590 | free_conf(conf); |
a64c876f | 7591 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7592 | } |
7593 | ||
849674e4 | 7594 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7595 | { |
d1688a6d | 7596 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7597 | int i; |
7598 | ||
9d8f0363 | 7599 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7600 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7601 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7602 | rcu_read_lock(); |
7603 | for (i = 0; i < conf->raid_disks; i++) { | |
7604 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7605 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7606 | } | |
7607 | rcu_read_unlock(); | |
1da177e4 | 7608 | seq_printf (seq, "]"); |
1da177e4 LT |
7609 | } |
7610 | ||
d1688a6d | 7611 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7612 | { |
7613 | int i; | |
7614 | struct disk_info *tmp; | |
7615 | ||
cc6167b4 | 7616 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7617 | if (!conf) { |
cc6167b4 | 7618 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7619 | return; |
7620 | } | |
cc6167b4 | 7621 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7622 | conf->raid_disks, |
7623 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7624 | |
7625 | for (i = 0; i < conf->raid_disks; i++) { | |
7626 | char b[BDEVNAME_SIZE]; | |
7627 | tmp = conf->disks + i; | |
7628 | if (tmp->rdev) | |
cc6167b4 | 7629 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7630 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7631 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7632 | } |
7633 | } | |
7634 | ||
fd01b88c | 7635 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7636 | { |
7637 | int i; | |
d1688a6d | 7638 | struct r5conf *conf = mddev->private; |
1da177e4 | 7639 | struct disk_info *tmp; |
6b965620 N |
7640 | int count = 0; |
7641 | unsigned long flags; | |
1da177e4 LT |
7642 | |
7643 | for (i = 0; i < conf->raid_disks; i++) { | |
7644 | tmp = conf->disks + i; | |
dd054fce N |
7645 | if (tmp->replacement |
7646 | && tmp->replacement->recovery_offset == MaxSector | |
7647 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7648 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7649 | /* Replacement has just become active. */ | |
7650 | if (!tmp->rdev | |
7651 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7652 | count++; | |
7653 | if (tmp->rdev) { | |
7654 | /* Replaced device not technically faulty, | |
7655 | * but we need to be sure it gets removed | |
7656 | * and never re-added. | |
7657 | */ | |
7658 | set_bit(Faulty, &tmp->rdev->flags); | |
7659 | sysfs_notify_dirent_safe( | |
7660 | tmp->rdev->sysfs_state); | |
7661 | } | |
7662 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7663 | } else if (tmp->rdev | |
70fffd0b | 7664 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7665 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7666 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7667 | count++; |
43c73ca4 | 7668 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7669 | } |
7670 | } | |
6b965620 | 7671 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7672 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7673 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7674 | print_raid5_conf(conf); |
6b965620 | 7675 | return count; |
1da177e4 LT |
7676 | } |
7677 | ||
b8321b68 | 7678 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7679 | { |
d1688a6d | 7680 | struct r5conf *conf = mddev->private; |
1da177e4 | 7681 | int err = 0; |
b8321b68 | 7682 | int number = rdev->raid_disk; |
657e3e4d | 7683 | struct md_rdev **rdevp; |
1da177e4 LT |
7684 | struct disk_info *p = conf->disks + number; |
7685 | ||
7686 | print_raid5_conf(conf); | |
f6b6ec5c | 7687 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
c2bb6242 | 7688 | /* |
f6b6ec5c SL |
7689 | * we can't wait pending write here, as this is called in |
7690 | * raid5d, wait will deadlock. | |
84dd97a6 N |
7691 | * neilb: there is no locking about new writes here, |
7692 | * so this cannot be safe. | |
c2bb6242 | 7693 | */ |
70d466f7 SL |
7694 | if (atomic_read(&conf->active_stripes) || |
7695 | atomic_read(&conf->r5c_cached_full_stripes) || | |
7696 | atomic_read(&conf->r5c_cached_partial_stripes)) { | |
f6b6ec5c | 7697 | return -EBUSY; |
84dd97a6 | 7698 | } |
ff875738 | 7699 | log_exit(conf); |
f6b6ec5c | 7700 | return 0; |
c2bb6242 | 7701 | } |
657e3e4d N |
7702 | if (rdev == p->rdev) |
7703 | rdevp = &p->rdev; | |
7704 | else if (rdev == p->replacement) | |
7705 | rdevp = &p->replacement; | |
7706 | else | |
7707 | return 0; | |
7708 | ||
7709 | if (number >= conf->raid_disks && | |
7710 | conf->reshape_progress == MaxSector) | |
7711 | clear_bit(In_sync, &rdev->flags); | |
7712 | ||
7713 | if (test_bit(In_sync, &rdev->flags) || | |
7714 | atomic_read(&rdev->nr_pending)) { | |
7715 | err = -EBUSY; | |
7716 | goto abort; | |
7717 | } | |
7718 | /* Only remove non-faulty devices if recovery | |
7719 | * isn't possible. | |
7720 | */ | |
7721 | if (!test_bit(Faulty, &rdev->flags) && | |
7722 | mddev->recovery_disabled != conf->recovery_disabled && | |
7723 | !has_failed(conf) && | |
dd054fce | 7724 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7725 | number < conf->raid_disks) { |
7726 | err = -EBUSY; | |
7727 | goto abort; | |
7728 | } | |
7729 | *rdevp = NULL; | |
d787be40 N |
7730 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7731 | synchronize_rcu(); | |
7732 | if (atomic_read(&rdev->nr_pending)) { | |
7733 | /* lost the race, try later */ | |
7734 | err = -EBUSY; | |
7735 | *rdevp = rdev; | |
7736 | } | |
7737 | } | |
6358c239 AP |
7738 | if (!err) { |
7739 | err = log_modify(conf, rdev, false); | |
7740 | if (err) | |
7741 | goto abort; | |
7742 | } | |
d787be40 | 7743 | if (p->replacement) { |
dd054fce N |
7744 | /* We must have just cleared 'rdev' */ |
7745 | p->rdev = p->replacement; | |
7746 | clear_bit(Replacement, &p->replacement->flags); | |
7747 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7748 | * but will never see neither - if they are careful | |
7749 | */ | |
7750 | p->replacement = NULL; | |
6358c239 AP |
7751 | |
7752 | if (!err) | |
7753 | err = log_modify(conf, p->rdev, true); | |
e5bc9c3c GJ |
7754 | } |
7755 | ||
7756 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7757 | abort: |
7758 | ||
7759 | print_raid5_conf(conf); | |
7760 | return err; | |
7761 | } | |
7762 | ||
fd01b88c | 7763 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7764 | { |
d1688a6d | 7765 | struct r5conf *conf = mddev->private; |
d9771f5e | 7766 | int ret, err = -EEXIST; |
1da177e4 LT |
7767 | int disk; |
7768 | struct disk_info *p; | |
6c2fce2e NB |
7769 | int first = 0; |
7770 | int last = conf->raid_disks - 1; | |
1da177e4 | 7771 | |
f6b6ec5c | 7772 | if (test_bit(Journal, &rdev->flags)) { |
f6b6ec5c SL |
7773 | if (conf->log) |
7774 | return -EBUSY; | |
7775 | ||
7776 | rdev->raid_disk = 0; | |
7777 | /* | |
7778 | * The array is in readonly mode if journal is missing, so no | |
7779 | * write requests running. We should be safe | |
7780 | */ | |
d9771f5e XN |
7781 | ret = log_init(conf, rdev, false); |
7782 | if (ret) | |
7783 | return ret; | |
7784 | ||
7785 | ret = r5l_start(conf->log); | |
7786 | if (ret) | |
7787 | return ret; | |
7788 | ||
f6b6ec5c SL |
7789 | return 0; |
7790 | } | |
7f0da59b N |
7791 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7792 | return -EBUSY; | |
7793 | ||
dc10c643 | 7794 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7795 | /* no point adding a device */ |
199050ea | 7796 | return -EINVAL; |
1da177e4 | 7797 | |
6c2fce2e NB |
7798 | if (rdev->raid_disk >= 0) |
7799 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7800 | |
7801 | /* | |
16a53ecc N |
7802 | * find the disk ... but prefer rdev->saved_raid_disk |
7803 | * if possible. | |
1da177e4 | 7804 | */ |
16a53ecc | 7805 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7806 | rdev->saved_raid_disk >= first && |
16a53ecc | 7807 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7808 | first = rdev->saved_raid_disk; |
7809 | ||
7810 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7811 | p = conf->disks + disk; |
7812 | if (p->rdev == NULL) { | |
b2d444d7 | 7813 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7814 | rdev->raid_disk = disk; |
72626685 N |
7815 | if (rdev->saved_raid_disk != disk) |
7816 | conf->fullsync = 1; | |
d6065f7b | 7817 | rcu_assign_pointer(p->rdev, rdev); |
6358c239 AP |
7818 | |
7819 | err = log_modify(conf, rdev, true); | |
7820 | ||
5cfb22a1 | 7821 | goto out; |
1da177e4 | 7822 | } |
5cfb22a1 N |
7823 | } |
7824 | for (disk = first; disk <= last; disk++) { | |
7825 | p = conf->disks + disk; | |
7bfec5f3 N |
7826 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7827 | p->replacement == NULL) { | |
7828 | clear_bit(In_sync, &rdev->flags); | |
7829 | set_bit(Replacement, &rdev->flags); | |
7830 | rdev->raid_disk = disk; | |
7831 | err = 0; | |
7832 | conf->fullsync = 1; | |
7833 | rcu_assign_pointer(p->replacement, rdev); | |
7834 | break; | |
7835 | } | |
7836 | } | |
5cfb22a1 | 7837 | out: |
1da177e4 | 7838 | print_raid5_conf(conf); |
199050ea | 7839 | return err; |
1da177e4 LT |
7840 | } |
7841 | ||
fd01b88c | 7842 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7843 | { |
7844 | /* no resync is happening, and there is enough space | |
7845 | * on all devices, so we can resize. | |
7846 | * We need to make sure resync covers any new space. | |
7847 | * If the array is shrinking we should possibly wait until | |
7848 | * any io in the removed space completes, but it hardly seems | |
7849 | * worth it. | |
7850 | */ | |
a4a6125a | 7851 | sector_t newsize; |
3cb5edf4 N |
7852 | struct r5conf *conf = mddev->private; |
7853 | ||
e254de6b | 7854 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 7855 | return -EINVAL; |
3cb5edf4 | 7856 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7857 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7858 | if (mddev->external_size && | |
7859 | mddev->array_sectors > newsize) | |
b522adcd | 7860 | return -EINVAL; |
a4a6125a | 7861 | if (mddev->bitmap) { |
e64e4018 | 7862 | int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0); |
a4a6125a N |
7863 | if (ret) |
7864 | return ret; | |
7865 | } | |
7866 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7867 | if (sectors > mddev->dev_sectors && |
7868 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7869 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7870 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7871 | } | |
58c0fed4 | 7872 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7873 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7874 | return 0; |
7875 | } | |
7876 | ||
fd01b88c | 7877 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7878 | { |
7879 | /* Can only proceed if there are plenty of stripe_heads. | |
7880 | * We need a minimum of one full stripe,, and for sensible progress | |
7881 | * it is best to have about 4 times that. | |
7882 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7883 | * allow for chunk sizes up to 256K, which is probably OK. | |
7884 | * If the chunk size is greater, user-space should request more | |
7885 | * stripe_heads first. | |
7886 | */ | |
d1688a6d | 7887 | struct r5conf *conf = mddev->private; |
c911c46c | 7888 | if (((mddev->chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4 |
edbe83ab | 7889 | > conf->min_nr_stripes || |
c911c46c | 7890 | ((mddev->new_chunk_sectors << 9) / RAID5_STRIPE_SIZE(conf)) * 4 |
edbe83ab | 7891 | > conf->min_nr_stripes) { |
cc6167b4 N |
7892 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7893 | mdname(mddev), | |
7894 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
c911c46c | 7895 | / RAID5_STRIPE_SIZE(conf))*4); |
01ee22b4 N |
7896 | return 0; |
7897 | } | |
7898 | return 1; | |
7899 | } | |
7900 | ||
fd01b88c | 7901 | static int check_reshape(struct mddev *mddev) |
29269553 | 7902 | { |
d1688a6d | 7903 | struct r5conf *conf = mddev->private; |
29269553 | 7904 | |
e254de6b | 7905 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 7906 | return -EINVAL; |
88ce4930 N |
7907 | if (mddev->delta_disks == 0 && |
7908 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7909 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7910 | return 0; /* nothing to do */ |
674806d6 | 7911 | if (has_failed(conf)) |
ec32a2bd | 7912 | return -EINVAL; |
fdcfbbb6 | 7913 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7914 | /* We might be able to shrink, but the devices must |
7915 | * be made bigger first. | |
7916 | * For raid6, 4 is the minimum size. | |
7917 | * Otherwise 2 is the minimum | |
7918 | */ | |
7919 | int min = 2; | |
7920 | if (mddev->level == 6) | |
7921 | min = 4; | |
7922 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7923 | return -EINVAL; | |
7924 | } | |
29269553 | 7925 | |
01ee22b4 | 7926 | if (!check_stripe_cache(mddev)) |
29269553 | 7927 | return -ENOSPC; |
29269553 | 7928 | |
738a2738 N |
7929 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7930 | mddev->delta_disks > 0) | |
7931 | if (resize_chunks(conf, | |
7932 | conf->previous_raid_disks | |
7933 | + max(0, mddev->delta_disks), | |
7934 | max(mddev->new_chunk_sectors, | |
7935 | mddev->chunk_sectors) | |
7936 | ) < 0) | |
7937 | return -ENOMEM; | |
845b9e22 AP |
7938 | |
7939 | if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size) | |
7940 | return 0; /* never bother to shrink */ | |
e56108d6 N |
7941 | return resize_stripes(conf, (conf->previous_raid_disks |
7942 | + mddev->delta_disks)); | |
63c70c4f N |
7943 | } |
7944 | ||
fd01b88c | 7945 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7946 | { |
d1688a6d | 7947 | struct r5conf *conf = mddev->private; |
3cb03002 | 7948 | struct md_rdev *rdev; |
63c70c4f | 7949 | int spares = 0; |
c04be0aa | 7950 | unsigned long flags; |
63c70c4f | 7951 | |
f416885e | 7952 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7953 | return -EBUSY; |
7954 | ||
01ee22b4 N |
7955 | if (!check_stripe_cache(mddev)) |
7956 | return -ENOSPC; | |
7957 | ||
30b67645 N |
7958 | if (has_failed(conf)) |
7959 | return -EINVAL; | |
7960 | ||
c6563a8c | 7961 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7962 | if (!test_bit(In_sync, &rdev->flags) |
7963 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7964 | spares++; |
c6563a8c | 7965 | } |
63c70c4f | 7966 | |
f416885e | 7967 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7968 | /* Not enough devices even to make a degraded array |
7969 | * of that size | |
7970 | */ | |
7971 | return -EINVAL; | |
7972 | ||
ec32a2bd N |
7973 | /* Refuse to reduce size of the array. Any reductions in |
7974 | * array size must be through explicit setting of array_size | |
7975 | * attribute. | |
7976 | */ | |
7977 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7978 | < mddev->array_sectors) { | |
cc6167b4 N |
7979 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7980 | mdname(mddev)); | |
ec32a2bd N |
7981 | return -EINVAL; |
7982 | } | |
7983 | ||
f6705578 | 7984 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7985 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7986 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7987 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7988 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7989 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7990 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7991 | conf->prev_algo = conf->algorithm; |
7992 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7993 | conf->generation++; |
7994 | /* Code that selects data_offset needs to see the generation update | |
7995 | * if reshape_progress has been set - so a memory barrier needed. | |
7996 | */ | |
7997 | smp_mb(); | |
2c810cdd | 7998 | if (mddev->reshape_backwards) |
fef9c61f N |
7999 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
8000 | else | |
8001 | conf->reshape_progress = 0; | |
8002 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 8003 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
8004 | spin_unlock_irq(&conf->device_lock); |
8005 | ||
4d77e3ba N |
8006 | /* Now make sure any requests that proceeded on the assumption |
8007 | * the reshape wasn't running - like Discard or Read - have | |
8008 | * completed. | |
8009 | */ | |
8010 | mddev_suspend(mddev); | |
8011 | mddev_resume(mddev); | |
8012 | ||
29269553 N |
8013 | /* Add some new drives, as many as will fit. |
8014 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
8015 | * Don't add devices if we are reducing the number of |
8016 | * devices in the array. This is because it is not possible | |
8017 | * to correctly record the "partially reconstructed" state of | |
8018 | * such devices during the reshape and confusion could result. | |
29269553 | 8019 | */ |
87a8dec9 | 8020 | if (mddev->delta_disks >= 0) { |
dafb20fa | 8021 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
8022 | if (rdev->raid_disk < 0 && |
8023 | !test_bit(Faulty, &rdev->flags)) { | |
8024 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 8025 | if (rdev->raid_disk |
9d4c7d87 | 8026 | >= conf->previous_raid_disks) |
87a8dec9 | 8027 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 8028 | else |
87a8dec9 | 8029 | rdev->recovery_offset = 0; |
36fad858 | 8030 | |
2aada5b1 DLM |
8031 | /* Failure here is OK */ |
8032 | sysfs_link_rdev(mddev, rdev); | |
50da0840 | 8033 | } |
87a8dec9 N |
8034 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
8035 | && !test_bit(Faulty, &rdev->flags)) { | |
8036 | /* This is a spare that was manually added */ | |
8037 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 8038 | } |
29269553 | 8039 | |
87a8dec9 N |
8040 | /* When a reshape changes the number of devices, |
8041 | * ->degraded is measured against the larger of the | |
8042 | * pre and post number of devices. | |
8043 | */ | |
ec32a2bd | 8044 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 8045 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
8046 | spin_unlock_irqrestore(&conf->device_lock, flags); |
8047 | } | |
63c70c4f | 8048 | mddev->raid_disks = conf->raid_disks; |
e516402c | 8049 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 8050 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 8051 | |
29269553 N |
8052 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
8053 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 8054 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
8055 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
8056 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
8057 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 8058 | "reshape"); |
29269553 N |
8059 | if (!mddev->sync_thread) { |
8060 | mddev->recovery = 0; | |
8061 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 8062 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 8063 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
8064 | mddev->new_chunk_sectors = |
8065 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
8066 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
8067 | rdev_for_each(rdev, mddev) |
8068 | rdev->new_data_offset = rdev->data_offset; | |
8069 | smp_wmb(); | |
ba8805b9 | 8070 | conf->generation --; |
fef9c61f | 8071 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 8072 | mddev->reshape_position = MaxSector; |
ba8805b9 | 8073 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
8074 | spin_unlock_irq(&conf->device_lock); |
8075 | return -EAGAIN; | |
8076 | } | |
c8f517c4 | 8077 | conf->reshape_checkpoint = jiffies; |
29269553 N |
8078 | md_wakeup_thread(mddev->sync_thread); |
8079 | md_new_event(mddev); | |
8080 | return 0; | |
8081 | } | |
29269553 | 8082 | |
ec32a2bd N |
8083 | /* This is called from the reshape thread and should make any |
8084 | * changes needed in 'conf' | |
8085 | */ | |
d1688a6d | 8086 | static void end_reshape(struct r5conf *conf) |
29269553 | 8087 | { |
29269553 | 8088 | |
f6705578 | 8089 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
db0505d3 | 8090 | struct md_rdev *rdev; |
f6705578 | 8091 | |
f6705578 | 8092 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 8093 | conf->previous_raid_disks = conf->raid_disks; |
b5d27718 | 8094 | md_finish_reshape(conf->mddev); |
05616be5 | 8095 | smp_wmb(); |
fef9c61f | 8096 | conf->reshape_progress = MaxSector; |
6cbd8148 | 8097 | conf->mddev->reshape_position = MaxSector; |
db0505d3 N |
8098 | rdev_for_each(rdev, conf->mddev) |
8099 | if (rdev->raid_disk >= 0 && | |
8100 | !test_bit(Journal, &rdev->flags) && | |
8101 | !test_bit(In_sync, &rdev->flags)) | |
8102 | rdev->recovery_offset = MaxSector; | |
f6705578 | 8103 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 8104 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
8105 | |
8106 | /* read-ahead size must cover two whole stripes, which is | |
8107 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
8108 | */ | |
4a5add49 | 8109 | if (conf->mddev->queue) { |
cea9c228 | 8110 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 8111 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 8112 | / PAGE_SIZE); |
dc3b17cc JK |
8113 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
8114 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 8115 | } |
29269553 | 8116 | } |
29269553 N |
8117 | } |
8118 | ||
ec32a2bd N |
8119 | /* This is called from the raid5d thread with mddev_lock held. |
8120 | * It makes config changes to the device. | |
8121 | */ | |
fd01b88c | 8122 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 8123 | { |
d1688a6d | 8124 | struct r5conf *conf = mddev->private; |
cea9c228 N |
8125 | |
8126 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8127 | ||
8876391e | 8128 | if (mddev->delta_disks <= 0) { |
ec32a2bd | 8129 | int d; |
908f4fbd | 8130 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 8131 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 8132 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
8133 | for (d = conf->raid_disks ; |
8134 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 8135 | d++) { |
3cb03002 | 8136 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
8137 | if (rdev) |
8138 | clear_bit(In_sync, &rdev->flags); | |
8139 | rdev = conf->disks[d].replacement; | |
8140 | if (rdev) | |
8141 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 8142 | } |
cea9c228 | 8143 | } |
88ce4930 | 8144 | mddev->layout = conf->algorithm; |
09c9e5fa | 8145 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
8146 | mddev->reshape_position = MaxSector; |
8147 | mddev->delta_disks = 0; | |
2c810cdd | 8148 | mddev->reshape_backwards = 0; |
cea9c228 N |
8149 | } |
8150 | } | |
8151 | ||
b03e0ccb | 8152 | static void raid5_quiesce(struct mddev *mddev, int quiesce) |
72626685 | 8153 | { |
d1688a6d | 8154 | struct r5conf *conf = mddev->private; |
72626685 | 8155 | |
b03e0ccb N |
8156 | if (quiesce) { |
8157 | /* stop all writes */ | |
566c09c5 | 8158 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
8159 | /* '2' tells resync/reshape to pause so that all |
8160 | * active stripes can drain | |
8161 | */ | |
a39f7afd | 8162 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 8163 | conf->quiesce = 2; |
b1b46486 | 8164 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
8165 | atomic_read(&conf->active_stripes) == 0 && |
8166 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
8167 | unlock_all_device_hash_locks_irq(conf), |
8168 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 8169 | conf->quiesce = 1; |
566c09c5 | 8170 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
8171 | /* allow reshape to continue */ |
8172 | wake_up(&conf->wait_for_overlap); | |
b03e0ccb N |
8173 | } else { |
8174 | /* re-enable writes */ | |
566c09c5 | 8175 | lock_all_device_hash_locks_irq(conf); |
72626685 | 8176 | conf->quiesce = 0; |
b1b46486 | 8177 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 8178 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 8179 | unlock_all_device_hash_locks_irq(conf); |
72626685 | 8180 | } |
1532d9e8 | 8181 | log_quiesce(conf, quiesce); |
72626685 | 8182 | } |
b15c2e57 | 8183 | |
fd01b88c | 8184 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 8185 | { |
e373ab10 | 8186 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 8187 | sector_t sectors; |
54071b38 | 8188 | |
f1b29bca | 8189 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 8190 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
8191 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
8192 | mdname(mddev)); | |
f1b29bca DW |
8193 | return ERR_PTR(-EINVAL); |
8194 | } | |
8195 | ||
e373ab10 N |
8196 | sectors = raid0_conf->strip_zone[0].zone_end; |
8197 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 8198 | mddev->dev_sectors = sectors; |
f1b29bca | 8199 | mddev->new_level = level; |
54071b38 TM |
8200 | mddev->new_layout = ALGORITHM_PARITY_N; |
8201 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
8202 | mddev->raid_disks += 1; | |
8203 | mddev->delta_disks = 1; | |
8204 | /* make sure it will be not marked as dirty */ | |
8205 | mddev->recovery_cp = MaxSector; | |
8206 | ||
8207 | return setup_conf(mddev); | |
8208 | } | |
8209 | ||
fd01b88c | 8210 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
8211 | { |
8212 | int chunksect; | |
6995f0b2 | 8213 | void *ret; |
d562b0c4 N |
8214 | |
8215 | if (mddev->raid_disks != 2 || | |
8216 | mddev->degraded > 1) | |
8217 | return ERR_PTR(-EINVAL); | |
8218 | ||
8219 | /* Should check if there are write-behind devices? */ | |
8220 | ||
8221 | chunksect = 64*2; /* 64K by default */ | |
8222 | ||
8223 | /* The array must be an exact multiple of chunksize */ | |
8224 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8225 | chunksect >>= 1; | |
8226 | ||
c911c46c | 8227 | if ((chunksect<<9) < RAID5_STRIPE_SIZE((struct r5conf *)mddev->private)) |
d562b0c4 N |
8228 | /* array size does not allow a suitable chunk size */ |
8229 | return ERR_PTR(-EINVAL); | |
8230 | ||
8231 | mddev->new_level = 5; | |
8232 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8233 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8234 | |
6995f0b2 | 8235 | ret = setup_conf(mddev); |
32cd7cbb | 8236 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8237 | mddev_clear_unsupported_flags(mddev, |
8238 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8239 | return ret; |
d562b0c4 N |
8240 | } |
8241 | ||
fd01b88c | 8242 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8243 | { |
8244 | int new_layout; | |
8245 | ||
8246 | switch (mddev->layout) { | |
8247 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8248 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8249 | break; | |
8250 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8251 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8252 | break; | |
8253 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8254 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8255 | break; | |
8256 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8257 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8258 | break; | |
8259 | case ALGORITHM_PARITY_0_6: | |
8260 | new_layout = ALGORITHM_PARITY_0; | |
8261 | break; | |
8262 | case ALGORITHM_PARITY_N: | |
8263 | new_layout = ALGORITHM_PARITY_N; | |
8264 | break; | |
8265 | default: | |
8266 | return ERR_PTR(-EINVAL); | |
8267 | } | |
8268 | mddev->new_level = 5; | |
8269 | mddev->new_layout = new_layout; | |
8270 | mddev->delta_disks = -1; | |
8271 | mddev->raid_disks -= 1; | |
8272 | return setup_conf(mddev); | |
8273 | } | |
8274 | ||
fd01b88c | 8275 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8276 | { |
88ce4930 N |
8277 | /* For a 2-drive array, the layout and chunk size can be changed |
8278 | * immediately as not restriping is needed. | |
8279 | * For larger arrays we record the new value - after validation | |
8280 | * to be used by a reshape pass. | |
b3546035 | 8281 | */ |
d1688a6d | 8282 | struct r5conf *conf = mddev->private; |
597a711b | 8283 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8284 | |
597a711b | 8285 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8286 | return -EINVAL; |
8287 | if (new_chunk > 0) { | |
0ba459d2 | 8288 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8289 | return -EINVAL; |
597a711b | 8290 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8291 | return -EINVAL; |
597a711b | 8292 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8293 | /* not factor of array size */ |
8294 | return -EINVAL; | |
8295 | } | |
8296 | ||
8297 | /* They look valid */ | |
8298 | ||
88ce4930 | 8299 | if (mddev->raid_disks == 2) { |
597a711b N |
8300 | /* can make the change immediately */ |
8301 | if (mddev->new_layout >= 0) { | |
8302 | conf->algorithm = mddev->new_layout; | |
8303 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8304 | } |
8305 | if (new_chunk > 0) { | |
597a711b N |
8306 | conf->chunk_sectors = new_chunk ; |
8307 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8308 | } |
2953079c | 8309 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8310 | md_wakeup_thread(mddev->thread); |
b3546035 | 8311 | } |
50ac168a | 8312 | return check_reshape(mddev); |
88ce4930 N |
8313 | } |
8314 | ||
fd01b88c | 8315 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8316 | { |
597a711b | 8317 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8318 | |
597a711b | 8319 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8320 | return -EINVAL; |
b3546035 | 8321 | if (new_chunk > 0) { |
0ba459d2 | 8322 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8323 | return -EINVAL; |
597a711b | 8324 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8325 | return -EINVAL; |
597a711b | 8326 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8327 | /* not factor of array size */ |
8328 | return -EINVAL; | |
b3546035 | 8329 | } |
88ce4930 N |
8330 | |
8331 | /* They look valid */ | |
50ac168a | 8332 | return check_reshape(mddev); |
b3546035 N |
8333 | } |
8334 | ||
fd01b88c | 8335 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8336 | { |
8337 | /* raid5 can take over: | |
f1b29bca | 8338 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8339 | * raid1 - if there are two drives. We need to know the chunk size |
8340 | * raid4 - trivial - just use a raid4 layout. | |
8341 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8342 | */ |
f1b29bca DW |
8343 | if (mddev->level == 0) |
8344 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8345 | if (mddev->level == 1) |
8346 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8347 | if (mddev->level == 4) { |
8348 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8349 | mddev->new_level = 5; | |
8350 | return setup_conf(mddev); | |
8351 | } | |
fc9739c6 N |
8352 | if (mddev->level == 6) |
8353 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8354 | |
8355 | return ERR_PTR(-EINVAL); | |
8356 | } | |
8357 | ||
fd01b88c | 8358 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8359 | { |
f1b29bca DW |
8360 | /* raid4 can take over: |
8361 | * raid0 - if there is only one strip zone | |
8362 | * raid5 - if layout is right | |
a78d38a1 | 8363 | */ |
f1b29bca DW |
8364 | if (mddev->level == 0) |
8365 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8366 | if (mddev->level == 5 && |
8367 | mddev->layout == ALGORITHM_PARITY_N) { | |
8368 | mddev->new_layout = 0; | |
8369 | mddev->new_level = 4; | |
8370 | return setup_conf(mddev); | |
8371 | } | |
8372 | return ERR_PTR(-EINVAL); | |
8373 | } | |
d562b0c4 | 8374 | |
84fc4b56 | 8375 | static struct md_personality raid5_personality; |
245f46c2 | 8376 | |
fd01b88c | 8377 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8378 | { |
8379 | /* Currently can only take over a raid5. We map the | |
8380 | * personality to an equivalent raid6 personality | |
8381 | * with the Q block at the end. | |
8382 | */ | |
8383 | int new_layout; | |
8384 | ||
8385 | if (mddev->pers != &raid5_personality) | |
8386 | return ERR_PTR(-EINVAL); | |
8387 | if (mddev->degraded > 1) | |
8388 | return ERR_PTR(-EINVAL); | |
8389 | if (mddev->raid_disks > 253) | |
8390 | return ERR_PTR(-EINVAL); | |
8391 | if (mddev->raid_disks < 3) | |
8392 | return ERR_PTR(-EINVAL); | |
8393 | ||
8394 | switch (mddev->layout) { | |
8395 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8396 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8397 | break; | |
8398 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8399 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8400 | break; | |
8401 | case ALGORITHM_LEFT_SYMMETRIC: | |
8402 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8403 | break; | |
8404 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8405 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8406 | break; | |
8407 | case ALGORITHM_PARITY_0: | |
8408 | new_layout = ALGORITHM_PARITY_0_6; | |
8409 | break; | |
8410 | case ALGORITHM_PARITY_N: | |
8411 | new_layout = ALGORITHM_PARITY_N; | |
8412 | break; | |
8413 | default: | |
8414 | return ERR_PTR(-EINVAL); | |
8415 | } | |
8416 | mddev->new_level = 6; | |
8417 | mddev->new_layout = new_layout; | |
8418 | mddev->delta_disks = 1; | |
8419 | mddev->raid_disks += 1; | |
8420 | return setup_conf(mddev); | |
8421 | } | |
8422 | ||
ba903a3e AP |
8423 | static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf) |
8424 | { | |
8425 | struct r5conf *conf; | |
8426 | int err; | |
8427 | ||
8428 | err = mddev_lock(mddev); | |
8429 | if (err) | |
8430 | return err; | |
8431 | conf = mddev->private; | |
8432 | if (!conf) { | |
8433 | mddev_unlock(mddev); | |
8434 | return -ENODEV; | |
8435 | } | |
8436 | ||
845b9e22 | 8437 | if (strncmp(buf, "ppl", 3) == 0) { |
0bb0c105 | 8438 | /* ppl only works with RAID 5 */ |
845b9e22 AP |
8439 | if (!raid5_has_ppl(conf) && conf->level == 5) { |
8440 | err = log_init(conf, NULL, true); | |
8441 | if (!err) { | |
8442 | err = resize_stripes(conf, conf->pool_size); | |
8443 | if (err) | |
8444 | log_exit(conf); | |
8445 | } | |
0bb0c105 SL |
8446 | } else |
8447 | err = -EINVAL; | |
8448 | } else if (strncmp(buf, "resync", 6) == 0) { | |
8449 | if (raid5_has_ppl(conf)) { | |
8450 | mddev_suspend(mddev); | |
8451 | log_exit(conf); | |
0bb0c105 | 8452 | mddev_resume(mddev); |
845b9e22 | 8453 | err = resize_stripes(conf, conf->pool_size); |
0bb0c105 SL |
8454 | } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) && |
8455 | r5l_log_disk_error(conf)) { | |
8456 | bool journal_dev_exists = false; | |
8457 | struct md_rdev *rdev; | |
8458 | ||
8459 | rdev_for_each(rdev, mddev) | |
8460 | if (test_bit(Journal, &rdev->flags)) { | |
8461 | journal_dev_exists = true; | |
8462 | break; | |
8463 | } | |
8464 | ||
8465 | if (!journal_dev_exists) { | |
8466 | mddev_suspend(mddev); | |
8467 | clear_bit(MD_HAS_JOURNAL, &mddev->flags); | |
8468 | mddev_resume(mddev); | |
8469 | } else /* need remove journal device first */ | |
8470 | err = -EBUSY; | |
8471 | } else | |
8472 | err = -EINVAL; | |
ba903a3e AP |
8473 | } else { |
8474 | err = -EINVAL; | |
8475 | } | |
8476 | ||
8477 | if (!err) | |
8478 | md_update_sb(mddev, 1); | |
8479 | ||
8480 | mddev_unlock(mddev); | |
8481 | ||
8482 | return err; | |
8483 | } | |
8484 | ||
d5d885fd SL |
8485 | static int raid5_start(struct mddev *mddev) |
8486 | { | |
8487 | struct r5conf *conf = mddev->private; | |
8488 | ||
8489 | return r5l_start(conf->log); | |
8490 | } | |
8491 | ||
84fc4b56 | 8492 | static struct md_personality raid6_personality = |
16a53ecc N |
8493 | { |
8494 | .name = "raid6", | |
8495 | .level = 6, | |
8496 | .owner = THIS_MODULE, | |
849674e4 SL |
8497 | .make_request = raid5_make_request, |
8498 | .run = raid5_run, | |
d5d885fd | 8499 | .start = raid5_start, |
afa0f557 | 8500 | .free = raid5_free, |
849674e4 SL |
8501 | .status = raid5_status, |
8502 | .error_handler = raid5_error, | |
16a53ecc N |
8503 | .hot_add_disk = raid5_add_disk, |
8504 | .hot_remove_disk= raid5_remove_disk, | |
8505 | .spare_active = raid5_spare_active, | |
849674e4 | 8506 | .sync_request = raid5_sync_request, |
16a53ecc | 8507 | .resize = raid5_resize, |
80c3a6ce | 8508 | .size = raid5_size, |
50ac168a | 8509 | .check_reshape = raid6_check_reshape, |
f416885e | 8510 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8511 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8512 | .quiesce = raid5_quiesce, |
245f46c2 | 8513 | .takeover = raid6_takeover, |
0bb0c105 | 8514 | .change_consistency_policy = raid5_change_consistency_policy, |
16a53ecc | 8515 | }; |
84fc4b56 | 8516 | static struct md_personality raid5_personality = |
1da177e4 LT |
8517 | { |
8518 | .name = "raid5", | |
2604b703 | 8519 | .level = 5, |
1da177e4 | 8520 | .owner = THIS_MODULE, |
849674e4 SL |
8521 | .make_request = raid5_make_request, |
8522 | .run = raid5_run, | |
d5d885fd | 8523 | .start = raid5_start, |
afa0f557 | 8524 | .free = raid5_free, |
849674e4 SL |
8525 | .status = raid5_status, |
8526 | .error_handler = raid5_error, | |
1da177e4 LT |
8527 | .hot_add_disk = raid5_add_disk, |
8528 | .hot_remove_disk= raid5_remove_disk, | |
8529 | .spare_active = raid5_spare_active, | |
849674e4 | 8530 | .sync_request = raid5_sync_request, |
1da177e4 | 8531 | .resize = raid5_resize, |
80c3a6ce | 8532 | .size = raid5_size, |
63c70c4f N |
8533 | .check_reshape = raid5_check_reshape, |
8534 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8535 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8536 | .quiesce = raid5_quiesce, |
d562b0c4 | 8537 | .takeover = raid5_takeover, |
ba903a3e | 8538 | .change_consistency_policy = raid5_change_consistency_policy, |
1da177e4 LT |
8539 | }; |
8540 | ||
84fc4b56 | 8541 | static struct md_personality raid4_personality = |
1da177e4 | 8542 | { |
2604b703 N |
8543 | .name = "raid4", |
8544 | .level = 4, | |
8545 | .owner = THIS_MODULE, | |
849674e4 SL |
8546 | .make_request = raid5_make_request, |
8547 | .run = raid5_run, | |
d5d885fd | 8548 | .start = raid5_start, |
afa0f557 | 8549 | .free = raid5_free, |
849674e4 SL |
8550 | .status = raid5_status, |
8551 | .error_handler = raid5_error, | |
2604b703 N |
8552 | .hot_add_disk = raid5_add_disk, |
8553 | .hot_remove_disk= raid5_remove_disk, | |
8554 | .spare_active = raid5_spare_active, | |
849674e4 | 8555 | .sync_request = raid5_sync_request, |
2604b703 | 8556 | .resize = raid5_resize, |
80c3a6ce | 8557 | .size = raid5_size, |
3d37890b N |
8558 | .check_reshape = raid5_check_reshape, |
8559 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8560 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8561 | .quiesce = raid5_quiesce, |
a78d38a1 | 8562 | .takeover = raid4_takeover, |
0bb0c105 | 8563 | .change_consistency_policy = raid5_change_consistency_policy, |
2604b703 N |
8564 | }; |
8565 | ||
8566 | static int __init raid5_init(void) | |
8567 | { | |
29c6d1bb SAS |
8568 | int ret; |
8569 | ||
851c30c9 SL |
8570 | raid5_wq = alloc_workqueue("raid5wq", |
8571 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8572 | if (!raid5_wq) | |
8573 | return -ENOMEM; | |
29c6d1bb SAS |
8574 | |
8575 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8576 | "md/raid5:prepare", | |
8577 | raid456_cpu_up_prepare, | |
8578 | raid456_cpu_dead); | |
8579 | if (ret) { | |
8580 | destroy_workqueue(raid5_wq); | |
8581 | return ret; | |
8582 | } | |
16a53ecc | 8583 | register_md_personality(&raid6_personality); |
2604b703 N |
8584 | register_md_personality(&raid5_personality); |
8585 | register_md_personality(&raid4_personality); | |
8586 | return 0; | |
1da177e4 LT |
8587 | } |
8588 | ||
2604b703 | 8589 | static void raid5_exit(void) |
1da177e4 | 8590 | { |
16a53ecc | 8591 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8592 | unregister_md_personality(&raid5_personality); |
8593 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8594 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8595 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8596 | } |
8597 | ||
8598 | module_init(raid5_init); | |
8599 | module_exit(raid5_exit); | |
8600 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8601 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8602 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8603 | MODULE_ALIAS("md-raid5"); |
8604 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8605 | MODULE_ALIAS("md-level-5"); |
8606 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8607 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8608 | MODULE_ALIAS("md-raid6"); | |
8609 | MODULE_ALIAS("md-level-6"); | |
8610 | ||
8611 | /* This used to be two separate modules, they were: */ | |
8612 | MODULE_ALIAS("raid5"); | |
8613 | MODULE_ALIAS("raid6"); |