]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
851c30c9 | 56 | #include <linux/nodemask.h> |
3f07c014 | 57 | |
a9add5d9 | 58 | #include <trace/events/block.h> |
aaf9f12e | 59 | #include <linux/list_sort.h> |
a9add5d9 | 60 | |
43b2e5d8 | 61 | #include "md.h" |
bff61975 | 62 | #include "raid5.h" |
54071b38 | 63 | #include "raid0.h" |
935fe098 | 64 | #include "md-bitmap.h" |
ff875738 | 65 | #include "raid5-log.h" |
72626685 | 66 | |
394ed8e4 SL |
67 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
68 | ||
851c30c9 SL |
69 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
70 | #define ANY_GROUP NUMA_NO_NODE | |
71 | ||
8e0e99ba N |
72 | static bool devices_handle_discard_safely = false; |
73 | module_param(devices_handle_discard_safely, bool, 0644); | |
74 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
75 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 76 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 77 | |
d1688a6d | 78 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
79 | { |
80 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
81 | return &conf->stripe_hashtbl[hash]; | |
82 | } | |
1da177e4 | 83 | |
566c09c5 SL |
84 | static inline int stripe_hash_locks_hash(sector_t sect) |
85 | { | |
86 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
87 | } | |
88 | ||
89 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
90 | { | |
91 | spin_lock_irq(conf->hash_locks + hash); | |
92 | spin_lock(&conf->device_lock); | |
93 | } | |
94 | ||
95 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
96 | { | |
97 | spin_unlock(&conf->device_lock); | |
98 | spin_unlock_irq(conf->hash_locks + hash); | |
99 | } | |
100 | ||
101 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
102 | { | |
103 | int i; | |
3d05f3ae | 104 | spin_lock_irq(conf->hash_locks); |
566c09c5 SL |
105 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) |
106 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
107 | spin_lock(&conf->device_lock); | |
108 | } | |
109 | ||
110 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
111 | { | |
112 | int i; | |
113 | spin_unlock(&conf->device_lock); | |
3d05f3ae JC |
114 | for (i = NR_STRIPE_HASH_LOCKS - 1; i; i--) |
115 | spin_unlock(conf->hash_locks + i); | |
116 | spin_unlock_irq(conf->hash_locks); | |
566c09c5 SL |
117 | } |
118 | ||
d0dabf7e N |
119 | /* Find first data disk in a raid6 stripe */ |
120 | static inline int raid6_d0(struct stripe_head *sh) | |
121 | { | |
67cc2b81 N |
122 | if (sh->ddf_layout) |
123 | /* ddf always start from first device */ | |
124 | return 0; | |
125 | /* md starts just after Q block */ | |
d0dabf7e N |
126 | if (sh->qd_idx == sh->disks - 1) |
127 | return 0; | |
128 | else | |
129 | return sh->qd_idx + 1; | |
130 | } | |
16a53ecc N |
131 | static inline int raid6_next_disk(int disk, int raid_disks) |
132 | { | |
133 | disk++; | |
134 | return (disk < raid_disks) ? disk : 0; | |
135 | } | |
a4456856 | 136 | |
d0dabf7e N |
137 | /* When walking through the disks in a raid5, starting at raid6_d0, |
138 | * We need to map each disk to a 'slot', where the data disks are slot | |
139 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
140 | * is raid_disks-1. This help does that mapping. | |
141 | */ | |
67cc2b81 N |
142 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
143 | int *count, int syndrome_disks) | |
d0dabf7e | 144 | { |
6629542e | 145 | int slot = *count; |
67cc2b81 | 146 | |
e4424fee | 147 | if (sh->ddf_layout) |
6629542e | 148 | (*count)++; |
d0dabf7e | 149 | if (idx == sh->pd_idx) |
67cc2b81 | 150 | return syndrome_disks; |
d0dabf7e | 151 | if (idx == sh->qd_idx) |
67cc2b81 | 152 | return syndrome_disks + 1; |
e4424fee | 153 | if (!sh->ddf_layout) |
6629542e | 154 | (*count)++; |
d0dabf7e N |
155 | return slot; |
156 | } | |
157 | ||
d1688a6d | 158 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 159 | |
600aa109 DW |
160 | static int stripe_operations_active(struct stripe_head *sh) |
161 | { | |
162 | return sh->check_state || sh->reconstruct_state || | |
163 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
164 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
165 | } | |
166 | ||
535ae4eb SL |
167 | static bool stripe_is_lowprio(struct stripe_head *sh) |
168 | { | |
169 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
170 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
171 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
172 | } | |
173 | ||
851c30c9 SL |
174 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
175 | { | |
176 | struct r5conf *conf = sh->raid_conf; | |
177 | struct r5worker_group *group; | |
bfc90cb0 | 178 | int thread_cnt; |
851c30c9 SL |
179 | int i, cpu = sh->cpu; |
180 | ||
181 | if (!cpu_online(cpu)) { | |
182 | cpu = cpumask_any(cpu_online_mask); | |
183 | sh->cpu = cpu; | |
184 | } | |
185 | ||
186 | if (list_empty(&sh->lru)) { | |
187 | struct r5worker_group *group; | |
188 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
189 | if (stripe_is_lowprio(sh)) |
190 | list_add_tail(&sh->lru, &group->loprio_list); | |
191 | else | |
192 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
193 | group->stripes_cnt++; |
194 | sh->group = group; | |
851c30c9 SL |
195 | } |
196 | ||
197 | if (conf->worker_cnt_per_group == 0) { | |
198 | md_wakeup_thread(conf->mddev->thread); | |
199 | return; | |
200 | } | |
201 | ||
202 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
203 | ||
bfc90cb0 SL |
204 | group->workers[0].working = true; |
205 | /* at least one worker should run to avoid race */ | |
206 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
207 | ||
208 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
209 | /* wakeup more workers */ | |
210 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
211 | if (group->workers[i].working == false) { | |
212 | group->workers[i].working = true; | |
213 | queue_work_on(sh->cpu, raid5_wq, | |
214 | &group->workers[i].work); | |
215 | thread_cnt--; | |
216 | } | |
217 | } | |
851c30c9 SL |
218 | } |
219 | ||
566c09c5 SL |
220 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
221 | struct list_head *temp_inactive_list) | |
1da177e4 | 222 | { |
1e6d690b SL |
223 | int i; |
224 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
225 | ||
4eb788df SL |
226 | BUG_ON(!list_empty(&sh->lru)); |
227 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
228 | |
229 | if (r5c_is_writeback(conf->log)) | |
230 | for (i = sh->disks; i--; ) | |
231 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
232 | injournal++; | |
a39f7afd | 233 | /* |
5ddf0440 SL |
234 | * In the following cases, the stripe cannot be released to cached |
235 | * lists. Therefore, we make the stripe write out and set | |
236 | * STRIPE_HANDLE: | |
237 | * 1. when quiesce in r5c write back; | |
238 | * 2. when resync is requested fot the stripe. | |
a39f7afd | 239 | */ |
5ddf0440 SL |
240 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) || |
241 | (conf->quiesce && r5c_is_writeback(conf->log) && | |
242 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0)) { | |
a39f7afd SL |
243 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) |
244 | r5c_make_stripe_write_out(sh); | |
245 | set_bit(STRIPE_HANDLE, &sh->state); | |
246 | } | |
1e6d690b | 247 | |
4eb788df SL |
248 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
249 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 250 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 251 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 252 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
253 | sh->bm_seq - conf->seq_write > 0) |
254 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
255 | else { | |
256 | clear_bit(STRIPE_DELAYED, &sh->state); | |
257 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 258 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
259 | if (stripe_is_lowprio(sh)) |
260 | list_add_tail(&sh->lru, | |
261 | &conf->loprio_list); | |
262 | else | |
263 | list_add_tail(&sh->lru, | |
264 | &conf->handle_list); | |
851c30c9 SL |
265 | } else { |
266 | raid5_wakeup_stripe_thread(sh); | |
267 | return; | |
268 | } | |
4eb788df SL |
269 | } |
270 | md_wakeup_thread(conf->mddev->thread); | |
271 | } else { | |
272 | BUG_ON(stripe_operations_active(sh)); | |
273 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
274 | if (atomic_dec_return(&conf->preread_active_stripes) | |
275 | < IO_THRESHOLD) | |
276 | md_wakeup_thread(conf->mddev->thread); | |
277 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
278 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
279 | if (!r5c_is_writeback(conf->log)) | |
280 | list_add_tail(&sh->lru, temp_inactive_list); | |
281 | else { | |
282 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
283 | if (injournal == 0) | |
284 | list_add_tail(&sh->lru, temp_inactive_list); | |
285 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
286 | /* full stripe */ | |
287 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
288 | atomic_inc(&conf->r5c_cached_full_stripes); | |
289 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
290 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
291 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 292 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
293 | } else |
294 | /* | |
295 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
296 | * r5c_try_caching_write(). No need to | |
297 | * set it again. | |
298 | */ | |
1e6d690b | 299 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
300 | } |
301 | } | |
1da177e4 LT |
302 | } |
303 | } | |
d0dabf7e | 304 | |
566c09c5 SL |
305 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
306 | struct list_head *temp_inactive_list) | |
4eb788df SL |
307 | { |
308 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
309 | do_release_stripe(conf, sh, temp_inactive_list); |
310 | } | |
311 | ||
312 | /* | |
313 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
314 | * | |
315 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
316 | * given time. Adding stripes only takes device lock, while deleting stripes | |
317 | * only takes hash lock. | |
318 | */ | |
319 | static void release_inactive_stripe_list(struct r5conf *conf, | |
320 | struct list_head *temp_inactive_list, | |
321 | int hash) | |
322 | { | |
323 | int size; | |
6ab2a4b8 | 324 | bool do_wakeup = false; |
566c09c5 SL |
325 | unsigned long flags; |
326 | ||
327 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
328 | size = NR_STRIPE_HASH_LOCKS; | |
329 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
330 | } else | |
331 | size = 1; | |
332 | while (size) { | |
333 | struct list_head *list = &temp_inactive_list[size - 1]; | |
334 | ||
335 | /* | |
6d036f7d | 336 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
337 | * remove stripes from the list |
338 | */ | |
339 | if (!list_empty_careful(list)) { | |
340 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
341 | if (list_empty(conf->inactive_list + hash) && |
342 | !list_empty(list)) | |
343 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 344 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 345 | do_wakeup = true; |
566c09c5 SL |
346 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
347 | } | |
348 | size--; | |
349 | hash--; | |
350 | } | |
351 | ||
352 | if (do_wakeup) { | |
6ab2a4b8 | 353 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
354 | if (atomic_read(&conf->active_stripes) == 0) |
355 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
356 | if (conf->retry_read_aligned) |
357 | md_wakeup_thread(conf->mddev->thread); | |
358 | } | |
4eb788df SL |
359 | } |
360 | ||
773ca82f | 361 | /* should hold conf->device_lock already */ |
566c09c5 SL |
362 | static int release_stripe_list(struct r5conf *conf, |
363 | struct list_head *temp_inactive_list) | |
773ca82f | 364 | { |
eae8263f | 365 | struct stripe_head *sh, *t; |
773ca82f SL |
366 | int count = 0; |
367 | struct llist_node *head; | |
368 | ||
369 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 370 | head = llist_reverse_order(head); |
eae8263f | 371 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
372 | int hash; |
373 | ||
773ca82f SL |
374 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
375 | smp_mb(); | |
376 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
377 | /* | |
378 | * Don't worry the bit is set here, because if the bit is set | |
379 | * again, the count is always > 1. This is true for | |
380 | * STRIPE_ON_UNPLUG_LIST bit too. | |
381 | */ | |
566c09c5 SL |
382 | hash = sh->hash_lock_index; |
383 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
384 | count++; |
385 | } | |
386 | ||
387 | return count; | |
388 | } | |
389 | ||
6d036f7d | 390 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 391 | { |
d1688a6d | 392 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 393 | unsigned long flags; |
566c09c5 SL |
394 | struct list_head list; |
395 | int hash; | |
773ca82f | 396 | bool wakeup; |
16a53ecc | 397 | |
cf170f3f ES |
398 | /* Avoid release_list until the last reference. |
399 | */ | |
400 | if (atomic_add_unless(&sh->count, -1, 1)) | |
401 | return; | |
402 | ||
ad4068de | 403 | if (unlikely(!conf->mddev->thread) || |
404 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
405 | goto slow_path; |
406 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
407 | if (wakeup) | |
408 | md_wakeup_thread(conf->mddev->thread); | |
409 | return; | |
410 | slow_path: | |
773ca82f | 411 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
685dbcaa | 412 | if (atomic_dec_and_lock_irqsave(&sh->count, &conf->device_lock, flags)) { |
566c09c5 SL |
413 | INIT_LIST_HEAD(&list); |
414 | hash = sh->hash_lock_index; | |
415 | do_release_stripe(conf, sh, &list); | |
08edaaa6 | 416 | spin_unlock_irqrestore(&conf->device_lock, flags); |
566c09c5 | 417 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df | 418 | } |
1da177e4 LT |
419 | } |
420 | ||
fccddba0 | 421 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 422 | { |
45b4233c DW |
423 | pr_debug("remove_hash(), stripe %llu\n", |
424 | (unsigned long long)sh->sector); | |
1da177e4 | 425 | |
fccddba0 | 426 | hlist_del_init(&sh->hash); |
1da177e4 LT |
427 | } |
428 | ||
d1688a6d | 429 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 430 | { |
fccddba0 | 431 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 432 | |
45b4233c DW |
433 | pr_debug("insert_hash(), stripe %llu\n", |
434 | (unsigned long long)sh->sector); | |
1da177e4 | 435 | |
fccddba0 | 436 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
437 | } |
438 | ||
1da177e4 | 439 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 440 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
441 | { |
442 | struct stripe_head *sh = NULL; | |
443 | struct list_head *first; | |
444 | ||
566c09c5 | 445 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 446 | goto out; |
566c09c5 | 447 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
448 | sh = list_entry(first, struct stripe_head, lru); |
449 | list_del_init(first); | |
450 | remove_hash(sh); | |
451 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 452 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
453 | if (list_empty(conf->inactive_list + hash)) |
454 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
455 | out: |
456 | return sh; | |
457 | } | |
458 | ||
e4e11e38 | 459 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
460 | { |
461 | struct page *p; | |
462 | int i; | |
e4e11e38 | 463 | int num = sh->raid_conf->pool_size; |
1da177e4 | 464 | |
e4e11e38 | 465 | for (i = 0; i < num ; i++) { |
d592a996 | 466 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
467 | p = sh->dev[i].page; |
468 | if (!p) | |
469 | continue; | |
470 | sh->dev[i].page = NULL; | |
2d1f3b5d | 471 | put_page(p); |
1da177e4 LT |
472 | } |
473 | } | |
474 | ||
a9683a79 | 475 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
476 | { |
477 | int i; | |
e4e11e38 | 478 | int num = sh->raid_conf->pool_size; |
1da177e4 | 479 | |
e4e11e38 | 480 | for (i = 0; i < num; i++) { |
1da177e4 LT |
481 | struct page *page; |
482 | ||
a9683a79 | 483 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
484 | return 1; |
485 | } | |
486 | sh->dev[i].page = page; | |
d592a996 | 487 | sh->dev[i].orig_page = page; |
1da177e4 | 488 | } |
3418d036 | 489 | |
1da177e4 LT |
490 | return 0; |
491 | } | |
492 | ||
d1688a6d | 493 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 494 | struct stripe_head *sh); |
1da177e4 | 495 | |
b5663ba4 | 496 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 497 | { |
d1688a6d | 498 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 499 | int i, seq; |
1da177e4 | 500 | |
78bafebd ES |
501 | BUG_ON(atomic_read(&sh->count) != 0); |
502 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 503 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 504 | BUG_ON(sh->batch_head); |
d84e0f10 | 505 | |
45b4233c | 506 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 507 | (unsigned long long)sector); |
566c09c5 SL |
508 | retry: |
509 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 510 | sh->generation = conf->generation - previous; |
b5663ba4 | 511 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 512 | sh->sector = sector; |
911d4ee8 | 513 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
514 | sh->state = 0; |
515 | ||
7ecaa1e6 | 516 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
517 | struct r5dev *dev = &sh->dev[i]; |
518 | ||
d84e0f10 | 519 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 520 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 521 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 522 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 523 | dev->read, dev->towrite, dev->written, |
1da177e4 | 524 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 525 | WARN_ON(1); |
1da177e4 LT |
526 | } |
527 | dev->flags = 0; | |
27a4ff8f | 528 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 | 529 | } |
566c09c5 SL |
530 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
531 | goto retry; | |
7a87f434 | 532 | sh->overwrite_disks = 0; |
1da177e4 | 533 | insert_hash(conf, sh); |
851c30c9 | 534 | sh->cpu = smp_processor_id(); |
da41ba65 | 535 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
536 | } |
537 | ||
d1688a6d | 538 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 539 | short generation) |
1da177e4 LT |
540 | { |
541 | struct stripe_head *sh; | |
542 | ||
45b4233c | 543 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 544 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 545 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 546 | return sh; |
45b4233c | 547 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
548 | return NULL; |
549 | } | |
550 | ||
674806d6 N |
551 | /* |
552 | * Need to check if array has failed when deciding whether to: | |
553 | * - start an array | |
554 | * - remove non-faulty devices | |
555 | * - add a spare | |
556 | * - allow a reshape | |
557 | * This determination is simple when no reshape is happening. | |
558 | * However if there is a reshape, we need to carefully check | |
559 | * both the before and after sections. | |
560 | * This is because some failed devices may only affect one | |
561 | * of the two sections, and some non-in_sync devices may | |
562 | * be insync in the section most affected by failed devices. | |
563 | */ | |
2e38a37f | 564 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 565 | { |
908f4fbd | 566 | int degraded, degraded2; |
674806d6 | 567 | int i; |
674806d6 N |
568 | |
569 | rcu_read_lock(); | |
570 | degraded = 0; | |
571 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 572 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
573 | if (rdev && test_bit(Faulty, &rdev->flags)) |
574 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
575 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
576 | degraded++; | |
577 | else if (test_bit(In_sync, &rdev->flags)) | |
578 | ; | |
579 | else | |
580 | /* not in-sync or faulty. | |
581 | * If the reshape increases the number of devices, | |
582 | * this is being recovered by the reshape, so | |
583 | * this 'previous' section is not in_sync. | |
584 | * If the number of devices is being reduced however, | |
585 | * the device can only be part of the array if | |
586 | * we are reverting a reshape, so this section will | |
587 | * be in-sync. | |
588 | */ | |
589 | if (conf->raid_disks >= conf->previous_raid_disks) | |
590 | degraded++; | |
591 | } | |
592 | rcu_read_unlock(); | |
908f4fbd N |
593 | if (conf->raid_disks == conf->previous_raid_disks) |
594 | return degraded; | |
674806d6 | 595 | rcu_read_lock(); |
908f4fbd | 596 | degraded2 = 0; |
674806d6 | 597 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 598 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
599 | if (rdev && test_bit(Faulty, &rdev->flags)) |
600 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 601 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 602 | degraded2++; |
674806d6 N |
603 | else if (test_bit(In_sync, &rdev->flags)) |
604 | ; | |
605 | else | |
606 | /* not in-sync or faulty. | |
607 | * If reshape increases the number of devices, this | |
608 | * section has already been recovered, else it | |
609 | * almost certainly hasn't. | |
610 | */ | |
611 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 612 | degraded2++; |
674806d6 N |
613 | } |
614 | rcu_read_unlock(); | |
908f4fbd N |
615 | if (degraded2 > degraded) |
616 | return degraded2; | |
617 | return degraded; | |
618 | } | |
619 | ||
620 | static int has_failed(struct r5conf *conf) | |
621 | { | |
622 | int degraded; | |
623 | ||
624 | if (conf->mddev->reshape_position == MaxSector) | |
625 | return conf->mddev->degraded > conf->max_degraded; | |
626 | ||
2e38a37f | 627 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
628 | if (degraded > conf->max_degraded) |
629 | return 1; | |
630 | return 0; | |
631 | } | |
632 | ||
6d036f7d SL |
633 | struct stripe_head * |
634 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
635 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
636 | { |
637 | struct stripe_head *sh; | |
566c09c5 | 638 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 639 | int inc_empty_inactive_list_flag; |
1da177e4 | 640 | |
45b4233c | 641 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 642 | |
566c09c5 | 643 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
644 | |
645 | do { | |
b1b46486 | 646 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 647 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 648 | *(conf->hash_locks + hash)); |
86b42c71 | 649 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 650 | if (!sh) { |
edbe83ab | 651 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 652 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
653 | if (!sh && !test_bit(R5_DID_ALLOC, |
654 | &conf->cache_state)) | |
edbe83ab N |
655 | set_bit(R5_ALLOC_MORE, |
656 | &conf->cache_state); | |
657 | } | |
1da177e4 LT |
658 | if (noblock && sh == NULL) |
659 | break; | |
a39f7afd SL |
660 | |
661 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 662 | if (!sh) { |
5423399a N |
663 | set_bit(R5_INACTIVE_BLOCKED, |
664 | &conf->cache_state); | |
a39f7afd | 665 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
666 | wait_event_lock_irq( |
667 | conf->wait_for_stripe, | |
566c09c5 SL |
668 | !list_empty(conf->inactive_list + hash) && |
669 | (atomic_read(&conf->active_stripes) | |
670 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
671 | || !test_bit(R5_INACTIVE_BLOCKED, |
672 | &conf->cache_state)), | |
6ab2a4b8 | 673 | *(conf->hash_locks + hash)); |
5423399a N |
674 | clear_bit(R5_INACTIVE_BLOCKED, |
675 | &conf->cache_state); | |
7da9d450 | 676 | } else { |
b5663ba4 | 677 | init_stripe(sh, sector, previous); |
7da9d450 N |
678 | atomic_inc(&sh->count); |
679 | } | |
e240c183 | 680 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 681 | spin_lock(&conf->device_lock); |
e240c183 | 682 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
683 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
684 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
685 | BUG_ON(list_empty(&sh->lru) && |
686 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
687 | inc_empty_inactive_list_flag = 0; |
688 | if (!list_empty(conf->inactive_list + hash)) | |
689 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 690 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
691 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
692 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
693 | if (sh->group) { |
694 | sh->group->stripes_cnt--; | |
695 | sh->group = NULL; | |
696 | } | |
1da177e4 | 697 | } |
7da9d450 | 698 | atomic_inc(&sh->count); |
6d183de4 | 699 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
700 | } |
701 | } while (sh == NULL); | |
702 | ||
566c09c5 | 703 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
704 | return sh; |
705 | } | |
706 | ||
7a87f434 | 707 | static bool is_full_stripe_write(struct stripe_head *sh) |
708 | { | |
709 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
710 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
711 | } | |
712 | ||
59fc630b | 713 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
714 | { | |
59fc630b | 715 | if (sh1 > sh2) { |
3d05f3ae | 716 | spin_lock_irq(&sh2->stripe_lock); |
59fc630b | 717 | spin_lock_nested(&sh1->stripe_lock, 1); |
718 | } else { | |
3d05f3ae | 719 | spin_lock_irq(&sh1->stripe_lock); |
59fc630b | 720 | spin_lock_nested(&sh2->stripe_lock, 1); |
721 | } | |
722 | } | |
723 | ||
724 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
725 | { | |
726 | spin_unlock(&sh1->stripe_lock); | |
3d05f3ae | 727 | spin_unlock_irq(&sh2->stripe_lock); |
59fc630b | 728 | } |
729 | ||
730 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
731 | static bool stripe_can_batch(struct stripe_head *sh) | |
732 | { | |
9c3e333d SL |
733 | struct r5conf *conf = sh->raid_conf; |
734 | ||
e254de6b | 735 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
9c3e333d | 736 | return false; |
59fc630b | 737 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 738 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 739 | is_full_stripe_write(sh); |
740 | } | |
741 | ||
742 | /* we only do back search */ | |
743 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
744 | { | |
745 | struct stripe_head *head; | |
746 | sector_t head_sector, tmp_sec; | |
747 | int hash; | |
748 | int dd_idx; | |
ff00d3b4 | 749 | int inc_empty_inactive_list_flag; |
59fc630b | 750 | |
59fc630b | 751 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
752 | tmp_sec = sh->sector; | |
753 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
754 | return; | |
755 | head_sector = sh->sector - STRIPE_SECTORS; | |
756 | ||
757 | hash = stripe_hash_locks_hash(head_sector); | |
758 | spin_lock_irq(conf->hash_locks + hash); | |
759 | head = __find_stripe(conf, head_sector, conf->generation); | |
760 | if (head && !atomic_inc_not_zero(&head->count)) { | |
761 | spin_lock(&conf->device_lock); | |
762 | if (!atomic_read(&head->count)) { | |
763 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
764 | atomic_inc(&conf->active_stripes); | |
765 | BUG_ON(list_empty(&head->lru) && | |
766 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
767 | inc_empty_inactive_list_flag = 0; |
768 | if (!list_empty(conf->inactive_list + hash)) | |
769 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 770 | list_del_init(&head->lru); |
ff00d3b4 ZL |
771 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
772 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 773 | if (head->group) { |
774 | head->group->stripes_cnt--; | |
775 | head->group = NULL; | |
776 | } | |
777 | } | |
778 | atomic_inc(&head->count); | |
779 | spin_unlock(&conf->device_lock); | |
780 | } | |
781 | spin_unlock_irq(conf->hash_locks + hash); | |
782 | ||
783 | if (!head) | |
784 | return; | |
785 | if (!stripe_can_batch(head)) | |
786 | goto out; | |
787 | ||
788 | lock_two_stripes(head, sh); | |
789 | /* clear_batch_ready clear the flag */ | |
790 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
791 | goto unlock_out; | |
792 | ||
793 | if (sh->batch_head) | |
794 | goto unlock_out; | |
795 | ||
796 | dd_idx = 0; | |
797 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
798 | dd_idx++; | |
1eff9d32 | 799 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 800 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 801 | goto unlock_out; |
802 | ||
803 | if (head->batch_head) { | |
804 | spin_lock(&head->batch_head->batch_lock); | |
805 | /* This batch list is already running */ | |
806 | if (!stripe_can_batch(head)) { | |
807 | spin_unlock(&head->batch_head->batch_lock); | |
808 | goto unlock_out; | |
809 | } | |
3664847d SL |
810 | /* |
811 | * We must assign batch_head of this stripe within the | |
812 | * batch_lock, otherwise clear_batch_ready of batch head | |
813 | * stripe could clear BATCH_READY bit of this stripe and | |
814 | * this stripe->batch_head doesn't get assigned, which | |
815 | * could confuse clear_batch_ready for this stripe | |
816 | */ | |
817 | sh->batch_head = head->batch_head; | |
59fc630b | 818 | |
819 | /* | |
820 | * at this point, head's BATCH_READY could be cleared, but we | |
821 | * can still add the stripe to batch list | |
822 | */ | |
823 | list_add(&sh->batch_list, &head->batch_list); | |
824 | spin_unlock(&head->batch_head->batch_lock); | |
59fc630b | 825 | } else { |
826 | head->batch_head = head; | |
827 | sh->batch_head = head->batch_head; | |
828 | spin_lock(&head->batch_lock); | |
829 | list_add_tail(&sh->batch_list, &head->batch_list); | |
830 | spin_unlock(&head->batch_lock); | |
831 | } | |
832 | ||
833 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
834 | if (atomic_dec_return(&conf->preread_active_stripes) | |
835 | < IO_THRESHOLD) | |
836 | md_wakeup_thread(conf->mddev->thread); | |
837 | ||
2b6b2457 N |
838 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
839 | int seq = sh->bm_seq; | |
840 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
841 | sh->batch_head->bm_seq > seq) | |
842 | seq = sh->batch_head->bm_seq; | |
843 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
844 | sh->batch_head->bm_seq = seq; | |
845 | } | |
846 | ||
59fc630b | 847 | atomic_inc(&sh->count); |
848 | unlock_out: | |
849 | unlock_two_stripes(head, sh); | |
850 | out: | |
6d036f7d | 851 | raid5_release_stripe(head); |
59fc630b | 852 | } |
853 | ||
05616be5 N |
854 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
855 | * in this stripe_head. | |
856 | */ | |
857 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
858 | { | |
859 | sector_t progress = conf->reshape_progress; | |
860 | /* Need a memory barrier to make sure we see the value | |
861 | * of conf->generation, or ->data_offset that was set before | |
862 | * reshape_progress was updated. | |
863 | */ | |
864 | smp_rmb(); | |
865 | if (progress == MaxSector) | |
866 | return 0; | |
867 | if (sh->generation == conf->generation - 1) | |
868 | return 0; | |
869 | /* We are in a reshape, and this is a new-generation stripe, | |
870 | * so use new_data_offset. | |
871 | */ | |
872 | return 1; | |
873 | } | |
874 | ||
aaf9f12e | 875 | static void dispatch_bio_list(struct bio_list *tmp) |
765d704d | 876 | { |
765d704d SL |
877 | struct bio *bio; |
878 | ||
aaf9f12e SL |
879 | while ((bio = bio_list_pop(tmp))) |
880 | generic_make_request(bio); | |
881 | } | |
882 | ||
883 | static int cmp_stripe(void *priv, struct list_head *a, struct list_head *b) | |
884 | { | |
885 | const struct r5pending_data *da = list_entry(a, | |
886 | struct r5pending_data, sibling); | |
887 | const struct r5pending_data *db = list_entry(b, | |
888 | struct r5pending_data, sibling); | |
889 | if (da->sector > db->sector) | |
890 | return 1; | |
891 | if (da->sector < db->sector) | |
892 | return -1; | |
893 | return 0; | |
894 | } | |
895 | ||
896 | static void dispatch_defer_bios(struct r5conf *conf, int target, | |
897 | struct bio_list *list) | |
898 | { | |
899 | struct r5pending_data *data; | |
900 | struct list_head *first, *next = NULL; | |
901 | int cnt = 0; | |
902 | ||
903 | if (conf->pending_data_cnt == 0) | |
904 | return; | |
905 | ||
906 | list_sort(NULL, &conf->pending_list, cmp_stripe); | |
907 | ||
908 | first = conf->pending_list.next; | |
909 | ||
910 | /* temporarily move the head */ | |
911 | if (conf->next_pending_data) | |
912 | list_move_tail(&conf->pending_list, | |
913 | &conf->next_pending_data->sibling); | |
914 | ||
915 | while (!list_empty(&conf->pending_list)) { | |
916 | data = list_first_entry(&conf->pending_list, | |
917 | struct r5pending_data, sibling); | |
918 | if (&data->sibling == first) | |
919 | first = data->sibling.next; | |
920 | next = data->sibling.next; | |
921 | ||
922 | bio_list_merge(list, &data->bios); | |
923 | list_move(&data->sibling, &conf->free_list); | |
924 | cnt++; | |
925 | if (cnt >= target) | |
926 | break; | |
927 | } | |
928 | conf->pending_data_cnt -= cnt; | |
929 | BUG_ON(conf->pending_data_cnt < 0 || cnt < target); | |
930 | ||
931 | if (next != &conf->pending_list) | |
932 | conf->next_pending_data = list_entry(next, | |
933 | struct r5pending_data, sibling); | |
934 | else | |
935 | conf->next_pending_data = NULL; | |
936 | /* list isn't empty */ | |
937 | if (first != &conf->pending_list) | |
938 | list_move_tail(&conf->pending_list, first); | |
939 | } | |
940 | ||
941 | static void flush_deferred_bios(struct r5conf *conf) | |
942 | { | |
943 | struct bio_list tmp = BIO_EMPTY_LIST; | |
944 | ||
945 | if (conf->pending_data_cnt == 0) | |
765d704d SL |
946 | return; |
947 | ||
765d704d | 948 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
949 | dispatch_defer_bios(conf, conf->pending_data_cnt, &tmp); |
950 | BUG_ON(conf->pending_data_cnt != 0); | |
765d704d SL |
951 | spin_unlock(&conf->pending_bios_lock); |
952 | ||
aaf9f12e | 953 | dispatch_bio_list(&tmp); |
765d704d SL |
954 | } |
955 | ||
aaf9f12e SL |
956 | static void defer_issue_bios(struct r5conf *conf, sector_t sector, |
957 | struct bio_list *bios) | |
765d704d | 958 | { |
aaf9f12e SL |
959 | struct bio_list tmp = BIO_EMPTY_LIST; |
960 | struct r5pending_data *ent; | |
961 | ||
765d704d | 962 | spin_lock(&conf->pending_bios_lock); |
aaf9f12e SL |
963 | ent = list_first_entry(&conf->free_list, struct r5pending_data, |
964 | sibling); | |
965 | list_move_tail(&ent->sibling, &conf->pending_list); | |
966 | ent->sector = sector; | |
967 | bio_list_init(&ent->bios); | |
968 | bio_list_merge(&ent->bios, bios); | |
969 | conf->pending_data_cnt++; | |
970 | if (conf->pending_data_cnt >= PENDING_IO_MAX) | |
971 | dispatch_defer_bios(conf, PENDING_IO_ONE_FLUSH, &tmp); | |
972 | ||
765d704d | 973 | spin_unlock(&conf->pending_bios_lock); |
aaf9f12e SL |
974 | |
975 | dispatch_bio_list(&tmp); | |
765d704d SL |
976 | } |
977 | ||
6712ecf8 | 978 | static void |
4246a0b6 | 979 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 980 | static void |
4246a0b6 | 981 | raid5_end_write_request(struct bio *bi); |
91c00924 | 982 | |
c4e5ac0a | 983 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 984 | { |
d1688a6d | 985 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 986 | int i, disks = sh->disks; |
59fc630b | 987 | struct stripe_head *head_sh = sh; |
aaf9f12e SL |
988 | struct bio_list pending_bios = BIO_EMPTY_LIST; |
989 | bool should_defer; | |
91c00924 DW |
990 | |
991 | might_sleep(); | |
992 | ||
ff875738 AP |
993 | if (log_stripe(sh, s) == 0) |
994 | return; | |
1e6d690b | 995 | |
aaf9f12e | 996 | should_defer = conf->batch_bio_dispatch && conf->group_cnt; |
1e6d690b | 997 | |
91c00924 | 998 | for (i = disks; i--; ) { |
796a5cf0 | 999 | int op, op_flags = 0; |
9a3e1101 | 1000 | int replace_only = 0; |
977df362 N |
1001 | struct bio *bi, *rbi; |
1002 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 1003 | |
1004 | sh = head_sh; | |
e9c7469b | 1005 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 1006 | op = REQ_OP_WRITE; |
e9c7469b | 1007 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 1008 | op_flags = REQ_FUA; |
9e444768 | 1009 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 1010 | op = REQ_OP_DISCARD; |
e9c7469b | 1011 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 1012 | op = REQ_OP_READ; |
9a3e1101 N |
1013 | else if (test_and_clear_bit(R5_WantReplace, |
1014 | &sh->dev[i].flags)) { | |
796a5cf0 | 1015 | op = REQ_OP_WRITE; |
9a3e1101 N |
1016 | replace_only = 1; |
1017 | } else | |
91c00924 | 1018 | continue; |
bc0934f0 | 1019 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 1020 | op_flags |= REQ_SYNC; |
91c00924 | 1021 | |
59fc630b | 1022 | again: |
91c00924 | 1023 | bi = &sh->dev[i].req; |
977df362 | 1024 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 1025 | |
91c00924 | 1026 | rcu_read_lock(); |
9a3e1101 | 1027 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
1028 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
1029 | rdev = rcu_dereference(conf->disks[i].rdev); | |
1030 | if (!rdev) { | |
1031 | rdev = rrdev; | |
1032 | rrdev = NULL; | |
1033 | } | |
796a5cf0 | 1034 | if (op_is_write(op)) { |
9a3e1101 N |
1035 | if (replace_only) |
1036 | rdev = NULL; | |
dd054fce N |
1037 | if (rdev == rrdev) |
1038 | /* We raced and saw duplicates */ | |
1039 | rrdev = NULL; | |
9a3e1101 | 1040 | } else { |
59fc630b | 1041 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
1042 | rdev = rrdev; |
1043 | rrdev = NULL; | |
1044 | } | |
977df362 | 1045 | |
91c00924 DW |
1046 | if (rdev && test_bit(Faulty, &rdev->flags)) |
1047 | rdev = NULL; | |
1048 | if (rdev) | |
1049 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
1050 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
1051 | rrdev = NULL; | |
1052 | if (rrdev) | |
1053 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1054 | rcu_read_unlock(); |
1055 | ||
73e92e51 | 1056 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1057 | * need to check for writes. We never accept write errors |
1058 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1059 | */ |
796a5cf0 | 1060 | while (op_is_write(op) && rdev && |
73e92e51 N |
1061 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1062 | sector_t first_bad; | |
1063 | int bad_sectors; | |
1064 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
1065 | &first_bad, &bad_sectors); | |
1066 | if (!bad) | |
1067 | break; | |
1068 | ||
1069 | if (bad < 0) { | |
1070 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1071 | if (!conf->mddev->external && | |
2953079c | 1072 | conf->mddev->sb_flags) { |
73e92e51 N |
1073 | /* It is very unlikely, but we might |
1074 | * still need to write out the | |
1075 | * bad block log - better give it | |
1076 | * a chance*/ | |
1077 | md_check_recovery(conf->mddev); | |
1078 | } | |
1850753d | 1079 | /* |
1080 | * Because md_wait_for_blocked_rdev | |
1081 | * will dec nr_pending, we must | |
1082 | * increment it first. | |
1083 | */ | |
1084 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1085 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1086 | } else { | |
1087 | /* Acknowledged bad block - skip the write */ | |
1088 | rdev_dec_pending(rdev, conf->mddev); | |
1089 | rdev = NULL; | |
1090 | } | |
1091 | } | |
1092 | ||
91c00924 | 1093 | if (rdev) { |
9a3e1101 N |
1094 | if (s->syncing || s->expanding || s->expanded |
1095 | || s->replacing) | |
91c00924 DW |
1096 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1097 | ||
2b7497f0 DW |
1098 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1099 | ||
74d46992 | 1100 | bio_set_dev(bi, rdev->bdev); |
796a5cf0 MC |
1101 | bio_set_op_attrs(bi, op, op_flags); |
1102 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1103 | ? raid5_end_write_request |
1104 | : raid5_end_read_request; | |
1105 | bi->bi_private = sh; | |
1106 | ||
6296b960 | 1107 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1108 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1109 | bi->bi_opf, i); |
91c00924 | 1110 | atomic_inc(&sh->count); |
59fc630b | 1111 | if (sh != head_sh) |
1112 | atomic_inc(&head_sh->count); | |
05616be5 | 1113 | if (use_new_offset(conf, sh)) |
4f024f37 | 1114 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1115 | + rdev->new_data_offset); |
1116 | else | |
4f024f37 | 1117 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1118 | + rdev->data_offset); |
59fc630b | 1119 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1120 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1121 | |
d592a996 SL |
1122 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1123 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1124 | |
1125 | if (!op_is_write(op) && | |
1126 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1127 | /* | |
1128 | * issuing read for a page in journal, this | |
1129 | * must be preparing for prexor in rmw; read | |
1130 | * the data into orig_page | |
1131 | */ | |
1132 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1133 | else | |
1134 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1135 | bi->bi_vcnt = 1; |
91c00924 DW |
1136 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1137 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1138 | bi->bi_iter.bi_size = STRIPE_SIZE; |
2cd259a7 MD |
1139 | bi->bi_write_hint = sh->dev[i].write_hint; |
1140 | if (!rrdev) | |
1141 | sh->dev[i].write_hint = RWF_WRITE_LIFE_NOT_SET; | |
37c61ff3 SL |
1142 | /* |
1143 | * If this is discard request, set bi_vcnt 0. We don't | |
1144 | * want to confuse SCSI because SCSI will replace payload | |
1145 | */ | |
796a5cf0 | 1146 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1147 | bi->bi_vcnt = 0; |
977df362 N |
1148 | if (rrdev) |
1149 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1150 | |
1151 | if (conf->mddev->gendisk) | |
74d46992 | 1152 | trace_block_bio_remap(bi->bi_disk->queue, |
e3620a3a JB |
1153 | bi, disk_devt(conf->mddev->gendisk), |
1154 | sh->dev[i].sector); | |
aaf9f12e SL |
1155 | if (should_defer && op_is_write(op)) |
1156 | bio_list_add(&pending_bios, bi); | |
1157 | else | |
1158 | generic_make_request(bi); | |
977df362 N |
1159 | } |
1160 | if (rrdev) { | |
9a3e1101 N |
1161 | if (s->syncing || s->expanding || s->expanded |
1162 | || s->replacing) | |
977df362 N |
1163 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1164 | ||
1165 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1166 | ||
74d46992 | 1167 | bio_set_dev(rbi, rrdev->bdev); |
796a5cf0 MC |
1168 | bio_set_op_attrs(rbi, op, op_flags); |
1169 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1170 | rbi->bi_end_io = raid5_end_write_request; |
1171 | rbi->bi_private = sh; | |
1172 | ||
6296b960 | 1173 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1174 | "replacement disc %d\n", |
1175 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1176 | rbi->bi_opf, i); |
977df362 | 1177 | atomic_inc(&sh->count); |
59fc630b | 1178 | if (sh != head_sh) |
1179 | atomic_inc(&head_sh->count); | |
05616be5 | 1180 | if (use_new_offset(conf, sh)) |
4f024f37 | 1181 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1182 | + rrdev->new_data_offset); |
1183 | else | |
4f024f37 | 1184 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1185 | + rrdev->data_offset); |
d592a996 SL |
1186 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1187 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1188 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1189 | rbi->bi_vcnt = 1; |
977df362 N |
1190 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1191 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1192 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
2cd259a7 MD |
1193 | rbi->bi_write_hint = sh->dev[i].write_hint; |
1194 | sh->dev[i].write_hint = RWF_WRITE_LIFE_NOT_SET; | |
37c61ff3 SL |
1195 | /* |
1196 | * If this is discard request, set bi_vcnt 0. We don't | |
1197 | * want to confuse SCSI because SCSI will replace payload | |
1198 | */ | |
796a5cf0 | 1199 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1200 | rbi->bi_vcnt = 0; |
e3620a3a | 1201 | if (conf->mddev->gendisk) |
74d46992 | 1202 | trace_block_bio_remap(rbi->bi_disk->queue, |
e3620a3a JB |
1203 | rbi, disk_devt(conf->mddev->gendisk), |
1204 | sh->dev[i].sector); | |
aaf9f12e SL |
1205 | if (should_defer && op_is_write(op)) |
1206 | bio_list_add(&pending_bios, rbi); | |
1207 | else | |
1208 | generic_make_request(rbi); | |
977df362 N |
1209 | } |
1210 | if (!rdev && !rrdev) { | |
796a5cf0 | 1211 | if (op_is_write(op)) |
91c00924 | 1212 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1213 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1214 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1215 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1216 | set_bit(STRIPE_HANDLE, &sh->state); | |
1217 | } | |
59fc630b | 1218 | |
1219 | if (!head_sh->batch_head) | |
1220 | continue; | |
1221 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1222 | batch_list); | |
1223 | if (sh != head_sh) | |
1224 | goto again; | |
91c00924 | 1225 | } |
aaf9f12e SL |
1226 | |
1227 | if (should_defer && !bio_list_empty(&pending_bios)) | |
1228 | defer_issue_bios(conf, head_sh->sector, &pending_bios); | |
91c00924 DW |
1229 | } |
1230 | ||
1231 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1232 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1233 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1234 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1235 | { |
7988613b KO |
1236 | struct bio_vec bvl; |
1237 | struct bvec_iter iter; | |
91c00924 | 1238 | struct page *bio_page; |
91c00924 | 1239 | int page_offset; |
a08abd8c | 1240 | struct async_submit_ctl submit; |
0403e382 | 1241 | enum async_tx_flags flags = 0; |
91c00924 | 1242 | |
4f024f37 KO |
1243 | if (bio->bi_iter.bi_sector >= sector) |
1244 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1245 | else |
4f024f37 | 1246 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1247 | |
0403e382 DW |
1248 | if (frombio) |
1249 | flags |= ASYNC_TX_FENCE; | |
1250 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1251 | ||
7988613b KO |
1252 | bio_for_each_segment(bvl, bio, iter) { |
1253 | int len = bvl.bv_len; | |
91c00924 DW |
1254 | int clen; |
1255 | int b_offset = 0; | |
1256 | ||
1257 | if (page_offset < 0) { | |
1258 | b_offset = -page_offset; | |
1259 | page_offset += b_offset; | |
1260 | len -= b_offset; | |
1261 | } | |
1262 | ||
1263 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1264 | clen = STRIPE_SIZE - page_offset; | |
1265 | else | |
1266 | clen = len; | |
1267 | ||
1268 | if (clen > 0) { | |
7988613b KO |
1269 | b_offset += bvl.bv_offset; |
1270 | bio_page = bvl.bv_page; | |
d592a996 SL |
1271 | if (frombio) { |
1272 | if (sh->raid_conf->skip_copy && | |
1273 | b_offset == 0 && page_offset == 0 && | |
1e6d690b SL |
1274 | clen == STRIPE_SIZE && |
1275 | !no_skipcopy) | |
d592a996 SL |
1276 | *page = bio_page; |
1277 | else | |
1278 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1279 | b_offset, clen, &submit); |
d592a996 SL |
1280 | } else |
1281 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1282 | page_offset, clen, &submit); |
91c00924 | 1283 | } |
a08abd8c DW |
1284 | /* chain the operations */ |
1285 | submit.depend_tx = tx; | |
1286 | ||
91c00924 DW |
1287 | if (clen < len) /* hit end of page */ |
1288 | break; | |
1289 | page_offset += len; | |
1290 | } | |
1291 | ||
1292 | return tx; | |
1293 | } | |
1294 | ||
1295 | static void ops_complete_biofill(void *stripe_head_ref) | |
1296 | { | |
1297 | struct stripe_head *sh = stripe_head_ref; | |
e4d84909 | 1298 | int i; |
91c00924 | 1299 | |
e46b272b | 1300 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1301 | (unsigned long long)sh->sector); |
1302 | ||
1303 | /* clear completed biofills */ | |
1304 | for (i = sh->disks; i--; ) { | |
1305 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1306 | |
1307 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1308 | /* and check if we need to reply to a read request, |
1309 | * new R5_Wantfill requests are held off until | |
83de75cc | 1310 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1311 | */ |
1312 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1313 | struct bio *rbi, *rbi2; |
91c00924 | 1314 | |
91c00924 DW |
1315 | BUG_ON(!dev->read); |
1316 | rbi = dev->read; | |
1317 | dev->read = NULL; | |
4f024f37 | 1318 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1319 | dev->sector + STRIPE_SECTORS) { |
1320 | rbi2 = r5_next_bio(rbi, dev->sector); | |
016c76ac | 1321 | bio_endio(rbi); |
91c00924 DW |
1322 | rbi = rbi2; |
1323 | } | |
1324 | } | |
1325 | } | |
83de75cc | 1326 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1327 | |
e4d84909 | 1328 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1329 | raid5_release_stripe(sh); |
91c00924 DW |
1330 | } |
1331 | ||
1332 | static void ops_run_biofill(struct stripe_head *sh) | |
1333 | { | |
1334 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1335 | struct async_submit_ctl submit; |
91c00924 DW |
1336 | int i; |
1337 | ||
59fc630b | 1338 | BUG_ON(sh->batch_head); |
e46b272b | 1339 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1340 | (unsigned long long)sh->sector); |
1341 | ||
1342 | for (i = sh->disks; i--; ) { | |
1343 | struct r5dev *dev = &sh->dev[i]; | |
1344 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1345 | struct bio *rbi; | |
b17459c0 | 1346 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1347 | dev->read = rbi = dev->toread; |
1348 | dev->toread = NULL; | |
b17459c0 | 1349 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1350 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1351 | dev->sector + STRIPE_SECTORS) { |
d592a996 | 1352 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1353 | dev->sector, tx, sh, 0); |
91c00924 DW |
1354 | rbi = r5_next_bio(rbi, dev->sector); |
1355 | } | |
1356 | } | |
1357 | } | |
1358 | ||
1359 | atomic_inc(&sh->count); | |
a08abd8c DW |
1360 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1361 | async_trigger_callback(&submit); | |
91c00924 DW |
1362 | } |
1363 | ||
4e7d2c0a | 1364 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1365 | { |
4e7d2c0a | 1366 | struct r5dev *tgt; |
91c00924 | 1367 | |
4e7d2c0a DW |
1368 | if (target < 0) |
1369 | return; | |
91c00924 | 1370 | |
4e7d2c0a | 1371 | tgt = &sh->dev[target]; |
91c00924 DW |
1372 | set_bit(R5_UPTODATE, &tgt->flags); |
1373 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1374 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1375 | } |
1376 | ||
ac6b53b6 | 1377 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1378 | { |
1379 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1380 | |
e46b272b | 1381 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1382 | (unsigned long long)sh->sector); |
1383 | ||
ac6b53b6 | 1384 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1385 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1386 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1387 | |
ecc65c9b DW |
1388 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1389 | if (sh->check_state == check_state_compute_run) | |
1390 | sh->check_state = check_state_compute_result; | |
91c00924 | 1391 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1392 | raid5_release_stripe(sh); |
91c00924 DW |
1393 | } |
1394 | ||
d6f38f31 | 1395 | /* return a pointer to the address conversion region of the scribble buffer */ |
b330e6a4 | 1396 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) |
d6f38f31 | 1397 | { |
b330e6a4 | 1398 | return percpu->scribble + i * percpu->scribble_obj_size; |
46d5b785 | 1399 | } |
1400 | ||
1401 | /* return a pointer to the address conversion region of the scribble buffer */ | |
b330e6a4 KO |
1402 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, |
1403 | struct raid5_percpu *percpu, int i) | |
46d5b785 | 1404 | { |
b330e6a4 | 1405 | return (void *) (to_addr_page(percpu, i) + sh->disks + 2); |
d6f38f31 DW |
1406 | } |
1407 | ||
1408 | static struct dma_async_tx_descriptor * | |
1409 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1410 | { |
91c00924 | 1411 | int disks = sh->disks; |
46d5b785 | 1412 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1413 | int target = sh->ops.target; |
1414 | struct r5dev *tgt = &sh->dev[target]; | |
1415 | struct page *xor_dest = tgt->page; | |
1416 | int count = 0; | |
1417 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1418 | struct async_submit_ctl submit; |
91c00924 DW |
1419 | int i; |
1420 | ||
59fc630b | 1421 | BUG_ON(sh->batch_head); |
1422 | ||
91c00924 | 1423 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1424 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1425 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1426 | ||
1427 | for (i = disks; i--; ) | |
1428 | if (i != target) | |
1429 | xor_srcs[count++] = sh->dev[i].page; | |
1430 | ||
1431 | atomic_inc(&sh->count); | |
1432 | ||
0403e382 | 1433 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1434 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1435 | if (unlikely(count == 1)) |
a08abd8c | 1436 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1437 | else |
a08abd8c | 1438 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1439 | |
91c00924 DW |
1440 | return tx; |
1441 | } | |
1442 | ||
ac6b53b6 DW |
1443 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1444 | * @srcs - (struct page *) array of size sh->disks | |
1445 | * @sh - stripe_head to parse | |
1446 | * | |
1447 | * Populates srcs in proper layout order for the stripe and returns the | |
1448 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1449 | * destination buffer is recorded in srcs[count] and the Q destination | |
1450 | * is recorded in srcs[count+1]]. | |
1451 | */ | |
584acdd4 MS |
1452 | static int set_syndrome_sources(struct page **srcs, |
1453 | struct stripe_head *sh, | |
1454 | int srctype) | |
ac6b53b6 DW |
1455 | { |
1456 | int disks = sh->disks; | |
1457 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1458 | int d0_idx = raid6_d0(sh); | |
1459 | int count; | |
1460 | int i; | |
1461 | ||
1462 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1463 | srcs[i] = NULL; |
ac6b53b6 DW |
1464 | |
1465 | count = 0; | |
1466 | i = d0_idx; | |
1467 | do { | |
1468 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1469 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1470 | |
584acdd4 MS |
1471 | if (i == sh->qd_idx || i == sh->pd_idx || |
1472 | (srctype == SYNDROME_SRC_ALL) || | |
1473 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1474 | (test_bit(R5_Wantdrain, &dev->flags) || |
1475 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1476 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1477 | (dev->written || |
1478 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1479 | if (test_bit(R5_InJournal, &dev->flags)) |
1480 | srcs[slot] = sh->dev[i].orig_page; | |
1481 | else | |
1482 | srcs[slot] = sh->dev[i].page; | |
1483 | } | |
ac6b53b6 DW |
1484 | i = raid6_next_disk(i, disks); |
1485 | } while (i != d0_idx); | |
ac6b53b6 | 1486 | |
e4424fee | 1487 | return syndrome_disks; |
ac6b53b6 DW |
1488 | } |
1489 | ||
1490 | static struct dma_async_tx_descriptor * | |
1491 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1492 | { | |
1493 | int disks = sh->disks; | |
46d5b785 | 1494 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1495 | int target; |
1496 | int qd_idx = sh->qd_idx; | |
1497 | struct dma_async_tx_descriptor *tx; | |
1498 | struct async_submit_ctl submit; | |
1499 | struct r5dev *tgt; | |
1500 | struct page *dest; | |
1501 | int i; | |
1502 | int count; | |
1503 | ||
59fc630b | 1504 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1505 | if (sh->ops.target < 0) |
1506 | target = sh->ops.target2; | |
1507 | else if (sh->ops.target2 < 0) | |
1508 | target = sh->ops.target; | |
91c00924 | 1509 | else |
ac6b53b6 DW |
1510 | /* we should only have one valid target */ |
1511 | BUG(); | |
1512 | BUG_ON(target < 0); | |
1513 | pr_debug("%s: stripe %llu block: %d\n", | |
1514 | __func__, (unsigned long long)sh->sector, target); | |
1515 | ||
1516 | tgt = &sh->dev[target]; | |
1517 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1518 | dest = tgt->page; | |
1519 | ||
1520 | atomic_inc(&sh->count); | |
1521 | ||
1522 | if (target == qd_idx) { | |
584acdd4 | 1523 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1524 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1525 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1526 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1527 | ops_complete_compute, sh, | |
46d5b785 | 1528 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1529 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1530 | } else { | |
1531 | /* Compute any data- or p-drive using XOR */ | |
1532 | count = 0; | |
1533 | for (i = disks; i-- ; ) { | |
1534 | if (i == target || i == qd_idx) | |
1535 | continue; | |
1536 | blocks[count++] = sh->dev[i].page; | |
1537 | } | |
1538 | ||
0403e382 DW |
1539 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1540 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1541 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1542 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1543 | } | |
91c00924 | 1544 | |
91c00924 DW |
1545 | return tx; |
1546 | } | |
1547 | ||
ac6b53b6 DW |
1548 | static struct dma_async_tx_descriptor * |
1549 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1550 | { | |
1551 | int i, count, disks = sh->disks; | |
1552 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1553 | int d0_idx = raid6_d0(sh); | |
1554 | int faila = -1, failb = -1; | |
1555 | int target = sh->ops.target; | |
1556 | int target2 = sh->ops.target2; | |
1557 | struct r5dev *tgt = &sh->dev[target]; | |
1558 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1559 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1560 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1561 | struct async_submit_ctl submit; |
1562 | ||
59fc630b | 1563 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1564 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1565 | __func__, (unsigned long long)sh->sector, target, target2); | |
1566 | BUG_ON(target < 0 || target2 < 0); | |
1567 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1568 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1569 | ||
6c910a78 | 1570 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1571 | * slot number conversion for 'faila' and 'failb' |
1572 | */ | |
1573 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1574 | blocks[i] = NULL; |
ac6b53b6 DW |
1575 | count = 0; |
1576 | i = d0_idx; | |
1577 | do { | |
1578 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1579 | ||
1580 | blocks[slot] = sh->dev[i].page; | |
1581 | ||
1582 | if (i == target) | |
1583 | faila = slot; | |
1584 | if (i == target2) | |
1585 | failb = slot; | |
1586 | i = raid6_next_disk(i, disks); | |
1587 | } while (i != d0_idx); | |
ac6b53b6 DW |
1588 | |
1589 | BUG_ON(faila == failb); | |
1590 | if (failb < faila) | |
1591 | swap(faila, failb); | |
1592 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1593 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1594 | ||
1595 | atomic_inc(&sh->count); | |
1596 | ||
1597 | if (failb == syndrome_disks+1) { | |
1598 | /* Q disk is one of the missing disks */ | |
1599 | if (faila == syndrome_disks) { | |
1600 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1601 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1602 | ops_complete_compute, sh, | |
46d5b785 | 1603 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1604 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1605 | STRIPE_SIZE, &submit); |
1606 | } else { | |
1607 | struct page *dest; | |
1608 | int data_target; | |
1609 | int qd_idx = sh->qd_idx; | |
1610 | ||
1611 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1612 | if (target == qd_idx) | |
1613 | data_target = target2; | |
1614 | else | |
1615 | data_target = target; | |
1616 | ||
1617 | count = 0; | |
1618 | for (i = disks; i-- ; ) { | |
1619 | if (i == data_target || i == qd_idx) | |
1620 | continue; | |
1621 | blocks[count++] = sh->dev[i].page; | |
1622 | } | |
1623 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1624 | init_async_submit(&submit, |
1625 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1626 | NULL, NULL, NULL, | |
46d5b785 | 1627 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1628 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1629 | &submit); | |
1630 | ||
584acdd4 | 1631 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1632 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1633 | ops_complete_compute, sh, | |
46d5b785 | 1634 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1635 | return async_gen_syndrome(blocks, 0, count+2, |
1636 | STRIPE_SIZE, &submit); | |
1637 | } | |
ac6b53b6 | 1638 | } else { |
6c910a78 DW |
1639 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1640 | ops_complete_compute, sh, | |
46d5b785 | 1641 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1642 | if (failb == syndrome_disks) { |
1643 | /* We're missing D+P. */ | |
1644 | return async_raid6_datap_recov(syndrome_disks+2, | |
1645 | STRIPE_SIZE, faila, | |
1646 | blocks, &submit); | |
1647 | } else { | |
1648 | /* We're missing D+D. */ | |
1649 | return async_raid6_2data_recov(syndrome_disks+2, | |
1650 | STRIPE_SIZE, faila, failb, | |
1651 | blocks, &submit); | |
1652 | } | |
ac6b53b6 DW |
1653 | } |
1654 | } | |
1655 | ||
91c00924 DW |
1656 | static void ops_complete_prexor(void *stripe_head_ref) |
1657 | { | |
1658 | struct stripe_head *sh = stripe_head_ref; | |
1659 | ||
e46b272b | 1660 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1661 | (unsigned long long)sh->sector); |
1e6d690b SL |
1662 | |
1663 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1664 | /* | |
1665 | * raid5-cache write back uses orig_page during prexor. | |
1666 | * After prexor, it is time to free orig_page | |
1667 | */ | |
1668 | r5c_release_extra_page(sh); | |
91c00924 DW |
1669 | } |
1670 | ||
1671 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1672 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1673 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1674 | { |
91c00924 | 1675 | int disks = sh->disks; |
46d5b785 | 1676 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1677 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1678 | struct async_submit_ctl submit; |
91c00924 DW |
1679 | |
1680 | /* existing parity data subtracted */ | |
1681 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1682 | ||
59fc630b | 1683 | BUG_ON(sh->batch_head); |
e46b272b | 1684 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1685 | (unsigned long long)sh->sector); |
1686 | ||
1687 | for (i = disks; i--; ) { | |
1688 | struct r5dev *dev = &sh->dev[i]; | |
1689 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1690 | if (test_bit(R5_InJournal, &dev->flags)) |
1691 | xor_srcs[count++] = dev->orig_page; | |
1692 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1693 | xor_srcs[count++] = dev->page; |
1694 | } | |
1695 | ||
0403e382 | 1696 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1697 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1698 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1699 | |
1700 | return tx; | |
1701 | } | |
1702 | ||
584acdd4 MS |
1703 | static struct dma_async_tx_descriptor * |
1704 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1705 | struct dma_async_tx_descriptor *tx) | |
1706 | { | |
1707 | struct page **blocks = to_addr_page(percpu, 0); | |
1708 | int count; | |
1709 | struct async_submit_ctl submit; | |
1710 | ||
1711 | pr_debug("%s: stripe %llu\n", __func__, | |
1712 | (unsigned long long)sh->sector); | |
1713 | ||
1714 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1715 | ||
1716 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1717 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1718 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1719 | ||
1720 | return tx; | |
1721 | } | |
1722 | ||
91c00924 | 1723 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1724 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1725 | { |
1e6d690b | 1726 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1727 | int disks = sh->disks; |
d8ee0728 | 1728 | int i; |
59fc630b | 1729 | struct stripe_head *head_sh = sh; |
91c00924 | 1730 | |
e46b272b | 1731 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1732 | (unsigned long long)sh->sector); |
1733 | ||
1734 | for (i = disks; i--; ) { | |
59fc630b | 1735 | struct r5dev *dev; |
91c00924 | 1736 | struct bio *chosen; |
91c00924 | 1737 | |
59fc630b | 1738 | sh = head_sh; |
1739 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1740 | struct bio *wbi; |
1741 | ||
59fc630b | 1742 | again: |
1743 | dev = &sh->dev[i]; | |
1e6d690b SL |
1744 | /* |
1745 | * clear R5_InJournal, so when rewriting a page in | |
1746 | * journal, it is not skipped by r5l_log_stripe() | |
1747 | */ | |
1748 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1749 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1750 | chosen = dev->towrite; |
1751 | dev->towrite = NULL; | |
7a87f434 | 1752 | sh->overwrite_disks = 0; |
91c00924 DW |
1753 | BUG_ON(dev->written); |
1754 | wbi = dev->written = chosen; | |
b17459c0 | 1755 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1756 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1757 | |
4f024f37 | 1758 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1759 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1760 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1761 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1762 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1763 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1764 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1765 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1766 | else { |
1767 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1768 | dev->sector, tx, sh, |
1769 | r5c_is_writeback(conf->log)); | |
1770 | if (dev->page != dev->orig_page && | |
1771 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1772 | set_bit(R5_SkipCopy, &dev->flags); |
1773 | clear_bit(R5_UPTODATE, &dev->flags); | |
1774 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1775 | } | |
1776 | } | |
91c00924 DW |
1777 | wbi = r5_next_bio(wbi, dev->sector); |
1778 | } | |
59fc630b | 1779 | |
1780 | if (head_sh->batch_head) { | |
1781 | sh = list_first_entry(&sh->batch_list, | |
1782 | struct stripe_head, | |
1783 | batch_list); | |
1784 | if (sh == head_sh) | |
1785 | continue; | |
1786 | goto again; | |
1787 | } | |
91c00924 DW |
1788 | } |
1789 | } | |
1790 | ||
1791 | return tx; | |
1792 | } | |
1793 | ||
ac6b53b6 | 1794 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1795 | { |
1796 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1797 | int disks = sh->disks; |
1798 | int pd_idx = sh->pd_idx; | |
1799 | int qd_idx = sh->qd_idx; | |
1800 | int i; | |
9e444768 | 1801 | bool fua = false, sync = false, discard = false; |
91c00924 | 1802 | |
e46b272b | 1803 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1804 | (unsigned long long)sh->sector); |
1805 | ||
bc0934f0 | 1806 | for (i = disks; i--; ) { |
e9c7469b | 1807 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1808 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1809 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1810 | } |
e9c7469b | 1811 | |
91c00924 DW |
1812 | for (i = disks; i--; ) { |
1813 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1814 | |
e9c7469b | 1815 | if (dev->written || i == pd_idx || i == qd_idx) { |
235b6003 | 1816 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) { |
9e444768 | 1817 | set_bit(R5_UPTODATE, &dev->flags); |
235b6003 N |
1818 | if (test_bit(STRIPE_EXPAND_READY, &sh->state)) |
1819 | set_bit(R5_Expanded, &dev->flags); | |
1820 | } | |
e9c7469b TH |
1821 | if (fua) |
1822 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1823 | if (sync) |
1824 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1825 | } |
91c00924 DW |
1826 | } |
1827 | ||
d8ee0728 DW |
1828 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1829 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1830 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1831 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1832 | else { | |
1833 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1834 | sh->reconstruct_state = reconstruct_state_result; | |
1835 | } | |
91c00924 DW |
1836 | |
1837 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1838 | raid5_release_stripe(sh); |
91c00924 DW |
1839 | } |
1840 | ||
1841 | static void | |
ac6b53b6 DW |
1842 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1843 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1844 | { |
91c00924 | 1845 | int disks = sh->disks; |
59fc630b | 1846 | struct page **xor_srcs; |
a08abd8c | 1847 | struct async_submit_ctl submit; |
59fc630b | 1848 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1849 | struct page *xor_dest; |
d8ee0728 | 1850 | int prexor = 0; |
91c00924 | 1851 | unsigned long flags; |
59fc630b | 1852 | int j = 0; |
1853 | struct stripe_head *head_sh = sh; | |
1854 | int last_stripe; | |
91c00924 | 1855 | |
e46b272b | 1856 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1857 | (unsigned long long)sh->sector); |
1858 | ||
620125f2 SL |
1859 | for (i = 0; i < sh->disks; i++) { |
1860 | if (pd_idx == i) | |
1861 | continue; | |
1862 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1863 | break; | |
1864 | } | |
1865 | if (i >= sh->disks) { | |
1866 | atomic_inc(&sh->count); | |
620125f2 SL |
1867 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1868 | ops_complete_reconstruct(sh); | |
1869 | return; | |
1870 | } | |
59fc630b | 1871 | again: |
1872 | count = 0; | |
1873 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1874 | /* check if prexor is active which means only process blocks |
1875 | * that are part of a read-modify-write (written) | |
1876 | */ | |
59fc630b | 1877 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1878 | prexor = 1; |
91c00924 DW |
1879 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1880 | for (i = disks; i--; ) { | |
1881 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1882 | if (head_sh->dev[i].written || |
1883 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1884 | xor_srcs[count++] = dev->page; |
1885 | } | |
1886 | } else { | |
1887 | xor_dest = sh->dev[pd_idx].page; | |
1888 | for (i = disks; i--; ) { | |
1889 | struct r5dev *dev = &sh->dev[i]; | |
1890 | if (i != pd_idx) | |
1891 | xor_srcs[count++] = dev->page; | |
1892 | } | |
1893 | } | |
1894 | ||
91c00924 DW |
1895 | /* 1/ if we prexor'd then the dest is reused as a source |
1896 | * 2/ if we did not prexor then we are redoing the parity | |
1897 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1898 | * for the synchronous xor case | |
1899 | */ | |
59fc630b | 1900 | last_stripe = !head_sh->batch_head || |
1901 | list_first_entry(&sh->batch_list, | |
1902 | struct stripe_head, batch_list) == head_sh; | |
1903 | if (last_stripe) { | |
1904 | flags = ASYNC_TX_ACK | | |
1905 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1906 | ||
1907 | atomic_inc(&head_sh->count); | |
1908 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1909 | to_addr_conv(sh, percpu, j)); | |
1910 | } else { | |
1911 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1912 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1913 | to_addr_conv(sh, percpu, j)); | |
1914 | } | |
91c00924 | 1915 | |
a08abd8c DW |
1916 | if (unlikely(count == 1)) |
1917 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1918 | else | |
1919 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1920 | if (!last_stripe) { |
1921 | j++; | |
1922 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1923 | batch_list); | |
1924 | goto again; | |
1925 | } | |
91c00924 DW |
1926 | } |
1927 | ||
ac6b53b6 DW |
1928 | static void |
1929 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1930 | struct dma_async_tx_descriptor *tx) | |
1931 | { | |
1932 | struct async_submit_ctl submit; | |
59fc630b | 1933 | struct page **blocks; |
1934 | int count, i, j = 0; | |
1935 | struct stripe_head *head_sh = sh; | |
1936 | int last_stripe; | |
584acdd4 MS |
1937 | int synflags; |
1938 | unsigned long txflags; | |
ac6b53b6 DW |
1939 | |
1940 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1941 | ||
620125f2 SL |
1942 | for (i = 0; i < sh->disks; i++) { |
1943 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1944 | continue; | |
1945 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1946 | break; | |
1947 | } | |
1948 | if (i >= sh->disks) { | |
1949 | atomic_inc(&sh->count); | |
620125f2 SL |
1950 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1951 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1952 | ops_complete_reconstruct(sh); | |
1953 | return; | |
1954 | } | |
1955 | ||
59fc630b | 1956 | again: |
1957 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1958 | |
1959 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1960 | synflags = SYNDROME_SRC_WRITTEN; | |
1961 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1962 | } else { | |
1963 | synflags = SYNDROME_SRC_ALL; | |
1964 | txflags = ASYNC_TX_ACK; | |
1965 | } | |
1966 | ||
1967 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1968 | last_stripe = !head_sh->batch_head || |
1969 | list_first_entry(&sh->batch_list, | |
1970 | struct stripe_head, batch_list) == head_sh; | |
1971 | ||
1972 | if (last_stripe) { | |
1973 | atomic_inc(&head_sh->count); | |
584acdd4 | 1974 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1975 | head_sh, to_addr_conv(sh, percpu, j)); |
1976 | } else | |
1977 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1978 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1979 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1980 | if (!last_stripe) { |
1981 | j++; | |
1982 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1983 | batch_list); | |
1984 | goto again; | |
1985 | } | |
91c00924 DW |
1986 | } |
1987 | ||
1988 | static void ops_complete_check(void *stripe_head_ref) | |
1989 | { | |
1990 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1991 | |
e46b272b | 1992 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1993 | (unsigned long long)sh->sector); |
1994 | ||
ecc65c9b | 1995 | sh->check_state = check_state_check_result; |
91c00924 | 1996 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1997 | raid5_release_stripe(sh); |
91c00924 DW |
1998 | } |
1999 | ||
ac6b53b6 | 2000 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 2001 | { |
91c00924 | 2002 | int disks = sh->disks; |
ac6b53b6 DW |
2003 | int pd_idx = sh->pd_idx; |
2004 | int qd_idx = sh->qd_idx; | |
2005 | struct page *xor_dest; | |
46d5b785 | 2006 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 2007 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 2008 | struct async_submit_ctl submit; |
ac6b53b6 DW |
2009 | int count; |
2010 | int i; | |
91c00924 | 2011 | |
e46b272b | 2012 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
2013 | (unsigned long long)sh->sector); |
2014 | ||
59fc630b | 2015 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
2016 | count = 0; |
2017 | xor_dest = sh->dev[pd_idx].page; | |
2018 | xor_srcs[count++] = xor_dest; | |
91c00924 | 2019 | for (i = disks; i--; ) { |
ac6b53b6 DW |
2020 | if (i == pd_idx || i == qd_idx) |
2021 | continue; | |
2022 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
2023 | } |
2024 | ||
d6f38f31 | 2025 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 2026 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 2027 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 2028 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 2029 | |
91c00924 | 2030 | atomic_inc(&sh->count); |
a08abd8c DW |
2031 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
2032 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
2033 | } |
2034 | ||
ac6b53b6 DW |
2035 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
2036 | { | |
46d5b785 | 2037 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
2038 | struct async_submit_ctl submit; |
2039 | int count; | |
2040 | ||
2041 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
2042 | (unsigned long long)sh->sector, checkp); | |
2043 | ||
59fc630b | 2044 | BUG_ON(sh->batch_head); |
584acdd4 | 2045 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
2046 | if (!checkp) |
2047 | srcs[count] = NULL; | |
91c00924 | 2048 | |
91c00924 | 2049 | atomic_inc(&sh->count); |
ac6b53b6 | 2050 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 2051 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
2052 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
2053 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
2054 | } |
2055 | ||
51acbcec | 2056 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
2057 | { |
2058 | int overlap_clear = 0, i, disks = sh->disks; | |
2059 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 2060 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2061 | int level = conf->level; |
d6f38f31 DW |
2062 | struct raid5_percpu *percpu; |
2063 | unsigned long cpu; | |
91c00924 | 2064 | |
d6f38f31 DW |
2065 | cpu = get_cpu(); |
2066 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2067 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2068 | ops_run_biofill(sh); |
2069 | overlap_clear++; | |
2070 | } | |
2071 | ||
7b3a871e | 2072 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2073 | if (level < 6) |
2074 | tx = ops_run_compute5(sh, percpu); | |
2075 | else { | |
2076 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2077 | tx = ops_run_compute6_1(sh, percpu); | |
2078 | else | |
2079 | tx = ops_run_compute6_2(sh, percpu); | |
2080 | } | |
2081 | /* terminate the chain if reconstruct is not set to be run */ | |
2082 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2083 | async_tx_ack(tx); |
2084 | } | |
91c00924 | 2085 | |
584acdd4 MS |
2086 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2087 | if (level < 6) | |
2088 | tx = ops_run_prexor5(sh, percpu, tx); | |
2089 | else | |
2090 | tx = ops_run_prexor6(sh, percpu, tx); | |
2091 | } | |
91c00924 | 2092 | |
ae1713e2 AP |
2093 | if (test_bit(STRIPE_OP_PARTIAL_PARITY, &ops_request)) |
2094 | tx = ops_run_partial_parity(sh, percpu, tx); | |
2095 | ||
600aa109 | 2096 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2097 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2098 | overlap_clear++; |
2099 | } | |
2100 | ||
ac6b53b6 DW |
2101 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2102 | if (level < 6) | |
2103 | ops_run_reconstruct5(sh, percpu, tx); | |
2104 | else | |
2105 | ops_run_reconstruct6(sh, percpu, tx); | |
2106 | } | |
91c00924 | 2107 | |
ac6b53b6 DW |
2108 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2109 | if (sh->check_state == check_state_run) | |
2110 | ops_run_check_p(sh, percpu); | |
2111 | else if (sh->check_state == check_state_run_q) | |
2112 | ops_run_check_pq(sh, percpu, 0); | |
2113 | else if (sh->check_state == check_state_run_pq) | |
2114 | ops_run_check_pq(sh, percpu, 1); | |
2115 | else | |
2116 | BUG(); | |
2117 | } | |
91c00924 | 2118 | |
59fc630b | 2119 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2120 | for (i = disks; i--; ) { |
2121 | struct r5dev *dev = &sh->dev[i]; | |
2122 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2123 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2124 | } | |
d6f38f31 | 2125 | put_cpu(); |
91c00924 DW |
2126 | } |
2127 | ||
845b9e22 AP |
2128 | static void free_stripe(struct kmem_cache *sc, struct stripe_head *sh) |
2129 | { | |
2130 | if (sh->ppl_page) | |
2131 | __free_page(sh->ppl_page); | |
2132 | kmem_cache_free(sc, sh); | |
2133 | } | |
2134 | ||
5f9d1fde | 2135 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
845b9e22 | 2136 | int disks, struct r5conf *conf) |
f18c1a35 N |
2137 | { |
2138 | struct stripe_head *sh; | |
5f9d1fde | 2139 | int i; |
f18c1a35 N |
2140 | |
2141 | sh = kmem_cache_zalloc(sc, gfp); | |
2142 | if (sh) { | |
2143 | spin_lock_init(&sh->stripe_lock); | |
2144 | spin_lock_init(&sh->batch_lock); | |
2145 | INIT_LIST_HEAD(&sh->batch_list); | |
2146 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2147 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2148 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2149 | atomic_set(&sh->count, 1); |
845b9e22 | 2150 | sh->raid_conf = conf; |
a39f7afd | 2151 | sh->log_start = MaxSector; |
5f9d1fde SL |
2152 | for (i = 0; i < disks; i++) { |
2153 | struct r5dev *dev = &sh->dev[i]; | |
2154 | ||
3a83f467 ML |
2155 | bio_init(&dev->req, &dev->vec, 1); |
2156 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2157 | } |
845b9e22 AP |
2158 | |
2159 | if (raid5_has_ppl(conf)) { | |
2160 | sh->ppl_page = alloc_page(gfp); | |
2161 | if (!sh->ppl_page) { | |
2162 | free_stripe(sc, sh); | |
2163 | sh = NULL; | |
2164 | } | |
2165 | } | |
f18c1a35 N |
2166 | } |
2167 | return sh; | |
2168 | } | |
486f0644 | 2169 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2170 | { |
2171 | struct stripe_head *sh; | |
f18c1a35 | 2172 | |
845b9e22 | 2173 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size, conf); |
3f294f4f N |
2174 | if (!sh) |
2175 | return 0; | |
6ce32846 | 2176 | |
a9683a79 | 2177 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2178 | shrink_buffers(sh); |
845b9e22 | 2179 | free_stripe(conf->slab_cache, sh); |
3f294f4f N |
2180 | return 0; |
2181 | } | |
486f0644 N |
2182 | sh->hash_lock_index = |
2183 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2184 | /* we just created an active stripe so... */ |
3f294f4f | 2185 | atomic_inc(&conf->active_stripes); |
59fc630b | 2186 | |
6d036f7d | 2187 | raid5_release_stripe(sh); |
486f0644 | 2188 | conf->max_nr_stripes++; |
3f294f4f N |
2189 | return 1; |
2190 | } | |
2191 | ||
d1688a6d | 2192 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2193 | { |
e18b890b | 2194 | struct kmem_cache *sc; |
53b8d89d | 2195 | size_t namelen = sizeof(conf->cache_name[0]); |
5e5e3e78 | 2196 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2197 | |
f4be6b43 | 2198 | if (conf->mddev->gendisk) |
53b8d89d | 2199 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 N |
2200 | "raid%d-%s", conf->level, mdname(conf->mddev)); |
2201 | else | |
53b8d89d | 2202 | snprintf(conf->cache_name[0], namelen, |
f4be6b43 | 2203 | "raid%d-%p", conf->level, conf->mddev); |
53b8d89d | 2204 | snprintf(conf->cache_name[1], namelen, "%.27s-alt", conf->cache_name[0]); |
f4be6b43 | 2205 | |
ad01c9e3 N |
2206 | conf->active_name = 0; |
2207 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2208 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2209 | 0, 0, NULL); |
1da177e4 LT |
2210 | if (!sc) |
2211 | return 1; | |
2212 | conf->slab_cache = sc; | |
ad01c9e3 | 2213 | conf->pool_size = devs; |
486f0644 N |
2214 | while (num--) |
2215 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2216 | return 1; |
486f0644 | 2217 | |
1da177e4 LT |
2218 | return 0; |
2219 | } | |
29269553 | 2220 | |
d6f38f31 DW |
2221 | /** |
2222 | * scribble_len - return the required size of the scribble region | |
2223 | * @num - total number of disks in the array | |
2224 | * | |
2225 | * The size must be enough to contain: | |
2226 | * 1/ a struct page pointer for each device in the array +2 | |
2227 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2228 | * (dma_map_page()) or page (page_address()) address. | |
2229 | * | |
2230 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2231 | * calculate over all devices (not just the data blocks), using zeros in place | |
2232 | * of the P and Q blocks. | |
2233 | */ | |
b330e6a4 KO |
2234 | static int scribble_alloc(struct raid5_percpu *percpu, |
2235 | int num, int cnt, gfp_t flags) | |
d6f38f31 | 2236 | { |
b330e6a4 KO |
2237 | size_t obj_size = |
2238 | sizeof(struct page *) * (num+2) + | |
2239 | sizeof(addr_conv_t) * (num+2); | |
2240 | void *scribble; | |
d6f38f31 | 2241 | |
b330e6a4 KO |
2242 | scribble = kvmalloc_array(cnt, obj_size, flags); |
2243 | if (!scribble) | |
2244 | return -ENOMEM; | |
2245 | ||
2246 | kvfree(percpu->scribble); | |
2247 | ||
2248 | percpu->scribble = scribble; | |
2249 | percpu->scribble_obj_size = obj_size; | |
2250 | return 0; | |
d6f38f31 DW |
2251 | } |
2252 | ||
738a2738 N |
2253 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2254 | { | |
2255 | unsigned long cpu; | |
2256 | int err = 0; | |
2257 | ||
27a353c0 SL |
2258 | /* |
2259 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2260 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2261 | * should equal to new_disks and new_sectors | |
2262 | */ | |
2263 | if (conf->scribble_disks >= new_disks && | |
2264 | conf->scribble_sectors >= new_sectors) | |
2265 | return 0; | |
738a2738 N |
2266 | mddev_suspend(conf->mddev); |
2267 | get_online_cpus(); | |
b330e6a4 | 2268 | |
738a2738 N |
2269 | for_each_present_cpu(cpu) { |
2270 | struct raid5_percpu *percpu; | |
738a2738 N |
2271 | |
2272 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
b330e6a4 KO |
2273 | err = scribble_alloc(percpu, new_disks, |
2274 | new_sectors / STRIPE_SECTORS, | |
2275 | GFP_NOIO); | |
2276 | if (err) | |
738a2738 | 2277 | break; |
738a2738 | 2278 | } |
b330e6a4 | 2279 | |
738a2738 N |
2280 | put_online_cpus(); |
2281 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2282 | if (!err) { |
2283 | conf->scribble_disks = new_disks; | |
2284 | conf->scribble_sectors = new_sectors; | |
2285 | } | |
738a2738 N |
2286 | return err; |
2287 | } | |
2288 | ||
d1688a6d | 2289 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2290 | { |
2291 | /* Make all the stripes able to hold 'newsize' devices. | |
2292 | * New slots in each stripe get 'page' set to a new page. | |
2293 | * | |
2294 | * This happens in stages: | |
2295 | * 1/ create a new kmem_cache and allocate the required number of | |
2296 | * stripe_heads. | |
83f0d77a | 2297 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2298 | * to the new stripe_heads. This will have the side effect of |
2299 | * freezing the array as once all stripe_heads have been collected, | |
2300 | * no IO will be possible. Old stripe heads are freed once their | |
2301 | * pages have been transferred over, and the old kmem_cache is | |
2302 | * freed when all stripes are done. | |
2303 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
3560741e | 2304 | * we simple return a failure status - no need to clean anything up. |
ad01c9e3 N |
2305 | * 4/ allocate new pages for the new slots in the new stripe_heads. |
2306 | * If this fails, we don't bother trying the shrink the | |
2307 | * stripe_heads down again, we just leave them as they are. | |
2308 | * As each stripe_head is processed the new one is released into | |
2309 | * active service. | |
2310 | * | |
2311 | * Once step2 is started, we cannot afford to wait for a write, | |
2312 | * so we use GFP_NOIO allocations. | |
2313 | */ | |
2314 | struct stripe_head *osh, *nsh; | |
2315 | LIST_HEAD(newstripes); | |
2316 | struct disk_info *ndisks; | |
2214c260 | 2317 | int err = 0; |
e18b890b | 2318 | struct kmem_cache *sc; |
ad01c9e3 | 2319 | int i; |
566c09c5 | 2320 | int hash, cnt; |
ad01c9e3 | 2321 | |
2214c260 | 2322 | md_allow_write(conf->mddev); |
2a2275d6 | 2323 | |
ad01c9e3 N |
2324 | /* Step 1 */ |
2325 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2326 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2327 | 0, 0, NULL); |
ad01c9e3 N |
2328 | if (!sc) |
2329 | return -ENOMEM; | |
2330 | ||
2d5b569b N |
2331 | /* Need to ensure auto-resizing doesn't interfere */ |
2332 | mutex_lock(&conf->cache_size_mutex); | |
2333 | ||
ad01c9e3 | 2334 | for (i = conf->max_nr_stripes; i; i--) { |
845b9e22 | 2335 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize, conf); |
ad01c9e3 N |
2336 | if (!nsh) |
2337 | break; | |
2338 | ||
ad01c9e3 N |
2339 | list_add(&nsh->lru, &newstripes); |
2340 | } | |
2341 | if (i) { | |
2342 | /* didn't get enough, give up */ | |
2343 | while (!list_empty(&newstripes)) { | |
2344 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2345 | list_del(&nsh->lru); | |
845b9e22 | 2346 | free_stripe(sc, nsh); |
ad01c9e3 N |
2347 | } |
2348 | kmem_cache_destroy(sc); | |
2d5b569b | 2349 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2350 | return -ENOMEM; |
2351 | } | |
2352 | /* Step 2 - Must use GFP_NOIO now. | |
2353 | * OK, we have enough stripes, start collecting inactive | |
2354 | * stripes and copying them over | |
2355 | */ | |
566c09c5 SL |
2356 | hash = 0; |
2357 | cnt = 0; | |
ad01c9e3 | 2358 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2359 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2360 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2361 | !list_empty(conf->inactive_list + hash), |
2362 | unlock_device_hash_lock(conf, hash), | |
2363 | lock_device_hash_lock(conf, hash)); | |
2364 | osh = get_free_stripe(conf, hash); | |
2365 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2366 | |
d592a996 | 2367 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2368 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2369 | nsh->dev[i].orig_page = osh->dev[i].page; |
2370 | } | |
566c09c5 | 2371 | nsh->hash_lock_index = hash; |
845b9e22 | 2372 | free_stripe(conf->slab_cache, osh); |
566c09c5 SL |
2373 | cnt++; |
2374 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2375 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2376 | hash++; | |
2377 | cnt = 0; | |
2378 | } | |
ad01c9e3 N |
2379 | } |
2380 | kmem_cache_destroy(conf->slab_cache); | |
2381 | ||
2382 | /* Step 3. | |
2383 | * At this point, we are holding all the stripes so the array | |
2384 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2385 | * conf->disks and the scribble region |
ad01c9e3 | 2386 | */ |
6396bb22 | 2387 | ndisks = kcalloc(newsize, sizeof(struct disk_info), GFP_NOIO); |
ad01c9e3 | 2388 | if (ndisks) { |
d7bd398e | 2389 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2390 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2391 | |
2392 | for (i = conf->pool_size; i < newsize; i++) { | |
2393 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2394 | if (!ndisks[i].extra_page) | |
2395 | err = -ENOMEM; | |
2396 | } | |
2397 | ||
2398 | if (err) { | |
2399 | for (i = conf->pool_size; i < newsize; i++) | |
2400 | if (ndisks[i].extra_page) | |
2401 | put_page(ndisks[i].extra_page); | |
2402 | kfree(ndisks); | |
2403 | } else { | |
2404 | kfree(conf->disks); | |
2405 | conf->disks = ndisks; | |
2406 | } | |
ad01c9e3 N |
2407 | } else |
2408 | err = -ENOMEM; | |
2409 | ||
2d5b569b | 2410 | mutex_unlock(&conf->cache_size_mutex); |
583da48e DY |
2411 | |
2412 | conf->slab_cache = sc; | |
2413 | conf->active_name = 1-conf->active_name; | |
2414 | ||
ad01c9e3 N |
2415 | /* Step 4, return new stripes to service */ |
2416 | while(!list_empty(&newstripes)) { | |
2417 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2418 | list_del_init(&nsh->lru); | |
d6f38f31 | 2419 | |
ad01c9e3 N |
2420 | for (i=conf->raid_disks; i < newsize; i++) |
2421 | if (nsh->dev[i].page == NULL) { | |
2422 | struct page *p = alloc_page(GFP_NOIO); | |
2423 | nsh->dev[i].page = p; | |
d592a996 | 2424 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2425 | if (!p) |
2426 | err = -ENOMEM; | |
2427 | } | |
6d036f7d | 2428 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2429 | } |
2430 | /* critical section pass, GFP_NOIO no longer needed */ | |
2431 | ||
6e9eac2d N |
2432 | if (!err) |
2433 | conf->pool_size = newsize; | |
ad01c9e3 N |
2434 | return err; |
2435 | } | |
1da177e4 | 2436 | |
486f0644 | 2437 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2438 | { |
2439 | struct stripe_head *sh; | |
49895bcc | 2440 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2441 | |
566c09c5 SL |
2442 | spin_lock_irq(conf->hash_locks + hash); |
2443 | sh = get_free_stripe(conf, hash); | |
2444 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2445 | if (!sh) |
2446 | return 0; | |
78bafebd | 2447 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2448 | shrink_buffers(sh); |
845b9e22 | 2449 | free_stripe(conf->slab_cache, sh); |
3f294f4f | 2450 | atomic_dec(&conf->active_stripes); |
486f0644 | 2451 | conf->max_nr_stripes--; |
3f294f4f N |
2452 | return 1; |
2453 | } | |
2454 | ||
d1688a6d | 2455 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2456 | { |
486f0644 N |
2457 | while (conf->max_nr_stripes && |
2458 | drop_one_stripe(conf)) | |
2459 | ; | |
3f294f4f | 2460 | |
644df1a8 | 2461 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2462 | conf->slab_cache = NULL; |
2463 | } | |
2464 | ||
4246a0b6 | 2465 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2466 | { |
99c0fb5f | 2467 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2468 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2469 | int disks = sh->disks, i; |
d6950432 | 2470 | char b[BDEVNAME_SIZE]; |
dd054fce | 2471 | struct md_rdev *rdev = NULL; |
05616be5 | 2472 | sector_t s; |
1da177e4 LT |
2473 | |
2474 | for (i=0 ; i<disks; i++) | |
2475 | if (bi == &sh->dev[i].req) | |
2476 | break; | |
2477 | ||
4246a0b6 | 2478 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2479 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2480 | bi->bi_status); |
1da177e4 | 2481 | if (i == disks) { |
5f9d1fde | 2482 | bio_reset(bi); |
1da177e4 | 2483 | BUG(); |
6712ecf8 | 2484 | return; |
1da177e4 | 2485 | } |
14a75d3e | 2486 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2487 | /* If replacement finished while this request was outstanding, |
2488 | * 'replacement' might be NULL already. | |
2489 | * In that case it moved down to 'rdev'. | |
2490 | * rdev is not removed until all requests are finished. | |
2491 | */ | |
14a75d3e | 2492 | rdev = conf->disks[i].replacement; |
dd054fce | 2493 | if (!rdev) |
14a75d3e | 2494 | rdev = conf->disks[i].rdev; |
1da177e4 | 2495 | |
05616be5 N |
2496 | if (use_new_offset(conf, sh)) |
2497 | s = sh->sector + rdev->new_data_offset; | |
2498 | else | |
2499 | s = sh->sector + rdev->data_offset; | |
4e4cbee9 | 2500 | if (!bi->bi_status) { |
1da177e4 | 2501 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2502 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2503 | /* Note that this cannot happen on a |
2504 | * replacement device. We just fail those on | |
2505 | * any error | |
2506 | */ | |
cc6167b4 N |
2507 | pr_info_ratelimited( |
2508 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
8bda470e | 2509 | mdname(conf->mddev), STRIPE_SECTORS, |
05616be5 | 2510 | (unsigned long long)s, |
8bda470e | 2511 | bdevname(rdev->bdev, b)); |
ddd5115f | 2512 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2513 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2514 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2515 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2516 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2517 | ||
86aa1397 SL |
2518 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2519 | /* | |
2520 | * end read for a page in journal, this | |
2521 | * must be preparing for prexor in rmw | |
2522 | */ | |
2523 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2524 | ||
14a75d3e N |
2525 | if (atomic_read(&rdev->read_errors)) |
2526 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2527 | } else { |
14a75d3e | 2528 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2529 | int retry = 0; |
2e8ac303 | 2530 | int set_bad = 0; |
d6950432 | 2531 | |
1da177e4 | 2532 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2533 | atomic_inc(&rdev->read_errors); |
14a75d3e | 2534 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2535 | pr_warn_ratelimited( |
2536 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2537 | mdname(conf->mddev), |
05616be5 | 2538 | (unsigned long long)s, |
14a75d3e | 2539 | bdn); |
2e8ac303 | 2540 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2541 | set_bad = 1; | |
cc6167b4 N |
2542 | pr_warn_ratelimited( |
2543 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2544 | mdname(conf->mddev), |
05616be5 | 2545 | (unsigned long long)s, |
8bda470e | 2546 | bdn); |
2e8ac303 | 2547 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2548 | /* Oh, no!!! */ |
2e8ac303 | 2549 | set_bad = 1; |
cc6167b4 N |
2550 | pr_warn_ratelimited( |
2551 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2552 | mdname(conf->mddev), |
05616be5 | 2553 | (unsigned long long)s, |
8bda470e | 2554 | bdn); |
2e8ac303 | 2555 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2556 | > conf->max_nr_stripes) |
cc6167b4 | 2557 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2558 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2559 | else |
2560 | retry = 1; | |
edfa1f65 BY |
2561 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2562 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2563 | retry = 1; | |
ba22dcbf | 2564 | if (retry) |
3f9e7c14 | 2565 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2566 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2567 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2568 | } else | |
2569 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2570 | else { |
4e5314b5 N |
2571 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2572 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2573 | if (!(set_bad |
2574 | && test_bit(In_sync, &rdev->flags) | |
2575 | && rdev_set_badblocks( | |
2576 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2577 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2578 | } |
1da177e4 | 2579 | } |
14a75d3e | 2580 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2581 | bio_reset(bi); |
1da177e4 LT |
2582 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2583 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2584 | raid5_release_stripe(sh); |
1da177e4 LT |
2585 | } |
2586 | ||
4246a0b6 | 2587 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2588 | { |
99c0fb5f | 2589 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2590 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2591 | int disks = sh->disks, i; |
977df362 | 2592 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2593 | sector_t first_bad; |
2594 | int bad_sectors; | |
977df362 | 2595 | int replacement = 0; |
1da177e4 | 2596 | |
977df362 N |
2597 | for (i = 0 ; i < disks; i++) { |
2598 | if (bi == &sh->dev[i].req) { | |
2599 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2600 | break; |
977df362 N |
2601 | } |
2602 | if (bi == &sh->dev[i].rreq) { | |
2603 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2604 | if (rdev) |
2605 | replacement = 1; | |
2606 | else | |
2607 | /* rdev was removed and 'replacement' | |
2608 | * replaced it. rdev is not removed | |
2609 | * until all requests are finished. | |
2610 | */ | |
2611 | rdev = conf->disks[i].rdev; | |
977df362 N |
2612 | break; |
2613 | } | |
2614 | } | |
4246a0b6 | 2615 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2616 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4e4cbee9 | 2617 | bi->bi_status); |
1da177e4 | 2618 | if (i == disks) { |
5f9d1fde | 2619 | bio_reset(bi); |
1da177e4 | 2620 | BUG(); |
6712ecf8 | 2621 | return; |
1da177e4 LT |
2622 | } |
2623 | ||
977df362 | 2624 | if (replacement) { |
4e4cbee9 | 2625 | if (bi->bi_status) |
977df362 N |
2626 | md_error(conf->mddev, rdev); |
2627 | else if (is_badblock(rdev, sh->sector, | |
2628 | STRIPE_SECTORS, | |
2629 | &first_bad, &bad_sectors)) | |
2630 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2631 | } else { | |
4e4cbee9 | 2632 | if (bi->bi_status) { |
9f97e4b1 | 2633 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2634 | set_bit(WriteErrorSeen, &rdev->flags); |
2635 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2636 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2637 | set_bit(MD_RECOVERY_NEEDED, | |
2638 | &rdev->mddev->recovery); | |
977df362 N |
2639 | } else if (is_badblock(rdev, sh->sector, |
2640 | STRIPE_SECTORS, | |
c0b32972 | 2641 | &first_bad, &bad_sectors)) { |
977df362 | 2642 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2643 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2644 | /* That was a successful write so make | |
2645 | * sure it looks like we already did | |
2646 | * a re-write. | |
2647 | */ | |
2648 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2649 | } | |
977df362 N |
2650 | } |
2651 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2652 | |
4e4cbee9 | 2653 | if (sh->batch_head && bi->bi_status && !replacement) |
72ac7330 | 2654 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2655 | ||
c9445555 | 2656 | bio_reset(bi); |
977df362 N |
2657 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2658 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2659 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2660 | raid5_release_stripe(sh); |
59fc630b | 2661 | |
2662 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2663 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2664 | } |
2665 | ||
849674e4 | 2666 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2667 | { |
2668 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2669 | struct r5conf *conf = mddev->private; |
908f4fbd | 2670 | unsigned long flags; |
0c55e022 | 2671 | pr_debug("raid456: error called\n"); |
1da177e4 | 2672 | |
908f4fbd | 2673 | spin_lock_irqsave(&conf->device_lock, flags); |
fb73b357 MT |
2674 | |
2675 | if (test_bit(In_sync, &rdev->flags) && | |
2676 | mddev->degraded == conf->max_degraded) { | |
2677 | /* | |
2678 | * Don't allow to achieve failed state | |
2679 | * Don't try to recover this device | |
2680 | */ | |
2681 | conf->recovery_disabled = mddev->recovery_disabled; | |
2682 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2683 | return; | |
2684 | } | |
2685 | ||
aff69d89 | 2686 | set_bit(Faulty, &rdev->flags); |
908f4fbd | 2687 | clear_bit(In_sync, &rdev->flags); |
2e38a37f | 2688 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2689 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2690 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2691 | ||
de393cde | 2692 | set_bit(Blocked, &rdev->flags); |
2953079c SL |
2693 | set_mask_bits(&mddev->sb_flags, 0, |
2694 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2695 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2696 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2697 | mdname(mddev), | |
2698 | bdevname(rdev->bdev, b), | |
2699 | mdname(mddev), | |
2700 | conf->raid_disks - mddev->degraded); | |
70d466f7 | 2701 | r5c_update_on_rdev_error(mddev, rdev); |
16a53ecc | 2702 | } |
1da177e4 LT |
2703 | |
2704 | /* | |
2705 | * Input: a 'big' sector number, | |
2706 | * Output: index of the data and parity disk, and the sector # in them. | |
2707 | */ | |
6d036f7d SL |
2708 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2709 | int previous, int *dd_idx, | |
2710 | struct stripe_head *sh) | |
1da177e4 | 2711 | { |
6e3b96ed | 2712 | sector_t stripe, stripe2; |
35f2a591 | 2713 | sector_t chunk_number; |
1da177e4 | 2714 | unsigned int chunk_offset; |
911d4ee8 | 2715 | int pd_idx, qd_idx; |
67cc2b81 | 2716 | int ddf_layout = 0; |
1da177e4 | 2717 | sector_t new_sector; |
e183eaed N |
2718 | int algorithm = previous ? conf->prev_algo |
2719 | : conf->algorithm; | |
09c9e5fa AN |
2720 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2721 | : conf->chunk_sectors; | |
112bf897 N |
2722 | int raid_disks = previous ? conf->previous_raid_disks |
2723 | : conf->raid_disks; | |
2724 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2725 | |
2726 | /* First compute the information on this sector */ | |
2727 | ||
2728 | /* | |
2729 | * Compute the chunk number and the sector offset inside the chunk | |
2730 | */ | |
2731 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2732 | chunk_number = r_sector; | |
1da177e4 LT |
2733 | |
2734 | /* | |
2735 | * Compute the stripe number | |
2736 | */ | |
35f2a591 N |
2737 | stripe = chunk_number; |
2738 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2739 | stripe2 = stripe; |
1da177e4 LT |
2740 | /* |
2741 | * Select the parity disk based on the user selected algorithm. | |
2742 | */ | |
84789554 | 2743 | pd_idx = qd_idx = -1; |
16a53ecc N |
2744 | switch(conf->level) { |
2745 | case 4: | |
911d4ee8 | 2746 | pd_idx = data_disks; |
16a53ecc N |
2747 | break; |
2748 | case 5: | |
e183eaed | 2749 | switch (algorithm) { |
1da177e4 | 2750 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2751 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2752 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2753 | (*dd_idx)++; |
2754 | break; | |
2755 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2756 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2757 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2758 | (*dd_idx)++; |
2759 | break; | |
2760 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2761 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2762 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2763 | break; |
2764 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2765 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2766 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2767 | break; |
99c0fb5f N |
2768 | case ALGORITHM_PARITY_0: |
2769 | pd_idx = 0; | |
2770 | (*dd_idx)++; | |
2771 | break; | |
2772 | case ALGORITHM_PARITY_N: | |
2773 | pd_idx = data_disks; | |
2774 | break; | |
1da177e4 | 2775 | default: |
99c0fb5f | 2776 | BUG(); |
16a53ecc N |
2777 | } |
2778 | break; | |
2779 | case 6: | |
2780 | ||
e183eaed | 2781 | switch (algorithm) { |
16a53ecc | 2782 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2783 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2784 | qd_idx = pd_idx + 1; |
2785 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2786 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2787 | qd_idx = 0; |
2788 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2789 | (*dd_idx) += 2; /* D D P Q D */ |
2790 | break; | |
2791 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2792 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2793 | qd_idx = pd_idx + 1; |
2794 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2795 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2796 | qd_idx = 0; |
2797 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2798 | (*dd_idx) += 2; /* D D P Q D */ |
2799 | break; | |
2800 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2801 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2802 | qd_idx = (pd_idx + 1) % raid_disks; |
2803 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2804 | break; |
2805 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2806 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2807 | qd_idx = (pd_idx + 1) % raid_disks; |
2808 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2809 | break; |
99c0fb5f N |
2810 | |
2811 | case ALGORITHM_PARITY_0: | |
2812 | pd_idx = 0; | |
2813 | qd_idx = 1; | |
2814 | (*dd_idx) += 2; | |
2815 | break; | |
2816 | case ALGORITHM_PARITY_N: | |
2817 | pd_idx = data_disks; | |
2818 | qd_idx = data_disks + 1; | |
2819 | break; | |
2820 | ||
2821 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2822 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2823 | * of blocks for computing Q is different. | |
2824 | */ | |
6e3b96ed | 2825 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2826 | qd_idx = pd_idx + 1; |
2827 | if (pd_idx == raid_disks-1) { | |
2828 | (*dd_idx)++; /* Q D D D P */ | |
2829 | qd_idx = 0; | |
2830 | } else if (*dd_idx >= pd_idx) | |
2831 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2832 | ddf_layout = 1; |
99c0fb5f N |
2833 | break; |
2834 | ||
2835 | case ALGORITHM_ROTATING_N_RESTART: | |
2836 | /* Same a left_asymmetric, by first stripe is | |
2837 | * D D D P Q rather than | |
2838 | * Q D D D P | |
2839 | */ | |
6e3b96ed N |
2840 | stripe2 += 1; |
2841 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2842 | qd_idx = pd_idx + 1; |
2843 | if (pd_idx == raid_disks-1) { | |
2844 | (*dd_idx)++; /* Q D D D P */ | |
2845 | qd_idx = 0; | |
2846 | } else if (*dd_idx >= pd_idx) | |
2847 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2848 | ddf_layout = 1; |
99c0fb5f N |
2849 | break; |
2850 | ||
2851 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2852 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2853 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2854 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2855 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2856 | ddf_layout = 1; |
99c0fb5f N |
2857 | break; |
2858 | ||
2859 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2860 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2861 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2862 | if (*dd_idx >= pd_idx) |
2863 | (*dd_idx)++; | |
2864 | qd_idx = raid_disks - 1; | |
2865 | break; | |
2866 | ||
2867 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2868 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2869 | if (*dd_idx >= pd_idx) |
2870 | (*dd_idx)++; | |
2871 | qd_idx = raid_disks - 1; | |
2872 | break; | |
2873 | ||
2874 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2875 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2876 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2877 | qd_idx = raid_disks - 1; | |
2878 | break; | |
2879 | ||
2880 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2881 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2882 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2883 | qd_idx = raid_disks - 1; | |
2884 | break; | |
2885 | ||
2886 | case ALGORITHM_PARITY_0_6: | |
2887 | pd_idx = 0; | |
2888 | (*dd_idx)++; | |
2889 | qd_idx = raid_disks - 1; | |
2890 | break; | |
2891 | ||
16a53ecc | 2892 | default: |
99c0fb5f | 2893 | BUG(); |
16a53ecc N |
2894 | } |
2895 | break; | |
1da177e4 LT |
2896 | } |
2897 | ||
911d4ee8 N |
2898 | if (sh) { |
2899 | sh->pd_idx = pd_idx; | |
2900 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2901 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2902 | } |
1da177e4 LT |
2903 | /* |
2904 | * Finally, compute the new sector number | |
2905 | */ | |
2906 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2907 | return new_sector; | |
2908 | } | |
2909 | ||
6d036f7d | 2910 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2911 | { |
d1688a6d | 2912 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2913 | int raid_disks = sh->disks; |
2914 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2915 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2916 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2917 | : conf->chunk_sectors; | |
e183eaed N |
2918 | int algorithm = previous ? conf->prev_algo |
2919 | : conf->algorithm; | |
1da177e4 LT |
2920 | sector_t stripe; |
2921 | int chunk_offset; | |
35f2a591 N |
2922 | sector_t chunk_number; |
2923 | int dummy1, dd_idx = i; | |
1da177e4 | 2924 | sector_t r_sector; |
911d4ee8 | 2925 | struct stripe_head sh2; |
1da177e4 LT |
2926 | |
2927 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2928 | stripe = new_sector; | |
1da177e4 | 2929 | |
16a53ecc N |
2930 | if (i == sh->pd_idx) |
2931 | return 0; | |
2932 | switch(conf->level) { | |
2933 | case 4: break; | |
2934 | case 5: | |
e183eaed | 2935 | switch (algorithm) { |
1da177e4 LT |
2936 | case ALGORITHM_LEFT_ASYMMETRIC: |
2937 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2938 | if (i > sh->pd_idx) | |
2939 | i--; | |
2940 | break; | |
2941 | case ALGORITHM_LEFT_SYMMETRIC: | |
2942 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2943 | if (i < sh->pd_idx) | |
2944 | i += raid_disks; | |
2945 | i -= (sh->pd_idx + 1); | |
2946 | break; | |
99c0fb5f N |
2947 | case ALGORITHM_PARITY_0: |
2948 | i -= 1; | |
2949 | break; | |
2950 | case ALGORITHM_PARITY_N: | |
2951 | break; | |
1da177e4 | 2952 | default: |
99c0fb5f | 2953 | BUG(); |
16a53ecc N |
2954 | } |
2955 | break; | |
2956 | case 6: | |
d0dabf7e | 2957 | if (i == sh->qd_idx) |
16a53ecc | 2958 | return 0; /* It is the Q disk */ |
e183eaed | 2959 | switch (algorithm) { |
16a53ecc N |
2960 | case ALGORITHM_LEFT_ASYMMETRIC: |
2961 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2962 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2963 | case ALGORITHM_ROTATING_N_RESTART: | |
2964 | if (sh->pd_idx == raid_disks-1) | |
2965 | i--; /* Q D D D P */ | |
16a53ecc N |
2966 | else if (i > sh->pd_idx) |
2967 | i -= 2; /* D D P Q D */ | |
2968 | break; | |
2969 | case ALGORITHM_LEFT_SYMMETRIC: | |
2970 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2971 | if (sh->pd_idx == raid_disks-1) | |
2972 | i--; /* Q D D D P */ | |
2973 | else { | |
2974 | /* D D P Q D */ | |
2975 | if (i < sh->pd_idx) | |
2976 | i += raid_disks; | |
2977 | i -= (sh->pd_idx + 2); | |
2978 | } | |
2979 | break; | |
99c0fb5f N |
2980 | case ALGORITHM_PARITY_0: |
2981 | i -= 2; | |
2982 | break; | |
2983 | case ALGORITHM_PARITY_N: | |
2984 | break; | |
2985 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2986 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2987 | if (sh->pd_idx == 0) |
2988 | i--; /* P D D D Q */ | |
e4424fee N |
2989 | else { |
2990 | /* D D Q P D */ | |
2991 | if (i < sh->pd_idx) | |
2992 | i += raid_disks; | |
2993 | i -= (sh->pd_idx + 1); | |
2994 | } | |
99c0fb5f N |
2995 | break; |
2996 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2997 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2998 | if (i > sh->pd_idx) | |
2999 | i--; | |
3000 | break; | |
3001 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
3002 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
3003 | if (i < sh->pd_idx) | |
3004 | i += data_disks + 1; | |
3005 | i -= (sh->pd_idx + 1); | |
3006 | break; | |
3007 | case ALGORITHM_PARITY_0_6: | |
3008 | i -= 1; | |
3009 | break; | |
16a53ecc | 3010 | default: |
99c0fb5f | 3011 | BUG(); |
16a53ecc N |
3012 | } |
3013 | break; | |
1da177e4 LT |
3014 | } |
3015 | ||
3016 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 3017 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 3018 | |
112bf897 | 3019 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 3020 | previous, &dummy1, &sh2); |
911d4ee8 N |
3021 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
3022 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
3023 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
3024 | mdname(conf->mddev)); | |
1da177e4 LT |
3025 | return 0; |
3026 | } | |
3027 | return r_sector; | |
3028 | } | |
3029 | ||
07e83364 SL |
3030 | /* |
3031 | * There are cases where we want handle_stripe_dirtying() and | |
3032 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
3033 | * | |
3034 | * This function checks whether we want to delay the towrite. Specifically, | |
3035 | * we delay the towrite when: | |
3036 | * | |
3037 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
3038 | * stripe has data in journal (for other devices). | |
3039 | * | |
3040 | * In this case, when reading data for the non-overwrite dev, it is | |
3041 | * necessary to handle complex rmw of write back cache (prexor with | |
3042 | * orig_page, and xor with page). To keep read path simple, we would | |
3043 | * like to flush data in journal to RAID disks first, so complex rmw | |
3044 | * is handled in the write patch (handle_stripe_dirtying). | |
3045 | * | |
39b99586 SL |
3046 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
3047 | * | |
3048 | * It is important to be able to flush all stripes in raid5-cache. | |
3049 | * Therefore, we need reserve some space on the journal device for | |
3050 | * these flushes. If flush operation includes pending writes to the | |
3051 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
3052 | * for the flush out. If we exclude these pending writes from flush | |
3053 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
3054 | * Therefore, excluding pending writes in these cases enables more | |
3055 | * efficient use of the journal device. | |
3056 | * | |
3057 | * Note: To make sure the stripe makes progress, we only delay | |
3058 | * towrite for stripes with data already in journal (injournal > 0). | |
3059 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
3060 | * no_space_stripes list. | |
3061 | * | |
70d466f7 SL |
3062 | * 3. during journal failure |
3063 | * In journal failure, we try to flush all cached data to raid disks | |
3064 | * based on data in stripe cache. The array is read-only to upper | |
3065 | * layers, so we would skip all pending writes. | |
3066 | * | |
07e83364 | 3067 | */ |
39b99586 SL |
3068 | static inline bool delay_towrite(struct r5conf *conf, |
3069 | struct r5dev *dev, | |
3070 | struct stripe_head_state *s) | |
07e83364 | 3071 | { |
39b99586 SL |
3072 | /* case 1 above */ |
3073 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
3074 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
3075 | return true; | |
3076 | /* case 2 above */ | |
3077 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
3078 | s->injournal > 0) | |
3079 | return true; | |
70d466f7 SL |
3080 | /* case 3 above */ |
3081 | if (s->log_failed && s->injournal) | |
3082 | return true; | |
39b99586 | 3083 | return false; |
07e83364 SL |
3084 | } |
3085 | ||
600aa109 | 3086 | static void |
c0f7bddb | 3087 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3088 | int rcw, int expand) |
e33129d8 | 3089 | { |
584acdd4 | 3090 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3091 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3092 | int level = conf->level; |
e33129d8 DW |
3093 | |
3094 | if (rcw) { | |
1e6d690b SL |
3095 | /* |
3096 | * In some cases, handle_stripe_dirtying initially decided to | |
3097 | * run rmw and allocates extra page for prexor. However, rcw is | |
3098 | * cheaper later on. We need to free the extra page now, | |
3099 | * because we won't be able to do that in ops_complete_prexor(). | |
3100 | */ | |
3101 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3102 | |
3103 | for (i = disks; i--; ) { | |
3104 | struct r5dev *dev = &sh->dev[i]; | |
3105 | ||
39b99586 | 3106 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3107 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3108 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3109 | if (!expand) |
3110 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3111 | s->locked++; |
1e6d690b SL |
3112 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3113 | set_bit(R5_LOCKED, &dev->flags); | |
3114 | s->locked++; | |
e33129d8 DW |
3115 | } |
3116 | } | |
ce7d363a N |
3117 | /* if we are not expanding this is a proper write request, and |
3118 | * there will be bios with new data to be drained into the | |
3119 | * stripe cache | |
3120 | */ | |
3121 | if (!expand) { | |
3122 | if (!s->locked) | |
3123 | /* False alarm, nothing to do */ | |
3124 | return; | |
3125 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3126 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3127 | } else | |
3128 | sh->reconstruct_state = reconstruct_state_run; | |
3129 | ||
3130 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3131 | ||
c0f7bddb | 3132 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3133 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3134 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3135 | } else { |
3136 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3137 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3138 | BUG_ON(level == 6 && |
3139 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3140 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3141 | |
e33129d8 DW |
3142 | for (i = disks; i--; ) { |
3143 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3144 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3145 | continue; |
3146 | ||
e33129d8 DW |
3147 | if (dev->towrite && |
3148 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3149 | test_bit(R5_Wantcompute, &dev->flags))) { |
3150 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3151 | set_bit(R5_LOCKED, &dev->flags); |
3152 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3153 | s->locked++; |
1e6d690b SL |
3154 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3155 | set_bit(R5_LOCKED, &dev->flags); | |
3156 | s->locked++; | |
e33129d8 DW |
3157 | } |
3158 | } | |
ce7d363a N |
3159 | if (!s->locked) |
3160 | /* False alarm - nothing to do */ | |
3161 | return; | |
3162 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3163 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3164 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3165 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3166 | } |
3167 | ||
c0f7bddb | 3168 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3169 | * are in flight |
3170 | */ | |
3171 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3172 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3173 | s->locked++; |
e33129d8 | 3174 | |
c0f7bddb YT |
3175 | if (level == 6) { |
3176 | int qd_idx = sh->qd_idx; | |
3177 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3178 | ||
3179 | set_bit(R5_LOCKED, &dev->flags); | |
3180 | clear_bit(R5_UPTODATE, &dev->flags); | |
3181 | s->locked++; | |
3182 | } | |
3183 | ||
845b9e22 | 3184 | if (raid5_has_ppl(sh->raid_conf) && sh->ppl_page && |
3418d036 AP |
3185 | test_bit(STRIPE_OP_BIODRAIN, &s->ops_request) && |
3186 | !test_bit(STRIPE_FULL_WRITE, &sh->state) && | |
3187 | test_bit(R5_Insync, &sh->dev[pd_idx].flags)) | |
3188 | set_bit(STRIPE_OP_PARTIAL_PARITY, &s->ops_request); | |
3189 | ||
600aa109 | 3190 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3191 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3192 | s->locked, s->ops_request); |
e33129d8 | 3193 | } |
16a53ecc | 3194 | |
1da177e4 LT |
3195 | /* |
3196 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3197 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3198 | * The bi_next chain must be in order. |
3199 | */ | |
da41ba65 | 3200 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3201 | int forwrite, int previous) | |
1da177e4 LT |
3202 | { |
3203 | struct bio **bip; | |
d1688a6d | 3204 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3205 | int firstwrite=0; |
1da177e4 | 3206 | |
cbe47ec5 | 3207 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3208 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3209 | (unsigned long long)sh->sector); |
3210 | ||
b17459c0 | 3211 | spin_lock_irq(&sh->stripe_lock); |
2cd259a7 | 3212 | sh->dev[dd_idx].write_hint = bi->bi_write_hint; |
59fc630b | 3213 | /* Don't allow new IO added to stripes in batch list */ |
3214 | if (sh->batch_head) | |
3215 | goto overlap; | |
72626685 | 3216 | if (forwrite) { |
1da177e4 | 3217 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3218 | if (*bip == NULL) |
72626685 N |
3219 | firstwrite = 1; |
3220 | } else | |
1da177e4 | 3221 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3222 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3223 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3224 | goto overlap; |
3225 | bip = & (*bip)->bi_next; | |
3226 | } | |
4f024f37 | 3227 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3228 | goto overlap; |
3229 | ||
3418d036 AP |
3230 | if (forwrite && raid5_has_ppl(conf)) { |
3231 | /* | |
3232 | * With PPL only writes to consecutive data chunks within a | |
3233 | * stripe are allowed because for a single stripe_head we can | |
3234 | * only have one PPL entry at a time, which describes one data | |
3235 | * range. Not really an overlap, but wait_for_overlap can be | |
3236 | * used to handle this. | |
3237 | */ | |
3238 | sector_t sector; | |
3239 | sector_t first = 0; | |
3240 | sector_t last = 0; | |
3241 | int count = 0; | |
3242 | int i; | |
3243 | ||
3244 | for (i = 0; i < sh->disks; i++) { | |
3245 | if (i != sh->pd_idx && | |
3246 | (i == dd_idx || sh->dev[i].towrite)) { | |
3247 | sector = sh->dev[i].sector; | |
3248 | if (count == 0 || sector < first) | |
3249 | first = sector; | |
3250 | if (sector > last) | |
3251 | last = sector; | |
3252 | count++; | |
3253 | } | |
3254 | } | |
3255 | ||
3256 | if (first + conf->chunk_sectors * (count - 1) != last) | |
3257 | goto overlap; | |
3258 | } | |
3259 | ||
da41ba65 | 3260 | if (!forwrite || previous) |
3261 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3262 | ||
78bafebd | 3263 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3264 | if (*bip) |
3265 | bi->bi_next = *bip; | |
3266 | *bip = bi; | |
016c76ac | 3267 | bio_inc_remaining(bi); |
49728050 | 3268 | md_write_inc(conf->mddev, bi); |
72626685 | 3269 | |
1da177e4 LT |
3270 | if (forwrite) { |
3271 | /* check if page is covered */ | |
3272 | sector_t sector = sh->dev[dd_idx].sector; | |
3273 | for (bi=sh->dev[dd_idx].towrite; | |
3274 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3275 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3276 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3277 | if (bio_end_sector(bi) >= sector) |
3278 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3279 | } |
3280 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3281 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3282 | sh->overwrite_disks++; | |
1da177e4 | 3283 | } |
cbe47ec5 N |
3284 | |
3285 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3286 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3287 | (unsigned long long)sh->sector, dd_idx); |
3288 | ||
3289 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3290 | /* Cannot hold spinlock over bitmap_startwrite, |
3291 | * but must ensure this isn't added to a batch until | |
3292 | * we have added to the bitmap and set bm_seq. | |
3293 | * So set STRIPE_BITMAP_PENDING to prevent | |
3294 | * batching. | |
3295 | * If multiple add_stripe_bio() calls race here they | |
3296 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3297 | * to complete "bitmap_startwrite" gets to set | |
3298 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3299 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3300 | * any more. | |
3301 | */ | |
3302 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3303 | spin_unlock_irq(&sh->stripe_lock); | |
e64e4018 AS |
3304 | md_bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3305 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3306 | spin_lock_irq(&sh->stripe_lock); |
3307 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3308 | if (!sh->batch_head) { | |
3309 | sh->bm_seq = conf->seq_flush+1; | |
3310 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3311 | } | |
cbe47ec5 | 3312 | } |
d0852df5 | 3313 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3314 | |
3315 | if (stripe_can_batch(sh)) | |
3316 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3317 | return 1; |
3318 | ||
3319 | overlap: | |
3320 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3321 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3322 | return 0; |
3323 | } | |
3324 | ||
d1688a6d | 3325 | static void end_reshape(struct r5conf *conf); |
29269553 | 3326 | |
d1688a6d | 3327 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3328 | struct stripe_head *sh) |
ccfcc3c1 | 3329 | { |
784052ec | 3330 | int sectors_per_chunk = |
09c9e5fa | 3331 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3332 | int dd_idx; |
2d2063ce | 3333 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3334 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3335 | |
112bf897 N |
3336 | raid5_compute_sector(conf, |
3337 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3338 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3339 | previous, |
911d4ee8 | 3340 | &dd_idx, sh); |
ccfcc3c1 N |
3341 | } |
3342 | ||
a4456856 | 3343 | static void |
d1688a6d | 3344 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
bd83d0a2 | 3345 | struct stripe_head_state *s, int disks) |
a4456856 DW |
3346 | { |
3347 | int i; | |
59fc630b | 3348 | BUG_ON(sh->batch_head); |
a4456856 DW |
3349 | for (i = disks; i--; ) { |
3350 | struct bio *bi; | |
3351 | int bitmap_end = 0; | |
3352 | ||
3353 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3354 | struct md_rdev *rdev; |
a4456856 DW |
3355 | rcu_read_lock(); |
3356 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3357 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3358 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3359 | atomic_inc(&rdev->nr_pending); |
3360 | else | |
3361 | rdev = NULL; | |
a4456856 | 3362 | rcu_read_unlock(); |
7f0da59b N |
3363 | if (rdev) { |
3364 | if (!rdev_set_badblocks( | |
3365 | rdev, | |
3366 | sh->sector, | |
3367 | STRIPE_SECTORS, 0)) | |
3368 | md_error(conf->mddev, rdev); | |
3369 | rdev_dec_pending(rdev, conf->mddev); | |
3370 | } | |
a4456856 | 3371 | } |
b17459c0 | 3372 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3373 | /* fail all writes first */ |
3374 | bi = sh->dev[i].towrite; | |
3375 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3376 | sh->overwrite_disks = 0; |
b17459c0 | 3377 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3378 | if (bi) |
a4456856 | 3379 | bitmap_end = 1; |
a4456856 | 3380 | |
ff875738 | 3381 | log_stripe_write_finished(sh); |
0576b1c6 | 3382 | |
a4456856 DW |
3383 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3384 | wake_up(&conf->wait_for_overlap); | |
3385 | ||
4f024f37 | 3386 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3387 | sh->dev[i].sector + STRIPE_SECTORS) { |
3388 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3389 | |
49728050 | 3390 | md_write_end(conf->mddev); |
6308d8e3 | 3391 | bio_io_error(bi); |
a4456856 DW |
3392 | bi = nextbi; |
3393 | } | |
7eaf7e8e | 3394 | if (bitmap_end) |
e64e4018 AS |
3395 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3396 | STRIPE_SECTORS, 0, 0); | |
7eaf7e8e | 3397 | bitmap_end = 0; |
a4456856 DW |
3398 | /* and fail all 'written' */ |
3399 | bi = sh->dev[i].written; | |
3400 | sh->dev[i].written = NULL; | |
d592a996 SL |
3401 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3402 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3403 | sh->dev[i].page = sh->dev[i].orig_page; | |
3404 | } | |
3405 | ||
a4456856 | 3406 | if (bi) bitmap_end = 1; |
4f024f37 | 3407 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3408 | sh->dev[i].sector + STRIPE_SECTORS) { |
3409 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3410 | |
49728050 | 3411 | md_write_end(conf->mddev); |
6308d8e3 | 3412 | bio_io_error(bi); |
a4456856 DW |
3413 | bi = bi2; |
3414 | } | |
3415 | ||
b5e98d65 DW |
3416 | /* fail any reads if this device is non-operational and |
3417 | * the data has not reached the cache yet. | |
3418 | */ | |
3419 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3420 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3421 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3422 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3423 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3424 | bi = sh->dev[i].toread; |
3425 | sh->dev[i].toread = NULL; | |
143c4d05 | 3426 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3427 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3428 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3429 | if (bi) |
3430 | s->to_read--; | |
4f024f37 | 3431 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3432 | sh->dev[i].sector + STRIPE_SECTORS) { |
3433 | struct bio *nextbi = | |
3434 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 | 3435 | |
6308d8e3 | 3436 | bio_io_error(bi); |
a4456856 DW |
3437 | bi = nextbi; |
3438 | } | |
3439 | } | |
a4456856 | 3440 | if (bitmap_end) |
e64e4018 AS |
3441 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3442 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3443 | /* If we were in the middle of a write the parity block might |
3444 | * still be locked - so just clear all R5_LOCKED flags | |
3445 | */ | |
3446 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3447 | } |
ebda780b SL |
3448 | s->to_write = 0; |
3449 | s->written = 0; | |
a4456856 | 3450 | |
8b3e6cdc DW |
3451 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3452 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3453 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3454 | } |
3455 | ||
7f0da59b | 3456 | static void |
d1688a6d | 3457 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3458 | struct stripe_head_state *s) |
3459 | { | |
3460 | int abort = 0; | |
3461 | int i; | |
3462 | ||
59fc630b | 3463 | BUG_ON(sh->batch_head); |
7f0da59b | 3464 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3465 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3466 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3467 | s->syncing = 0; |
9a3e1101 | 3468 | s->replacing = 0; |
7f0da59b | 3469 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3470 | * Don't even need to abort as that is handled elsewhere |
3471 | * if needed, and not always wanted e.g. if there is a known | |
3472 | * bad block here. | |
9a3e1101 | 3473 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3474 | * non-sync devices, or abort the recovery |
3475 | */ | |
18b9837e N |
3476 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3477 | /* During recovery devices cannot be removed, so | |
3478 | * locking and refcounting of rdevs is not needed | |
3479 | */ | |
e50d3992 | 3480 | rcu_read_lock(); |
18b9837e | 3481 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3482 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3483 | if (rdev |
3484 | && !test_bit(Faulty, &rdev->flags) | |
3485 | && !test_bit(In_sync, &rdev->flags) | |
3486 | && !rdev_set_badblocks(rdev, sh->sector, | |
3487 | STRIPE_SECTORS, 0)) | |
3488 | abort = 1; | |
e50d3992 | 3489 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3490 | if (rdev |
3491 | && !test_bit(Faulty, &rdev->flags) | |
3492 | && !test_bit(In_sync, &rdev->flags) | |
3493 | && !rdev_set_badblocks(rdev, sh->sector, | |
3494 | STRIPE_SECTORS, 0)) | |
3495 | abort = 1; | |
3496 | } | |
e50d3992 | 3497 | rcu_read_unlock(); |
18b9837e N |
3498 | if (abort) |
3499 | conf->recovery_disabled = | |
3500 | conf->mddev->recovery_disabled; | |
7f0da59b | 3501 | } |
18b9837e | 3502 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3503 | } |
3504 | ||
9a3e1101 N |
3505 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3506 | { | |
3507 | struct md_rdev *rdev; | |
3508 | int rv = 0; | |
3f232d6a N |
3509 | |
3510 | rcu_read_lock(); | |
3511 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3512 | if (rdev |
3513 | && !test_bit(Faulty, &rdev->flags) | |
3514 | && !test_bit(In_sync, &rdev->flags) | |
3515 | && (rdev->recovery_offset <= sh->sector | |
3516 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3517 | rv = 1; | |
3f232d6a | 3518 | rcu_read_unlock(); |
9a3e1101 N |
3519 | return rv; |
3520 | } | |
3521 | ||
2c58f06e N |
3522 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3523 | int disk_idx, int disks) | |
a4456856 | 3524 | { |
5599becc | 3525 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3526 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3527 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3528 | int i; |
5599becc | 3529 | |
a79cfe12 N |
3530 | |
3531 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3532 | test_bit(R5_UPTODATE, &dev->flags)) | |
3533 | /* No point reading this as we already have it or have | |
3534 | * decided to get it. | |
3535 | */ | |
3536 | return 0; | |
3537 | ||
3538 | if (dev->toread || | |
3539 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3540 | /* We need this block to directly satisfy a request */ | |
3541 | return 1; | |
3542 | ||
3543 | if (s->syncing || s->expanding || | |
3544 | (s->replacing && want_replace(sh, disk_idx))) | |
3545 | /* When syncing, or expanding we read everything. | |
3546 | * When replacing, we need the replaced block. | |
3547 | */ | |
3548 | return 1; | |
3549 | ||
3550 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3551 | (s->failed >= 2 && fdev[1]->toread)) | |
3552 | /* If we want to read from a failed device, then | |
3553 | * we need to actually read every other device. | |
3554 | */ | |
3555 | return 1; | |
3556 | ||
a9d56950 N |
3557 | /* Sometimes neither read-modify-write nor reconstruct-write |
3558 | * cycles can work. In those cases we read every block we | |
3559 | * can. Then the parity-update is certain to have enough to | |
3560 | * work with. | |
3561 | * This can only be a problem when we need to write something, | |
3562 | * and some device has failed. If either of those tests | |
3563 | * fail we need look no further. | |
3564 | */ | |
3565 | if (!s->failed || !s->to_write) | |
3566 | return 0; | |
3567 | ||
3568 | if (test_bit(R5_Insync, &dev->flags) && | |
3569 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3570 | /* Pre-reads at not permitted until after short delay | |
3571 | * to gather multiple requests. However if this | |
3560741e | 3572 | * device is no Insync, the block could only be computed |
a9d56950 N |
3573 | * and there is no need to delay that. |
3574 | */ | |
3575 | return 0; | |
ea664c82 | 3576 | |
36707bb2 | 3577 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3578 | if (fdev[i]->towrite && |
3579 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3580 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3581 | /* If we have a partial write to a failed | |
3582 | * device, then we will need to reconstruct | |
3583 | * the content of that device, so all other | |
3584 | * devices must be read. | |
3585 | */ | |
3586 | return 1; | |
3587 | } | |
3588 | ||
3589 | /* If we are forced to do a reconstruct-write, either because | |
3590 | * the current RAID6 implementation only supports that, or | |
3560741e | 3591 | * because parity cannot be trusted and we are currently |
ea664c82 N |
3592 | * recovering it, there is extra need to be careful. |
3593 | * If one of the devices that we would need to read, because | |
3594 | * it is not being overwritten (and maybe not written at all) | |
3595 | * is missing/faulty, then we need to read everything we can. | |
3596 | */ | |
3597 | if (sh->raid_conf->level != 6 && | |
3598 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3599 | /* reconstruct-write isn't being forced */ | |
3600 | return 0; | |
36707bb2 | 3601 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3602 | if (s->failed_num[i] != sh->pd_idx && |
3603 | s->failed_num[i] != sh->qd_idx && | |
3604 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3605 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3606 | return 1; | |
3607 | } | |
3608 | ||
2c58f06e N |
3609 | return 0; |
3610 | } | |
3611 | ||
ba02684d SL |
3612 | /* fetch_block - checks the given member device to see if its data needs |
3613 | * to be read or computed to satisfy a request. | |
3614 | * | |
3615 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3616 | * 0 to tell the loop in handle_stripe_fill to continue | |
3617 | */ | |
2c58f06e N |
3618 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3619 | int disk_idx, int disks) | |
3620 | { | |
3621 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3622 | ||
3623 | /* is the data in this block needed, and can we get it? */ | |
3624 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3625 | /* we would like to get this block, possibly by computing it, |
3626 | * otherwise read it if the backing disk is insync | |
3627 | */ | |
3628 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3629 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3630 | BUG_ON(sh->batch_head); |
7471fb77 N |
3631 | |
3632 | /* | |
3633 | * In the raid6 case if the only non-uptodate disk is P | |
3634 | * then we already trusted P to compute the other failed | |
3635 | * drives. It is safe to compute rather than re-read P. | |
3636 | * In other cases we only compute blocks from failed | |
3637 | * devices, otherwise check/repair might fail to detect | |
3638 | * a real inconsistency. | |
3639 | */ | |
3640 | ||
5599becc | 3641 | if ((s->uptodate == disks - 1) && |
7471fb77 | 3642 | ((sh->qd_idx >= 0 && sh->pd_idx == disk_idx) || |
f2b3b44d | 3643 | (s->failed && (disk_idx == s->failed_num[0] || |
7471fb77 | 3644 | disk_idx == s->failed_num[1])))) { |
5599becc YT |
3645 | /* have disk failed, and we're requested to fetch it; |
3646 | * do compute it | |
a4456856 | 3647 | */ |
5599becc YT |
3648 | pr_debug("Computing stripe %llu block %d\n", |
3649 | (unsigned long long)sh->sector, disk_idx); | |
3650 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3651 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3652 | set_bit(R5_Wantcompute, &dev->flags); | |
3653 | sh->ops.target = disk_idx; | |
3654 | sh->ops.target2 = -1; /* no 2nd target */ | |
3655 | s->req_compute = 1; | |
93b3dbce N |
3656 | /* Careful: from this point on 'uptodate' is in the eye |
3657 | * of raid_run_ops which services 'compute' operations | |
3658 | * before writes. R5_Wantcompute flags a block that will | |
3659 | * be R5_UPTODATE by the time it is needed for a | |
3660 | * subsequent operation. | |
3661 | */ | |
5599becc YT |
3662 | s->uptodate++; |
3663 | return 1; | |
3664 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3665 | /* Computing 2-failure is *very* expensive; only | |
3666 | * do it if failed >= 2 | |
3667 | */ | |
3668 | int other; | |
3669 | for (other = disks; other--; ) { | |
3670 | if (other == disk_idx) | |
3671 | continue; | |
3672 | if (!test_bit(R5_UPTODATE, | |
3673 | &sh->dev[other].flags)) | |
3674 | break; | |
a4456856 | 3675 | } |
5599becc YT |
3676 | BUG_ON(other < 0); |
3677 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3678 | (unsigned long long)sh->sector, | |
3679 | disk_idx, other); | |
3680 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3681 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3682 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3683 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3684 | sh->ops.target = disk_idx; | |
3685 | sh->ops.target2 = other; | |
3686 | s->uptodate += 2; | |
3687 | s->req_compute = 1; | |
3688 | return 1; | |
3689 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3690 | set_bit(R5_LOCKED, &dev->flags); | |
3691 | set_bit(R5_Wantread, &dev->flags); | |
3692 | s->locked++; | |
3693 | pr_debug("Reading block %d (sync=%d)\n", | |
3694 | disk_idx, s->syncing); | |
a4456856 DW |
3695 | } |
3696 | } | |
5599becc YT |
3697 | |
3698 | return 0; | |
3699 | } | |
3700 | ||
3701 | /** | |
93b3dbce | 3702 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3703 | */ |
93b3dbce N |
3704 | static void handle_stripe_fill(struct stripe_head *sh, |
3705 | struct stripe_head_state *s, | |
3706 | int disks) | |
5599becc YT |
3707 | { |
3708 | int i; | |
3709 | ||
3710 | /* look for blocks to read/compute, skip this if a compute | |
3711 | * is already in flight, or if the stripe contents are in the | |
3712 | * midst of changing due to a write | |
3713 | */ | |
3714 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3715 | !sh->reconstruct_state) { |
3716 | ||
3717 | /* | |
3718 | * For degraded stripe with data in journal, do not handle | |
3719 | * read requests yet, instead, flush the stripe to raid | |
3720 | * disks first, this avoids handling complex rmw of write | |
3721 | * back cache (prexor with orig_page, and then xor with | |
3722 | * page) in the read path | |
3723 | */ | |
3724 | if (s->injournal && s->failed) { | |
3725 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3726 | r5c_make_stripe_write_out(sh); | |
3727 | goto out; | |
3728 | } | |
3729 | ||
5599becc | 3730 | for (i = disks; i--; ) |
93b3dbce | 3731 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3732 | break; |
07e83364 SL |
3733 | } |
3734 | out: | |
a4456856 DW |
3735 | set_bit(STRIPE_HANDLE, &sh->state); |
3736 | } | |
3737 | ||
787b76fa N |
3738 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3739 | unsigned long handle_flags); | |
1fe797e6 | 3740 | /* handle_stripe_clean_event |
a4456856 DW |
3741 | * any written block on an uptodate or failed drive can be returned. |
3742 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3743 | * never LOCKED, so we don't need to test 'failed' directly. | |
3744 | */ | |
d1688a6d | 3745 | static void handle_stripe_clean_event(struct r5conf *conf, |
bd83d0a2 | 3746 | struct stripe_head *sh, int disks) |
a4456856 DW |
3747 | { |
3748 | int i; | |
3749 | struct r5dev *dev; | |
f8dfcffd | 3750 | int discard_pending = 0; |
59fc630b | 3751 | struct stripe_head *head_sh = sh; |
3752 | bool do_endio = false; | |
a4456856 DW |
3753 | |
3754 | for (i = disks; i--; ) | |
3755 | if (sh->dev[i].written) { | |
3756 | dev = &sh->dev[i]; | |
3757 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3758 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3759 | test_bit(R5_Discard, &dev->flags) || |
3760 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3761 | /* We can return any write requests */ |
3762 | struct bio *wbi, *wbi2; | |
45b4233c | 3763 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3764 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3765 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3766 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3767 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3768 | } |
59fc630b | 3769 | do_endio = true; |
3770 | ||
3771 | returnbi: | |
3772 | dev->page = dev->orig_page; | |
a4456856 DW |
3773 | wbi = dev->written; |
3774 | dev->written = NULL; | |
4f024f37 | 3775 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3776 | dev->sector + STRIPE_SECTORS) { |
3777 | wbi2 = r5_next_bio(wbi, dev->sector); | |
49728050 | 3778 | md_write_end(conf->mddev); |
016c76ac | 3779 | bio_endio(wbi); |
a4456856 DW |
3780 | wbi = wbi2; |
3781 | } | |
e64e4018 AS |
3782 | md_bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3783 | STRIPE_SECTORS, | |
3784 | !test_bit(STRIPE_DEGRADED, &sh->state), | |
3785 | 0); | |
59fc630b | 3786 | if (head_sh->batch_head) { |
3787 | sh = list_first_entry(&sh->batch_list, | |
3788 | struct stripe_head, | |
3789 | batch_list); | |
3790 | if (sh != head_sh) { | |
3791 | dev = &sh->dev[i]; | |
3792 | goto returnbi; | |
3793 | } | |
3794 | } | |
3795 | sh = head_sh; | |
3796 | dev = &sh->dev[i]; | |
f8dfcffd N |
3797 | } else if (test_bit(R5_Discard, &dev->flags)) |
3798 | discard_pending = 1; | |
3799 | } | |
f6bed0ef | 3800 | |
ff875738 | 3801 | log_stripe_write_finished(sh); |
0576b1c6 | 3802 | |
f8dfcffd N |
3803 | if (!discard_pending && |
3804 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3805 | int hash; |
f8dfcffd N |
3806 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3807 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3808 | if (sh->qd_idx >= 0) { | |
3809 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3810 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3811 | } | |
3812 | /* now that discard is done we can proceed with any sync */ | |
3813 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3814 | /* |
3815 | * SCSI discard will change some bio fields and the stripe has | |
3816 | * no updated data, so remove it from hash list and the stripe | |
3817 | * will be reinitialized | |
3818 | */ | |
59fc630b | 3819 | unhash: |
b8a9d66d RG |
3820 | hash = sh->hash_lock_index; |
3821 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3822 | remove_hash(sh); |
b8a9d66d | 3823 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3824 | if (head_sh->batch_head) { |
3825 | sh = list_first_entry(&sh->batch_list, | |
3826 | struct stripe_head, batch_list); | |
3827 | if (sh != head_sh) | |
3828 | goto unhash; | |
3829 | } | |
59fc630b | 3830 | sh = head_sh; |
3831 | ||
f8dfcffd N |
3832 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3833 | set_bit(STRIPE_HANDLE, &sh->state); | |
3834 | ||
3835 | } | |
8b3e6cdc DW |
3836 | |
3837 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3838 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3839 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3840 | |
787b76fa N |
3841 | if (head_sh->batch_head && do_endio) |
3842 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3843 | } |
3844 | ||
86aa1397 SL |
3845 | /* |
3846 | * For RMW in write back cache, we need extra page in prexor to store the | |
3847 | * old data. This page is stored in dev->orig_page. | |
3848 | * | |
3849 | * This function checks whether we have data for prexor. The exact logic | |
3850 | * is: | |
3851 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3852 | */ | |
3853 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3854 | { | |
3855 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3856 | (!test_bit(R5_InJournal, &dev->flags) || | |
3857 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3858 | } | |
3859 | ||
d7bd398e SL |
3860 | static int handle_stripe_dirtying(struct r5conf *conf, |
3861 | struct stripe_head *sh, | |
3862 | struct stripe_head_state *s, | |
3863 | int disks) | |
a4456856 DW |
3864 | { |
3865 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3866 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3867 | ||
584acdd4 | 3868 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3869 | * If yes, then the array is dirty (after unclean shutdown or |
3870 | * initial creation), so parity in some stripes might be inconsistent. | |
3871 | * In this case, we need to always do reconstruct-write, to ensure | |
3872 | * that in case of drive failure or read-error correction, we | |
3873 | * generate correct data from the parity. | |
3874 | */ | |
584acdd4 | 3875 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3876 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3877 | s->failed == 0)) { | |
a7854487 | 3878 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3879 | * look like rcw is cheaper |
3880 | */ | |
3881 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3882 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3883 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3884 | (unsigned long long)sh->sector); |
c8ac1803 | 3885 | } else for (i = disks; i--; ) { |
a4456856 DW |
3886 | /* would I have to read this buffer for read_modify_write */ |
3887 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3888 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3889 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3890 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3891 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3892 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3893 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3894 | if (test_bit(R5_Insync, &dev->flags)) |
3895 | rmw++; | |
3896 | else | |
3897 | rmw += 2*disks; /* cannot read it */ | |
3898 | } | |
3899 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3900 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3901 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3902 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3903 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3904 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3905 | if (test_bit(R5_Insync, &dev->flags)) |
3906 | rcw++; | |
a4456856 DW |
3907 | else |
3908 | rcw += 2*disks; | |
3909 | } | |
3910 | } | |
1e6d690b | 3911 | |
39b99586 SL |
3912 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3913 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3914 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3915 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3916 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3917 | if (conf->mddev->queue) |
3918 | blk_add_trace_msg(conf->mddev->queue, | |
3919 | "raid5 rmw %llu %d", | |
3920 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3921 | for (i = disks; i--; ) { |
3922 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3923 | if (test_bit(R5_InJournal, &dev->flags) && |
3924 | dev->page == dev->orig_page && | |
3925 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3926 | /* alloc page for prexor */ | |
d7bd398e SL |
3927 | struct page *p = alloc_page(GFP_NOIO); |
3928 | ||
3929 | if (p) { | |
3930 | dev->orig_page = p; | |
3931 | continue; | |
3932 | } | |
3933 | ||
3934 | /* | |
3935 | * alloc_page() failed, try use | |
3936 | * disk_info->extra_page | |
3937 | */ | |
3938 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3939 | &conf->cache_state)) { | |
3940 | r5c_use_extra_page(sh); | |
3941 | break; | |
3942 | } | |
1e6d690b | 3943 | |
d7bd398e SL |
3944 | /* extra_page in use, add to delayed_list */ |
3945 | set_bit(STRIPE_DELAYED, &sh->state); | |
3946 | s->waiting_extra_page = 1; | |
3947 | return -EAGAIN; | |
1e6d690b | 3948 | } |
d7bd398e | 3949 | } |
1e6d690b | 3950 | |
d7bd398e SL |
3951 | for (i = disks; i--; ) { |
3952 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3953 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3954 | i == sh->pd_idx || i == sh->qd_idx || |
3955 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3956 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3957 | !(uptodate_for_rmw(dev) || |
1e6d690b | 3958 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 3959 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3960 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3961 | &sh->state)) { | |
3962 | pr_debug("Read_old block %d for r-m-w\n", | |
3963 | i); | |
a4456856 DW |
3964 | set_bit(R5_LOCKED, &dev->flags); |
3965 | set_bit(R5_Wantread, &dev->flags); | |
3966 | s->locked++; | |
3967 | } else { | |
3968 | set_bit(STRIPE_DELAYED, &sh->state); | |
3969 | set_bit(STRIPE_HANDLE, &sh->state); | |
3970 | } | |
3971 | } | |
3972 | } | |
a9add5d9 | 3973 | } |
41257580 | 3974 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3975 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3976 | int qread =0; |
c8ac1803 | 3977 | rcw = 0; |
a4456856 DW |
3978 | for (i = disks; i--; ) { |
3979 | struct r5dev *dev = &sh->dev[i]; | |
3980 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3981 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3982 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3983 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3984 | test_bit(R5_Wantcompute, &dev->flags))) { |
3985 | rcw++; | |
67f45548 N |
3986 | if (test_bit(R5_Insync, &dev->flags) && |
3987 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3988 | &sh->state)) { | |
45b4233c | 3989 | pr_debug("Read_old block " |
a4456856 DW |
3990 | "%d for Reconstruct\n", i); |
3991 | set_bit(R5_LOCKED, &dev->flags); | |
3992 | set_bit(R5_Wantread, &dev->flags); | |
3993 | s->locked++; | |
a9add5d9 | 3994 | qread++; |
a4456856 DW |
3995 | } else { |
3996 | set_bit(STRIPE_DELAYED, &sh->state); | |
3997 | set_bit(STRIPE_HANDLE, &sh->state); | |
3998 | } | |
3999 | } | |
4000 | } | |
e3620a3a | 4001 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
4002 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
4003 | (unsigned long long)sh->sector, | |
4004 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 4005 | } |
b1b02fe9 N |
4006 | |
4007 | if (rcw > disks && rmw > disks && | |
4008 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4009 | set_bit(STRIPE_DELAYED, &sh->state); | |
4010 | ||
a4456856 DW |
4011 | /* now if nothing is locked, and if we have enough data, |
4012 | * we can start a write request | |
4013 | */ | |
f38e1219 DW |
4014 | /* since handle_stripe can be called at any time we need to handle the |
4015 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
4016 | * subsequent call wants to start a write request. raid_run_ops only |
4017 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
4018 | * simultaneously. If this is not the case then new writes need to be |
4019 | * held off until the compute completes. | |
4020 | */ | |
976ea8d4 DW |
4021 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
4022 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 4023 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 4024 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 4025 | return 0; |
a4456856 DW |
4026 | } |
4027 | ||
d1688a6d | 4028 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
4029 | struct stripe_head_state *s, int disks) |
4030 | { | |
ecc65c9b | 4031 | struct r5dev *dev = NULL; |
bd2ab670 | 4032 | |
59fc630b | 4033 | BUG_ON(sh->batch_head); |
a4456856 | 4034 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 4035 | |
ecc65c9b DW |
4036 | switch (sh->check_state) { |
4037 | case check_state_idle: | |
4038 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 4039 | if (s->failed == 0) { |
bd2ab670 | 4040 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
4041 | sh->check_state = check_state_run; |
4042 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 4043 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 4044 | s->uptodate--; |
ecc65c9b | 4045 | break; |
bd2ab670 | 4046 | } |
f2b3b44d | 4047 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
4048 | /* fall through */ |
4049 | case check_state_compute_result: | |
4050 | sh->check_state = check_state_idle; | |
4051 | if (!dev) | |
4052 | dev = &sh->dev[sh->pd_idx]; | |
4053 | ||
4054 | /* check that a write has not made the stripe insync */ | |
4055 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4056 | break; | |
c8894419 | 4057 | |
a4456856 | 4058 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
4059 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
4060 | BUG_ON(s->uptodate != disks); | |
4061 | ||
4062 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 4063 | s->locked++; |
a4456856 | 4064 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 4065 | |
a4456856 | 4066 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 4067 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
4068 | break; |
4069 | case check_state_run: | |
4070 | break; /* we will be called again upon completion */ | |
4071 | case check_state_check_result: | |
4072 | sh->check_state = check_state_idle; | |
4073 | ||
4074 | /* if a failure occurred during the check operation, leave | |
4075 | * STRIPE_INSYNC not set and let the stripe be handled again | |
4076 | */ | |
4077 | if (s->failed) | |
4078 | break; | |
4079 | ||
4080 | /* handle a successful check operation, if parity is correct | |
4081 | * we are done. Otherwise update the mismatch count and repair | |
4082 | * parity if !MD_RECOVERY_CHECK | |
4083 | */ | |
ad283ea4 | 4084 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
4085 | /* parity is correct (on disc, |
4086 | * not in buffer any more) | |
4087 | */ | |
4088 | set_bit(STRIPE_INSYNC, &sh->state); | |
4089 | else { | |
7f7583d4 | 4090 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
e1539036 | 4091 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
ecc65c9b DW |
4092 | /* don't try to repair!! */ |
4093 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4094 | pr_warn_ratelimited("%s: mismatch sector in range " |
4095 | "%llu-%llu\n", mdname(conf->mddev), | |
4096 | (unsigned long long) sh->sector, | |
4097 | (unsigned long long) sh->sector + | |
4098 | STRIPE_SECTORS); | |
4099 | } else { | |
ecc65c9b | 4100 | sh->check_state = check_state_compute_run; |
976ea8d4 | 4101 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
4102 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
4103 | set_bit(R5_Wantcompute, | |
4104 | &sh->dev[sh->pd_idx].flags); | |
4105 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 4106 | sh->ops.target2 = -1; |
ecc65c9b DW |
4107 | s->uptodate++; |
4108 | } | |
4109 | } | |
4110 | break; | |
4111 | case check_state_compute_run: | |
4112 | break; | |
4113 | default: | |
cc6167b4 | 4114 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
4115 | __func__, sh->check_state, |
4116 | (unsigned long long) sh->sector); | |
4117 | BUG(); | |
a4456856 DW |
4118 | } |
4119 | } | |
4120 | ||
d1688a6d | 4121 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4122 | struct stripe_head_state *s, |
f2b3b44d | 4123 | int disks) |
a4456856 | 4124 | { |
a4456856 | 4125 | int pd_idx = sh->pd_idx; |
34e04e87 | 4126 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4127 | struct r5dev *dev; |
a4456856 | 4128 | |
59fc630b | 4129 | BUG_ON(sh->batch_head); |
a4456856 DW |
4130 | set_bit(STRIPE_HANDLE, &sh->state); |
4131 | ||
4132 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4133 | |
a4456856 DW |
4134 | /* Want to check and possibly repair P and Q. |
4135 | * However there could be one 'failed' device, in which | |
4136 | * case we can only check one of them, possibly using the | |
4137 | * other to generate missing data | |
4138 | */ | |
4139 | ||
d82dfee0 DW |
4140 | switch (sh->check_state) { |
4141 | case check_state_idle: | |
4142 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4143 | if (s->failed == s->q_failed) { |
d82dfee0 | 4144 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4145 | * makes sense to check P (If anything else were failed, |
4146 | * we would have used P to recreate it). | |
4147 | */ | |
d82dfee0 | 4148 | sh->check_state = check_state_run; |
a4456856 | 4149 | } |
f2b3b44d | 4150 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4151 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4152 | * anything, so it makes sense to check it |
4153 | */ | |
d82dfee0 DW |
4154 | if (sh->check_state == check_state_run) |
4155 | sh->check_state = check_state_run_pq; | |
4156 | else | |
4157 | sh->check_state = check_state_run_q; | |
a4456856 | 4158 | } |
a4456856 | 4159 | |
d82dfee0 DW |
4160 | /* discard potentially stale zero_sum_result */ |
4161 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4162 | |
d82dfee0 DW |
4163 | if (sh->check_state == check_state_run) { |
4164 | /* async_xor_zero_sum destroys the contents of P */ | |
4165 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4166 | s->uptodate--; | |
a4456856 | 4167 | } |
d82dfee0 DW |
4168 | if (sh->check_state >= check_state_run && |
4169 | sh->check_state <= check_state_run_pq) { | |
4170 | /* async_syndrome_zero_sum preserves P and Q, so | |
4171 | * no need to mark them !uptodate here | |
4172 | */ | |
4173 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4174 | break; | |
a4456856 DW |
4175 | } |
4176 | ||
d82dfee0 DW |
4177 | /* we have 2-disk failure */ |
4178 | BUG_ON(s->failed != 2); | |
4179 | /* fall through */ | |
4180 | case check_state_compute_result: | |
4181 | sh->check_state = check_state_idle; | |
a4456856 | 4182 | |
d82dfee0 DW |
4183 | /* check that a write has not made the stripe insync */ |
4184 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4185 | break; | |
a4456856 DW |
4186 | |
4187 | /* now write out any block on a failed drive, | |
d82dfee0 | 4188 | * or P or Q if they were recomputed |
a4456856 | 4189 | */ |
d82dfee0 | 4190 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 4191 | if (s->failed == 2) { |
f2b3b44d | 4192 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4193 | s->locked++; |
4194 | set_bit(R5_LOCKED, &dev->flags); | |
4195 | set_bit(R5_Wantwrite, &dev->flags); | |
4196 | } | |
4197 | if (s->failed >= 1) { | |
f2b3b44d | 4198 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4199 | s->locked++; |
4200 | set_bit(R5_LOCKED, &dev->flags); | |
4201 | set_bit(R5_Wantwrite, &dev->flags); | |
4202 | } | |
d82dfee0 | 4203 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4204 | dev = &sh->dev[pd_idx]; |
4205 | s->locked++; | |
4206 | set_bit(R5_LOCKED, &dev->flags); | |
4207 | set_bit(R5_Wantwrite, &dev->flags); | |
4208 | } | |
d82dfee0 | 4209 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4210 | dev = &sh->dev[qd_idx]; |
4211 | s->locked++; | |
4212 | set_bit(R5_LOCKED, &dev->flags); | |
4213 | set_bit(R5_Wantwrite, &dev->flags); | |
4214 | } | |
4215 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
4216 | ||
4217 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4218 | break; |
4219 | case check_state_run: | |
4220 | case check_state_run_q: | |
4221 | case check_state_run_pq: | |
4222 | break; /* we will be called again upon completion */ | |
4223 | case check_state_check_result: | |
4224 | sh->check_state = check_state_idle; | |
4225 | ||
4226 | /* handle a successful check operation, if parity is correct | |
4227 | * we are done. Otherwise update the mismatch count and repair | |
4228 | * parity if !MD_RECOVERY_CHECK | |
4229 | */ | |
4230 | if (sh->ops.zero_sum_result == 0) { | |
4231 | /* both parities are correct */ | |
4232 | if (!s->failed) | |
4233 | set_bit(STRIPE_INSYNC, &sh->state); | |
4234 | else { | |
4235 | /* in contrast to the raid5 case we can validate | |
4236 | * parity, but still have a failure to write | |
4237 | * back | |
4238 | */ | |
4239 | sh->check_state = check_state_compute_result; | |
4240 | /* Returning at this point means that we may go | |
4241 | * off and bring p and/or q uptodate again so | |
4242 | * we make sure to check zero_sum_result again | |
4243 | * to verify if p or q need writeback | |
4244 | */ | |
4245 | } | |
4246 | } else { | |
7f7583d4 | 4247 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
e1539036 | 4248 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) { |
d82dfee0 DW |
4249 | /* don't try to repair!! */ |
4250 | set_bit(STRIPE_INSYNC, &sh->state); | |
e1539036 N |
4251 | pr_warn_ratelimited("%s: mismatch sector in range " |
4252 | "%llu-%llu\n", mdname(conf->mddev), | |
4253 | (unsigned long long) sh->sector, | |
4254 | (unsigned long long) sh->sector + | |
4255 | STRIPE_SECTORS); | |
4256 | } else { | |
d82dfee0 DW |
4257 | int *target = &sh->ops.target; |
4258 | ||
4259 | sh->ops.target = -1; | |
4260 | sh->ops.target2 = -1; | |
4261 | sh->check_state = check_state_compute_run; | |
4262 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4263 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4264 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4265 | set_bit(R5_Wantcompute, | |
4266 | &sh->dev[pd_idx].flags); | |
4267 | *target = pd_idx; | |
4268 | target = &sh->ops.target2; | |
4269 | s->uptodate++; | |
4270 | } | |
4271 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4272 | set_bit(R5_Wantcompute, | |
4273 | &sh->dev[qd_idx].flags); | |
4274 | *target = qd_idx; | |
4275 | s->uptodate++; | |
4276 | } | |
4277 | } | |
4278 | } | |
4279 | break; | |
4280 | case check_state_compute_run: | |
4281 | break; | |
4282 | default: | |
cc6167b4 N |
4283 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4284 | __func__, sh->check_state, | |
4285 | (unsigned long long) sh->sector); | |
d82dfee0 | 4286 | BUG(); |
a4456856 DW |
4287 | } |
4288 | } | |
4289 | ||
d1688a6d | 4290 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4291 | { |
4292 | int i; | |
4293 | ||
4294 | /* We have read all the blocks in this stripe and now we need to | |
4295 | * copy some of them into a target stripe for expand. | |
4296 | */ | |
f0a50d37 | 4297 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4298 | BUG_ON(sh->batch_head); |
a4456856 DW |
4299 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4300 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4301 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4302 | int dd_idx, j; |
a4456856 | 4303 | struct stripe_head *sh2; |
a08abd8c | 4304 | struct async_submit_ctl submit; |
a4456856 | 4305 | |
6d036f7d | 4306 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4307 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4308 | &dd_idx, NULL); | |
6d036f7d | 4309 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4310 | if (sh2 == NULL) |
4311 | /* so far only the early blocks of this stripe | |
4312 | * have been requested. When later blocks | |
4313 | * get requested, we will try again | |
4314 | */ | |
4315 | continue; | |
4316 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4317 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4318 | /* must have already done this block */ | |
6d036f7d | 4319 | raid5_release_stripe(sh2); |
a4456856 DW |
4320 | continue; |
4321 | } | |
f0a50d37 DW |
4322 | |
4323 | /* place all the copies on one channel */ | |
a08abd8c | 4324 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4325 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 4326 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 4327 | &submit); |
f0a50d37 | 4328 | |
a4456856 DW |
4329 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4330 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4331 | for (j = 0; j < conf->raid_disks; j++) | |
4332 | if (j != sh2->pd_idx && | |
86c374ba | 4333 | j != sh2->qd_idx && |
a4456856 DW |
4334 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4335 | break; | |
4336 | if (j == conf->raid_disks) { | |
4337 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4338 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4339 | } | |
6d036f7d | 4340 | raid5_release_stripe(sh2); |
f0a50d37 | 4341 | |
a4456856 | 4342 | } |
a2e08551 | 4343 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4344 | async_tx_quiesce(&tx); |
a4456856 | 4345 | } |
1da177e4 LT |
4346 | |
4347 | /* | |
4348 | * handle_stripe - do things to a stripe. | |
4349 | * | |
9a3e1101 N |
4350 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4351 | * state of various bits to see what needs to be done. | |
1da177e4 | 4352 | * Possible results: |
9a3e1101 N |
4353 | * return some read requests which now have data |
4354 | * return some write requests which are safely on storage | |
1da177e4 LT |
4355 | * schedule a read on some buffers |
4356 | * schedule a write of some buffers | |
4357 | * return confirmation of parity correctness | |
4358 | * | |
1da177e4 | 4359 | */ |
a4456856 | 4360 | |
acfe726b | 4361 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4362 | { |
d1688a6d | 4363 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4364 | int disks = sh->disks; |
474af965 N |
4365 | struct r5dev *dev; |
4366 | int i; | |
9a3e1101 | 4367 | int do_recovery = 0; |
1da177e4 | 4368 | |
acfe726b N |
4369 | memset(s, 0, sizeof(*s)); |
4370 | ||
dabc4ec6 | 4371 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4372 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4373 | s->failed_num[0] = -1; |
4374 | s->failed_num[1] = -1; | |
6e74a9cf | 4375 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4376 | |
acfe726b | 4377 | /* Now to look around and see what can be done */ |
1da177e4 | 4378 | rcu_read_lock(); |
16a53ecc | 4379 | for (i=disks; i--; ) { |
3cb03002 | 4380 | struct md_rdev *rdev; |
31c176ec N |
4381 | sector_t first_bad; |
4382 | int bad_sectors; | |
4383 | int is_bad = 0; | |
acfe726b | 4384 | |
16a53ecc | 4385 | dev = &sh->dev[i]; |
1da177e4 | 4386 | |
45b4233c | 4387 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4388 | i, dev->flags, |
4389 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4390 | /* maybe we can reply to a read |
4391 | * | |
4392 | * new wantfill requests are only permitted while | |
4393 | * ops_complete_biofill is guaranteed to be inactive | |
4394 | */ | |
4395 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4396 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4397 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4398 | |
16a53ecc | 4399 | /* now count some things */ |
cc94015a N |
4400 | if (test_bit(R5_LOCKED, &dev->flags)) |
4401 | s->locked++; | |
4402 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4403 | s->uptodate++; | |
2d6e4ecc | 4404 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4405 | s->compute++; |
4406 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4407 | } |
1da177e4 | 4408 | |
acfe726b | 4409 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4410 | s->to_fill++; |
acfe726b | 4411 | else if (dev->toread) |
cc94015a | 4412 | s->to_read++; |
16a53ecc | 4413 | if (dev->towrite) { |
cc94015a | 4414 | s->to_write++; |
16a53ecc | 4415 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4416 | s->non_overwrite++; |
16a53ecc | 4417 | } |
a4456856 | 4418 | if (dev->written) |
cc94015a | 4419 | s->written++; |
14a75d3e N |
4420 | /* Prefer to use the replacement for reads, but only |
4421 | * if it is recovered enough and has no bad blocks. | |
4422 | */ | |
4423 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4424 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4425 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4426 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4427 | &first_bad, &bad_sectors)) | |
4428 | set_bit(R5_ReadRepl, &dev->flags); | |
4429 | else { | |
e6030cb0 | 4430 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4431 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4432 | else |
4433 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4434 | rdev = rcu_dereference(conf->disks[i].rdev); |
4435 | clear_bit(R5_ReadRepl, &dev->flags); | |
4436 | } | |
9283d8c5 N |
4437 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4438 | rdev = NULL; | |
31c176ec N |
4439 | if (rdev) { |
4440 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4441 | &first_bad, &bad_sectors); | |
4442 | if (s->blocked_rdev == NULL | |
4443 | && (test_bit(Blocked, &rdev->flags) | |
4444 | || is_bad < 0)) { | |
4445 | if (is_bad < 0) | |
4446 | set_bit(BlockedBadBlocks, | |
4447 | &rdev->flags); | |
4448 | s->blocked_rdev = rdev; | |
4449 | atomic_inc(&rdev->nr_pending); | |
4450 | } | |
6bfe0b49 | 4451 | } |
415e72d0 N |
4452 | clear_bit(R5_Insync, &dev->flags); |
4453 | if (!rdev) | |
4454 | /* Not in-sync */; | |
31c176ec N |
4455 | else if (is_bad) { |
4456 | /* also not in-sync */ | |
18b9837e N |
4457 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4458 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4459 | /* treat as in-sync, but with a read error |
4460 | * which we can now try to correct | |
4461 | */ | |
4462 | set_bit(R5_Insync, &dev->flags); | |
4463 | set_bit(R5_ReadError, &dev->flags); | |
4464 | } | |
4465 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4466 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4467 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4468 | /* in sync if before recovery_offset */ |
30d7a483 N |
4469 | set_bit(R5_Insync, &dev->flags); |
4470 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4471 | test_bit(R5_Expanded, &dev->flags)) | |
4472 | /* If we've reshaped into here, we assume it is Insync. | |
4473 | * We will shortly update recovery_offset to make | |
4474 | * it official. | |
4475 | */ | |
4476 | set_bit(R5_Insync, &dev->flags); | |
4477 | ||
1cc03eb9 | 4478 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4479 | /* This flag does not apply to '.replacement' |
4480 | * only to .rdev, so make sure to check that*/ | |
4481 | struct md_rdev *rdev2 = rcu_dereference( | |
4482 | conf->disks[i].rdev); | |
4483 | if (rdev2 == rdev) | |
4484 | clear_bit(R5_Insync, &dev->flags); | |
4485 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4486 | s->handle_bad_blocks = 1; |
14a75d3e | 4487 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4488 | } else |
4489 | clear_bit(R5_WriteError, &dev->flags); | |
4490 | } | |
1cc03eb9 | 4491 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4492 | /* This flag does not apply to '.replacement' |
4493 | * only to .rdev, so make sure to check that*/ | |
4494 | struct md_rdev *rdev2 = rcu_dereference( | |
4495 | conf->disks[i].rdev); | |
4496 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4497 | s->handle_bad_blocks = 1; |
14a75d3e | 4498 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4499 | } else |
4500 | clear_bit(R5_MadeGood, &dev->flags); | |
4501 | } | |
977df362 N |
4502 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4503 | struct md_rdev *rdev2 = rcu_dereference( | |
4504 | conf->disks[i].replacement); | |
4505 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4506 | s->handle_bad_blocks = 1; | |
4507 | atomic_inc(&rdev2->nr_pending); | |
4508 | } else | |
4509 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4510 | } | |
415e72d0 | 4511 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4512 | /* The ReadError flag will just be confusing now */ |
4513 | clear_bit(R5_ReadError, &dev->flags); | |
4514 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4515 | } |
415e72d0 N |
4516 | if (test_bit(R5_ReadError, &dev->flags)) |
4517 | clear_bit(R5_Insync, &dev->flags); | |
4518 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4519 | if (s->failed < 2) |
4520 | s->failed_num[s->failed] = i; | |
4521 | s->failed++; | |
9a3e1101 N |
4522 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4523 | do_recovery = 1; | |
d63e2fc8 BC |
4524 | else if (!rdev) { |
4525 | rdev = rcu_dereference( | |
4526 | conf->disks[i].replacement); | |
4527 | if (rdev && !test_bit(Faulty, &rdev->flags)) | |
4528 | do_recovery = 1; | |
4529 | } | |
415e72d0 | 4530 | } |
2ded3703 SL |
4531 | |
4532 | if (test_bit(R5_InJournal, &dev->flags)) | |
4533 | s->injournal++; | |
1e6d690b SL |
4534 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4535 | s->just_cached++; | |
1da177e4 | 4536 | } |
9a3e1101 N |
4537 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4538 | /* If there is a failed device being replaced, | |
4539 | * we must be recovering. | |
4540 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4541 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4542 | * else we can only be replacing |
4543 | * sync and recovery both need to read all devices, and so | |
4544 | * use the same flag. | |
4545 | */ | |
4546 | if (do_recovery || | |
c6d2e084 | 4547 | sh->sector >= conf->mddev->recovery_cp || |
4548 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4549 | s->syncing = 1; |
4550 | else | |
4551 | s->replacing = 1; | |
4552 | } | |
1da177e4 | 4553 | rcu_read_unlock(); |
cc94015a N |
4554 | } |
4555 | ||
59fc630b | 4556 | static int clear_batch_ready(struct stripe_head *sh) |
4557 | { | |
b15a9dbd N |
4558 | /* Return '1' if this is a member of batch, or |
4559 | * '0' if it is a lone stripe or a head which can now be | |
4560 | * handled. | |
4561 | */ | |
59fc630b | 4562 | struct stripe_head *tmp; |
4563 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4564 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4565 | spin_lock(&sh->stripe_lock); |
4566 | if (!sh->batch_head) { | |
4567 | spin_unlock(&sh->stripe_lock); | |
4568 | return 0; | |
4569 | } | |
4570 | ||
4571 | /* | |
4572 | * this stripe could be added to a batch list before we check | |
4573 | * BATCH_READY, skips it | |
4574 | */ | |
4575 | if (sh->batch_head != sh) { | |
4576 | spin_unlock(&sh->stripe_lock); | |
4577 | return 1; | |
4578 | } | |
4579 | spin_lock(&sh->batch_lock); | |
4580 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4581 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4582 | spin_unlock(&sh->batch_lock); | |
4583 | spin_unlock(&sh->stripe_lock); | |
4584 | ||
4585 | /* | |
4586 | * BATCH_READY is cleared, no new stripes can be added. | |
4587 | * batch_list can be accessed without lock | |
4588 | */ | |
4589 | return 0; | |
4590 | } | |
4591 | ||
3960ce79 N |
4592 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4593 | unsigned long handle_flags) | |
72ac7330 | 4594 | { |
4e3d62ff | 4595 | struct stripe_head *sh, *next; |
72ac7330 | 4596 | int i; |
fb642b92 | 4597 | int do_wakeup = 0; |
72ac7330 | 4598 | |
bb27051f N |
4599 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4600 | ||
72ac7330 | 4601 | list_del_init(&sh->batch_list); |
4602 | ||
fb3229d5 | 4603 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4604 | (1 << STRIPE_SYNCING) | |
4605 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4606 | (1 << STRIPE_DELAYED) | |
4607 | (1 << STRIPE_BIT_DELAY) | | |
4608 | (1 << STRIPE_FULL_WRITE) | | |
4609 | (1 << STRIPE_BIOFILL_RUN) | | |
4610 | (1 << STRIPE_COMPUTE_RUN) | | |
4611 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4612 | (1 << STRIPE_DISCARD) | | |
4613 | (1 << STRIPE_BATCH_READY) | | |
4614 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4615 | (1 << STRIPE_BITMAP_PENDING)), |
4616 | "stripe state: %lx\n", sh->state); | |
4617 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4618 | (1 << STRIPE_REPLACED)), | |
4619 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4620 | |
4621 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4622 | (1 << STRIPE_PREREAD_ACTIVE) | |
184a09eb DY |
4623 | (1 << STRIPE_DEGRADED) | |
4624 | (1 << STRIPE_ON_UNPLUG_LIST)), | |
1b956f7a N |
4625 | head_sh->state & (1 << STRIPE_INSYNC)); |
4626 | ||
72ac7330 | 4627 | sh->check_state = head_sh->check_state; |
4628 | sh->reconstruct_state = head_sh->reconstruct_state; | |
448ec638 AC |
4629 | spin_lock_irq(&sh->stripe_lock); |
4630 | sh->batch_head = NULL; | |
4631 | spin_unlock_irq(&sh->stripe_lock); | |
fb642b92 N |
4632 | for (i = 0; i < sh->disks; i++) { |
4633 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4634 | do_wakeup = 1; | |
72ac7330 | 4635 | sh->dev[i].flags = head_sh->dev[i].flags & |
4636 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4637 | } |
3960ce79 N |
4638 | if (handle_flags == 0 || |
4639 | sh->state & handle_flags) | |
4640 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4641 | raid5_release_stripe(sh); |
72ac7330 | 4642 | } |
fb642b92 N |
4643 | spin_lock_irq(&head_sh->stripe_lock); |
4644 | head_sh->batch_head = NULL; | |
4645 | spin_unlock_irq(&head_sh->stripe_lock); | |
4646 | for (i = 0; i < head_sh->disks; i++) | |
4647 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4648 | do_wakeup = 1; | |
3960ce79 N |
4649 | if (head_sh->state & handle_flags) |
4650 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4651 | |
4652 | if (do_wakeup) | |
4653 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4654 | } |
4655 | ||
cc94015a N |
4656 | static void handle_stripe(struct stripe_head *sh) |
4657 | { | |
4658 | struct stripe_head_state s; | |
d1688a6d | 4659 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4660 | int i; |
84789554 N |
4661 | int prexor; |
4662 | int disks = sh->disks; | |
474af965 | 4663 | struct r5dev *pdev, *qdev; |
cc94015a N |
4664 | |
4665 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4666 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4667 | /* already being handled, ensure it gets handled |
4668 | * again when current action finishes */ | |
4669 | set_bit(STRIPE_HANDLE, &sh->state); | |
4670 | return; | |
4671 | } | |
4672 | ||
59fc630b | 4673 | if (clear_batch_ready(sh) ) { |
4674 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4675 | return; | |
4676 | } | |
4677 | ||
4e3d62ff | 4678 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4679 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4680 | |
dabc4ec6 | 4681 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd | 4682 | spin_lock(&sh->stripe_lock); |
5ddf0440 SL |
4683 | /* |
4684 | * Cannot process 'sync' concurrently with 'discard'. | |
4685 | * Flush data in r5cache before 'sync'. | |
4686 | */ | |
4687 | if (!test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state) && | |
4688 | !test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) && | |
4689 | !test_bit(STRIPE_DISCARD, &sh->state) && | |
f8dfcffd N |
4690 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { |
4691 | set_bit(STRIPE_SYNCING, &sh->state); | |
4692 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4693 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4694 | } |
4695 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4696 | } |
4697 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4698 | ||
4699 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4700 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4701 | (unsigned long long)sh->sector, sh->state, | |
4702 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4703 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4704 | |
acfe726b | 4705 | analyse_stripe(sh, &s); |
c5a31000 | 4706 | |
b70abcb2 SL |
4707 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4708 | goto finish; | |
4709 | ||
16d997b7 N |
4710 | if (s.handle_bad_blocks || |
4711 | test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { | |
bc2607f3 N |
4712 | set_bit(STRIPE_HANDLE, &sh->state); |
4713 | goto finish; | |
4714 | } | |
4715 | ||
474af965 N |
4716 | if (unlikely(s.blocked_rdev)) { |
4717 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4718 | s.replacing || s.to_write || s.written) { |
474af965 N |
4719 | set_bit(STRIPE_HANDLE, &sh->state); |
4720 | goto finish; | |
4721 | } | |
4722 | /* There is nothing for the blocked_rdev to block */ | |
4723 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4724 | s.blocked_rdev = NULL; | |
4725 | } | |
4726 | ||
4727 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4728 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4729 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4730 | } | |
4731 | ||
4732 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4733 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4734 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4735 | s.failed_num[0], s.failed_num[1]); | |
70d466f7 SL |
4736 | /* |
4737 | * check if the array has lost more than max_degraded devices and, | |
474af965 | 4738 | * if so, some requests might need to be failed. |
70d466f7 SL |
4739 | * |
4740 | * When journal device failed (log_failed), we will only process | |
4741 | * the stripe if there is data need write to raid disks | |
474af965 | 4742 | */ |
70d466f7 SL |
4743 | if (s.failed > conf->max_degraded || |
4744 | (s.log_failed && s.injournal == 0)) { | |
9a3f530f N |
4745 | sh->check_state = 0; |
4746 | sh->reconstruct_state = 0; | |
626f2092 | 4747 | break_stripe_batch_list(sh, 0); |
9a3f530f | 4748 | if (s.to_read+s.to_write+s.written) |
bd83d0a2 | 4749 | handle_failed_stripe(conf, sh, &s, disks); |
9a3e1101 | 4750 | if (s.syncing + s.replacing) |
9a3f530f N |
4751 | handle_failed_sync(conf, sh, &s); |
4752 | } | |
474af965 | 4753 | |
84789554 N |
4754 | /* Now we check to see if any write operations have recently |
4755 | * completed | |
4756 | */ | |
4757 | prexor = 0; | |
4758 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4759 | prexor = 1; | |
4760 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4761 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4762 | sh->reconstruct_state = reconstruct_state_idle; | |
4763 | ||
4764 | /* All the 'written' buffers and the parity block are ready to | |
4765 | * be written back to disk | |
4766 | */ | |
9e444768 SL |
4767 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4768 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4769 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4770 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4771 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4772 | for (i = disks; i--; ) { |
4773 | struct r5dev *dev = &sh->dev[i]; | |
4774 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4775 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4776 | dev->written || test_bit(R5_InJournal, |
4777 | &dev->flags))) { | |
84789554 N |
4778 | pr_debug("Writing block %d\n", i); |
4779 | set_bit(R5_Wantwrite, &dev->flags); | |
4780 | if (prexor) | |
4781 | continue; | |
9c4bdf69 N |
4782 | if (s.failed > 1) |
4783 | continue; | |
84789554 N |
4784 | if (!test_bit(R5_Insync, &dev->flags) || |
4785 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4786 | s.failed == 0)) | |
4787 | set_bit(STRIPE_INSYNC, &sh->state); | |
4788 | } | |
4789 | } | |
4790 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4791 | s.dec_preread_active = 1; | |
4792 | } | |
4793 | ||
ef5b7c69 N |
4794 | /* |
4795 | * might be able to return some write requests if the parity blocks | |
4796 | * are safe, or on a failed drive | |
4797 | */ | |
4798 | pdev = &sh->dev[sh->pd_idx]; | |
4799 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4800 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4801 | qdev = &sh->dev[sh->qd_idx]; | |
4802 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4803 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4804 | || conf->level < 6; | |
4805 | ||
4806 | if (s.written && | |
4807 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4808 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4809 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4810 | test_bit(R5_Discard, &pdev->flags))))) && | |
4811 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4812 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4813 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4814 | test_bit(R5_Discard, &qdev->flags)))))) | |
bd83d0a2 | 4815 | handle_stripe_clean_event(conf, sh, disks); |
ef5b7c69 | 4816 | |
1e6d690b | 4817 | if (s.just_cached) |
bd83d0a2 | 4818 | r5c_handle_cached_data_endio(conf, sh, disks); |
ff875738 | 4819 | log_stripe_write_finished(sh); |
1e6d690b | 4820 | |
ef5b7c69 N |
4821 | /* Now we might consider reading some blocks, either to check/generate |
4822 | * parity, or to satisfy requests | |
4823 | * or to load a block that is being partially written. | |
4824 | */ | |
4825 | if (s.to_read || s.non_overwrite | |
4826 | || (conf->level == 6 && s.to_write && s.failed) | |
4827 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4828 | || s.replacing | |
4829 | || s.expanding) | |
4830 | handle_stripe_fill(sh, &s, disks); | |
4831 | ||
2ded3703 SL |
4832 | /* |
4833 | * When the stripe finishes full journal write cycle (write to journal | |
4834 | * and raid disk), this is the clean up procedure so it is ready for | |
4835 | * next operation. | |
4836 | */ | |
4837 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4838 | ||
4839 | /* | |
4840 | * Now to consider new write requests, cache write back and what else, | |
4841 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4842 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4843 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4844 | * block. | |
2ded3703 | 4845 | * 3/ A r5c cache log write is in flight. |
84789554 | 4846 | */ |
2ded3703 SL |
4847 | |
4848 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4849 | if (!r5c_is_writeback(conf->log)) { | |
4850 | if (s.to_write) | |
4851 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4852 | } else { /* write back cache */ | |
4853 | int ret = 0; | |
4854 | ||
4855 | /* First, try handle writes in caching phase */ | |
4856 | if (s.to_write) | |
4857 | ret = r5c_try_caching_write(conf, sh, &s, | |
4858 | disks); | |
4859 | /* | |
4860 | * If caching phase failed: ret == -EAGAIN | |
4861 | * OR | |
4862 | * stripe under reclaim: !caching && injournal | |
4863 | * | |
4864 | * fall back to handle_stripe_dirtying() | |
4865 | */ | |
4866 | if (ret == -EAGAIN || | |
4867 | /* stripe under reclaim: !caching && injournal */ | |
4868 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4869 | s.injournal > 0)) { |
4870 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4871 | disks); | |
4872 | if (ret == -EAGAIN) | |
4873 | goto finish; | |
4874 | } | |
2ded3703 SL |
4875 | } |
4876 | } | |
84789554 N |
4877 | |
4878 | /* maybe we need to check and possibly fix the parity for this stripe | |
4879 | * Any reads will already have been scheduled, so we just see if enough | |
4880 | * data is available. The parity check is held off while parity | |
4881 | * dependent operations are in flight. | |
4882 | */ | |
4883 | if (sh->check_state || | |
4884 | (s.syncing && s.locked == 0 && | |
4885 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4886 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4887 | if (conf->level == 6) | |
4888 | handle_parity_checks6(conf, sh, &s, disks); | |
4889 | else | |
4890 | handle_parity_checks5(conf, sh, &s, disks); | |
4891 | } | |
c5a31000 | 4892 | |
f94c0b66 N |
4893 | if ((s.replacing || s.syncing) && s.locked == 0 |
4894 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4895 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4896 | /* Write out to replacement devices where possible */ |
4897 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4898 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4899 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4900 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4901 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4902 | s.locked++; | |
4903 | } | |
f94c0b66 N |
4904 | if (s.replacing) |
4905 | set_bit(STRIPE_INSYNC, &sh->state); | |
4906 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4907 | } |
4908 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4909 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4910 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4911 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4912 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4913 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4914 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4915 | } |
4916 | ||
4917 | /* If the failed drives are just a ReadError, then we might need | |
4918 | * to progress the repair/check process | |
4919 | */ | |
4920 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4921 | for (i = 0; i < s.failed; i++) { | |
4922 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4923 | if (test_bit(R5_ReadError, &dev->flags) | |
4924 | && !test_bit(R5_LOCKED, &dev->flags) | |
4925 | && test_bit(R5_UPTODATE, &dev->flags) | |
4926 | ) { | |
4927 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4928 | set_bit(R5_Wantwrite, &dev->flags); | |
4929 | set_bit(R5_ReWrite, &dev->flags); | |
4930 | set_bit(R5_LOCKED, &dev->flags); | |
4931 | s.locked++; | |
4932 | } else { | |
4933 | /* let's read it back */ | |
4934 | set_bit(R5_Wantread, &dev->flags); | |
4935 | set_bit(R5_LOCKED, &dev->flags); | |
4936 | s.locked++; | |
4937 | } | |
4938 | } | |
4939 | } | |
4940 | ||
3687c061 N |
4941 | /* Finish reconstruct operations initiated by the expansion process */ |
4942 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4943 | struct stripe_head *sh_src | |
6d036f7d | 4944 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4945 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4946 | /* sh cannot be written until sh_src has been read. | |
4947 | * so arrange for sh to be delayed a little | |
4948 | */ | |
4949 | set_bit(STRIPE_DELAYED, &sh->state); | |
4950 | set_bit(STRIPE_HANDLE, &sh->state); | |
4951 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4952 | &sh_src->state)) | |
4953 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4954 | raid5_release_stripe(sh_src); |
3687c061 N |
4955 | goto finish; |
4956 | } | |
4957 | if (sh_src) | |
6d036f7d | 4958 | raid5_release_stripe(sh_src); |
3687c061 N |
4959 | |
4960 | sh->reconstruct_state = reconstruct_state_idle; | |
4961 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4962 | for (i = conf->raid_disks; i--; ) { | |
4963 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4964 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4965 | s.locked++; | |
4966 | } | |
4967 | } | |
f416885e | 4968 | |
3687c061 N |
4969 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4970 | !sh->reconstruct_state) { | |
4971 | /* Need to write out all blocks after computing parity */ | |
4972 | sh->disks = conf->raid_disks; | |
4973 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4974 | schedule_reconstruction(sh, &s, 1, 1); | |
4975 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4976 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4977 | atomic_dec(&conf->reshape_stripes); | |
4978 | wake_up(&conf->wait_for_overlap); | |
4979 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4980 | } | |
4981 | ||
4982 | if (s.expanding && s.locked == 0 && | |
4983 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4984 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4985 | |
3687c061 | 4986 | finish: |
6bfe0b49 | 4987 | /* wait for this device to become unblocked */ |
5f066c63 N |
4988 | if (unlikely(s.blocked_rdev)) { |
4989 | if (conf->mddev->external) | |
4990 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4991 | conf->mddev); | |
4992 | else | |
4993 | /* Internal metadata will immediately | |
4994 | * be written by raid5d, so we don't | |
4995 | * need to wait here. | |
4996 | */ | |
4997 | rdev_dec_pending(s.blocked_rdev, | |
4998 | conf->mddev); | |
4999 | } | |
6bfe0b49 | 5000 | |
bc2607f3 N |
5001 | if (s.handle_bad_blocks) |
5002 | for (i = disks; i--; ) { | |
3cb03002 | 5003 | struct md_rdev *rdev; |
bc2607f3 N |
5004 | struct r5dev *dev = &sh->dev[i]; |
5005 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
5006 | /* We own a safe reference to the rdev */ | |
5007 | rdev = conf->disks[i].rdev; | |
5008 | if (!rdev_set_badblocks(rdev, sh->sector, | |
5009 | STRIPE_SECTORS, 0)) | |
5010 | md_error(conf->mddev, rdev); | |
5011 | rdev_dec_pending(rdev, conf->mddev); | |
5012 | } | |
b84db560 N |
5013 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
5014 | rdev = conf->disks[i].rdev; | |
5015 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 5016 | STRIPE_SECTORS, 0); |
b84db560 N |
5017 | rdev_dec_pending(rdev, conf->mddev); |
5018 | } | |
977df362 N |
5019 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
5020 | rdev = conf->disks[i].replacement; | |
dd054fce N |
5021 | if (!rdev) |
5022 | /* rdev have been moved down */ | |
5023 | rdev = conf->disks[i].rdev; | |
977df362 | 5024 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 5025 | STRIPE_SECTORS, 0); |
977df362 N |
5026 | rdev_dec_pending(rdev, conf->mddev); |
5027 | } | |
bc2607f3 N |
5028 | } |
5029 | ||
6c0069c0 YT |
5030 | if (s.ops_request) |
5031 | raid_run_ops(sh, s.ops_request); | |
5032 | ||
f0e43bcd | 5033 | ops_run_io(sh, &s); |
16a53ecc | 5034 | |
c5709ef6 | 5035 | if (s.dec_preread_active) { |
729a1866 | 5036 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 5037 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
5038 | * have actually been submitted. |
5039 | */ | |
5040 | atomic_dec(&conf->preread_active_stripes); | |
5041 | if (atomic_read(&conf->preread_active_stripes) < | |
5042 | IO_THRESHOLD) | |
5043 | md_wakeup_thread(conf->mddev->thread); | |
5044 | } | |
5045 | ||
257a4b42 | 5046 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
5047 | } |
5048 | ||
d1688a6d | 5049 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
5050 | { |
5051 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
5052 | while (!list_empty(&conf->delayed_list)) { | |
5053 | struct list_head *l = conf->delayed_list.next; | |
5054 | struct stripe_head *sh; | |
5055 | sh = list_entry(l, struct stripe_head, lru); | |
5056 | list_del_init(l); | |
5057 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5058 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5059 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 5060 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 5061 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 5062 | } |
482c0834 | 5063 | } |
16a53ecc N |
5064 | } |
5065 | ||
566c09c5 SL |
5066 | static void activate_bit_delay(struct r5conf *conf, |
5067 | struct list_head *temp_inactive_list) | |
16a53ecc N |
5068 | { |
5069 | /* device_lock is held */ | |
5070 | struct list_head head; | |
5071 | list_add(&head, &conf->bitmap_list); | |
5072 | list_del_init(&conf->bitmap_list); | |
5073 | while (!list_empty(&head)) { | |
5074 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 5075 | int hash; |
16a53ecc N |
5076 | list_del_init(&sh->lru); |
5077 | atomic_inc(&sh->count); | |
566c09c5 SL |
5078 | hash = sh->hash_lock_index; |
5079 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
5080 | } |
5081 | } | |
5082 | ||
5c675f83 | 5083 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 5084 | { |
d1688a6d | 5085 | struct r5conf *conf = mddev->private; |
f022b2fd N |
5086 | |
5087 | /* No difference between reads and writes. Just check | |
5088 | * how busy the stripe_cache is | |
5089 | */ | |
3fa841d7 | 5090 | |
5423399a | 5091 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd | 5092 | return 1; |
a39f7afd SL |
5093 | |
5094 | /* Also checks whether there is pressure on r5cache log space */ | |
5095 | if (test_bit(R5C_LOG_TIGHT, &conf->cache_state)) | |
5096 | return 1; | |
f022b2fd N |
5097 | if (conf->quiesce) |
5098 | return 1; | |
4bda556a | 5099 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
5100 | return 1; |
5101 | ||
5102 | return 0; | |
5103 | } | |
5104 | ||
fd01b88c | 5105 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 5106 | { |
3cb5edf4 | 5107 | struct r5conf *conf = mddev->private; |
10433d04 | 5108 | sector_t sector = bio->bi_iter.bi_sector; |
3cb5edf4 | 5109 | unsigned int chunk_sectors; |
aa8b57aa | 5110 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 5111 | |
10433d04 CH |
5112 | WARN_ON_ONCE(bio->bi_partno); |
5113 | ||
3cb5edf4 | 5114 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
5115 | return chunk_sectors >= |
5116 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
5117 | } | |
5118 | ||
46031f9a RBJ |
5119 | /* |
5120 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
5121 | * later sampled by raid5d. | |
5122 | */ | |
d1688a6d | 5123 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
5124 | { |
5125 | unsigned long flags; | |
5126 | ||
5127 | spin_lock_irqsave(&conf->device_lock, flags); | |
5128 | ||
5129 | bi->bi_next = conf->retry_read_aligned_list; | |
5130 | conf->retry_read_aligned_list = bi; | |
5131 | ||
5132 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
5133 | md_wakeup_thread(conf->mddev->thread); | |
5134 | } | |
5135 | ||
0472a42b N |
5136 | static struct bio *remove_bio_from_retry(struct r5conf *conf, |
5137 | unsigned int *offset) | |
46031f9a RBJ |
5138 | { |
5139 | struct bio *bi; | |
5140 | ||
5141 | bi = conf->retry_read_aligned; | |
5142 | if (bi) { | |
0472a42b | 5143 | *offset = conf->retry_read_offset; |
46031f9a RBJ |
5144 | conf->retry_read_aligned = NULL; |
5145 | return bi; | |
5146 | } | |
5147 | bi = conf->retry_read_aligned_list; | |
5148 | if(bi) { | |
387bb173 | 5149 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5150 | bi->bi_next = NULL; |
0472a42b | 5151 | *offset = 0; |
46031f9a RBJ |
5152 | } |
5153 | ||
5154 | return bi; | |
5155 | } | |
5156 | ||
f679623f RBJ |
5157 | /* |
5158 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5159 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5160 | * first). | |
5161 | * If the read failed.. | |
5162 | */ | |
4246a0b6 | 5163 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5164 | { |
5165 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5166 | struct mddev *mddev; |
d1688a6d | 5167 | struct r5conf *conf; |
3cb03002 | 5168 | struct md_rdev *rdev; |
4e4cbee9 | 5169 | blk_status_t error = bi->bi_status; |
46031f9a | 5170 | |
f679623f | 5171 | bio_put(bi); |
46031f9a | 5172 | |
46031f9a RBJ |
5173 | rdev = (void*)raid_bi->bi_next; |
5174 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5175 | mddev = rdev->mddev; |
5176 | conf = mddev->private; | |
46031f9a RBJ |
5177 | |
5178 | rdev_dec_pending(rdev, conf->mddev); | |
5179 | ||
9b81c842 | 5180 | if (!error) { |
4246a0b6 | 5181 | bio_endio(raid_bi); |
46031f9a | 5182 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5183 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5184 | return; |
46031f9a RBJ |
5185 | } |
5186 | ||
45b4233c | 5187 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5188 | |
5189 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5190 | } |
5191 | ||
7ef6b12a | 5192 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5193 | { |
d1688a6d | 5194 | struct r5conf *conf = mddev->private; |
8553fe7e | 5195 | int dd_idx; |
f679623f | 5196 | struct bio* align_bi; |
3cb03002 | 5197 | struct md_rdev *rdev; |
671488cc | 5198 | sector_t end_sector; |
f679623f RBJ |
5199 | |
5200 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5201 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5202 | return 0; |
5203 | } | |
5204 | /* | |
d7a10308 | 5205 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5206 | */ |
afeee514 | 5207 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, &mddev->bio_set); |
f679623f RBJ |
5208 | if (!align_bi) |
5209 | return 0; | |
5210 | /* | |
5211 | * set bi_end_io to a new function, and set bi_private to the | |
5212 | * original bio. | |
5213 | */ | |
5214 | align_bi->bi_end_io = raid5_align_endio; | |
5215 | align_bi->bi_private = raid_bio; | |
5216 | /* | |
5217 | * compute position | |
5218 | */ | |
4f024f37 KO |
5219 | align_bi->bi_iter.bi_sector = |
5220 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5221 | 0, &dd_idx, NULL); | |
f679623f | 5222 | |
f73a1c7d | 5223 | end_sector = bio_end_sector(align_bi); |
f679623f | 5224 | rcu_read_lock(); |
671488cc N |
5225 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5226 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5227 | rdev->recovery_offset < end_sector) { | |
5228 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5229 | if (rdev && | |
5230 | (test_bit(Faulty, &rdev->flags) || | |
5231 | !(test_bit(In_sync, &rdev->flags) || | |
5232 | rdev->recovery_offset >= end_sector))) | |
5233 | rdev = NULL; | |
5234 | } | |
03b047f4 SL |
5235 | |
5236 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5237 | rcu_read_unlock(); | |
5238 | bio_put(align_bi); | |
5239 | return 0; | |
5240 | } | |
5241 | ||
671488cc | 5242 | if (rdev) { |
31c176ec N |
5243 | sector_t first_bad; |
5244 | int bad_sectors; | |
5245 | ||
f679623f RBJ |
5246 | atomic_inc(&rdev->nr_pending); |
5247 | rcu_read_unlock(); | |
46031f9a | 5248 | raid_bio->bi_next = (void*)rdev; |
74d46992 | 5249 | bio_set_dev(align_bi, rdev->bdev); |
b7c44ed9 | 5250 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 5251 | |
7140aafc | 5252 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5253 | bio_sectors(align_bi), |
31c176ec | 5254 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5255 | bio_put(align_bi); |
5256 | rdev_dec_pending(rdev, mddev); | |
5257 | return 0; | |
5258 | } | |
5259 | ||
6c0544e2 | 5260 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5261 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5262 | |
46031f9a | 5263 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5264 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5265 | conf->quiesce == 0, |
eed8c02e | 5266 | conf->device_lock); |
46031f9a RBJ |
5267 | atomic_inc(&conf->active_aligned_reads); |
5268 | spin_unlock_irq(&conf->device_lock); | |
5269 | ||
e3620a3a | 5270 | if (mddev->gendisk) |
74d46992 | 5271 | trace_block_bio_remap(align_bi->bi_disk->queue, |
e3620a3a | 5272 | align_bi, disk_devt(mddev->gendisk), |
4f024f37 | 5273 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
5274 | generic_make_request(align_bi); |
5275 | return 1; | |
5276 | } else { | |
5277 | rcu_read_unlock(); | |
46031f9a | 5278 | bio_put(align_bi); |
f679623f RBJ |
5279 | return 0; |
5280 | } | |
5281 | } | |
5282 | ||
7ef6b12a ML |
5283 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5284 | { | |
5285 | struct bio *split; | |
dd7a8f5d N |
5286 | sector_t sector = raid_bio->bi_iter.bi_sector; |
5287 | unsigned chunk_sects = mddev->chunk_sectors; | |
5288 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
7ef6b12a | 5289 | |
dd7a8f5d N |
5290 | if (sectors < bio_sectors(raid_bio)) { |
5291 | struct r5conf *conf = mddev->private; | |
afeee514 | 5292 | split = bio_split(raid_bio, sectors, GFP_NOIO, &conf->bio_split); |
dd7a8f5d N |
5293 | bio_chain(split, raid_bio); |
5294 | generic_make_request(raid_bio); | |
5295 | raid_bio = split; | |
5296 | } | |
7ef6b12a | 5297 | |
dd7a8f5d N |
5298 | if (!raid5_read_one_chunk(mddev, raid_bio)) |
5299 | return raid_bio; | |
7ef6b12a ML |
5300 | |
5301 | return NULL; | |
5302 | } | |
5303 | ||
8b3e6cdc DW |
5304 | /* __get_priority_stripe - get the next stripe to process |
5305 | * | |
5306 | * Full stripe writes are allowed to pass preread active stripes up until | |
5307 | * the bypass_threshold is exceeded. In general the bypass_count | |
5308 | * increments when the handle_list is handled before the hold_list; however, it | |
5309 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5310 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5311 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5312 | * handle_list. | |
5313 | */ | |
851c30c9 | 5314 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5315 | { |
535ae4eb | 5316 | struct stripe_head *sh, *tmp; |
851c30c9 | 5317 | struct list_head *handle_list = NULL; |
535ae4eb | 5318 | struct r5worker_group *wg; |
70d466f7 SL |
5319 | bool second_try = !r5c_is_writeback(conf->log) && |
5320 | !r5l_log_disk_error(conf); | |
5321 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state) || | |
5322 | r5l_log_disk_error(conf); | |
851c30c9 | 5323 | |
535ae4eb SL |
5324 | again: |
5325 | wg = NULL; | |
5326 | sh = NULL; | |
851c30c9 | 5327 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5328 | handle_list = try_loprio ? &conf->loprio_list : |
5329 | &conf->handle_list; | |
851c30c9 | 5330 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5331 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5332 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5333 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5334 | } else { |
5335 | int i; | |
5336 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5337 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5338 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5339 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5340 | if (!list_empty(handle_list)) |
5341 | break; | |
5342 | } | |
5343 | } | |
8b3e6cdc DW |
5344 | |
5345 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5346 | __func__, | |
851c30c9 | 5347 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5348 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5349 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5350 | ||
851c30c9 SL |
5351 | if (!list_empty(handle_list)) { |
5352 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5353 | |
5354 | if (list_empty(&conf->hold_list)) | |
5355 | conf->bypass_count = 0; | |
5356 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5357 | if (conf->hold_list.next == conf->last_hold) | |
5358 | conf->bypass_count++; | |
5359 | else { | |
5360 | conf->last_hold = conf->hold_list.next; | |
5361 | conf->bypass_count -= conf->bypass_threshold; | |
5362 | if (conf->bypass_count < 0) | |
5363 | conf->bypass_count = 0; | |
5364 | } | |
5365 | } | |
5366 | } else if (!list_empty(&conf->hold_list) && | |
5367 | ((conf->bypass_threshold && | |
5368 | conf->bypass_count > conf->bypass_threshold) || | |
5369 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5370 | |
5371 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5372 | if (conf->worker_cnt_per_group == 0 || | |
5373 | group == ANY_GROUP || | |
5374 | !cpu_online(tmp->cpu) || | |
5375 | cpu_to_group(tmp->cpu) == group) { | |
5376 | sh = tmp; | |
5377 | break; | |
5378 | } | |
5379 | } | |
5380 | ||
5381 | if (sh) { | |
5382 | conf->bypass_count -= conf->bypass_threshold; | |
5383 | if (conf->bypass_count < 0) | |
5384 | conf->bypass_count = 0; | |
5385 | } | |
bfc90cb0 | 5386 | wg = NULL; |
851c30c9 SL |
5387 | } |
5388 | ||
535ae4eb SL |
5389 | if (!sh) { |
5390 | if (second_try) | |
5391 | return NULL; | |
5392 | second_try = true; | |
5393 | try_loprio = !try_loprio; | |
5394 | goto again; | |
5395 | } | |
8b3e6cdc | 5396 | |
bfc90cb0 SL |
5397 | if (wg) { |
5398 | wg->stripes_cnt--; | |
5399 | sh->group = NULL; | |
5400 | } | |
8b3e6cdc | 5401 | list_del_init(&sh->lru); |
c7a6d35e | 5402 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5403 | return sh; |
5404 | } | |
f679623f | 5405 | |
8811b596 SL |
5406 | struct raid5_plug_cb { |
5407 | struct blk_plug_cb cb; | |
5408 | struct list_head list; | |
566c09c5 | 5409 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5410 | }; |
5411 | ||
5412 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5413 | { | |
5414 | struct raid5_plug_cb *cb = container_of( | |
5415 | blk_cb, struct raid5_plug_cb, cb); | |
5416 | struct stripe_head *sh; | |
5417 | struct mddev *mddev = cb->cb.data; | |
5418 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5419 | int cnt = 0; |
566c09c5 | 5420 | int hash; |
8811b596 SL |
5421 | |
5422 | if (cb->list.next && !list_empty(&cb->list)) { | |
5423 | spin_lock_irq(&conf->device_lock); | |
5424 | while (!list_empty(&cb->list)) { | |
5425 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5426 | list_del_init(&sh->lru); | |
5427 | /* | |
5428 | * avoid race release_stripe_plug() sees | |
5429 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5430 | * is still in our list | |
5431 | */ | |
4e857c58 | 5432 | smp_mb__before_atomic(); |
8811b596 | 5433 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5434 | /* |
5435 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5436 | * case, the count is always > 1 here | |
5437 | */ | |
566c09c5 SL |
5438 | hash = sh->hash_lock_index; |
5439 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5440 | cnt++; |
8811b596 SL |
5441 | } |
5442 | spin_unlock_irq(&conf->device_lock); | |
5443 | } | |
566c09c5 SL |
5444 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5445 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5446 | if (mddev->queue) |
5447 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5448 | kfree(cb); |
5449 | } | |
5450 | ||
5451 | static void release_stripe_plug(struct mddev *mddev, | |
5452 | struct stripe_head *sh) | |
5453 | { | |
5454 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5455 | raid5_unplug, mddev, | |
5456 | sizeof(struct raid5_plug_cb)); | |
5457 | struct raid5_plug_cb *cb; | |
5458 | ||
5459 | if (!blk_cb) { | |
6d036f7d | 5460 | raid5_release_stripe(sh); |
8811b596 SL |
5461 | return; |
5462 | } | |
5463 | ||
5464 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5465 | ||
566c09c5 SL |
5466 | if (cb->list.next == NULL) { |
5467 | int i; | |
8811b596 | 5468 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5469 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5470 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5471 | } | |
8811b596 SL |
5472 | |
5473 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5474 | list_add_tail(&sh->lru, &cb->list); | |
5475 | else | |
6d036f7d | 5476 | raid5_release_stripe(sh); |
8811b596 SL |
5477 | } |
5478 | ||
620125f2 SL |
5479 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5480 | { | |
5481 | struct r5conf *conf = mddev->private; | |
5482 | sector_t logical_sector, last_sector; | |
5483 | struct stripe_head *sh; | |
620125f2 SL |
5484 | int stripe_sectors; |
5485 | ||
5486 | if (mddev->reshape_position != MaxSector) | |
5487 | /* Skip discard while reshape is happening */ | |
5488 | return; | |
5489 | ||
4f024f37 KO |
5490 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5491 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5492 | |
5493 | bi->bi_next = NULL; | |
620125f2 SL |
5494 | |
5495 | stripe_sectors = conf->chunk_sectors * | |
5496 | (conf->raid_disks - conf->max_degraded); | |
5497 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5498 | stripe_sectors); | |
5499 | sector_div(last_sector, stripe_sectors); | |
5500 | ||
5501 | logical_sector *= conf->chunk_sectors; | |
5502 | last_sector *= conf->chunk_sectors; | |
5503 | ||
5504 | for (; logical_sector < last_sector; | |
5505 | logical_sector += STRIPE_SECTORS) { | |
5506 | DEFINE_WAIT(w); | |
5507 | int d; | |
5508 | again: | |
6d036f7d | 5509 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5510 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5511 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5512 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5513 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5514 | raid5_release_stripe(sh); |
f8dfcffd N |
5515 | schedule(); |
5516 | goto again; | |
5517 | } | |
5518 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5519 | spin_lock_irq(&sh->stripe_lock); |
5520 | for (d = 0; d < conf->raid_disks; d++) { | |
5521 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5522 | continue; | |
5523 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5524 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5525 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5526 | raid5_release_stripe(sh); |
620125f2 SL |
5527 | schedule(); |
5528 | goto again; | |
5529 | } | |
5530 | } | |
f8dfcffd | 5531 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5532 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5533 | sh->overwrite_disks = 0; |
620125f2 SL |
5534 | for (d = 0; d < conf->raid_disks; d++) { |
5535 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5536 | continue; | |
5537 | sh->dev[d].towrite = bi; | |
5538 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
016c76ac | 5539 | bio_inc_remaining(bi); |
49728050 | 5540 | md_write_inc(mddev, bi); |
7a87f434 | 5541 | sh->overwrite_disks++; |
620125f2 SL |
5542 | } |
5543 | spin_unlock_irq(&sh->stripe_lock); | |
5544 | if (conf->mddev->bitmap) { | |
5545 | for (d = 0; | |
5546 | d < conf->raid_disks - conf->max_degraded; | |
5547 | d++) | |
e64e4018 AS |
5548 | md_bitmap_startwrite(mddev->bitmap, |
5549 | sh->sector, | |
5550 | STRIPE_SECTORS, | |
5551 | 0); | |
620125f2 SL |
5552 | sh->bm_seq = conf->seq_flush + 1; |
5553 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5554 | } | |
5555 | ||
5556 | set_bit(STRIPE_HANDLE, &sh->state); | |
5557 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5558 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5559 | atomic_inc(&conf->preread_active_stripes); | |
5560 | release_stripe_plug(mddev, sh); | |
5561 | } | |
5562 | ||
016c76ac | 5563 | bio_endio(bi); |
620125f2 SL |
5564 | } |
5565 | ||
cc27b0c7 | 5566 | static bool raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5567 | { |
d1688a6d | 5568 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5569 | int dd_idx; |
1da177e4 LT |
5570 | sector_t new_sector; |
5571 | sector_t logical_sector, last_sector; | |
5572 | struct stripe_head *sh; | |
a362357b | 5573 | const int rw = bio_data_dir(bi); |
27c0f68f SL |
5574 | DEFINE_WAIT(w); |
5575 | bool do_prepare; | |
3bddb7f8 | 5576 | bool do_flush = false; |
1da177e4 | 5577 | |
1eff9d32 | 5578 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
1532d9e8 | 5579 | int ret = log_handle_flush_request(conf, bi); |
828cbe98 SL |
5580 | |
5581 | if (ret == 0) | |
cc27b0c7 | 5582 | return true; |
828cbe98 SL |
5583 | if (ret == -ENODEV) { |
5584 | md_flush_request(mddev, bi); | |
cc27b0c7 | 5585 | return true; |
828cbe98 SL |
5586 | } |
5587 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5588 | /* |
5589 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5590 | * we need to flush journal device | |
5591 | */ | |
5592 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5593 | } |
5594 | ||
cc27b0c7 N |
5595 | if (!md_write_start(mddev, bi)) |
5596 | return false; | |
9ffc8f7c EM |
5597 | /* |
5598 | * If array is degraded, better not do chunk aligned read because | |
5599 | * later we might have to read it again in order to reconstruct | |
5600 | * data on failed drives. | |
5601 | */ | |
5602 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5603 | mddev->reshape_position == MaxSector) { |
5604 | bi = chunk_aligned_read(mddev, bi); | |
5605 | if (!bi) | |
cc27b0c7 | 5606 | return true; |
7ef6b12a | 5607 | } |
52488615 | 5608 | |
796a5cf0 | 5609 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 | 5610 | make_discard_request(mddev, bi); |
cc27b0c7 N |
5611 | md_write_end(mddev); |
5612 | return true; | |
620125f2 SL |
5613 | } |
5614 | ||
4f024f37 | 5615 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5616 | last_sector = bio_end_sector(bi); |
1da177e4 | 5617 | bi->bi_next = NULL; |
06d91a5f | 5618 | |
27c0f68f | 5619 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5620 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5621 | int previous; |
c46501b2 | 5622 | int seq; |
b578d55f | 5623 | |
27c0f68f | 5624 | do_prepare = false; |
7ecaa1e6 | 5625 | retry: |
c46501b2 | 5626 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5627 | previous = 0; |
27c0f68f SL |
5628 | if (do_prepare) |
5629 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5630 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5631 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5632 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5633 | * 64bit on a 32bit platform, and so it might be |
5634 | * possible to see a half-updated value | |
aeb878b0 | 5635 | * Of course reshape_progress could change after |
df8e7f76 N |
5636 | * the lock is dropped, so once we get a reference |
5637 | * to the stripe that we think it is, we will have | |
5638 | * to check again. | |
5639 | */ | |
7ecaa1e6 | 5640 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5641 | if (mddev->reshape_backwards |
fef9c61f N |
5642 | ? logical_sector < conf->reshape_progress |
5643 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5644 | previous = 1; |
5645 | } else { | |
2c810cdd | 5646 | if (mddev->reshape_backwards |
fef9c61f N |
5647 | ? logical_sector < conf->reshape_safe |
5648 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5649 | spin_unlock_irq(&conf->device_lock); |
5650 | schedule(); | |
27c0f68f | 5651 | do_prepare = true; |
b578d55f N |
5652 | goto retry; |
5653 | } | |
5654 | } | |
7ecaa1e6 N |
5655 | spin_unlock_irq(&conf->device_lock); |
5656 | } | |
16a53ecc | 5657 | |
112bf897 N |
5658 | new_sector = raid5_compute_sector(conf, logical_sector, |
5659 | previous, | |
911d4ee8 | 5660 | &dd_idx, NULL); |
849674e4 | 5661 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5662 | (unsigned long long)new_sector, |
1da177e4 LT |
5663 | (unsigned long long)logical_sector); |
5664 | ||
6d036f7d | 5665 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5666 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5667 | if (sh) { |
b0f9ec04 | 5668 | if (unlikely(previous)) { |
7ecaa1e6 | 5669 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5670 | * stripe, so we must do the range check again. |
5671 | * Expansion could still move past after this | |
5672 | * test, but as we are holding a reference to | |
5673 | * 'sh', we know that if that happens, | |
5674 | * STRIPE_EXPANDING will get set and the expansion | |
5675 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5676 | */ |
5677 | int must_retry = 0; | |
5678 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5679 | if (mddev->reshape_backwards |
b0f9ec04 N |
5680 | ? logical_sector >= conf->reshape_progress |
5681 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5682 | /* mismatch, need to try again */ |
5683 | must_retry = 1; | |
5684 | spin_unlock_irq(&conf->device_lock); | |
5685 | if (must_retry) { | |
6d036f7d | 5686 | raid5_release_stripe(sh); |
7a3ab908 | 5687 | schedule(); |
27c0f68f | 5688 | do_prepare = true; |
7ecaa1e6 N |
5689 | goto retry; |
5690 | } | |
5691 | } | |
c46501b2 N |
5692 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5693 | /* Might have got the wrong stripe_head | |
5694 | * by accident | |
5695 | */ | |
6d036f7d | 5696 | raid5_release_stripe(sh); |
c46501b2 N |
5697 | goto retry; |
5698 | } | |
e62e58a5 | 5699 | |
7ecaa1e6 | 5700 | if (test_bit(STRIPE_EXPANDING, &sh->state) || |
da41ba65 | 5701 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5702 | /* Stripe is busy expanding or |
5703 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5704 | * and wait a while |
5705 | */ | |
482c0834 | 5706 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5707 | raid5_release_stripe(sh); |
1da177e4 | 5708 | schedule(); |
27c0f68f | 5709 | do_prepare = true; |
1da177e4 LT |
5710 | goto retry; |
5711 | } | |
3bddb7f8 SL |
5712 | if (do_flush) { |
5713 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5714 | /* we only need flush for one stripe */ | |
5715 | do_flush = false; | |
5716 | } | |
5717 | ||
6ed3003c N |
5718 | set_bit(STRIPE_HANDLE, &sh->state); |
5719 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5720 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5721 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5722 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5723 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5724 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5725 | } else { |
5726 | /* cannot get stripe for read-ahead, just give-up */ | |
4e4cbee9 | 5727 | bi->bi_status = BLK_STS_IOERR; |
1da177e4 LT |
5728 | break; |
5729 | } | |
1da177e4 | 5730 | } |
27c0f68f | 5731 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5732 | |
49728050 N |
5733 | if (rw == WRITE) |
5734 | md_write_end(mddev); | |
016c76ac | 5735 | bio_endio(bi); |
cc27b0c7 | 5736 | return true; |
1da177e4 LT |
5737 | } |
5738 | ||
fd01b88c | 5739 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5740 | |
fd01b88c | 5741 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5742 | { |
52c03291 N |
5743 | /* reshaping is quite different to recovery/resync so it is |
5744 | * handled quite separately ... here. | |
5745 | * | |
5746 | * On each call to sync_request, we gather one chunk worth of | |
5747 | * destination stripes and flag them as expanding. | |
5748 | * Then we find all the source stripes and request reads. | |
5749 | * As the reads complete, handle_stripe will copy the data | |
5750 | * into the destination stripe and release that stripe. | |
5751 | */ | |
d1688a6d | 5752 | struct r5conf *conf = mddev->private; |
1da177e4 | 5753 | struct stripe_head *sh; |
db0505d3 | 5754 | struct md_rdev *rdev; |
ccfcc3c1 | 5755 | sector_t first_sector, last_sector; |
f416885e N |
5756 | int raid_disks = conf->previous_raid_disks; |
5757 | int data_disks = raid_disks - conf->max_degraded; | |
5758 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5759 | int i; |
5760 | int dd_idx; | |
c8f517c4 | 5761 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5762 | sector_t stripe_addr; |
7a661381 | 5763 | int reshape_sectors; |
ab69ae12 | 5764 | struct list_head stripes; |
92140480 | 5765 | sector_t retn; |
52c03291 | 5766 | |
fef9c61f N |
5767 | if (sector_nr == 0) { |
5768 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5769 | if (mddev->reshape_backwards && |
fef9c61f N |
5770 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5771 | sector_nr = raid5_size(mddev, 0, 0) | |
5772 | - conf->reshape_progress; | |
6cbd8148 N |
5773 | } else if (mddev->reshape_backwards && |
5774 | conf->reshape_progress == MaxSector) { | |
5775 | /* shouldn't happen, but just in case, finish up.*/ | |
5776 | sector_nr = MaxSector; | |
2c810cdd | 5777 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5778 | conf->reshape_progress > 0) |
5779 | sector_nr = conf->reshape_progress; | |
f416885e | 5780 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5781 | if (sector_nr) { |
8dee7211 N |
5782 | mddev->curr_resync_completed = sector_nr; |
5783 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5784 | *skipped = 1; |
92140480 N |
5785 | retn = sector_nr; |
5786 | goto finish; | |
fef9c61f | 5787 | } |
52c03291 N |
5788 | } |
5789 | ||
7a661381 N |
5790 | /* We need to process a full chunk at a time. |
5791 | * If old and new chunk sizes differ, we need to process the | |
5792 | * largest of these | |
5793 | */ | |
3cb5edf4 N |
5794 | |
5795 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5796 | |
b5254dd5 N |
5797 | /* We update the metadata at least every 10 seconds, or when |
5798 | * the data about to be copied would over-write the source of | |
5799 | * the data at the front of the range. i.e. one new_stripe | |
5800 | * along from reshape_progress new_maps to after where | |
5801 | * reshape_safe old_maps to | |
52c03291 | 5802 | */ |
fef9c61f | 5803 | writepos = conf->reshape_progress; |
f416885e | 5804 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5805 | readpos = conf->reshape_progress; |
5806 | sector_div(readpos, data_disks); | |
fef9c61f | 5807 | safepos = conf->reshape_safe; |
f416885e | 5808 | sector_div(safepos, data_disks); |
2c810cdd | 5809 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5810 | BUG_ON(writepos < reshape_sectors); |
5811 | writepos -= reshape_sectors; | |
c8f517c4 | 5812 | readpos += reshape_sectors; |
7a661381 | 5813 | safepos += reshape_sectors; |
fef9c61f | 5814 | } else { |
7a661381 | 5815 | writepos += reshape_sectors; |
c74c0d76 N |
5816 | /* readpos and safepos are worst-case calculations. |
5817 | * A negative number is overly pessimistic, and causes | |
5818 | * obvious problems for unsigned storage. So clip to 0. | |
5819 | */ | |
ed37d83e N |
5820 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5821 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5822 | } |
52c03291 | 5823 | |
b5254dd5 N |
5824 | /* Having calculated the 'writepos' possibly use it |
5825 | * to set 'stripe_addr' which is where we will write to. | |
5826 | */ | |
5827 | if (mddev->reshape_backwards) { | |
5828 | BUG_ON(conf->reshape_progress == 0); | |
5829 | stripe_addr = writepos; | |
5830 | BUG_ON((mddev->dev_sectors & | |
5831 | ~((sector_t)reshape_sectors - 1)) | |
5832 | - reshape_sectors - stripe_addr | |
5833 | != sector_nr); | |
5834 | } else { | |
5835 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5836 | stripe_addr = sector_nr; | |
5837 | } | |
5838 | ||
c8f517c4 N |
5839 | /* 'writepos' is the most advanced device address we might write. |
5840 | * 'readpos' is the least advanced device address we might read. | |
5841 | * 'safepos' is the least address recorded in the metadata as having | |
5842 | * been reshaped. | |
b5254dd5 N |
5843 | * If there is a min_offset_diff, these are adjusted either by |
5844 | * increasing the safepos/readpos if diff is negative, or | |
5845 | * increasing writepos if diff is positive. | |
5846 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5847 | * ensure safety in the face of a crash - that must be done by userspace |
5848 | * making a backup of the data. So in that case there is no particular | |
5849 | * rush to update metadata. | |
5850 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5851 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5852 | * we can be safe in the event of a crash. | |
5853 | * So we insist on updating metadata if safepos is behind writepos and | |
5854 | * readpos is beyond writepos. | |
5855 | * In any case, update the metadata every 10 seconds. | |
5856 | * Maybe that number should be configurable, but I'm not sure it is | |
5857 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5858 | */ | |
b5254dd5 N |
5859 | if (conf->min_offset_diff < 0) { |
5860 | safepos += -conf->min_offset_diff; | |
5861 | readpos += -conf->min_offset_diff; | |
5862 | } else | |
5863 | writepos += conf->min_offset_diff; | |
5864 | ||
2c810cdd | 5865 | if ((mddev->reshape_backwards |
c8f517c4 N |
5866 | ? (safepos > writepos && readpos < writepos) |
5867 | : (safepos < writepos && readpos > writepos)) || | |
5868 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5869 | /* Cannot proceed until we've updated the superblock... */ |
5870 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5871 | atomic_read(&conf->reshape_stripes)==0 |
5872 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5873 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5874 | return 0; | |
fef9c61f | 5875 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5876 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
5877 | if (!mddev->reshape_backwards) |
5878 | /* Can update recovery_offset */ | |
5879 | rdev_for_each(rdev, mddev) | |
5880 | if (rdev->raid_disk >= 0 && | |
5881 | !test_bit(Journal, &rdev->flags) && | |
5882 | !test_bit(In_sync, &rdev->flags) && | |
5883 | rdev->recovery_offset < sector_nr) | |
5884 | rdev->recovery_offset = sector_nr; | |
5885 | ||
c8f517c4 | 5886 | conf->reshape_checkpoint = jiffies; |
2953079c | 5887 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5888 | md_wakeup_thread(mddev->thread); |
2953079c | 5889 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5890 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5891 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5892 | return 0; | |
52c03291 | 5893 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5894 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5895 | spin_unlock_irq(&conf->device_lock); |
5896 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5897 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5898 | } |
5899 | ||
ab69ae12 | 5900 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5901 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5902 | int j; |
a9f326eb | 5903 | int skipped_disk = 0; |
6d036f7d | 5904 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5905 | set_bit(STRIPE_EXPANDING, &sh->state); |
5906 | atomic_inc(&conf->reshape_stripes); | |
5907 | /* If any of this stripe is beyond the end of the old | |
5908 | * array, then we need to zero those blocks | |
5909 | */ | |
5910 | for (j=sh->disks; j--;) { | |
5911 | sector_t s; | |
5912 | if (j == sh->pd_idx) | |
5913 | continue; | |
f416885e | 5914 | if (conf->level == 6 && |
d0dabf7e | 5915 | j == sh->qd_idx) |
f416885e | 5916 | continue; |
6d036f7d | 5917 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5918 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5919 | skipped_disk = 1; |
52c03291 N |
5920 | continue; |
5921 | } | |
5922 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5923 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5924 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5925 | } | |
a9f326eb | 5926 | if (!skipped_disk) { |
52c03291 N |
5927 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5928 | set_bit(STRIPE_HANDLE, &sh->state); | |
5929 | } | |
ab69ae12 | 5930 | list_add(&sh->lru, &stripes); |
52c03291 N |
5931 | } |
5932 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5933 | if (mddev->reshape_backwards) |
7a661381 | 5934 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5935 | else |
7a661381 | 5936 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5937 | spin_unlock_irq(&conf->device_lock); |
5938 | /* Ok, those stripe are ready. We can start scheduling | |
5939 | * reads on the source stripes. | |
5940 | * The source stripes are determined by mapping the first and last | |
5941 | * block on the destination stripes. | |
5942 | */ | |
52c03291 | 5943 | first_sector = |
ec32a2bd | 5944 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5945 | 1, &dd_idx, NULL); |
52c03291 | 5946 | last_sector = |
0e6e0271 | 5947 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5948 | * new_data_disks - 1), |
911d4ee8 | 5949 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5950 | if (last_sector >= mddev->dev_sectors) |
5951 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5952 | while (first_sector <= last_sector) { |
6d036f7d | 5953 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5954 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5955 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5956 | raid5_release_stripe(sh); |
52c03291 N |
5957 | first_sector += STRIPE_SECTORS; |
5958 | } | |
ab69ae12 N |
5959 | /* Now that the sources are clearly marked, we can release |
5960 | * the destination stripes | |
5961 | */ | |
5962 | while (!list_empty(&stripes)) { | |
5963 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5964 | list_del_init(&sh->lru); | |
6d036f7d | 5965 | raid5_release_stripe(sh); |
ab69ae12 | 5966 | } |
c6207277 N |
5967 | /* If this takes us to the resync_max point where we have to pause, |
5968 | * then we need to write out the superblock. | |
5969 | */ | |
7a661381 | 5970 | sector_nr += reshape_sectors; |
92140480 N |
5971 | retn = reshape_sectors; |
5972 | finish: | |
c5e19d90 N |
5973 | if (mddev->curr_resync_completed > mddev->resync_max || |
5974 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5975 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 N |
5976 | /* Cannot proceed until we've updated the superblock... */ |
5977 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5978 | atomic_read(&conf->reshape_stripes) == 0 |
5979 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5980 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5981 | goto ret; | |
fef9c61f | 5982 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5983 | mddev->curr_resync_completed = sector_nr; |
db0505d3 N |
5984 | if (!mddev->reshape_backwards) |
5985 | /* Can update recovery_offset */ | |
5986 | rdev_for_each(rdev, mddev) | |
5987 | if (rdev->raid_disk >= 0 && | |
5988 | !test_bit(Journal, &rdev->flags) && | |
5989 | !test_bit(In_sync, &rdev->flags) && | |
5990 | rdev->recovery_offset < sector_nr) | |
5991 | rdev->recovery_offset = sector_nr; | |
c8f517c4 | 5992 | conf->reshape_checkpoint = jiffies; |
2953079c | 5993 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
5994 | md_wakeup_thread(mddev->thread); |
5995 | wait_event(mddev->sb_wait, | |
2953079c | 5996 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
5997 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5998 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5999 | goto ret; | |
c6207277 | 6000 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 6001 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
6002 | spin_unlock_irq(&conf->device_lock); |
6003 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 6004 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 6005 | } |
c91abf5a | 6006 | ret: |
92140480 | 6007 | return retn; |
52c03291 N |
6008 | } |
6009 | ||
849674e4 SL |
6010 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
6011 | int *skipped) | |
52c03291 | 6012 | { |
d1688a6d | 6013 | struct r5conf *conf = mddev->private; |
52c03291 | 6014 | struct stripe_head *sh; |
58c0fed4 | 6015 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 6016 | sector_t sync_blocks; |
16a53ecc N |
6017 | int still_degraded = 0; |
6018 | int i; | |
1da177e4 | 6019 | |
72626685 | 6020 | if (sector_nr >= max_sector) { |
1da177e4 | 6021 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 6022 | |
29269553 N |
6023 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
6024 | end_reshape(conf); | |
6025 | return 0; | |
6026 | } | |
72626685 N |
6027 | |
6028 | if (mddev->curr_resync < max_sector) /* aborted */ | |
e64e4018 AS |
6029 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
6030 | &sync_blocks, 1); | |
16a53ecc | 6031 | else /* completed sync */ |
72626685 | 6032 | conf->fullsync = 0; |
e64e4018 | 6033 | md_bitmap_close_sync(mddev->bitmap); |
72626685 | 6034 | |
1da177e4 LT |
6035 | return 0; |
6036 | } | |
ccfcc3c1 | 6037 | |
64bd660b N |
6038 | /* Allow raid5_quiesce to complete */ |
6039 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
6040 | ||
52c03291 N |
6041 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
6042 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 6043 | |
c6207277 N |
6044 | /* No need to check resync_max as we never do more than one |
6045 | * stripe, and as resync_max will always be on a chunk boundary, | |
6046 | * if the check in md_do_sync didn't fire, there is no chance | |
6047 | * of overstepping resync_max here | |
6048 | */ | |
6049 | ||
16a53ecc | 6050 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
6051 | * to resync, then assert that we are finished, because there is |
6052 | * nothing we can do. | |
6053 | */ | |
3285edf1 | 6054 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 6055 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 6056 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 6057 | *skipped = 1; |
1da177e4 LT |
6058 | return rv; |
6059 | } | |
6f608040 | 6060 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
6061 | !conf->fullsync && | |
e64e4018 | 6062 | !md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && |
6f608040 | 6063 | sync_blocks >= STRIPE_SECTORS) { |
72626685 N |
6064 | /* we can skip this block, and probably more */ |
6065 | sync_blocks /= STRIPE_SECTORS; | |
6066 | *skipped = 1; | |
6067 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
6068 | } | |
1da177e4 | 6069 | |
e64e4018 | 6070 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 6071 | |
6d036f7d | 6072 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 6073 | if (sh == NULL) { |
6d036f7d | 6074 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 6075 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 6076 | * is trying to get access |
1da177e4 | 6077 | */ |
66c006a5 | 6078 | schedule_timeout_uninterruptible(1); |
1da177e4 | 6079 | } |
16a53ecc | 6080 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
6081 | * Note in case of > 1 drive failures it's possible we're rebuilding |
6082 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 6083 | */ |
16d9cfab EM |
6084 | rcu_read_lock(); |
6085 | for (i = 0; i < conf->raid_disks; i++) { | |
6aa7de05 | 6086 | struct md_rdev *rdev = READ_ONCE(conf->disks[i].rdev); |
16d9cfab EM |
6087 | |
6088 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 6089 | still_degraded = 1; |
16d9cfab EM |
6090 | } |
6091 | rcu_read_unlock(); | |
16a53ecc | 6092 | |
e64e4018 | 6093 | md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); |
16a53ecc | 6094 | |
83206d66 | 6095 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 6096 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 6097 | |
6d036f7d | 6098 | raid5_release_stripe(sh); |
1da177e4 LT |
6099 | |
6100 | return STRIPE_SECTORS; | |
6101 | } | |
6102 | ||
0472a42b N |
6103 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio, |
6104 | unsigned int offset) | |
46031f9a RBJ |
6105 | { |
6106 | /* We may not be able to submit a whole bio at once as there | |
6107 | * may not be enough stripe_heads available. | |
6108 | * We cannot pre-allocate enough stripe_heads as we may need | |
6109 | * more than exist in the cache (if we allow ever large chunks). | |
6110 | * So we do one stripe head at a time and record in | |
6111 | * ->bi_hw_segments how many have been done. | |
6112 | * | |
6113 | * We *know* that this entire raid_bio is in one chunk, so | |
6114 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
6115 | */ | |
6116 | struct stripe_head *sh; | |
911d4ee8 | 6117 | int dd_idx; |
46031f9a RBJ |
6118 | sector_t sector, logical_sector, last_sector; |
6119 | int scnt = 0; | |
46031f9a RBJ |
6120 | int handled = 0; |
6121 | ||
4f024f37 KO |
6122 | logical_sector = raid_bio->bi_iter.bi_sector & |
6123 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 6124 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6125 | 0, &dd_idx, NULL); |
f73a1c7d | 6126 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6127 | |
6128 | for (; logical_sector < last_sector; | |
387bb173 NB |
6129 | logical_sector += STRIPE_SECTORS, |
6130 | sector += STRIPE_SECTORS, | |
6131 | scnt++) { | |
46031f9a | 6132 | |
0472a42b | 6133 | if (scnt < offset) |
46031f9a RBJ |
6134 | /* already done this stripe */ |
6135 | continue; | |
6136 | ||
6d036f7d | 6137 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6138 | |
6139 | if (!sh) { | |
6140 | /* failed to get a stripe - must wait */ | |
46031f9a | 6141 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6142 | conf->retry_read_offset = scnt; |
46031f9a RBJ |
6143 | return handled; |
6144 | } | |
6145 | ||
da41ba65 | 6146 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6147 | raid5_release_stripe(sh); |
387bb173 | 6148 | conf->retry_read_aligned = raid_bio; |
0472a42b | 6149 | conf->retry_read_offset = scnt; |
387bb173 NB |
6150 | return handled; |
6151 | } | |
6152 | ||
3f9e7c14 | 6153 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6154 | handle_stripe(sh); |
6d036f7d | 6155 | raid5_release_stripe(sh); |
46031f9a RBJ |
6156 | handled++; |
6157 | } | |
016c76ac N |
6158 | |
6159 | bio_endio(raid_bio); | |
6160 | ||
46031f9a | 6161 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6162 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6163 | return handled; |
6164 | } | |
6165 | ||
bfc90cb0 | 6166 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6167 | struct r5worker *worker, |
6168 | struct list_head *temp_inactive_list) | |
46a06401 SL |
6169 | { |
6170 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6171 | int i, batch_size = 0, hash; |
6172 | bool release_inactive = false; | |
46a06401 SL |
6173 | |
6174 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6175 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6176 | batch[batch_size++] = sh; |
6177 | ||
566c09c5 SL |
6178 | if (batch_size == 0) { |
6179 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6180 | if (!list_empty(temp_inactive_list + i)) | |
6181 | break; | |
a8c34f91 SL |
6182 | if (i == NR_STRIPE_HASH_LOCKS) { |
6183 | spin_unlock_irq(&conf->device_lock); | |
1532d9e8 | 6184 | log_flush_stripe_to_raid(conf); |
a8c34f91 | 6185 | spin_lock_irq(&conf->device_lock); |
566c09c5 | 6186 | return batch_size; |
a8c34f91 | 6187 | } |
566c09c5 SL |
6188 | release_inactive = true; |
6189 | } | |
46a06401 SL |
6190 | spin_unlock_irq(&conf->device_lock); |
6191 | ||
566c09c5 SL |
6192 | release_inactive_stripe_list(conf, temp_inactive_list, |
6193 | NR_STRIPE_HASH_LOCKS); | |
6194 | ||
a8c34f91 | 6195 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6196 | if (release_inactive) { |
6197 | spin_lock_irq(&conf->device_lock); | |
6198 | return 0; | |
6199 | } | |
6200 | ||
46a06401 SL |
6201 | for (i = 0; i < batch_size; i++) |
6202 | handle_stripe(batch[i]); | |
ff875738 | 6203 | log_write_stripe_run(conf); |
46a06401 SL |
6204 | |
6205 | cond_resched(); | |
6206 | ||
6207 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6208 | for (i = 0; i < batch_size; i++) { |
6209 | hash = batch[i]->hash_lock_index; | |
6210 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6211 | } | |
46a06401 SL |
6212 | return batch_size; |
6213 | } | |
46031f9a | 6214 | |
851c30c9 SL |
6215 | static void raid5_do_work(struct work_struct *work) |
6216 | { | |
6217 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6218 | struct r5worker_group *group = worker->group; | |
6219 | struct r5conf *conf = group->conf; | |
16d997b7 | 6220 | struct mddev *mddev = conf->mddev; |
851c30c9 SL |
6221 | int group_id = group - conf->worker_groups; |
6222 | int handled; | |
6223 | struct blk_plug plug; | |
6224 | ||
6225 | pr_debug("+++ raid5worker active\n"); | |
6226 | ||
6227 | blk_start_plug(&plug); | |
6228 | handled = 0; | |
6229 | spin_lock_irq(&conf->device_lock); | |
6230 | while (1) { | |
6231 | int batch_size, released; | |
6232 | ||
566c09c5 | 6233 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6234 | |
566c09c5 SL |
6235 | batch_size = handle_active_stripes(conf, group_id, worker, |
6236 | worker->temp_inactive_list); | |
bfc90cb0 | 6237 | worker->working = false; |
851c30c9 SL |
6238 | if (!batch_size && !released) |
6239 | break; | |
6240 | handled += batch_size; | |
16d997b7 N |
6241 | wait_event_lock_irq(mddev->sb_wait, |
6242 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags), | |
6243 | conf->device_lock); | |
851c30c9 SL |
6244 | } |
6245 | pr_debug("%d stripes handled\n", handled); | |
6246 | ||
6247 | spin_unlock_irq(&conf->device_lock); | |
7e96d559 | 6248 | |
9c72a18e SL |
6249 | flush_deferred_bios(conf); |
6250 | ||
6251 | r5l_flush_stripe_to_raid(conf->log); | |
6252 | ||
7e96d559 | 6253 | async_tx_issue_pending_all(); |
851c30c9 SL |
6254 | blk_finish_plug(&plug); |
6255 | ||
6256 | pr_debug("--- raid5worker inactive\n"); | |
6257 | } | |
6258 | ||
1da177e4 LT |
6259 | /* |
6260 | * This is our raid5 kernel thread. | |
6261 | * | |
6262 | * We scan the hash table for stripes which can be handled now. | |
6263 | * During the scan, completed stripes are saved for us by the interrupt | |
6264 | * handler, so that they will not have to wait for our next wakeup. | |
6265 | */ | |
4ed8731d | 6266 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6267 | { |
4ed8731d | 6268 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6269 | struct r5conf *conf = mddev->private; |
1da177e4 | 6270 | int handled; |
e1dfa0a2 | 6271 | struct blk_plug plug; |
1da177e4 | 6272 | |
45b4233c | 6273 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6274 | |
6275 | md_check_recovery(mddev); | |
1da177e4 | 6276 | |
e1dfa0a2 | 6277 | blk_start_plug(&plug); |
1da177e4 LT |
6278 | handled = 0; |
6279 | spin_lock_irq(&conf->device_lock); | |
6280 | while (1) { | |
46031f9a | 6281 | struct bio *bio; |
773ca82f | 6282 | int batch_size, released; |
0472a42b | 6283 | unsigned int offset; |
773ca82f | 6284 | |
566c09c5 | 6285 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6286 | if (released) |
6287 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6288 | |
0021b7bc | 6289 | if ( |
7c13edc8 N |
6290 | !list_empty(&conf->bitmap_list)) { |
6291 | /* Now is a good time to flush some bitmap updates */ | |
6292 | conf->seq_flush++; | |
700e432d | 6293 | spin_unlock_irq(&conf->device_lock); |
e64e4018 | 6294 | md_bitmap_unplug(mddev->bitmap); |
700e432d | 6295 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6296 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6297 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6298 | } |
0021b7bc | 6299 | raid5_activate_delayed(conf); |
72626685 | 6300 | |
0472a42b | 6301 | while ((bio = remove_bio_from_retry(conf, &offset))) { |
46031f9a RBJ |
6302 | int ok; |
6303 | spin_unlock_irq(&conf->device_lock); | |
0472a42b | 6304 | ok = retry_aligned_read(conf, bio, offset); |
46031f9a RBJ |
6305 | spin_lock_irq(&conf->device_lock); |
6306 | if (!ok) | |
6307 | break; | |
6308 | handled++; | |
6309 | } | |
6310 | ||
566c09c5 SL |
6311 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6312 | conf->temp_inactive_list); | |
773ca82f | 6313 | if (!batch_size && !released) |
1da177e4 | 6314 | break; |
46a06401 | 6315 | handled += batch_size; |
1da177e4 | 6316 | |
2953079c | 6317 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6318 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6319 | md_check_recovery(mddev); |
46a06401 SL |
6320 | spin_lock_irq(&conf->device_lock); |
6321 | } | |
1da177e4 | 6322 | } |
45b4233c | 6323 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6324 | |
6325 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6326 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6327 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6328 | grow_one_stripe(conf, __GFP_NOWARN); |
6329 | /* Set flag even if allocation failed. This helps | |
6330 | * slow down allocation requests when mem is short | |
6331 | */ | |
6332 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6333 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6334 | } |
1da177e4 | 6335 | |
765d704d SL |
6336 | flush_deferred_bios(conf); |
6337 | ||
0576b1c6 SL |
6338 | r5l_flush_stripe_to_raid(conf->log); |
6339 | ||
c9f21aaf | 6340 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6341 | blk_finish_plug(&plug); |
1da177e4 | 6342 | |
45b4233c | 6343 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6344 | } |
6345 | ||
3f294f4f | 6346 | static ssize_t |
fd01b88c | 6347 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6348 | { |
7b1485ba N |
6349 | struct r5conf *conf; |
6350 | int ret = 0; | |
6351 | spin_lock(&mddev->lock); | |
6352 | conf = mddev->private; | |
96de1e66 | 6353 | if (conf) |
edbe83ab | 6354 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6355 | spin_unlock(&mddev->lock); |
6356 | return ret; | |
3f294f4f N |
6357 | } |
6358 | ||
c41d4ac4 | 6359 | int |
fd01b88c | 6360 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6361 | { |
483cbbed | 6362 | int result = 0; |
d1688a6d | 6363 | struct r5conf *conf = mddev->private; |
b5470dc5 | 6364 | |
c41d4ac4 | 6365 | if (size <= 16 || size > 32768) |
3f294f4f | 6366 | return -EINVAL; |
486f0644 | 6367 | |
edbe83ab | 6368 | conf->min_nr_stripes = size; |
2d5b569b | 6369 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6370 | while (size < conf->max_nr_stripes && |
6371 | drop_one_stripe(conf)) | |
6372 | ; | |
2d5b569b | 6373 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6374 | |
2214c260 | 6375 | md_allow_write(mddev); |
486f0644 | 6376 | |
2d5b569b | 6377 | mutex_lock(&conf->cache_size_mutex); |
486f0644 | 6378 | while (size > conf->max_nr_stripes) |
483cbbed AN |
6379 | if (!grow_one_stripe(conf, GFP_KERNEL)) { |
6380 | conf->min_nr_stripes = conf->max_nr_stripes; | |
6381 | result = -ENOMEM; | |
486f0644 | 6382 | break; |
483cbbed | 6383 | } |
2d5b569b | 6384 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6385 | |
483cbbed | 6386 | return result; |
c41d4ac4 N |
6387 | } |
6388 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6389 | ||
6390 | static ssize_t | |
fd01b88c | 6391 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6392 | { |
6791875e | 6393 | struct r5conf *conf; |
c41d4ac4 N |
6394 | unsigned long new; |
6395 | int err; | |
6396 | ||
6397 | if (len >= PAGE_SIZE) | |
6398 | return -EINVAL; | |
b29bebd6 | 6399 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6400 | return -EINVAL; |
6791875e | 6401 | err = mddev_lock(mddev); |
c41d4ac4 N |
6402 | if (err) |
6403 | return err; | |
6791875e N |
6404 | conf = mddev->private; |
6405 | if (!conf) | |
6406 | err = -ENODEV; | |
6407 | else | |
6408 | err = raid5_set_cache_size(mddev, new); | |
6409 | mddev_unlock(mddev); | |
6410 | ||
6411 | return err ?: len; | |
3f294f4f | 6412 | } |
007583c9 | 6413 | |
96de1e66 N |
6414 | static struct md_sysfs_entry |
6415 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6416 | raid5_show_stripe_cache_size, | |
6417 | raid5_store_stripe_cache_size); | |
3f294f4f | 6418 | |
d06f191f MS |
6419 | static ssize_t |
6420 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6421 | { | |
6422 | struct r5conf *conf = mddev->private; | |
6423 | if (conf) | |
6424 | return sprintf(page, "%d\n", conf->rmw_level); | |
6425 | else | |
6426 | return 0; | |
6427 | } | |
6428 | ||
6429 | static ssize_t | |
6430 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6431 | { | |
6432 | struct r5conf *conf = mddev->private; | |
6433 | unsigned long new; | |
6434 | ||
6435 | if (!conf) | |
6436 | return -ENODEV; | |
6437 | ||
6438 | if (len >= PAGE_SIZE) | |
6439 | return -EINVAL; | |
6440 | ||
6441 | if (kstrtoul(page, 10, &new)) | |
6442 | return -EINVAL; | |
6443 | ||
6444 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6445 | return -EINVAL; | |
6446 | ||
6447 | if (new != PARITY_DISABLE_RMW && | |
6448 | new != PARITY_ENABLE_RMW && | |
6449 | new != PARITY_PREFER_RMW) | |
6450 | return -EINVAL; | |
6451 | ||
6452 | conf->rmw_level = new; | |
6453 | return len; | |
6454 | } | |
6455 | ||
6456 | static struct md_sysfs_entry | |
6457 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6458 | raid5_show_rmw_level, | |
6459 | raid5_store_rmw_level); | |
6460 | ||
6461 | ||
8b3e6cdc | 6462 | static ssize_t |
fd01b88c | 6463 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6464 | { |
7b1485ba N |
6465 | struct r5conf *conf; |
6466 | int ret = 0; | |
6467 | spin_lock(&mddev->lock); | |
6468 | conf = mddev->private; | |
8b3e6cdc | 6469 | if (conf) |
7b1485ba N |
6470 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6471 | spin_unlock(&mddev->lock); | |
6472 | return ret; | |
8b3e6cdc DW |
6473 | } |
6474 | ||
6475 | static ssize_t | |
fd01b88c | 6476 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6477 | { |
6791875e | 6478 | struct r5conf *conf; |
4ef197d8 | 6479 | unsigned long new; |
6791875e N |
6480 | int err; |
6481 | ||
8b3e6cdc DW |
6482 | if (len >= PAGE_SIZE) |
6483 | return -EINVAL; | |
b29bebd6 | 6484 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6485 | return -EINVAL; |
6791875e N |
6486 | |
6487 | err = mddev_lock(mddev); | |
6488 | if (err) | |
6489 | return err; | |
6490 | conf = mddev->private; | |
6491 | if (!conf) | |
6492 | err = -ENODEV; | |
edbe83ab | 6493 | else if (new > conf->min_nr_stripes) |
6791875e N |
6494 | err = -EINVAL; |
6495 | else | |
6496 | conf->bypass_threshold = new; | |
6497 | mddev_unlock(mddev); | |
6498 | return err ?: len; | |
8b3e6cdc DW |
6499 | } |
6500 | ||
6501 | static struct md_sysfs_entry | |
6502 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6503 | S_IRUGO | S_IWUSR, | |
6504 | raid5_show_preread_threshold, | |
6505 | raid5_store_preread_threshold); | |
6506 | ||
d592a996 SL |
6507 | static ssize_t |
6508 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6509 | { | |
7b1485ba N |
6510 | struct r5conf *conf; |
6511 | int ret = 0; | |
6512 | spin_lock(&mddev->lock); | |
6513 | conf = mddev->private; | |
d592a996 | 6514 | if (conf) |
7b1485ba N |
6515 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6516 | spin_unlock(&mddev->lock); | |
6517 | return ret; | |
d592a996 SL |
6518 | } |
6519 | ||
6520 | static ssize_t | |
6521 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6522 | { | |
6791875e | 6523 | struct r5conf *conf; |
d592a996 | 6524 | unsigned long new; |
6791875e N |
6525 | int err; |
6526 | ||
d592a996 SL |
6527 | if (len >= PAGE_SIZE) |
6528 | return -EINVAL; | |
d592a996 SL |
6529 | if (kstrtoul(page, 10, &new)) |
6530 | return -EINVAL; | |
6531 | new = !!new; | |
6791875e N |
6532 | |
6533 | err = mddev_lock(mddev); | |
6534 | if (err) | |
6535 | return err; | |
6536 | conf = mddev->private; | |
6537 | if (!conf) | |
6538 | err = -ENODEV; | |
6539 | else if (new != conf->skip_copy) { | |
6540 | mddev_suspend(mddev); | |
6541 | conf->skip_copy = new; | |
6542 | if (new) | |
dc3b17cc | 6543 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6544 | BDI_CAP_STABLE_WRITES; |
6545 | else | |
dc3b17cc | 6546 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6547 | ~BDI_CAP_STABLE_WRITES; |
6548 | mddev_resume(mddev); | |
6549 | } | |
6550 | mddev_unlock(mddev); | |
6551 | return err ?: len; | |
d592a996 SL |
6552 | } |
6553 | ||
6554 | static struct md_sysfs_entry | |
6555 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6556 | raid5_show_skip_copy, | |
6557 | raid5_store_skip_copy); | |
6558 | ||
3f294f4f | 6559 | static ssize_t |
fd01b88c | 6560 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6561 | { |
d1688a6d | 6562 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6563 | if (conf) |
6564 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6565 | else | |
6566 | return 0; | |
3f294f4f N |
6567 | } |
6568 | ||
96de1e66 N |
6569 | static struct md_sysfs_entry |
6570 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6571 | |
b721420e SL |
6572 | static ssize_t |
6573 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6574 | { | |
7b1485ba N |
6575 | struct r5conf *conf; |
6576 | int ret = 0; | |
6577 | spin_lock(&mddev->lock); | |
6578 | conf = mddev->private; | |
b721420e | 6579 | if (conf) |
7b1485ba N |
6580 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6581 | spin_unlock(&mddev->lock); | |
6582 | return ret; | |
b721420e SL |
6583 | } |
6584 | ||
60aaf933 | 6585 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6586 | int *group_cnt, | |
6587 | int *worker_cnt_per_group, | |
6588 | struct r5worker_group **worker_groups); | |
b721420e SL |
6589 | static ssize_t |
6590 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6591 | { | |
6791875e | 6592 | struct r5conf *conf; |
7d5d7b50 | 6593 | unsigned int new; |
b721420e | 6594 | int err; |
60aaf933 | 6595 | struct r5worker_group *new_groups, *old_groups; |
6596 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6597 | |
6598 | if (len >= PAGE_SIZE) | |
6599 | return -EINVAL; | |
7d5d7b50 SL |
6600 | if (kstrtouint(page, 10, &new)) |
6601 | return -EINVAL; | |
6602 | /* 8192 should be big enough */ | |
6603 | if (new > 8192) | |
b721420e SL |
6604 | return -EINVAL; |
6605 | ||
6791875e N |
6606 | err = mddev_lock(mddev); |
6607 | if (err) | |
6608 | return err; | |
6609 | conf = mddev->private; | |
6610 | if (!conf) | |
6611 | err = -ENODEV; | |
6612 | else if (new != conf->worker_cnt_per_group) { | |
6613 | mddev_suspend(mddev); | |
b721420e | 6614 | |
6791875e N |
6615 | old_groups = conf->worker_groups; |
6616 | if (old_groups) | |
6617 | flush_workqueue(raid5_wq); | |
d206dcfa | 6618 | |
6791875e N |
6619 | err = alloc_thread_groups(conf, new, |
6620 | &group_cnt, &worker_cnt_per_group, | |
6621 | &new_groups); | |
6622 | if (!err) { | |
6623 | spin_lock_irq(&conf->device_lock); | |
6624 | conf->group_cnt = group_cnt; | |
6625 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6626 | conf->worker_groups = new_groups; | |
6627 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6628 | |
6791875e N |
6629 | if (old_groups) |
6630 | kfree(old_groups[0].workers); | |
6631 | kfree(old_groups); | |
6632 | } | |
6633 | mddev_resume(mddev); | |
b721420e | 6634 | } |
6791875e | 6635 | mddev_unlock(mddev); |
b721420e | 6636 | |
6791875e | 6637 | return err ?: len; |
b721420e SL |
6638 | } |
6639 | ||
6640 | static struct md_sysfs_entry | |
6641 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6642 | raid5_show_group_thread_cnt, | |
6643 | raid5_store_group_thread_cnt); | |
6644 | ||
007583c9 | 6645 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6646 | &raid5_stripecache_size.attr, |
6647 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6648 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6649 | &raid5_group_thread_cnt.attr, |
d592a996 | 6650 | &raid5_skip_copy.attr, |
d06f191f | 6651 | &raid5_rmw_level.attr, |
2c7da14b | 6652 | &r5c_journal_mode.attr, |
3f294f4f N |
6653 | NULL, |
6654 | }; | |
007583c9 N |
6655 | static struct attribute_group raid5_attrs_group = { |
6656 | .name = NULL, | |
6657 | .attrs = raid5_attrs, | |
3f294f4f N |
6658 | }; |
6659 | ||
60aaf933 | 6660 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6661 | int *group_cnt, | |
6662 | int *worker_cnt_per_group, | |
6663 | struct r5worker_group **worker_groups) | |
851c30c9 | 6664 | { |
566c09c5 | 6665 | int i, j, k; |
851c30c9 SL |
6666 | ssize_t size; |
6667 | struct r5worker *workers; | |
6668 | ||
60aaf933 | 6669 | *worker_cnt_per_group = cnt; |
851c30c9 | 6670 | if (cnt == 0) { |
60aaf933 | 6671 | *group_cnt = 0; |
6672 | *worker_groups = NULL; | |
851c30c9 SL |
6673 | return 0; |
6674 | } | |
60aaf933 | 6675 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6676 | size = sizeof(struct r5worker) * cnt; |
6396bb22 KC |
6677 | workers = kcalloc(size, *group_cnt, GFP_NOIO); |
6678 | *worker_groups = kcalloc(*group_cnt, sizeof(struct r5worker_group), | |
6679 | GFP_NOIO); | |
60aaf933 | 6680 | if (!*worker_groups || !workers) { |
851c30c9 | 6681 | kfree(workers); |
60aaf933 | 6682 | kfree(*worker_groups); |
851c30c9 SL |
6683 | return -ENOMEM; |
6684 | } | |
6685 | ||
60aaf933 | 6686 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6687 | struct r5worker_group *group; |
6688 | ||
0c775d52 | 6689 | group = &(*worker_groups)[i]; |
851c30c9 | 6690 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6691 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6692 | group->conf = conf; |
6693 | group->workers = workers + i * cnt; | |
6694 | ||
6695 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6696 | struct r5worker *worker = group->workers + j; |
6697 | worker->group = group; | |
6698 | INIT_WORK(&worker->work, raid5_do_work); | |
6699 | ||
6700 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6701 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6702 | } |
6703 | } | |
6704 | ||
6705 | return 0; | |
6706 | } | |
6707 | ||
6708 | static void free_thread_groups(struct r5conf *conf) | |
6709 | { | |
6710 | if (conf->worker_groups) | |
6711 | kfree(conf->worker_groups[0].workers); | |
6712 | kfree(conf->worker_groups); | |
6713 | conf->worker_groups = NULL; | |
6714 | } | |
6715 | ||
80c3a6ce | 6716 | static sector_t |
fd01b88c | 6717 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6718 | { |
d1688a6d | 6719 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6720 | |
6721 | if (!sectors) | |
6722 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6723 | if (!raid_disks) |
7ec05478 | 6724 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6725 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6726 | |
3cb5edf4 N |
6727 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6728 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6729 | return sectors * (raid_disks - conf->max_degraded); |
6730 | } | |
6731 | ||
789b5e03 ON |
6732 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6733 | { | |
6734 | safe_put_page(percpu->spare_page); | |
789b5e03 | 6735 | percpu->spare_page = NULL; |
b330e6a4 | 6736 | kvfree(percpu->scribble); |
789b5e03 ON |
6737 | percpu->scribble = NULL; |
6738 | } | |
6739 | ||
6740 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6741 | { | |
b330e6a4 | 6742 | if (conf->level == 6 && !percpu->spare_page) { |
789b5e03 | 6743 | percpu->spare_page = alloc_page(GFP_KERNEL); |
b330e6a4 KO |
6744 | if (!percpu->spare_page) |
6745 | return -ENOMEM; | |
6746 | } | |
6747 | ||
6748 | if (scribble_alloc(percpu, | |
6749 | max(conf->raid_disks, | |
6750 | conf->previous_raid_disks), | |
6751 | max(conf->chunk_sectors, | |
6752 | conf->prev_chunk_sectors) | |
6753 | / STRIPE_SECTORS, | |
6754 | GFP_KERNEL)) { | |
789b5e03 ON |
6755 | free_scratch_buffer(conf, percpu); |
6756 | return -ENOMEM; | |
6757 | } | |
6758 | ||
6759 | return 0; | |
6760 | } | |
6761 | ||
29c6d1bb | 6762 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6763 | { |
29c6d1bb SAS |
6764 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6765 | ||
6766 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6767 | return 0; | |
6768 | } | |
36d1c647 | 6769 | |
29c6d1bb SAS |
6770 | static void raid5_free_percpu(struct r5conf *conf) |
6771 | { | |
36d1c647 DW |
6772 | if (!conf->percpu) |
6773 | return; | |
6774 | ||
29c6d1bb | 6775 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6776 | free_percpu(conf->percpu); |
6777 | } | |
6778 | ||
d1688a6d | 6779 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6780 | { |
d7bd398e SL |
6781 | int i; |
6782 | ||
ff875738 AP |
6783 | log_exit(conf); |
6784 | ||
565e0450 | 6785 | unregister_shrinker(&conf->shrinker); |
851c30c9 | 6786 | free_thread_groups(conf); |
95fc17aa | 6787 | shrink_stripes(conf); |
36d1c647 | 6788 | raid5_free_percpu(conf); |
d7bd398e SL |
6789 | for (i = 0; i < conf->pool_size; i++) |
6790 | if (conf->disks[i].extra_page) | |
6791 | put_page(conf->disks[i].extra_page); | |
95fc17aa | 6792 | kfree(conf->disks); |
afeee514 | 6793 | bioset_exit(&conf->bio_split); |
95fc17aa | 6794 | kfree(conf->stripe_hashtbl); |
aaf9f12e | 6795 | kfree(conf->pending_data); |
95fc17aa DW |
6796 | kfree(conf); |
6797 | } | |
6798 | ||
29c6d1bb | 6799 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6800 | { |
29c6d1bb | 6801 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6802 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6803 | ||
29c6d1bb | 6804 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6805 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6806 | __func__, cpu); | |
29c6d1bb | 6807 | return -ENOMEM; |
36d1c647 | 6808 | } |
29c6d1bb | 6809 | return 0; |
36d1c647 | 6810 | } |
36d1c647 | 6811 | |
d1688a6d | 6812 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6813 | { |
789b5e03 | 6814 | int err = 0; |
36d1c647 | 6815 | |
789b5e03 ON |
6816 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6817 | if (!conf->percpu) | |
36d1c647 | 6818 | return -ENOMEM; |
789b5e03 | 6819 | |
29c6d1bb | 6820 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6821 | if (!err) { |
6822 | conf->scribble_disks = max(conf->raid_disks, | |
6823 | conf->previous_raid_disks); | |
6824 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6825 | conf->prev_chunk_sectors); | |
6826 | } | |
36d1c647 DW |
6827 | return err; |
6828 | } | |
6829 | ||
edbe83ab N |
6830 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6831 | struct shrink_control *sc) | |
6832 | { | |
6833 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6834 | unsigned long ret = SHRINK_STOP; |
6835 | ||
6836 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6837 | ret= 0; | |
49895bcc N |
6838 | while (ret < sc->nr_to_scan && |
6839 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6840 | if (drop_one_stripe(conf) == 0) { |
6841 | ret = SHRINK_STOP; | |
6842 | break; | |
6843 | } | |
6844 | ret++; | |
6845 | } | |
6846 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6847 | } |
6848 | return ret; | |
6849 | } | |
6850 | ||
6851 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6852 | struct shrink_control *sc) | |
6853 | { | |
6854 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6855 | ||
6856 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6857 | /* unlikely, but not impossible */ | |
6858 | return 0; | |
6859 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6860 | } | |
6861 | ||
d1688a6d | 6862 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6863 | { |
d1688a6d | 6864 | struct r5conf *conf; |
5e5e3e78 | 6865 | int raid_disk, memory, max_disks; |
3cb03002 | 6866 | struct md_rdev *rdev; |
1da177e4 | 6867 | struct disk_info *disk; |
0232605d | 6868 | char pers_name[6]; |
566c09c5 | 6869 | int i; |
60aaf933 | 6870 | int group_cnt, worker_cnt_per_group; |
6871 | struct r5worker_group *new_group; | |
afeee514 | 6872 | int ret; |
1da177e4 | 6873 | |
91adb564 N |
6874 | if (mddev->new_level != 5 |
6875 | && mddev->new_level != 4 | |
6876 | && mddev->new_level != 6) { | |
cc6167b4 N |
6877 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6878 | mdname(mddev), mddev->new_level); | |
91adb564 | 6879 | return ERR_PTR(-EIO); |
1da177e4 | 6880 | } |
91adb564 N |
6881 | if ((mddev->new_level == 5 |
6882 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6883 | (mddev->new_level == 6 | |
6884 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6885 | pr_warn("md/raid:%s: layout %d not supported\n", |
6886 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6887 | return ERR_PTR(-EIO); |
99c0fb5f | 6888 | } |
91adb564 | 6889 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6890 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6891 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6892 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6893 | } |
6894 | ||
664e7c41 AN |
6895 | if (!mddev->new_chunk_sectors || |
6896 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6897 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6898 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6899 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6900 | return ERR_PTR(-EINVAL); |
f6705578 N |
6901 | } |
6902 | ||
d1688a6d | 6903 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6904 | if (conf == NULL) |
1da177e4 | 6905 | goto abort; |
aaf9f12e SL |
6906 | INIT_LIST_HEAD(&conf->free_list); |
6907 | INIT_LIST_HEAD(&conf->pending_list); | |
6396bb22 KC |
6908 | conf->pending_data = kcalloc(PENDING_IO_MAX, |
6909 | sizeof(struct r5pending_data), | |
6910 | GFP_KERNEL); | |
aaf9f12e SL |
6911 | if (!conf->pending_data) |
6912 | goto abort; | |
6913 | for (i = 0; i < PENDING_IO_MAX; i++) | |
6914 | list_add(&conf->pending_data[i].sibling, &conf->free_list); | |
851c30c9 | 6915 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6916 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6917 | &new_group)) { | |
6918 | conf->group_cnt = group_cnt; | |
6919 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6920 | conf->worker_groups = new_group; | |
6921 | } else | |
851c30c9 | 6922 | goto abort; |
f5efd45a | 6923 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6924 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6925 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6926 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6927 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6928 | init_waitqueue_head(&conf->wait_for_overlap); |
6929 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 6930 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
6931 | INIT_LIST_HEAD(&conf->hold_list); |
6932 | INIT_LIST_HEAD(&conf->delayed_list); | |
6933 | INIT_LIST_HEAD(&conf->bitmap_list); | |
773ca82f | 6934 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6935 | atomic_set(&conf->active_stripes, 0); |
6936 | atomic_set(&conf->preread_active_stripes, 0); | |
6937 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
6938 | spin_lock_init(&conf->pending_bios_lock); |
6939 | conf->batch_bio_dispatch = true; | |
6940 | rdev_for_each(rdev, mddev) { | |
6941 | if (test_bit(Journal, &rdev->flags)) | |
6942 | continue; | |
6943 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
6944 | conf->batch_bio_dispatch = false; | |
6945 | break; | |
6946 | } | |
6947 | } | |
6948 | ||
f5efd45a | 6949 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 6950 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6951 | |
6952 | conf->raid_disks = mddev->raid_disks; | |
6953 | if (mddev->reshape_position == MaxSector) | |
6954 | conf->previous_raid_disks = mddev->raid_disks; | |
6955 | else | |
f6705578 | 6956 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6957 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6958 | |
6396bb22 | 6959 | conf->disks = kcalloc(max_disks, sizeof(struct disk_info), |
b55e6bfc | 6960 | GFP_KERNEL); |
d7bd398e | 6961 | |
b55e6bfc N |
6962 | if (!conf->disks) |
6963 | goto abort; | |
9ffae0cf | 6964 | |
d7bd398e SL |
6965 | for (i = 0; i < max_disks; i++) { |
6966 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
6967 | if (!conf->disks[i].extra_page) | |
6968 | goto abort; | |
6969 | } | |
6970 | ||
afeee514 KO |
6971 | ret = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
6972 | if (ret) | |
dd7a8f5d | 6973 | goto abort; |
1da177e4 LT |
6974 | conf->mddev = mddev; |
6975 | ||
fccddba0 | 6976 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6977 | goto abort; |
1da177e4 | 6978 | |
566c09c5 SL |
6979 | /* We init hash_locks[0] separately to that it can be used |
6980 | * as the reference lock in the spin_lock_nest_lock() call | |
6981 | * in lock_all_device_hash_locks_irq in order to convince | |
6982 | * lockdep that we know what we are doing. | |
6983 | */ | |
6984 | spin_lock_init(conf->hash_locks); | |
6985 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6986 | spin_lock_init(conf->hash_locks + i); | |
6987 | ||
6988 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6989 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6990 | ||
6991 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6992 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6993 | ||
1e6d690b SL |
6994 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
6995 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
6996 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
6997 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
6998 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
6999 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 7000 | |
36d1c647 | 7001 | conf->level = mddev->new_level; |
46d5b785 | 7002 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
7003 | if (raid5_alloc_percpu(conf) != 0) |
7004 | goto abort; | |
7005 | ||
0c55e022 | 7006 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 7007 | |
dafb20fa | 7008 | rdev_for_each(rdev, mddev) { |
1da177e4 | 7009 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 7010 | if (raid_disk >= max_disks |
f2076e7d | 7011 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
7012 | continue; |
7013 | disk = conf->disks + raid_disk; | |
7014 | ||
17045f52 N |
7015 | if (test_bit(Replacement, &rdev->flags)) { |
7016 | if (disk->replacement) | |
7017 | goto abort; | |
7018 | disk->replacement = rdev; | |
7019 | } else { | |
7020 | if (disk->rdev) | |
7021 | goto abort; | |
7022 | disk->rdev = rdev; | |
7023 | } | |
1da177e4 | 7024 | |
b2d444d7 | 7025 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 7026 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
7027 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
7028 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 7029 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
7030 | /* Cannot rely on bitmap to complete recovery */ |
7031 | conf->fullsync = 1; | |
1da177e4 LT |
7032 | } |
7033 | ||
91adb564 | 7034 | conf->level = mddev->new_level; |
584acdd4 | 7035 | if (conf->level == 6) { |
16a53ecc | 7036 | conf->max_degraded = 2; |
584acdd4 MS |
7037 | if (raid6_call.xor_syndrome) |
7038 | conf->rmw_level = PARITY_ENABLE_RMW; | |
7039 | else | |
7040 | conf->rmw_level = PARITY_DISABLE_RMW; | |
7041 | } else { | |
16a53ecc | 7042 | conf->max_degraded = 1; |
584acdd4 MS |
7043 | conf->rmw_level = PARITY_ENABLE_RMW; |
7044 | } | |
91adb564 | 7045 | conf->algorithm = mddev->new_layout; |
fef9c61f | 7046 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 7047 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 7048 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 7049 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
7050 | } else { |
7051 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
7052 | conf->prev_algo = conf->algorithm; | |
e183eaed | 7053 | } |
1da177e4 | 7054 | |
edbe83ab | 7055 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
7056 | if (mddev->reshape_position != MaxSector) { |
7057 | int stripes = max_t(int, | |
7058 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
7059 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
7060 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
7061 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 7062 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
7063 | mdname(mddev), conf->min_nr_stripes); |
7064 | } | |
edbe83ab | 7065 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 7066 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 7067 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 7068 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
7069 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
7070 | mdname(mddev), memory); | |
91adb564 N |
7071 | goto abort; |
7072 | } else | |
cc6167b4 | 7073 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
7074 | /* |
7075 | * Losing a stripe head costs more than the time to refill it, | |
7076 | * it reduces the queue depth and so can hurt throughput. | |
7077 | * So set it rather large, scaled by number of devices. | |
7078 | */ | |
7079 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
7080 | conf->shrinker.scan_objects = raid5_cache_scan; | |
7081 | conf->shrinker.count_objects = raid5_cache_count; | |
7082 | conf->shrinker.batch = 128; | |
7083 | conf->shrinker.flags = 0; | |
6a0f53ff | 7084 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
7085 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
7086 | mdname(mddev)); | |
6a0f53ff CY |
7087 | goto abort; |
7088 | } | |
1da177e4 | 7089 | |
0232605d N |
7090 | sprintf(pers_name, "raid%d", mddev->new_level); |
7091 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 7092 | if (!conf->thread) { |
cc6167b4 N |
7093 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
7094 | mdname(mddev)); | |
16a53ecc N |
7095 | goto abort; |
7096 | } | |
91adb564 N |
7097 | |
7098 | return conf; | |
7099 | ||
7100 | abort: | |
7101 | if (conf) { | |
95fc17aa | 7102 | free_conf(conf); |
91adb564 N |
7103 | return ERR_PTR(-EIO); |
7104 | } else | |
7105 | return ERR_PTR(-ENOMEM); | |
7106 | } | |
7107 | ||
c148ffdc N |
7108 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
7109 | { | |
7110 | switch (algo) { | |
7111 | case ALGORITHM_PARITY_0: | |
7112 | if (raid_disk < max_degraded) | |
7113 | return 1; | |
7114 | break; | |
7115 | case ALGORITHM_PARITY_N: | |
7116 | if (raid_disk >= raid_disks - max_degraded) | |
7117 | return 1; | |
7118 | break; | |
7119 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 7120 | if (raid_disk == 0 || |
c148ffdc N |
7121 | raid_disk == raid_disks - 1) |
7122 | return 1; | |
7123 | break; | |
7124 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
7125 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7126 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7127 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7128 | if (raid_disk == raid_disks - 1) | |
7129 | return 1; | |
7130 | } | |
7131 | return 0; | |
7132 | } | |
7133 | ||
849674e4 | 7134 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7135 | { |
d1688a6d | 7136 | struct r5conf *conf; |
9f7c2220 | 7137 | int working_disks = 0; |
c148ffdc | 7138 | int dirty_parity_disks = 0; |
3cb03002 | 7139 | struct md_rdev *rdev; |
713cf5a6 | 7140 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7141 | sector_t reshape_offset = 0; |
17045f52 | 7142 | int i; |
b5254dd5 N |
7143 | long long min_offset_diff = 0; |
7144 | int first = 1; | |
91adb564 | 7145 | |
a415c0f1 N |
7146 | if (mddev_init_writes_pending(mddev) < 0) |
7147 | return -ENOMEM; | |
7148 | ||
8c6ac868 | 7149 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7150 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7151 | mdname(mddev)); | |
b5254dd5 N |
7152 | |
7153 | rdev_for_each(rdev, mddev) { | |
7154 | long long diff; | |
713cf5a6 | 7155 | |
f2076e7d | 7156 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7157 | journal_dev = rdev; |
f2076e7d SL |
7158 | continue; |
7159 | } | |
b5254dd5 N |
7160 | if (rdev->raid_disk < 0) |
7161 | continue; | |
7162 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7163 | if (first) { | |
7164 | min_offset_diff = diff; | |
7165 | first = 0; | |
7166 | } else if (mddev->reshape_backwards && | |
7167 | diff < min_offset_diff) | |
7168 | min_offset_diff = diff; | |
7169 | else if (!mddev->reshape_backwards && | |
7170 | diff > min_offset_diff) | |
7171 | min_offset_diff = diff; | |
7172 | } | |
7173 | ||
230b55fa N |
7174 | if ((test_bit(MD_HAS_JOURNAL, &mddev->flags) || journal_dev) && |
7175 | (mddev->bitmap_info.offset || mddev->bitmap_info.file)) { | |
7176 | pr_notice("md/raid:%s: array cannot have both journal and bitmap\n", | |
7177 | mdname(mddev)); | |
7178 | return -EINVAL; | |
7179 | } | |
7180 | ||
91adb564 N |
7181 | if (mddev->reshape_position != MaxSector) { |
7182 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7183 | * Difficulties arise if the stripe we would write to |
7184 | * next is at or after the stripe we would read from next. | |
7185 | * For a reshape that changes the number of devices, this | |
7186 | * is only possible for a very short time, and mdadm makes | |
7187 | * sure that time appears to have past before assembling | |
7188 | * the array. So we fail if that time hasn't passed. | |
7189 | * For a reshape that keeps the number of devices the same | |
7190 | * mdadm must be monitoring the reshape can keeping the | |
7191 | * critical areas read-only and backed up. It will start | |
7192 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7193 | */ |
7194 | sector_t here_new, here_old; | |
7195 | int old_disks; | |
18b00334 | 7196 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7197 | int chunk_sectors; |
7198 | int new_data_disks; | |
91adb564 | 7199 | |
713cf5a6 | 7200 | if (journal_dev) { |
cc6167b4 N |
7201 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7202 | mdname(mddev)); | |
713cf5a6 SL |
7203 | return -EINVAL; |
7204 | } | |
7205 | ||
88ce4930 | 7206 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7207 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7208 | mdname(mddev)); | |
91adb564 N |
7209 | return -EINVAL; |
7210 | } | |
91adb564 N |
7211 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7212 | /* reshape_position must be on a new-stripe boundary, and one | |
7213 | * further up in new geometry must map after here in old | |
7214 | * geometry. | |
05256d98 N |
7215 | * If the chunk sizes are different, then as we perform reshape |
7216 | * in units of the largest of the two, reshape_position needs | |
7217 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7218 | */ |
7219 | here_new = mddev->reshape_position; | |
05256d98 N |
7220 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7221 | new_data_disks = mddev->raid_disks - max_degraded; | |
7222 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7223 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7224 | mdname(mddev)); | |
91adb564 N |
7225 | return -EINVAL; |
7226 | } | |
05256d98 | 7227 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7228 | /* here_new is the stripe we will write to */ |
7229 | here_old = mddev->reshape_position; | |
05256d98 | 7230 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7231 | /* here_old is the first stripe that we might need to read |
7232 | * from */ | |
67ac6011 N |
7233 | if (mddev->delta_disks == 0) { |
7234 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7235 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7236 | * and taking constant backups. |
7237 | * mdadm always starts a situation like this in | |
7238 | * readonly mode so it can take control before | |
7239 | * allowing any writes. So just check for that. | |
7240 | */ | |
b5254dd5 N |
7241 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7242 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7243 | /* not really in-place - so OK */; | |
7244 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7245 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7246 | mdname(mddev)); | |
67ac6011 N |
7247 | return -EINVAL; |
7248 | } | |
2c810cdd | 7249 | } else if (mddev->reshape_backwards |
05256d98 N |
7250 | ? (here_new * chunk_sectors + min_offset_diff <= |
7251 | here_old * chunk_sectors) | |
7252 | : (here_new * chunk_sectors >= | |
7253 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7254 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7255 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7256 | mdname(mddev)); | |
91adb564 N |
7257 | return -EINVAL; |
7258 | } | |
cc6167b4 | 7259 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7260 | /* OK, we should be able to continue; */ |
7261 | } else { | |
7262 | BUG_ON(mddev->level != mddev->new_level); | |
7263 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7264 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7265 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7266 | } |
91adb564 | 7267 | |
3418d036 AP |
7268 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && |
7269 | test_bit(MD_HAS_PPL, &mddev->flags)) { | |
7270 | pr_warn("md/raid:%s: using journal device and PPL not allowed - disabling PPL\n", | |
7271 | mdname(mddev)); | |
7272 | clear_bit(MD_HAS_PPL, &mddev->flags); | |
ddc08823 | 7273 | clear_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags); |
3418d036 AP |
7274 | } |
7275 | ||
245f46c2 N |
7276 | if (mddev->private == NULL) |
7277 | conf = setup_conf(mddev); | |
7278 | else | |
7279 | conf = mddev->private; | |
7280 | ||
91adb564 N |
7281 | if (IS_ERR(conf)) |
7282 | return PTR_ERR(conf); | |
7283 | ||
486b0f7b SL |
7284 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7285 | if (!journal_dev) { | |
cc6167b4 N |
7286 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7287 | mdname(mddev)); | |
486b0f7b SL |
7288 | mddev->ro = 1; |
7289 | set_disk_ro(mddev->gendisk, 1); | |
7290 | } else if (mddev->recovery_cp == MaxSector) | |
7291 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7292 | } |
7293 | ||
b5254dd5 | 7294 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7295 | mddev->thread = conf->thread; |
7296 | conf->thread = NULL; | |
7297 | mddev->private = conf; | |
7298 | ||
17045f52 N |
7299 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7300 | i++) { | |
7301 | rdev = conf->disks[i].rdev; | |
7302 | if (!rdev && conf->disks[i].replacement) { | |
7303 | /* The replacement is all we have yet */ | |
7304 | rdev = conf->disks[i].replacement; | |
7305 | conf->disks[i].replacement = NULL; | |
7306 | clear_bit(Replacement, &rdev->flags); | |
7307 | conf->disks[i].rdev = rdev; | |
7308 | } | |
7309 | if (!rdev) | |
c148ffdc | 7310 | continue; |
17045f52 N |
7311 | if (conf->disks[i].replacement && |
7312 | conf->reshape_progress != MaxSector) { | |
7313 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7314 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7315 | goto abort; |
7316 | } | |
2f115882 | 7317 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7318 | working_disks++; |
2f115882 N |
7319 | continue; |
7320 | } | |
c148ffdc N |
7321 | /* This disc is not fully in-sync. However if it |
7322 | * just stored parity (beyond the recovery_offset), | |
7323 | * when we don't need to be concerned about the | |
7324 | * array being dirty. | |
7325 | * When reshape goes 'backwards', we never have | |
7326 | * partially completed devices, so we only need | |
7327 | * to worry about reshape going forwards. | |
7328 | */ | |
7329 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7330 | if (mddev->major_version == 0 && | |
7331 | mddev->minor_version > 90) | |
7332 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7333 | |
c148ffdc N |
7334 | if (rdev->recovery_offset < reshape_offset) { |
7335 | /* We need to check old and new layout */ | |
7336 | if (!only_parity(rdev->raid_disk, | |
7337 | conf->algorithm, | |
7338 | conf->raid_disks, | |
7339 | conf->max_degraded)) | |
7340 | continue; | |
7341 | } | |
7342 | if (!only_parity(rdev->raid_disk, | |
7343 | conf->prev_algo, | |
7344 | conf->previous_raid_disks, | |
7345 | conf->max_degraded)) | |
7346 | continue; | |
7347 | dirty_parity_disks++; | |
7348 | } | |
91adb564 | 7349 | |
17045f52 N |
7350 | /* |
7351 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7352 | */ | |
2e38a37f | 7353 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7354 | |
674806d6 | 7355 | if (has_failed(conf)) { |
cc6167b4 | 7356 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7357 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7358 | goto abort; |
7359 | } | |
7360 | ||
91adb564 | 7361 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7362 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7363 | mddev->resync_max_sectors = mddev->dev_sectors; |
7364 | ||
c148ffdc | 7365 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7366 | mddev->recovery_cp != MaxSector) { |
4536bf9b AP |
7367 | if (test_bit(MD_HAS_PPL, &mddev->flags)) |
7368 | pr_crit("md/raid:%s: starting dirty degraded array with PPL.\n", | |
7369 | mdname(mddev)); | |
7370 | else if (mddev->ok_start_degraded) | |
cc6167b4 N |
7371 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7372 | mdname(mddev)); | |
6ff8d8ec | 7373 | else { |
cc6167b4 N |
7374 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7375 | mdname(mddev)); | |
6ff8d8ec N |
7376 | goto abort; |
7377 | } | |
1da177e4 LT |
7378 | } |
7379 | ||
cc6167b4 N |
7380 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7381 | mdname(mddev), conf->level, | |
7382 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7383 | mddev->new_layout); | |
1da177e4 LT |
7384 | |
7385 | print_raid5_conf(conf); | |
7386 | ||
fef9c61f | 7387 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7388 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7389 | atomic_set(&conf->reshape_stripes, 0); |
7390 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7391 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7392 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7393 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7394 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7395 | "reshape"); |
f6705578 N |
7396 | } |
7397 | ||
1da177e4 | 7398 | /* Ok, everything is just fine now */ |
a64c876f N |
7399 | if (mddev->to_remove == &raid5_attrs_group) |
7400 | mddev->to_remove = NULL; | |
00bcb4ac N |
7401 | else if (mddev->kobj.sd && |
7402 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7403 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7404 | mdname(mddev)); | |
4a5add49 | 7405 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7406 | |
4a5add49 | 7407 | if (mddev->queue) { |
9f7c2220 | 7408 | int chunk_size; |
4a5add49 N |
7409 | /* read-ahead size must cover two whole stripes, which |
7410 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7411 | * number of raid devices | |
7412 | */ | |
7413 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7414 | int stripe = data_disks * | |
7415 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7416 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7417 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7418 | |
9f7c2220 N |
7419 | chunk_size = mddev->chunk_sectors << 9; |
7420 | blk_queue_io_min(mddev->queue, chunk_size); | |
7421 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7422 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7423 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7424 | /* |
7425 | * We can only discard a whole stripe. It doesn't make sense to | |
7426 | * discard data disk but write parity disk | |
7427 | */ | |
7428 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7429 | /* Round up to power of 2, as discard handling |
7430 | * currently assumes that */ | |
7431 | while ((stripe-1) & stripe) | |
7432 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7433 | mddev->queue->limits.discard_alignment = stripe; |
7434 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 | 7435 | |
5026d7a9 | 7436 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
3deff1a7 | 7437 | blk_queue_max_write_zeroes_sectors(mddev->queue, 0); |
5026d7a9 | 7438 | |
05616be5 | 7439 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7440 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7441 | rdev->data_offset << 9); | |
05616be5 N |
7442 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7443 | rdev->new_data_offset << 9); | |
7444 | } | |
620125f2 | 7445 | |
48920ff2 CH |
7446 | /* |
7447 | * zeroing is required, otherwise data | |
7448 | * could be lost. Consider a scenario: discard a stripe | |
7449 | * (the stripe could be inconsistent if | |
7450 | * discard_zeroes_data is 0); write one disk of the | |
7451 | * stripe (the stripe could be inconsistent again | |
7452 | * depending on which disks are used to calculate | |
7453 | * parity); the disk is broken; The stripe data of this | |
7454 | * disk is lost. | |
7455 | * | |
7456 | * We only allow DISCARD if the sysadmin has confirmed that | |
7457 | * only safe devices are in use by setting a module parameter. | |
7458 | * A better idea might be to turn DISCARD into WRITE_ZEROES | |
7459 | * requests, as that is required to be safe. | |
7460 | */ | |
7461 | if (devices_handle_discard_safely && | |
e7597e69 JS |
7462 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7463 | mddev->queue->limits.discard_granularity >= stripe) | |
8b904b5b | 7464 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, |
620125f2 SL |
7465 | mddev->queue); |
7466 | else | |
8b904b5b | 7467 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, |
620125f2 | 7468 | mddev->queue); |
1dffdddd SL |
7469 | |
7470 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7471 | } |
23032a0e | 7472 | |
845b9e22 | 7473 | if (log_init(conf, journal_dev, raid5_has_ppl(conf))) |
ff875738 | 7474 | goto abort; |
5c7e81c3 | 7475 | |
1da177e4 LT |
7476 | return 0; |
7477 | abort: | |
01f96c0a | 7478 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7479 | print_raid5_conf(conf); |
7480 | free_conf(conf); | |
1da177e4 | 7481 | mddev->private = NULL; |
cc6167b4 | 7482 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7483 | return -EIO; |
7484 | } | |
7485 | ||
afa0f557 | 7486 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7487 | { |
afa0f557 | 7488 | struct r5conf *conf = priv; |
1da177e4 | 7489 | |
95fc17aa | 7490 | free_conf(conf); |
a64c876f | 7491 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7492 | } |
7493 | ||
849674e4 | 7494 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7495 | { |
d1688a6d | 7496 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7497 | int i; |
7498 | ||
9d8f0363 | 7499 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7500 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7501 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7502 | rcu_read_lock(); |
7503 | for (i = 0; i < conf->raid_disks; i++) { | |
7504 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7505 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7506 | } | |
7507 | rcu_read_unlock(); | |
1da177e4 | 7508 | seq_printf (seq, "]"); |
1da177e4 LT |
7509 | } |
7510 | ||
d1688a6d | 7511 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7512 | { |
7513 | int i; | |
7514 | struct disk_info *tmp; | |
7515 | ||
cc6167b4 | 7516 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7517 | if (!conf) { |
cc6167b4 | 7518 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7519 | return; |
7520 | } | |
cc6167b4 | 7521 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7522 | conf->raid_disks, |
7523 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7524 | |
7525 | for (i = 0; i < conf->raid_disks; i++) { | |
7526 | char b[BDEVNAME_SIZE]; | |
7527 | tmp = conf->disks + i; | |
7528 | if (tmp->rdev) | |
cc6167b4 | 7529 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7530 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7531 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7532 | } |
7533 | } | |
7534 | ||
fd01b88c | 7535 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7536 | { |
7537 | int i; | |
d1688a6d | 7538 | struct r5conf *conf = mddev->private; |
1da177e4 | 7539 | struct disk_info *tmp; |
6b965620 N |
7540 | int count = 0; |
7541 | unsigned long flags; | |
1da177e4 LT |
7542 | |
7543 | for (i = 0; i < conf->raid_disks; i++) { | |
7544 | tmp = conf->disks + i; | |
dd054fce N |
7545 | if (tmp->replacement |
7546 | && tmp->replacement->recovery_offset == MaxSector | |
7547 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7548 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7549 | /* Replacement has just become active. */ | |
7550 | if (!tmp->rdev | |
7551 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7552 | count++; | |
7553 | if (tmp->rdev) { | |
7554 | /* Replaced device not technically faulty, | |
7555 | * but we need to be sure it gets removed | |
7556 | * and never re-added. | |
7557 | */ | |
7558 | set_bit(Faulty, &tmp->rdev->flags); | |
7559 | sysfs_notify_dirent_safe( | |
7560 | tmp->rdev->sysfs_state); | |
7561 | } | |
7562 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7563 | } else if (tmp->rdev | |
70fffd0b | 7564 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7565 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7566 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7567 | count++; |
43c73ca4 | 7568 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7569 | } |
7570 | } | |
6b965620 | 7571 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7572 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7573 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7574 | print_raid5_conf(conf); |
6b965620 | 7575 | return count; |
1da177e4 LT |
7576 | } |
7577 | ||
b8321b68 | 7578 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7579 | { |
d1688a6d | 7580 | struct r5conf *conf = mddev->private; |
1da177e4 | 7581 | int err = 0; |
b8321b68 | 7582 | int number = rdev->raid_disk; |
657e3e4d | 7583 | struct md_rdev **rdevp; |
1da177e4 LT |
7584 | struct disk_info *p = conf->disks + number; |
7585 | ||
7586 | print_raid5_conf(conf); | |
f6b6ec5c | 7587 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
c2bb6242 | 7588 | /* |
f6b6ec5c SL |
7589 | * we can't wait pending write here, as this is called in |
7590 | * raid5d, wait will deadlock. | |
84dd97a6 N |
7591 | * neilb: there is no locking about new writes here, |
7592 | * so this cannot be safe. | |
c2bb6242 | 7593 | */ |
70d466f7 SL |
7594 | if (atomic_read(&conf->active_stripes) || |
7595 | atomic_read(&conf->r5c_cached_full_stripes) || | |
7596 | atomic_read(&conf->r5c_cached_partial_stripes)) { | |
f6b6ec5c | 7597 | return -EBUSY; |
84dd97a6 | 7598 | } |
ff875738 | 7599 | log_exit(conf); |
f6b6ec5c | 7600 | return 0; |
c2bb6242 | 7601 | } |
657e3e4d N |
7602 | if (rdev == p->rdev) |
7603 | rdevp = &p->rdev; | |
7604 | else if (rdev == p->replacement) | |
7605 | rdevp = &p->replacement; | |
7606 | else | |
7607 | return 0; | |
7608 | ||
7609 | if (number >= conf->raid_disks && | |
7610 | conf->reshape_progress == MaxSector) | |
7611 | clear_bit(In_sync, &rdev->flags); | |
7612 | ||
7613 | if (test_bit(In_sync, &rdev->flags) || | |
7614 | atomic_read(&rdev->nr_pending)) { | |
7615 | err = -EBUSY; | |
7616 | goto abort; | |
7617 | } | |
7618 | /* Only remove non-faulty devices if recovery | |
7619 | * isn't possible. | |
7620 | */ | |
7621 | if (!test_bit(Faulty, &rdev->flags) && | |
7622 | mddev->recovery_disabled != conf->recovery_disabled && | |
7623 | !has_failed(conf) && | |
dd054fce | 7624 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7625 | number < conf->raid_disks) { |
7626 | err = -EBUSY; | |
7627 | goto abort; | |
7628 | } | |
7629 | *rdevp = NULL; | |
d787be40 N |
7630 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7631 | synchronize_rcu(); | |
7632 | if (atomic_read(&rdev->nr_pending)) { | |
7633 | /* lost the race, try later */ | |
7634 | err = -EBUSY; | |
7635 | *rdevp = rdev; | |
7636 | } | |
7637 | } | |
6358c239 AP |
7638 | if (!err) { |
7639 | err = log_modify(conf, rdev, false); | |
7640 | if (err) | |
7641 | goto abort; | |
7642 | } | |
d787be40 | 7643 | if (p->replacement) { |
dd054fce N |
7644 | /* We must have just cleared 'rdev' */ |
7645 | p->rdev = p->replacement; | |
7646 | clear_bit(Replacement, &p->replacement->flags); | |
7647 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7648 | * but will never see neither - if they are careful | |
7649 | */ | |
7650 | p->replacement = NULL; | |
6358c239 AP |
7651 | |
7652 | if (!err) | |
7653 | err = log_modify(conf, p->rdev, true); | |
e5bc9c3c GJ |
7654 | } |
7655 | ||
7656 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7657 | abort: |
7658 | ||
7659 | print_raid5_conf(conf); | |
7660 | return err; | |
7661 | } | |
7662 | ||
fd01b88c | 7663 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7664 | { |
d1688a6d | 7665 | struct r5conf *conf = mddev->private; |
199050ea | 7666 | int err = -EEXIST; |
1da177e4 LT |
7667 | int disk; |
7668 | struct disk_info *p; | |
6c2fce2e NB |
7669 | int first = 0; |
7670 | int last = conf->raid_disks - 1; | |
1da177e4 | 7671 | |
f6b6ec5c | 7672 | if (test_bit(Journal, &rdev->flags)) { |
f6b6ec5c SL |
7673 | if (conf->log) |
7674 | return -EBUSY; | |
7675 | ||
7676 | rdev->raid_disk = 0; | |
7677 | /* | |
7678 | * The array is in readonly mode if journal is missing, so no | |
7679 | * write requests running. We should be safe | |
7680 | */ | |
845b9e22 | 7681 | log_init(conf, rdev, false); |
f6b6ec5c SL |
7682 | return 0; |
7683 | } | |
7f0da59b N |
7684 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7685 | return -EBUSY; | |
7686 | ||
dc10c643 | 7687 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7688 | /* no point adding a device */ |
199050ea | 7689 | return -EINVAL; |
1da177e4 | 7690 | |
6c2fce2e NB |
7691 | if (rdev->raid_disk >= 0) |
7692 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7693 | |
7694 | /* | |
16a53ecc N |
7695 | * find the disk ... but prefer rdev->saved_raid_disk |
7696 | * if possible. | |
1da177e4 | 7697 | */ |
16a53ecc | 7698 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7699 | rdev->saved_raid_disk >= first && |
16a53ecc | 7700 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7701 | first = rdev->saved_raid_disk; |
7702 | ||
7703 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7704 | p = conf->disks + disk; |
7705 | if (p->rdev == NULL) { | |
b2d444d7 | 7706 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7707 | rdev->raid_disk = disk; |
72626685 N |
7708 | if (rdev->saved_raid_disk != disk) |
7709 | conf->fullsync = 1; | |
d6065f7b | 7710 | rcu_assign_pointer(p->rdev, rdev); |
6358c239 AP |
7711 | |
7712 | err = log_modify(conf, rdev, true); | |
7713 | ||
5cfb22a1 | 7714 | goto out; |
1da177e4 | 7715 | } |
5cfb22a1 N |
7716 | } |
7717 | for (disk = first; disk <= last; disk++) { | |
7718 | p = conf->disks + disk; | |
7bfec5f3 N |
7719 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7720 | p->replacement == NULL) { | |
7721 | clear_bit(In_sync, &rdev->flags); | |
7722 | set_bit(Replacement, &rdev->flags); | |
7723 | rdev->raid_disk = disk; | |
7724 | err = 0; | |
7725 | conf->fullsync = 1; | |
7726 | rcu_assign_pointer(p->replacement, rdev); | |
7727 | break; | |
7728 | } | |
7729 | } | |
5cfb22a1 | 7730 | out: |
1da177e4 | 7731 | print_raid5_conf(conf); |
199050ea | 7732 | return err; |
1da177e4 LT |
7733 | } |
7734 | ||
fd01b88c | 7735 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7736 | { |
7737 | /* no resync is happening, and there is enough space | |
7738 | * on all devices, so we can resize. | |
7739 | * We need to make sure resync covers any new space. | |
7740 | * If the array is shrinking we should possibly wait until | |
7741 | * any io in the removed space completes, but it hardly seems | |
7742 | * worth it. | |
7743 | */ | |
a4a6125a | 7744 | sector_t newsize; |
3cb5edf4 N |
7745 | struct r5conf *conf = mddev->private; |
7746 | ||
e254de6b | 7747 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 7748 | return -EINVAL; |
3cb5edf4 | 7749 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7750 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7751 | if (mddev->external_size && | |
7752 | mddev->array_sectors > newsize) | |
b522adcd | 7753 | return -EINVAL; |
a4a6125a | 7754 | if (mddev->bitmap) { |
e64e4018 | 7755 | int ret = md_bitmap_resize(mddev->bitmap, sectors, 0, 0); |
a4a6125a N |
7756 | if (ret) |
7757 | return ret; | |
7758 | } | |
7759 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7760 | if (sectors > mddev->dev_sectors && |
7761 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7762 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7763 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7764 | } | |
58c0fed4 | 7765 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7766 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7767 | return 0; |
7768 | } | |
7769 | ||
fd01b88c | 7770 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7771 | { |
7772 | /* Can only proceed if there are plenty of stripe_heads. | |
7773 | * We need a minimum of one full stripe,, and for sensible progress | |
7774 | * it is best to have about 4 times that. | |
7775 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7776 | * allow for chunk sizes up to 256K, which is probably OK. | |
7777 | * If the chunk size is greater, user-space should request more | |
7778 | * stripe_heads first. | |
7779 | */ | |
d1688a6d | 7780 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7781 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7782 | > conf->min_nr_stripes || |
01ee22b4 | 7783 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7784 | > conf->min_nr_stripes) { |
cc6167b4 N |
7785 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7786 | mdname(mddev), | |
7787 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
7788 | / STRIPE_SIZE)*4); | |
01ee22b4 N |
7789 | return 0; |
7790 | } | |
7791 | return 1; | |
7792 | } | |
7793 | ||
fd01b88c | 7794 | static int check_reshape(struct mddev *mddev) |
29269553 | 7795 | { |
d1688a6d | 7796 | struct r5conf *conf = mddev->private; |
29269553 | 7797 | |
e254de6b | 7798 | if (raid5_has_log(conf) || raid5_has_ppl(conf)) |
713cf5a6 | 7799 | return -EINVAL; |
88ce4930 N |
7800 | if (mddev->delta_disks == 0 && |
7801 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7802 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7803 | return 0; /* nothing to do */ |
674806d6 | 7804 | if (has_failed(conf)) |
ec32a2bd | 7805 | return -EINVAL; |
fdcfbbb6 | 7806 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7807 | /* We might be able to shrink, but the devices must |
7808 | * be made bigger first. | |
7809 | * For raid6, 4 is the minimum size. | |
7810 | * Otherwise 2 is the minimum | |
7811 | */ | |
7812 | int min = 2; | |
7813 | if (mddev->level == 6) | |
7814 | min = 4; | |
7815 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7816 | return -EINVAL; | |
7817 | } | |
29269553 | 7818 | |
01ee22b4 | 7819 | if (!check_stripe_cache(mddev)) |
29269553 | 7820 | return -ENOSPC; |
29269553 | 7821 | |
738a2738 N |
7822 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7823 | mddev->delta_disks > 0) | |
7824 | if (resize_chunks(conf, | |
7825 | conf->previous_raid_disks | |
7826 | + max(0, mddev->delta_disks), | |
7827 | max(mddev->new_chunk_sectors, | |
7828 | mddev->chunk_sectors) | |
7829 | ) < 0) | |
7830 | return -ENOMEM; | |
845b9e22 AP |
7831 | |
7832 | if (conf->previous_raid_disks + mddev->delta_disks <= conf->pool_size) | |
7833 | return 0; /* never bother to shrink */ | |
e56108d6 N |
7834 | return resize_stripes(conf, (conf->previous_raid_disks |
7835 | + mddev->delta_disks)); | |
63c70c4f N |
7836 | } |
7837 | ||
fd01b88c | 7838 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7839 | { |
d1688a6d | 7840 | struct r5conf *conf = mddev->private; |
3cb03002 | 7841 | struct md_rdev *rdev; |
63c70c4f | 7842 | int spares = 0; |
c04be0aa | 7843 | unsigned long flags; |
63c70c4f | 7844 | |
f416885e | 7845 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7846 | return -EBUSY; |
7847 | ||
01ee22b4 N |
7848 | if (!check_stripe_cache(mddev)) |
7849 | return -ENOSPC; | |
7850 | ||
30b67645 N |
7851 | if (has_failed(conf)) |
7852 | return -EINVAL; | |
7853 | ||
c6563a8c | 7854 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7855 | if (!test_bit(In_sync, &rdev->flags) |
7856 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7857 | spares++; |
c6563a8c | 7858 | } |
63c70c4f | 7859 | |
f416885e | 7860 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7861 | /* Not enough devices even to make a degraded array |
7862 | * of that size | |
7863 | */ | |
7864 | return -EINVAL; | |
7865 | ||
ec32a2bd N |
7866 | /* Refuse to reduce size of the array. Any reductions in |
7867 | * array size must be through explicit setting of array_size | |
7868 | * attribute. | |
7869 | */ | |
7870 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7871 | < mddev->array_sectors) { | |
cc6167b4 N |
7872 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7873 | mdname(mddev)); | |
ec32a2bd N |
7874 | return -EINVAL; |
7875 | } | |
7876 | ||
f6705578 | 7877 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7878 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7879 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7880 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7881 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7882 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7883 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7884 | conf->prev_algo = conf->algorithm; |
7885 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7886 | conf->generation++; |
7887 | /* Code that selects data_offset needs to see the generation update | |
7888 | * if reshape_progress has been set - so a memory barrier needed. | |
7889 | */ | |
7890 | smp_mb(); | |
2c810cdd | 7891 | if (mddev->reshape_backwards) |
fef9c61f N |
7892 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7893 | else | |
7894 | conf->reshape_progress = 0; | |
7895 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7896 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7897 | spin_unlock_irq(&conf->device_lock); |
7898 | ||
4d77e3ba N |
7899 | /* Now make sure any requests that proceeded on the assumption |
7900 | * the reshape wasn't running - like Discard or Read - have | |
7901 | * completed. | |
7902 | */ | |
7903 | mddev_suspend(mddev); | |
7904 | mddev_resume(mddev); | |
7905 | ||
29269553 N |
7906 | /* Add some new drives, as many as will fit. |
7907 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7908 | * Don't add devices if we are reducing the number of |
7909 | * devices in the array. This is because it is not possible | |
7910 | * to correctly record the "partially reconstructed" state of | |
7911 | * such devices during the reshape and confusion could result. | |
29269553 | 7912 | */ |
87a8dec9 | 7913 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7914 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7915 | if (rdev->raid_disk < 0 && |
7916 | !test_bit(Faulty, &rdev->flags)) { | |
7917 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7918 | if (rdev->raid_disk |
9d4c7d87 | 7919 | >= conf->previous_raid_disks) |
87a8dec9 | 7920 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7921 | else |
87a8dec9 | 7922 | rdev->recovery_offset = 0; |
36fad858 NK |
7923 | |
7924 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7925 | /* Failure here is OK */; |
50da0840 | 7926 | } |
87a8dec9 N |
7927 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7928 | && !test_bit(Faulty, &rdev->flags)) { | |
7929 | /* This is a spare that was manually added */ | |
7930 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7931 | } |
29269553 | 7932 | |
87a8dec9 N |
7933 | /* When a reshape changes the number of devices, |
7934 | * ->degraded is measured against the larger of the | |
7935 | * pre and post number of devices. | |
7936 | */ | |
ec32a2bd | 7937 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7938 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
7939 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7940 | } | |
63c70c4f | 7941 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7942 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 7943 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 7944 | |
29269553 N |
7945 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7946 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7947 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7948 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7949 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7950 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7951 | "reshape"); |
29269553 N |
7952 | if (!mddev->sync_thread) { |
7953 | mddev->recovery = 0; | |
7954 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7955 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7956 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7957 | mddev->new_chunk_sectors = |
7958 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7959 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7960 | rdev_for_each(rdev, mddev) |
7961 | rdev->new_data_offset = rdev->data_offset; | |
7962 | smp_wmb(); | |
ba8805b9 | 7963 | conf->generation --; |
fef9c61f | 7964 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7965 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7966 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7967 | spin_unlock_irq(&conf->device_lock); |
7968 | return -EAGAIN; | |
7969 | } | |
c8f517c4 | 7970 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7971 | md_wakeup_thread(mddev->sync_thread); |
7972 | md_new_event(mddev); | |
7973 | return 0; | |
7974 | } | |
29269553 | 7975 | |
ec32a2bd N |
7976 | /* This is called from the reshape thread and should make any |
7977 | * changes needed in 'conf' | |
7978 | */ | |
d1688a6d | 7979 | static void end_reshape(struct r5conf *conf) |
29269553 | 7980 | { |
29269553 | 7981 | |
f6705578 | 7982 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
db0505d3 | 7983 | struct md_rdev *rdev; |
f6705578 | 7984 | |
f6705578 | 7985 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7986 | conf->previous_raid_disks = conf->raid_disks; |
b5d27718 | 7987 | md_finish_reshape(conf->mddev); |
05616be5 | 7988 | smp_wmb(); |
fef9c61f | 7989 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7990 | conf->mddev->reshape_position = MaxSector; |
db0505d3 N |
7991 | rdev_for_each(rdev, conf->mddev) |
7992 | if (rdev->raid_disk >= 0 && | |
7993 | !test_bit(Journal, &rdev->flags) && | |
7994 | !test_bit(In_sync, &rdev->flags)) | |
7995 | rdev->recovery_offset = MaxSector; | |
f6705578 | 7996 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7997 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7998 | |
7999 | /* read-ahead size must cover two whole stripes, which is | |
8000 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
8001 | */ | |
4a5add49 | 8002 | if (conf->mddev->queue) { |
cea9c228 | 8003 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 8004 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 8005 | / PAGE_SIZE); |
dc3b17cc JK |
8006 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
8007 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 8008 | } |
29269553 | 8009 | } |
29269553 N |
8010 | } |
8011 | ||
ec32a2bd N |
8012 | /* This is called from the raid5d thread with mddev_lock held. |
8013 | * It makes config changes to the device. | |
8014 | */ | |
fd01b88c | 8015 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 8016 | { |
d1688a6d | 8017 | struct r5conf *conf = mddev->private; |
cea9c228 N |
8018 | |
8019 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
8020 | ||
8876391e | 8021 | if (mddev->delta_disks <= 0) { |
ec32a2bd | 8022 | int d; |
908f4fbd | 8023 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 8024 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 8025 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
8026 | for (d = conf->raid_disks ; |
8027 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 8028 | d++) { |
3cb03002 | 8029 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
8030 | if (rdev) |
8031 | clear_bit(In_sync, &rdev->flags); | |
8032 | rdev = conf->disks[d].replacement; | |
8033 | if (rdev) | |
8034 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 8035 | } |
cea9c228 | 8036 | } |
88ce4930 | 8037 | mddev->layout = conf->algorithm; |
09c9e5fa | 8038 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
8039 | mddev->reshape_position = MaxSector; |
8040 | mddev->delta_disks = 0; | |
2c810cdd | 8041 | mddev->reshape_backwards = 0; |
cea9c228 N |
8042 | } |
8043 | } | |
8044 | ||
b03e0ccb | 8045 | static void raid5_quiesce(struct mddev *mddev, int quiesce) |
72626685 | 8046 | { |
d1688a6d | 8047 | struct r5conf *conf = mddev->private; |
72626685 | 8048 | |
b03e0ccb N |
8049 | if (quiesce) { |
8050 | /* stop all writes */ | |
566c09c5 | 8051 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
8052 | /* '2' tells resync/reshape to pause so that all |
8053 | * active stripes can drain | |
8054 | */ | |
a39f7afd | 8055 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 8056 | conf->quiesce = 2; |
b1b46486 | 8057 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
8058 | atomic_read(&conf->active_stripes) == 0 && |
8059 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
8060 | unlock_all_device_hash_locks_irq(conf), |
8061 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 8062 | conf->quiesce = 1; |
566c09c5 | 8063 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
8064 | /* allow reshape to continue */ |
8065 | wake_up(&conf->wait_for_overlap); | |
b03e0ccb N |
8066 | } else { |
8067 | /* re-enable writes */ | |
566c09c5 | 8068 | lock_all_device_hash_locks_irq(conf); |
72626685 | 8069 | conf->quiesce = 0; |
b1b46486 | 8070 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 8071 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 8072 | unlock_all_device_hash_locks_irq(conf); |
72626685 | 8073 | } |
1532d9e8 | 8074 | log_quiesce(conf, quiesce); |
72626685 | 8075 | } |
b15c2e57 | 8076 | |
fd01b88c | 8077 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 8078 | { |
e373ab10 | 8079 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 8080 | sector_t sectors; |
54071b38 | 8081 | |
f1b29bca | 8082 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 8083 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
8084 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
8085 | mdname(mddev)); | |
f1b29bca DW |
8086 | return ERR_PTR(-EINVAL); |
8087 | } | |
8088 | ||
e373ab10 N |
8089 | sectors = raid0_conf->strip_zone[0].zone_end; |
8090 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 8091 | mddev->dev_sectors = sectors; |
f1b29bca | 8092 | mddev->new_level = level; |
54071b38 TM |
8093 | mddev->new_layout = ALGORITHM_PARITY_N; |
8094 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
8095 | mddev->raid_disks += 1; | |
8096 | mddev->delta_disks = 1; | |
8097 | /* make sure it will be not marked as dirty */ | |
8098 | mddev->recovery_cp = MaxSector; | |
8099 | ||
8100 | return setup_conf(mddev); | |
8101 | } | |
8102 | ||
fd01b88c | 8103 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
8104 | { |
8105 | int chunksect; | |
6995f0b2 | 8106 | void *ret; |
d562b0c4 N |
8107 | |
8108 | if (mddev->raid_disks != 2 || | |
8109 | mddev->degraded > 1) | |
8110 | return ERR_PTR(-EINVAL); | |
8111 | ||
8112 | /* Should check if there are write-behind devices? */ | |
8113 | ||
8114 | chunksect = 64*2; /* 64K by default */ | |
8115 | ||
8116 | /* The array must be an exact multiple of chunksize */ | |
8117 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8118 | chunksect >>= 1; | |
8119 | ||
8120 | if ((chunksect<<9) < STRIPE_SIZE) | |
8121 | /* array size does not allow a suitable chunk size */ | |
8122 | return ERR_PTR(-EINVAL); | |
8123 | ||
8124 | mddev->new_level = 5; | |
8125 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8126 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8127 | |
6995f0b2 | 8128 | ret = setup_conf(mddev); |
32cd7cbb | 8129 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8130 | mddev_clear_unsupported_flags(mddev, |
8131 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8132 | return ret; |
d562b0c4 N |
8133 | } |
8134 | ||
fd01b88c | 8135 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8136 | { |
8137 | int new_layout; | |
8138 | ||
8139 | switch (mddev->layout) { | |
8140 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8141 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8142 | break; | |
8143 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8144 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8145 | break; | |
8146 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8147 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8148 | break; | |
8149 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8150 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8151 | break; | |
8152 | case ALGORITHM_PARITY_0_6: | |
8153 | new_layout = ALGORITHM_PARITY_0; | |
8154 | break; | |
8155 | case ALGORITHM_PARITY_N: | |
8156 | new_layout = ALGORITHM_PARITY_N; | |
8157 | break; | |
8158 | default: | |
8159 | return ERR_PTR(-EINVAL); | |
8160 | } | |
8161 | mddev->new_level = 5; | |
8162 | mddev->new_layout = new_layout; | |
8163 | mddev->delta_disks = -1; | |
8164 | mddev->raid_disks -= 1; | |
8165 | return setup_conf(mddev); | |
8166 | } | |
8167 | ||
fd01b88c | 8168 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8169 | { |
88ce4930 N |
8170 | /* For a 2-drive array, the layout and chunk size can be changed |
8171 | * immediately as not restriping is needed. | |
8172 | * For larger arrays we record the new value - after validation | |
8173 | * to be used by a reshape pass. | |
b3546035 | 8174 | */ |
d1688a6d | 8175 | struct r5conf *conf = mddev->private; |
597a711b | 8176 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8177 | |
597a711b | 8178 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8179 | return -EINVAL; |
8180 | if (new_chunk > 0) { | |
0ba459d2 | 8181 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8182 | return -EINVAL; |
597a711b | 8183 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8184 | return -EINVAL; |
597a711b | 8185 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8186 | /* not factor of array size */ |
8187 | return -EINVAL; | |
8188 | } | |
8189 | ||
8190 | /* They look valid */ | |
8191 | ||
88ce4930 | 8192 | if (mddev->raid_disks == 2) { |
597a711b N |
8193 | /* can make the change immediately */ |
8194 | if (mddev->new_layout >= 0) { | |
8195 | conf->algorithm = mddev->new_layout; | |
8196 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8197 | } |
8198 | if (new_chunk > 0) { | |
597a711b N |
8199 | conf->chunk_sectors = new_chunk ; |
8200 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8201 | } |
2953079c | 8202 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8203 | md_wakeup_thread(mddev->thread); |
b3546035 | 8204 | } |
50ac168a | 8205 | return check_reshape(mddev); |
88ce4930 N |
8206 | } |
8207 | ||
fd01b88c | 8208 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8209 | { |
597a711b | 8210 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8211 | |
597a711b | 8212 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8213 | return -EINVAL; |
b3546035 | 8214 | if (new_chunk > 0) { |
0ba459d2 | 8215 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8216 | return -EINVAL; |
597a711b | 8217 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8218 | return -EINVAL; |
597a711b | 8219 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8220 | /* not factor of array size */ |
8221 | return -EINVAL; | |
b3546035 | 8222 | } |
88ce4930 N |
8223 | |
8224 | /* They look valid */ | |
50ac168a | 8225 | return check_reshape(mddev); |
b3546035 N |
8226 | } |
8227 | ||
fd01b88c | 8228 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8229 | { |
8230 | /* raid5 can take over: | |
f1b29bca | 8231 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8232 | * raid1 - if there are two drives. We need to know the chunk size |
8233 | * raid4 - trivial - just use a raid4 layout. | |
8234 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8235 | */ |
f1b29bca DW |
8236 | if (mddev->level == 0) |
8237 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8238 | if (mddev->level == 1) |
8239 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8240 | if (mddev->level == 4) { |
8241 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8242 | mddev->new_level = 5; | |
8243 | return setup_conf(mddev); | |
8244 | } | |
fc9739c6 N |
8245 | if (mddev->level == 6) |
8246 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8247 | |
8248 | return ERR_PTR(-EINVAL); | |
8249 | } | |
8250 | ||
fd01b88c | 8251 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8252 | { |
f1b29bca DW |
8253 | /* raid4 can take over: |
8254 | * raid0 - if there is only one strip zone | |
8255 | * raid5 - if layout is right | |
a78d38a1 | 8256 | */ |
f1b29bca DW |
8257 | if (mddev->level == 0) |
8258 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8259 | if (mddev->level == 5 && |
8260 | mddev->layout == ALGORITHM_PARITY_N) { | |
8261 | mddev->new_layout = 0; | |
8262 | mddev->new_level = 4; | |
8263 | return setup_conf(mddev); | |
8264 | } | |
8265 | return ERR_PTR(-EINVAL); | |
8266 | } | |
d562b0c4 | 8267 | |
84fc4b56 | 8268 | static struct md_personality raid5_personality; |
245f46c2 | 8269 | |
fd01b88c | 8270 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8271 | { |
8272 | /* Currently can only take over a raid5. We map the | |
8273 | * personality to an equivalent raid6 personality | |
8274 | * with the Q block at the end. | |
8275 | */ | |
8276 | int new_layout; | |
8277 | ||
8278 | if (mddev->pers != &raid5_personality) | |
8279 | return ERR_PTR(-EINVAL); | |
8280 | if (mddev->degraded > 1) | |
8281 | return ERR_PTR(-EINVAL); | |
8282 | if (mddev->raid_disks > 253) | |
8283 | return ERR_PTR(-EINVAL); | |
8284 | if (mddev->raid_disks < 3) | |
8285 | return ERR_PTR(-EINVAL); | |
8286 | ||
8287 | switch (mddev->layout) { | |
8288 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8289 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8290 | break; | |
8291 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8292 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8293 | break; | |
8294 | case ALGORITHM_LEFT_SYMMETRIC: | |
8295 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8296 | break; | |
8297 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8298 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8299 | break; | |
8300 | case ALGORITHM_PARITY_0: | |
8301 | new_layout = ALGORITHM_PARITY_0_6; | |
8302 | break; | |
8303 | case ALGORITHM_PARITY_N: | |
8304 | new_layout = ALGORITHM_PARITY_N; | |
8305 | break; | |
8306 | default: | |
8307 | return ERR_PTR(-EINVAL); | |
8308 | } | |
8309 | mddev->new_level = 6; | |
8310 | mddev->new_layout = new_layout; | |
8311 | mddev->delta_disks = 1; | |
8312 | mddev->raid_disks += 1; | |
8313 | return setup_conf(mddev); | |
8314 | } | |
8315 | ||
ba903a3e AP |
8316 | static int raid5_change_consistency_policy(struct mddev *mddev, const char *buf) |
8317 | { | |
8318 | struct r5conf *conf; | |
8319 | int err; | |
8320 | ||
8321 | err = mddev_lock(mddev); | |
8322 | if (err) | |
8323 | return err; | |
8324 | conf = mddev->private; | |
8325 | if (!conf) { | |
8326 | mddev_unlock(mddev); | |
8327 | return -ENODEV; | |
8328 | } | |
8329 | ||
845b9e22 | 8330 | if (strncmp(buf, "ppl", 3) == 0) { |
0bb0c105 | 8331 | /* ppl only works with RAID 5 */ |
845b9e22 AP |
8332 | if (!raid5_has_ppl(conf) && conf->level == 5) { |
8333 | err = log_init(conf, NULL, true); | |
8334 | if (!err) { | |
8335 | err = resize_stripes(conf, conf->pool_size); | |
8336 | if (err) | |
8337 | log_exit(conf); | |
8338 | } | |
0bb0c105 SL |
8339 | } else |
8340 | err = -EINVAL; | |
8341 | } else if (strncmp(buf, "resync", 6) == 0) { | |
8342 | if (raid5_has_ppl(conf)) { | |
8343 | mddev_suspend(mddev); | |
8344 | log_exit(conf); | |
0bb0c105 | 8345 | mddev_resume(mddev); |
845b9e22 | 8346 | err = resize_stripes(conf, conf->pool_size); |
0bb0c105 SL |
8347 | } else if (test_bit(MD_HAS_JOURNAL, &conf->mddev->flags) && |
8348 | r5l_log_disk_error(conf)) { | |
8349 | bool journal_dev_exists = false; | |
8350 | struct md_rdev *rdev; | |
8351 | ||
8352 | rdev_for_each(rdev, mddev) | |
8353 | if (test_bit(Journal, &rdev->flags)) { | |
8354 | journal_dev_exists = true; | |
8355 | break; | |
8356 | } | |
8357 | ||
8358 | if (!journal_dev_exists) { | |
8359 | mddev_suspend(mddev); | |
8360 | clear_bit(MD_HAS_JOURNAL, &mddev->flags); | |
8361 | mddev_resume(mddev); | |
8362 | } else /* need remove journal device first */ | |
8363 | err = -EBUSY; | |
8364 | } else | |
8365 | err = -EINVAL; | |
ba903a3e AP |
8366 | } else { |
8367 | err = -EINVAL; | |
8368 | } | |
8369 | ||
8370 | if (!err) | |
8371 | md_update_sb(mddev, 1); | |
8372 | ||
8373 | mddev_unlock(mddev); | |
8374 | ||
8375 | return err; | |
8376 | } | |
8377 | ||
d5d885fd SL |
8378 | static int raid5_start(struct mddev *mddev) |
8379 | { | |
8380 | struct r5conf *conf = mddev->private; | |
8381 | ||
8382 | return r5l_start(conf->log); | |
8383 | } | |
8384 | ||
84fc4b56 | 8385 | static struct md_personality raid6_personality = |
16a53ecc N |
8386 | { |
8387 | .name = "raid6", | |
8388 | .level = 6, | |
8389 | .owner = THIS_MODULE, | |
849674e4 SL |
8390 | .make_request = raid5_make_request, |
8391 | .run = raid5_run, | |
d5d885fd | 8392 | .start = raid5_start, |
afa0f557 | 8393 | .free = raid5_free, |
849674e4 SL |
8394 | .status = raid5_status, |
8395 | .error_handler = raid5_error, | |
16a53ecc N |
8396 | .hot_add_disk = raid5_add_disk, |
8397 | .hot_remove_disk= raid5_remove_disk, | |
8398 | .spare_active = raid5_spare_active, | |
849674e4 | 8399 | .sync_request = raid5_sync_request, |
16a53ecc | 8400 | .resize = raid5_resize, |
80c3a6ce | 8401 | .size = raid5_size, |
50ac168a | 8402 | .check_reshape = raid6_check_reshape, |
f416885e | 8403 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8404 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8405 | .quiesce = raid5_quiesce, |
245f46c2 | 8406 | .takeover = raid6_takeover, |
5c675f83 | 8407 | .congested = raid5_congested, |
0bb0c105 | 8408 | .change_consistency_policy = raid5_change_consistency_policy, |
16a53ecc | 8409 | }; |
84fc4b56 | 8410 | static struct md_personality raid5_personality = |
1da177e4 LT |
8411 | { |
8412 | .name = "raid5", | |
2604b703 | 8413 | .level = 5, |
1da177e4 | 8414 | .owner = THIS_MODULE, |
849674e4 SL |
8415 | .make_request = raid5_make_request, |
8416 | .run = raid5_run, | |
d5d885fd | 8417 | .start = raid5_start, |
afa0f557 | 8418 | .free = raid5_free, |
849674e4 SL |
8419 | .status = raid5_status, |
8420 | .error_handler = raid5_error, | |
1da177e4 LT |
8421 | .hot_add_disk = raid5_add_disk, |
8422 | .hot_remove_disk= raid5_remove_disk, | |
8423 | .spare_active = raid5_spare_active, | |
849674e4 | 8424 | .sync_request = raid5_sync_request, |
1da177e4 | 8425 | .resize = raid5_resize, |
80c3a6ce | 8426 | .size = raid5_size, |
63c70c4f N |
8427 | .check_reshape = raid5_check_reshape, |
8428 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8429 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8430 | .quiesce = raid5_quiesce, |
d562b0c4 | 8431 | .takeover = raid5_takeover, |
5c675f83 | 8432 | .congested = raid5_congested, |
ba903a3e | 8433 | .change_consistency_policy = raid5_change_consistency_policy, |
1da177e4 LT |
8434 | }; |
8435 | ||
84fc4b56 | 8436 | static struct md_personality raid4_personality = |
1da177e4 | 8437 | { |
2604b703 N |
8438 | .name = "raid4", |
8439 | .level = 4, | |
8440 | .owner = THIS_MODULE, | |
849674e4 SL |
8441 | .make_request = raid5_make_request, |
8442 | .run = raid5_run, | |
d5d885fd | 8443 | .start = raid5_start, |
afa0f557 | 8444 | .free = raid5_free, |
849674e4 SL |
8445 | .status = raid5_status, |
8446 | .error_handler = raid5_error, | |
2604b703 N |
8447 | .hot_add_disk = raid5_add_disk, |
8448 | .hot_remove_disk= raid5_remove_disk, | |
8449 | .spare_active = raid5_spare_active, | |
849674e4 | 8450 | .sync_request = raid5_sync_request, |
2604b703 | 8451 | .resize = raid5_resize, |
80c3a6ce | 8452 | .size = raid5_size, |
3d37890b N |
8453 | .check_reshape = raid5_check_reshape, |
8454 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8455 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8456 | .quiesce = raid5_quiesce, |
a78d38a1 | 8457 | .takeover = raid4_takeover, |
5c675f83 | 8458 | .congested = raid5_congested, |
0bb0c105 | 8459 | .change_consistency_policy = raid5_change_consistency_policy, |
2604b703 N |
8460 | }; |
8461 | ||
8462 | static int __init raid5_init(void) | |
8463 | { | |
29c6d1bb SAS |
8464 | int ret; |
8465 | ||
851c30c9 SL |
8466 | raid5_wq = alloc_workqueue("raid5wq", |
8467 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8468 | if (!raid5_wq) | |
8469 | return -ENOMEM; | |
29c6d1bb SAS |
8470 | |
8471 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8472 | "md/raid5:prepare", | |
8473 | raid456_cpu_up_prepare, | |
8474 | raid456_cpu_dead); | |
8475 | if (ret) { | |
8476 | destroy_workqueue(raid5_wq); | |
8477 | return ret; | |
8478 | } | |
16a53ecc | 8479 | register_md_personality(&raid6_personality); |
2604b703 N |
8480 | register_md_personality(&raid5_personality); |
8481 | register_md_personality(&raid4_personality); | |
8482 | return 0; | |
1da177e4 LT |
8483 | } |
8484 | ||
2604b703 | 8485 | static void raid5_exit(void) |
1da177e4 | 8486 | { |
16a53ecc | 8487 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8488 | unregister_md_personality(&raid5_personality); |
8489 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8490 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8491 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8492 | } |
8493 | ||
8494 | module_init(raid5_init); | |
8495 | module_exit(raid5_exit); | |
8496 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8497 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8498 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8499 | MODULE_ALIAS("md-raid5"); |
8500 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8501 | MODULE_ALIAS("md-level-5"); |
8502 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8503 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8504 | MODULE_ALIAS("md-raid6"); | |
8505 | MODULE_ALIAS("md-level-6"); | |
8506 | ||
8507 | /* This used to be two separate modules, they were: */ | |
8508 | MODULE_ALIAS("raid5"); | |
8509 | MODULE_ALIAS("raid6"); |