]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid5.c : Multiple Devices driver for Linux | |
3 | * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman | |
4 | * Copyright (C) 1999, 2000 Ingo Molnar | |
16a53ecc | 5 | * Copyright (C) 2002, 2003 H. Peter Anvin |
1da177e4 | 6 | * |
16a53ecc N |
7 | * RAID-4/5/6 management functions. |
8 | * Thanks to Penguin Computing for making the RAID-6 development possible | |
9 | * by donating a test server! | |
1da177e4 LT |
10 | * |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
ae3c20cc N |
21 | /* |
22 | * BITMAP UNPLUGGING: | |
23 | * | |
24 | * The sequencing for updating the bitmap reliably is a little | |
25 | * subtle (and I got it wrong the first time) so it deserves some | |
26 | * explanation. | |
27 | * | |
28 | * We group bitmap updates into batches. Each batch has a number. | |
29 | * We may write out several batches at once, but that isn't very important. | |
7c13edc8 N |
30 | * conf->seq_write is the number of the last batch successfully written. |
31 | * conf->seq_flush is the number of the last batch that was closed to | |
ae3c20cc N |
32 | * new additions. |
33 | * When we discover that we will need to write to any block in a stripe | |
34 | * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq | |
7c13edc8 | 35 | * the number of the batch it will be in. This is seq_flush+1. |
ae3c20cc N |
36 | * When we are ready to do a write, if that batch hasn't been written yet, |
37 | * we plug the array and queue the stripe for later. | |
38 | * When an unplug happens, we increment bm_flush, thus closing the current | |
39 | * batch. | |
40 | * When we notice that bm_flush > bm_write, we write out all pending updates | |
41 | * to the bitmap, and advance bm_write to where bm_flush was. | |
42 | * This may occasionally write a bit out twice, but is sure never to | |
43 | * miss any bits. | |
44 | */ | |
1da177e4 | 45 | |
bff61975 | 46 | #include <linux/blkdev.h> |
f6705578 | 47 | #include <linux/kthread.h> |
f701d589 | 48 | #include <linux/raid/pq.h> |
91c00924 | 49 | #include <linux/async_tx.h> |
056075c7 | 50 | #include <linux/module.h> |
07a3b417 | 51 | #include <linux/async.h> |
bff61975 | 52 | #include <linux/seq_file.h> |
36d1c647 | 53 | #include <linux/cpu.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
8bda470e | 55 | #include <linux/ratelimit.h> |
851c30c9 | 56 | #include <linux/nodemask.h> |
46d5b785 | 57 | #include <linux/flex_array.h> |
3f07c014 IM |
58 | #include <linux/sched/signal.h> |
59 | ||
a9add5d9 N |
60 | #include <trace/events/block.h> |
61 | ||
43b2e5d8 | 62 | #include "md.h" |
bff61975 | 63 | #include "raid5.h" |
54071b38 | 64 | #include "raid0.h" |
ef740c37 | 65 | #include "bitmap.h" |
72626685 | 66 | |
394ed8e4 SL |
67 | #define UNSUPPORTED_MDDEV_FLAGS (1L << MD_FAILFAST_SUPPORTED) |
68 | ||
851c30c9 SL |
69 | #define cpu_to_group(cpu) cpu_to_node(cpu) |
70 | #define ANY_GROUP NUMA_NO_NODE | |
71 | ||
8e0e99ba N |
72 | static bool devices_handle_discard_safely = false; |
73 | module_param(devices_handle_discard_safely, bool, 0644); | |
74 | MODULE_PARM_DESC(devices_handle_discard_safely, | |
75 | "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions"); | |
851c30c9 | 76 | static struct workqueue_struct *raid5_wq; |
1da177e4 | 77 | |
d1688a6d | 78 | static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect) |
db298e19 N |
79 | { |
80 | int hash = (sect >> STRIPE_SHIFT) & HASH_MASK; | |
81 | return &conf->stripe_hashtbl[hash]; | |
82 | } | |
1da177e4 | 83 | |
566c09c5 SL |
84 | static inline int stripe_hash_locks_hash(sector_t sect) |
85 | { | |
86 | return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK; | |
87 | } | |
88 | ||
89 | static inline void lock_device_hash_lock(struct r5conf *conf, int hash) | |
90 | { | |
91 | spin_lock_irq(conf->hash_locks + hash); | |
92 | spin_lock(&conf->device_lock); | |
93 | } | |
94 | ||
95 | static inline void unlock_device_hash_lock(struct r5conf *conf, int hash) | |
96 | { | |
97 | spin_unlock(&conf->device_lock); | |
98 | spin_unlock_irq(conf->hash_locks + hash); | |
99 | } | |
100 | ||
101 | static inline void lock_all_device_hash_locks_irq(struct r5conf *conf) | |
102 | { | |
103 | int i; | |
104 | local_irq_disable(); | |
105 | spin_lock(conf->hash_locks); | |
106 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
107 | spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks); | |
108 | spin_lock(&conf->device_lock); | |
109 | } | |
110 | ||
111 | static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf) | |
112 | { | |
113 | int i; | |
114 | spin_unlock(&conf->device_lock); | |
115 | for (i = NR_STRIPE_HASH_LOCKS; i; i--) | |
116 | spin_unlock(conf->hash_locks + i - 1); | |
117 | local_irq_enable(); | |
118 | } | |
119 | ||
d0dabf7e N |
120 | /* Find first data disk in a raid6 stripe */ |
121 | static inline int raid6_d0(struct stripe_head *sh) | |
122 | { | |
67cc2b81 N |
123 | if (sh->ddf_layout) |
124 | /* ddf always start from first device */ | |
125 | return 0; | |
126 | /* md starts just after Q block */ | |
d0dabf7e N |
127 | if (sh->qd_idx == sh->disks - 1) |
128 | return 0; | |
129 | else | |
130 | return sh->qd_idx + 1; | |
131 | } | |
16a53ecc N |
132 | static inline int raid6_next_disk(int disk, int raid_disks) |
133 | { | |
134 | disk++; | |
135 | return (disk < raid_disks) ? disk : 0; | |
136 | } | |
a4456856 | 137 | |
d0dabf7e N |
138 | /* When walking through the disks in a raid5, starting at raid6_d0, |
139 | * We need to map each disk to a 'slot', where the data disks are slot | |
140 | * 0 .. raid_disks-3, the parity disk is raid_disks-2 and the Q disk | |
141 | * is raid_disks-1. This help does that mapping. | |
142 | */ | |
67cc2b81 N |
143 | static int raid6_idx_to_slot(int idx, struct stripe_head *sh, |
144 | int *count, int syndrome_disks) | |
d0dabf7e | 145 | { |
6629542e | 146 | int slot = *count; |
67cc2b81 | 147 | |
e4424fee | 148 | if (sh->ddf_layout) |
6629542e | 149 | (*count)++; |
d0dabf7e | 150 | if (idx == sh->pd_idx) |
67cc2b81 | 151 | return syndrome_disks; |
d0dabf7e | 152 | if (idx == sh->qd_idx) |
67cc2b81 | 153 | return syndrome_disks + 1; |
e4424fee | 154 | if (!sh->ddf_layout) |
6629542e | 155 | (*count)++; |
d0dabf7e N |
156 | return slot; |
157 | } | |
158 | ||
34a6f80e | 159 | static void return_io(struct bio_list *return_bi) |
a4456856 | 160 | { |
34a6f80e N |
161 | struct bio *bi; |
162 | while ((bi = bio_list_pop(return_bi)) != NULL) { | |
4f024f37 | 163 | bi->bi_iter.bi_size = 0; |
0a82a8d1 LT |
164 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
165 | bi, 0); | |
4246a0b6 | 166 | bio_endio(bi); |
a4456856 DW |
167 | } |
168 | } | |
169 | ||
d1688a6d | 170 | static void print_raid5_conf (struct r5conf *conf); |
1da177e4 | 171 | |
600aa109 DW |
172 | static int stripe_operations_active(struct stripe_head *sh) |
173 | { | |
174 | return sh->check_state || sh->reconstruct_state || | |
175 | test_bit(STRIPE_BIOFILL_RUN, &sh->state) || | |
176 | test_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
177 | } | |
178 | ||
535ae4eb SL |
179 | static bool stripe_is_lowprio(struct stripe_head *sh) |
180 | { | |
181 | return (test_bit(STRIPE_R5C_FULL_STRIPE, &sh->state) || | |
182 | test_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) && | |
183 | !test_bit(STRIPE_R5C_CACHING, &sh->state); | |
184 | } | |
185 | ||
851c30c9 SL |
186 | static void raid5_wakeup_stripe_thread(struct stripe_head *sh) |
187 | { | |
188 | struct r5conf *conf = sh->raid_conf; | |
189 | struct r5worker_group *group; | |
bfc90cb0 | 190 | int thread_cnt; |
851c30c9 SL |
191 | int i, cpu = sh->cpu; |
192 | ||
193 | if (!cpu_online(cpu)) { | |
194 | cpu = cpumask_any(cpu_online_mask); | |
195 | sh->cpu = cpu; | |
196 | } | |
197 | ||
198 | if (list_empty(&sh->lru)) { | |
199 | struct r5worker_group *group; | |
200 | group = conf->worker_groups + cpu_to_group(cpu); | |
535ae4eb SL |
201 | if (stripe_is_lowprio(sh)) |
202 | list_add_tail(&sh->lru, &group->loprio_list); | |
203 | else | |
204 | list_add_tail(&sh->lru, &group->handle_list); | |
bfc90cb0 SL |
205 | group->stripes_cnt++; |
206 | sh->group = group; | |
851c30c9 SL |
207 | } |
208 | ||
209 | if (conf->worker_cnt_per_group == 0) { | |
210 | md_wakeup_thread(conf->mddev->thread); | |
211 | return; | |
212 | } | |
213 | ||
214 | group = conf->worker_groups + cpu_to_group(sh->cpu); | |
215 | ||
bfc90cb0 SL |
216 | group->workers[0].working = true; |
217 | /* at least one worker should run to avoid race */ | |
218 | queue_work_on(sh->cpu, raid5_wq, &group->workers[0].work); | |
219 | ||
220 | thread_cnt = group->stripes_cnt / MAX_STRIPE_BATCH - 1; | |
221 | /* wakeup more workers */ | |
222 | for (i = 1; i < conf->worker_cnt_per_group && thread_cnt > 0; i++) { | |
223 | if (group->workers[i].working == false) { | |
224 | group->workers[i].working = true; | |
225 | queue_work_on(sh->cpu, raid5_wq, | |
226 | &group->workers[i].work); | |
227 | thread_cnt--; | |
228 | } | |
229 | } | |
851c30c9 SL |
230 | } |
231 | ||
566c09c5 SL |
232 | static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh, |
233 | struct list_head *temp_inactive_list) | |
1da177e4 | 234 | { |
1e6d690b SL |
235 | int i; |
236 | int injournal = 0; /* number of date pages with R5_InJournal */ | |
237 | ||
4eb788df SL |
238 | BUG_ON(!list_empty(&sh->lru)); |
239 | BUG_ON(atomic_read(&conf->active_stripes)==0); | |
1e6d690b SL |
240 | |
241 | if (r5c_is_writeback(conf->log)) | |
242 | for (i = sh->disks; i--; ) | |
243 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) | |
244 | injournal++; | |
a39f7afd SL |
245 | /* |
246 | * When quiesce in r5c write back, set STRIPE_HANDLE for stripes with | |
247 | * data in journal, so they are not released to cached lists | |
248 | */ | |
249 | if (conf->quiesce && r5c_is_writeback(conf->log) && | |
250 | !test_bit(STRIPE_HANDLE, &sh->state) && injournal != 0) { | |
251 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
252 | r5c_make_stripe_write_out(sh); | |
253 | set_bit(STRIPE_HANDLE, &sh->state); | |
254 | } | |
1e6d690b | 255 | |
4eb788df SL |
256 | if (test_bit(STRIPE_HANDLE, &sh->state)) { |
257 | if (test_bit(STRIPE_DELAYED, &sh->state) && | |
ad3ab8b6 | 258 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
4eb788df | 259 | list_add_tail(&sh->lru, &conf->delayed_list); |
ad3ab8b6 | 260 | else if (test_bit(STRIPE_BIT_DELAY, &sh->state) && |
4eb788df SL |
261 | sh->bm_seq - conf->seq_write > 0) |
262 | list_add_tail(&sh->lru, &conf->bitmap_list); | |
263 | else { | |
264 | clear_bit(STRIPE_DELAYED, &sh->state); | |
265 | clear_bit(STRIPE_BIT_DELAY, &sh->state); | |
851c30c9 | 266 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
267 | if (stripe_is_lowprio(sh)) |
268 | list_add_tail(&sh->lru, | |
269 | &conf->loprio_list); | |
270 | else | |
271 | list_add_tail(&sh->lru, | |
272 | &conf->handle_list); | |
851c30c9 SL |
273 | } else { |
274 | raid5_wakeup_stripe_thread(sh); | |
275 | return; | |
276 | } | |
4eb788df SL |
277 | } |
278 | md_wakeup_thread(conf->mddev->thread); | |
279 | } else { | |
280 | BUG_ON(stripe_operations_active(sh)); | |
281 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
282 | if (atomic_dec_return(&conf->preread_active_stripes) | |
283 | < IO_THRESHOLD) | |
284 | md_wakeup_thread(conf->mddev->thread); | |
285 | atomic_dec(&conf->active_stripes); | |
1e6d690b SL |
286 | if (!test_bit(STRIPE_EXPANDING, &sh->state)) { |
287 | if (!r5c_is_writeback(conf->log)) | |
288 | list_add_tail(&sh->lru, temp_inactive_list); | |
289 | else { | |
290 | WARN_ON(test_bit(R5_InJournal, &sh->dev[sh->pd_idx].flags)); | |
291 | if (injournal == 0) | |
292 | list_add_tail(&sh->lru, temp_inactive_list); | |
293 | else if (injournal == conf->raid_disks - conf->max_degraded) { | |
294 | /* full stripe */ | |
295 | if (!test_and_set_bit(STRIPE_R5C_FULL_STRIPE, &sh->state)) | |
296 | atomic_inc(&conf->r5c_cached_full_stripes); | |
297 | if (test_and_clear_bit(STRIPE_R5C_PARTIAL_STRIPE, &sh->state)) | |
298 | atomic_dec(&conf->r5c_cached_partial_stripes); | |
299 | list_add_tail(&sh->lru, &conf->r5c_full_stripe_list); | |
a39f7afd | 300 | r5c_check_cached_full_stripe(conf); |
03b047f4 SL |
301 | } else |
302 | /* | |
303 | * STRIPE_R5C_PARTIAL_STRIPE is set in | |
304 | * r5c_try_caching_write(). No need to | |
305 | * set it again. | |
306 | */ | |
1e6d690b | 307 | list_add_tail(&sh->lru, &conf->r5c_partial_stripe_list); |
1e6d690b SL |
308 | } |
309 | } | |
1da177e4 LT |
310 | } |
311 | } | |
d0dabf7e | 312 | |
566c09c5 SL |
313 | static void __release_stripe(struct r5conf *conf, struct stripe_head *sh, |
314 | struct list_head *temp_inactive_list) | |
4eb788df SL |
315 | { |
316 | if (atomic_dec_and_test(&sh->count)) | |
566c09c5 SL |
317 | do_release_stripe(conf, sh, temp_inactive_list); |
318 | } | |
319 | ||
320 | /* | |
321 | * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list | |
322 | * | |
323 | * Be careful: Only one task can add/delete stripes from temp_inactive_list at | |
324 | * given time. Adding stripes only takes device lock, while deleting stripes | |
325 | * only takes hash lock. | |
326 | */ | |
327 | static void release_inactive_stripe_list(struct r5conf *conf, | |
328 | struct list_head *temp_inactive_list, | |
329 | int hash) | |
330 | { | |
331 | int size; | |
6ab2a4b8 | 332 | bool do_wakeup = false; |
566c09c5 SL |
333 | unsigned long flags; |
334 | ||
335 | if (hash == NR_STRIPE_HASH_LOCKS) { | |
336 | size = NR_STRIPE_HASH_LOCKS; | |
337 | hash = NR_STRIPE_HASH_LOCKS - 1; | |
338 | } else | |
339 | size = 1; | |
340 | while (size) { | |
341 | struct list_head *list = &temp_inactive_list[size - 1]; | |
342 | ||
343 | /* | |
6d036f7d | 344 | * We don't hold any lock here yet, raid5_get_active_stripe() might |
566c09c5 SL |
345 | * remove stripes from the list |
346 | */ | |
347 | if (!list_empty_careful(list)) { | |
348 | spin_lock_irqsave(conf->hash_locks + hash, flags); | |
4bda556a SL |
349 | if (list_empty(conf->inactive_list + hash) && |
350 | !list_empty(list)) | |
351 | atomic_dec(&conf->empty_inactive_list_nr); | |
566c09c5 | 352 | list_splice_tail_init(list, conf->inactive_list + hash); |
6ab2a4b8 | 353 | do_wakeup = true; |
566c09c5 SL |
354 | spin_unlock_irqrestore(conf->hash_locks + hash, flags); |
355 | } | |
356 | size--; | |
357 | hash--; | |
358 | } | |
359 | ||
360 | if (do_wakeup) { | |
6ab2a4b8 | 361 | wake_up(&conf->wait_for_stripe); |
b1b46486 YL |
362 | if (atomic_read(&conf->active_stripes) == 0) |
363 | wake_up(&conf->wait_for_quiescent); | |
566c09c5 SL |
364 | if (conf->retry_read_aligned) |
365 | md_wakeup_thread(conf->mddev->thread); | |
366 | } | |
4eb788df SL |
367 | } |
368 | ||
773ca82f | 369 | /* should hold conf->device_lock already */ |
566c09c5 SL |
370 | static int release_stripe_list(struct r5conf *conf, |
371 | struct list_head *temp_inactive_list) | |
773ca82f | 372 | { |
eae8263f | 373 | struct stripe_head *sh, *t; |
773ca82f SL |
374 | int count = 0; |
375 | struct llist_node *head; | |
376 | ||
377 | head = llist_del_all(&conf->released_stripes); | |
d265d9dc | 378 | head = llist_reverse_order(head); |
eae8263f | 379 | llist_for_each_entry_safe(sh, t, head, release_list) { |
566c09c5 SL |
380 | int hash; |
381 | ||
773ca82f SL |
382 | /* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */ |
383 | smp_mb(); | |
384 | clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); | |
385 | /* | |
386 | * Don't worry the bit is set here, because if the bit is set | |
387 | * again, the count is always > 1. This is true for | |
388 | * STRIPE_ON_UNPLUG_LIST bit too. | |
389 | */ | |
566c09c5 SL |
390 | hash = sh->hash_lock_index; |
391 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
773ca82f SL |
392 | count++; |
393 | } | |
394 | ||
395 | return count; | |
396 | } | |
397 | ||
6d036f7d | 398 | void raid5_release_stripe(struct stripe_head *sh) |
1da177e4 | 399 | { |
d1688a6d | 400 | struct r5conf *conf = sh->raid_conf; |
1da177e4 | 401 | unsigned long flags; |
566c09c5 SL |
402 | struct list_head list; |
403 | int hash; | |
773ca82f | 404 | bool wakeup; |
16a53ecc | 405 | |
cf170f3f ES |
406 | /* Avoid release_list until the last reference. |
407 | */ | |
408 | if (atomic_add_unless(&sh->count, -1, 1)) | |
409 | return; | |
410 | ||
ad4068de | 411 | if (unlikely(!conf->mddev->thread) || |
412 | test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) | |
773ca82f SL |
413 | goto slow_path; |
414 | wakeup = llist_add(&sh->release_list, &conf->released_stripes); | |
415 | if (wakeup) | |
416 | md_wakeup_thread(conf->mddev->thread); | |
417 | return; | |
418 | slow_path: | |
4eb788df | 419 | local_irq_save(flags); |
773ca82f | 420 | /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ |
4eb788df | 421 | if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { |
566c09c5 SL |
422 | INIT_LIST_HEAD(&list); |
423 | hash = sh->hash_lock_index; | |
424 | do_release_stripe(conf, sh, &list); | |
4eb788df | 425 | spin_unlock(&conf->device_lock); |
566c09c5 | 426 | release_inactive_stripe_list(conf, &list, hash); |
4eb788df SL |
427 | } |
428 | local_irq_restore(flags); | |
1da177e4 LT |
429 | } |
430 | ||
fccddba0 | 431 | static inline void remove_hash(struct stripe_head *sh) |
1da177e4 | 432 | { |
45b4233c DW |
433 | pr_debug("remove_hash(), stripe %llu\n", |
434 | (unsigned long long)sh->sector); | |
1da177e4 | 435 | |
fccddba0 | 436 | hlist_del_init(&sh->hash); |
1da177e4 LT |
437 | } |
438 | ||
d1688a6d | 439 | static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh) |
1da177e4 | 440 | { |
fccddba0 | 441 | struct hlist_head *hp = stripe_hash(conf, sh->sector); |
1da177e4 | 442 | |
45b4233c DW |
443 | pr_debug("insert_hash(), stripe %llu\n", |
444 | (unsigned long long)sh->sector); | |
1da177e4 | 445 | |
fccddba0 | 446 | hlist_add_head(&sh->hash, hp); |
1da177e4 LT |
447 | } |
448 | ||
1da177e4 | 449 | /* find an idle stripe, make sure it is unhashed, and return it. */ |
566c09c5 | 450 | static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash) |
1da177e4 LT |
451 | { |
452 | struct stripe_head *sh = NULL; | |
453 | struct list_head *first; | |
454 | ||
566c09c5 | 455 | if (list_empty(conf->inactive_list + hash)) |
1da177e4 | 456 | goto out; |
566c09c5 | 457 | first = (conf->inactive_list + hash)->next; |
1da177e4 LT |
458 | sh = list_entry(first, struct stripe_head, lru); |
459 | list_del_init(first); | |
460 | remove_hash(sh); | |
461 | atomic_inc(&conf->active_stripes); | |
566c09c5 | 462 | BUG_ON(hash != sh->hash_lock_index); |
4bda556a SL |
463 | if (list_empty(conf->inactive_list + hash)) |
464 | atomic_inc(&conf->empty_inactive_list_nr); | |
1da177e4 LT |
465 | out: |
466 | return sh; | |
467 | } | |
468 | ||
e4e11e38 | 469 | static void shrink_buffers(struct stripe_head *sh) |
1da177e4 LT |
470 | { |
471 | struct page *p; | |
472 | int i; | |
e4e11e38 | 473 | int num = sh->raid_conf->pool_size; |
1da177e4 | 474 | |
e4e11e38 | 475 | for (i = 0; i < num ; i++) { |
d592a996 | 476 | WARN_ON(sh->dev[i].page != sh->dev[i].orig_page); |
1da177e4 LT |
477 | p = sh->dev[i].page; |
478 | if (!p) | |
479 | continue; | |
480 | sh->dev[i].page = NULL; | |
2d1f3b5d | 481 | put_page(p); |
1da177e4 LT |
482 | } |
483 | } | |
484 | ||
a9683a79 | 485 | static int grow_buffers(struct stripe_head *sh, gfp_t gfp) |
1da177e4 LT |
486 | { |
487 | int i; | |
e4e11e38 | 488 | int num = sh->raid_conf->pool_size; |
1da177e4 | 489 | |
e4e11e38 | 490 | for (i = 0; i < num; i++) { |
1da177e4 LT |
491 | struct page *page; |
492 | ||
a9683a79 | 493 | if (!(page = alloc_page(gfp))) { |
1da177e4 LT |
494 | return 1; |
495 | } | |
496 | sh->dev[i].page = page; | |
d592a996 | 497 | sh->dev[i].orig_page = page; |
1da177e4 LT |
498 | } |
499 | return 0; | |
500 | } | |
501 | ||
784052ec | 502 | static void raid5_build_block(struct stripe_head *sh, int i, int previous); |
d1688a6d | 503 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 504 | struct stripe_head *sh); |
1da177e4 | 505 | |
b5663ba4 | 506 | static void init_stripe(struct stripe_head *sh, sector_t sector, int previous) |
1da177e4 | 507 | { |
d1688a6d | 508 | struct r5conf *conf = sh->raid_conf; |
566c09c5 | 509 | int i, seq; |
1da177e4 | 510 | |
78bafebd ES |
511 | BUG_ON(atomic_read(&sh->count) != 0); |
512 | BUG_ON(test_bit(STRIPE_HANDLE, &sh->state)); | |
600aa109 | 513 | BUG_ON(stripe_operations_active(sh)); |
59fc630b | 514 | BUG_ON(sh->batch_head); |
d84e0f10 | 515 | |
45b4233c | 516 | pr_debug("init_stripe called, stripe %llu\n", |
b8e6a15a | 517 | (unsigned long long)sector); |
566c09c5 SL |
518 | retry: |
519 | seq = read_seqcount_begin(&conf->gen_lock); | |
86b42c71 | 520 | sh->generation = conf->generation - previous; |
b5663ba4 | 521 | sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
1da177e4 | 522 | sh->sector = sector; |
911d4ee8 | 523 | stripe_set_idx(sector, conf, previous, sh); |
1da177e4 LT |
524 | sh->state = 0; |
525 | ||
7ecaa1e6 | 526 | for (i = sh->disks; i--; ) { |
1da177e4 LT |
527 | struct r5dev *dev = &sh->dev[i]; |
528 | ||
d84e0f10 | 529 | if (dev->toread || dev->read || dev->towrite || dev->written || |
1da177e4 | 530 | test_bit(R5_LOCKED, &dev->flags)) { |
cc6167b4 | 531 | pr_err("sector=%llx i=%d %p %p %p %p %d\n", |
1da177e4 | 532 | (unsigned long long)sh->sector, i, dev->toread, |
d84e0f10 | 533 | dev->read, dev->towrite, dev->written, |
1da177e4 | 534 | test_bit(R5_LOCKED, &dev->flags)); |
8cfa7b0f | 535 | WARN_ON(1); |
1da177e4 LT |
536 | } |
537 | dev->flags = 0; | |
784052ec | 538 | raid5_build_block(sh, i, previous); |
1da177e4 | 539 | } |
566c09c5 SL |
540 | if (read_seqcount_retry(&conf->gen_lock, seq)) |
541 | goto retry; | |
7a87f434 | 542 | sh->overwrite_disks = 0; |
1da177e4 | 543 | insert_hash(conf, sh); |
851c30c9 | 544 | sh->cpu = smp_processor_id(); |
da41ba65 | 545 | set_bit(STRIPE_BATCH_READY, &sh->state); |
1da177e4 LT |
546 | } |
547 | ||
d1688a6d | 548 | static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector, |
86b42c71 | 549 | short generation) |
1da177e4 LT |
550 | { |
551 | struct stripe_head *sh; | |
552 | ||
45b4233c | 553 | pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector); |
b67bfe0d | 554 | hlist_for_each_entry(sh, stripe_hash(conf, sector), hash) |
86b42c71 | 555 | if (sh->sector == sector && sh->generation == generation) |
1da177e4 | 556 | return sh; |
45b4233c | 557 | pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector); |
1da177e4 LT |
558 | return NULL; |
559 | } | |
560 | ||
674806d6 N |
561 | /* |
562 | * Need to check if array has failed when deciding whether to: | |
563 | * - start an array | |
564 | * - remove non-faulty devices | |
565 | * - add a spare | |
566 | * - allow a reshape | |
567 | * This determination is simple when no reshape is happening. | |
568 | * However if there is a reshape, we need to carefully check | |
569 | * both the before and after sections. | |
570 | * This is because some failed devices may only affect one | |
571 | * of the two sections, and some non-in_sync devices may | |
572 | * be insync in the section most affected by failed devices. | |
573 | */ | |
2e38a37f | 574 | int raid5_calc_degraded(struct r5conf *conf) |
674806d6 | 575 | { |
908f4fbd | 576 | int degraded, degraded2; |
674806d6 | 577 | int i; |
674806d6 N |
578 | |
579 | rcu_read_lock(); | |
580 | degraded = 0; | |
581 | for (i = 0; i < conf->previous_raid_disks; i++) { | |
3cb03002 | 582 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
583 | if (rdev && test_bit(Faulty, &rdev->flags)) |
584 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 N |
585 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
586 | degraded++; | |
587 | else if (test_bit(In_sync, &rdev->flags)) | |
588 | ; | |
589 | else | |
590 | /* not in-sync or faulty. | |
591 | * If the reshape increases the number of devices, | |
592 | * this is being recovered by the reshape, so | |
593 | * this 'previous' section is not in_sync. | |
594 | * If the number of devices is being reduced however, | |
595 | * the device can only be part of the array if | |
596 | * we are reverting a reshape, so this section will | |
597 | * be in-sync. | |
598 | */ | |
599 | if (conf->raid_disks >= conf->previous_raid_disks) | |
600 | degraded++; | |
601 | } | |
602 | rcu_read_unlock(); | |
908f4fbd N |
603 | if (conf->raid_disks == conf->previous_raid_disks) |
604 | return degraded; | |
674806d6 | 605 | rcu_read_lock(); |
908f4fbd | 606 | degraded2 = 0; |
674806d6 | 607 | for (i = 0; i < conf->raid_disks; i++) { |
3cb03002 | 608 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
e5c86471 N |
609 | if (rdev && test_bit(Faulty, &rdev->flags)) |
610 | rdev = rcu_dereference(conf->disks[i].replacement); | |
674806d6 | 611 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
908f4fbd | 612 | degraded2++; |
674806d6 N |
613 | else if (test_bit(In_sync, &rdev->flags)) |
614 | ; | |
615 | else | |
616 | /* not in-sync or faulty. | |
617 | * If reshape increases the number of devices, this | |
618 | * section has already been recovered, else it | |
619 | * almost certainly hasn't. | |
620 | */ | |
621 | if (conf->raid_disks <= conf->previous_raid_disks) | |
908f4fbd | 622 | degraded2++; |
674806d6 N |
623 | } |
624 | rcu_read_unlock(); | |
908f4fbd N |
625 | if (degraded2 > degraded) |
626 | return degraded2; | |
627 | return degraded; | |
628 | } | |
629 | ||
630 | static int has_failed(struct r5conf *conf) | |
631 | { | |
632 | int degraded; | |
633 | ||
634 | if (conf->mddev->reshape_position == MaxSector) | |
635 | return conf->mddev->degraded > conf->max_degraded; | |
636 | ||
2e38a37f | 637 | degraded = raid5_calc_degraded(conf); |
674806d6 N |
638 | if (degraded > conf->max_degraded) |
639 | return 1; | |
640 | return 0; | |
641 | } | |
642 | ||
6d036f7d SL |
643 | struct stripe_head * |
644 | raid5_get_active_stripe(struct r5conf *conf, sector_t sector, | |
645 | int previous, int noblock, int noquiesce) | |
1da177e4 LT |
646 | { |
647 | struct stripe_head *sh; | |
566c09c5 | 648 | int hash = stripe_hash_locks_hash(sector); |
ff00d3b4 | 649 | int inc_empty_inactive_list_flag; |
1da177e4 | 650 | |
45b4233c | 651 | pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector); |
1da177e4 | 652 | |
566c09c5 | 653 | spin_lock_irq(conf->hash_locks + hash); |
1da177e4 LT |
654 | |
655 | do { | |
b1b46486 | 656 | wait_event_lock_irq(conf->wait_for_quiescent, |
a8c906ca | 657 | conf->quiesce == 0 || noquiesce, |
566c09c5 | 658 | *(conf->hash_locks + hash)); |
86b42c71 | 659 | sh = __find_stripe(conf, sector, conf->generation - previous); |
1da177e4 | 660 | if (!sh) { |
edbe83ab | 661 | if (!test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) { |
566c09c5 | 662 | sh = get_free_stripe(conf, hash); |
713bc5c2 SL |
663 | if (!sh && !test_bit(R5_DID_ALLOC, |
664 | &conf->cache_state)) | |
edbe83ab N |
665 | set_bit(R5_ALLOC_MORE, |
666 | &conf->cache_state); | |
667 | } | |
1da177e4 LT |
668 | if (noblock && sh == NULL) |
669 | break; | |
a39f7afd SL |
670 | |
671 | r5c_check_stripe_cache_usage(conf); | |
1da177e4 | 672 | if (!sh) { |
5423399a N |
673 | set_bit(R5_INACTIVE_BLOCKED, |
674 | &conf->cache_state); | |
a39f7afd | 675 | r5l_wake_reclaim(conf->log, 0); |
6ab2a4b8 SL |
676 | wait_event_lock_irq( |
677 | conf->wait_for_stripe, | |
566c09c5 SL |
678 | !list_empty(conf->inactive_list + hash) && |
679 | (atomic_read(&conf->active_stripes) | |
680 | < (conf->max_nr_stripes * 3 / 4) | |
5423399a N |
681 | || !test_bit(R5_INACTIVE_BLOCKED, |
682 | &conf->cache_state)), | |
6ab2a4b8 | 683 | *(conf->hash_locks + hash)); |
5423399a N |
684 | clear_bit(R5_INACTIVE_BLOCKED, |
685 | &conf->cache_state); | |
7da9d450 | 686 | } else { |
b5663ba4 | 687 | init_stripe(sh, sector, previous); |
7da9d450 N |
688 | atomic_inc(&sh->count); |
689 | } | |
e240c183 | 690 | } else if (!atomic_inc_not_zero(&sh->count)) { |
6d183de4 | 691 | spin_lock(&conf->device_lock); |
e240c183 | 692 | if (!atomic_read(&sh->count)) { |
1da177e4 LT |
693 | if (!test_bit(STRIPE_HANDLE, &sh->state)) |
694 | atomic_inc(&conf->active_stripes); | |
5af9bef7 N |
695 | BUG_ON(list_empty(&sh->lru) && |
696 | !test_bit(STRIPE_EXPANDING, &sh->state)); | |
ff00d3b4 ZL |
697 | inc_empty_inactive_list_flag = 0; |
698 | if (!list_empty(conf->inactive_list + hash)) | |
699 | inc_empty_inactive_list_flag = 1; | |
16a53ecc | 700 | list_del_init(&sh->lru); |
ff00d3b4 ZL |
701 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
702 | atomic_inc(&conf->empty_inactive_list_nr); | |
bfc90cb0 SL |
703 | if (sh->group) { |
704 | sh->group->stripes_cnt--; | |
705 | sh->group = NULL; | |
706 | } | |
1da177e4 | 707 | } |
7da9d450 | 708 | atomic_inc(&sh->count); |
6d183de4 | 709 | spin_unlock(&conf->device_lock); |
1da177e4 LT |
710 | } |
711 | } while (sh == NULL); | |
712 | ||
566c09c5 | 713 | spin_unlock_irq(conf->hash_locks + hash); |
1da177e4 LT |
714 | return sh; |
715 | } | |
716 | ||
7a87f434 | 717 | static bool is_full_stripe_write(struct stripe_head *sh) |
718 | { | |
719 | BUG_ON(sh->overwrite_disks > (sh->disks - sh->raid_conf->max_degraded)); | |
720 | return sh->overwrite_disks == (sh->disks - sh->raid_conf->max_degraded); | |
721 | } | |
722 | ||
59fc630b | 723 | static void lock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) |
724 | { | |
725 | local_irq_disable(); | |
726 | if (sh1 > sh2) { | |
727 | spin_lock(&sh2->stripe_lock); | |
728 | spin_lock_nested(&sh1->stripe_lock, 1); | |
729 | } else { | |
730 | spin_lock(&sh1->stripe_lock); | |
731 | spin_lock_nested(&sh2->stripe_lock, 1); | |
732 | } | |
733 | } | |
734 | ||
735 | static void unlock_two_stripes(struct stripe_head *sh1, struct stripe_head *sh2) | |
736 | { | |
737 | spin_unlock(&sh1->stripe_lock); | |
738 | spin_unlock(&sh2->stripe_lock); | |
739 | local_irq_enable(); | |
740 | } | |
741 | ||
742 | /* Only freshly new full stripe normal write stripe can be added to a batch list */ | |
743 | static bool stripe_can_batch(struct stripe_head *sh) | |
744 | { | |
9c3e333d SL |
745 | struct r5conf *conf = sh->raid_conf; |
746 | ||
747 | if (conf->log) | |
748 | return false; | |
59fc630b | 749 | return test_bit(STRIPE_BATCH_READY, &sh->state) && |
d0852df5 | 750 | !test_bit(STRIPE_BITMAP_PENDING, &sh->state) && |
59fc630b | 751 | is_full_stripe_write(sh); |
752 | } | |
753 | ||
754 | /* we only do back search */ | |
755 | static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh) | |
756 | { | |
757 | struct stripe_head *head; | |
758 | sector_t head_sector, tmp_sec; | |
759 | int hash; | |
760 | int dd_idx; | |
ff00d3b4 | 761 | int inc_empty_inactive_list_flag; |
59fc630b | 762 | |
59fc630b | 763 | /* Don't cross chunks, so stripe pd_idx/qd_idx is the same */ |
764 | tmp_sec = sh->sector; | |
765 | if (!sector_div(tmp_sec, conf->chunk_sectors)) | |
766 | return; | |
767 | head_sector = sh->sector - STRIPE_SECTORS; | |
768 | ||
769 | hash = stripe_hash_locks_hash(head_sector); | |
770 | spin_lock_irq(conf->hash_locks + hash); | |
771 | head = __find_stripe(conf, head_sector, conf->generation); | |
772 | if (head && !atomic_inc_not_zero(&head->count)) { | |
773 | spin_lock(&conf->device_lock); | |
774 | if (!atomic_read(&head->count)) { | |
775 | if (!test_bit(STRIPE_HANDLE, &head->state)) | |
776 | atomic_inc(&conf->active_stripes); | |
777 | BUG_ON(list_empty(&head->lru) && | |
778 | !test_bit(STRIPE_EXPANDING, &head->state)); | |
ff00d3b4 ZL |
779 | inc_empty_inactive_list_flag = 0; |
780 | if (!list_empty(conf->inactive_list + hash)) | |
781 | inc_empty_inactive_list_flag = 1; | |
59fc630b | 782 | list_del_init(&head->lru); |
ff00d3b4 ZL |
783 | if (list_empty(conf->inactive_list + hash) && inc_empty_inactive_list_flag) |
784 | atomic_inc(&conf->empty_inactive_list_nr); | |
59fc630b | 785 | if (head->group) { |
786 | head->group->stripes_cnt--; | |
787 | head->group = NULL; | |
788 | } | |
789 | } | |
790 | atomic_inc(&head->count); | |
791 | spin_unlock(&conf->device_lock); | |
792 | } | |
793 | spin_unlock_irq(conf->hash_locks + hash); | |
794 | ||
795 | if (!head) | |
796 | return; | |
797 | if (!stripe_can_batch(head)) | |
798 | goto out; | |
799 | ||
800 | lock_two_stripes(head, sh); | |
801 | /* clear_batch_ready clear the flag */ | |
802 | if (!stripe_can_batch(head) || !stripe_can_batch(sh)) | |
803 | goto unlock_out; | |
804 | ||
805 | if (sh->batch_head) | |
806 | goto unlock_out; | |
807 | ||
808 | dd_idx = 0; | |
809 | while (dd_idx == sh->pd_idx || dd_idx == sh->qd_idx) | |
810 | dd_idx++; | |
1eff9d32 | 811 | if (head->dev[dd_idx].towrite->bi_opf != sh->dev[dd_idx].towrite->bi_opf || |
796a5cf0 | 812 | bio_op(head->dev[dd_idx].towrite) != bio_op(sh->dev[dd_idx].towrite)) |
59fc630b | 813 | goto unlock_out; |
814 | ||
815 | if (head->batch_head) { | |
816 | spin_lock(&head->batch_head->batch_lock); | |
817 | /* This batch list is already running */ | |
818 | if (!stripe_can_batch(head)) { | |
819 | spin_unlock(&head->batch_head->batch_lock); | |
820 | goto unlock_out; | |
821 | } | |
822 | ||
823 | /* | |
824 | * at this point, head's BATCH_READY could be cleared, but we | |
825 | * can still add the stripe to batch list | |
826 | */ | |
827 | list_add(&sh->batch_list, &head->batch_list); | |
828 | spin_unlock(&head->batch_head->batch_lock); | |
829 | ||
830 | sh->batch_head = head->batch_head; | |
831 | } else { | |
832 | head->batch_head = head; | |
833 | sh->batch_head = head->batch_head; | |
834 | spin_lock(&head->batch_lock); | |
835 | list_add_tail(&sh->batch_list, &head->batch_list); | |
836 | spin_unlock(&head->batch_lock); | |
837 | } | |
838 | ||
839 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
840 | if (atomic_dec_return(&conf->preread_active_stripes) | |
841 | < IO_THRESHOLD) | |
842 | md_wakeup_thread(conf->mddev->thread); | |
843 | ||
2b6b2457 N |
844 | if (test_and_clear_bit(STRIPE_BIT_DELAY, &sh->state)) { |
845 | int seq = sh->bm_seq; | |
846 | if (test_bit(STRIPE_BIT_DELAY, &sh->batch_head->state) && | |
847 | sh->batch_head->bm_seq > seq) | |
848 | seq = sh->batch_head->bm_seq; | |
849 | set_bit(STRIPE_BIT_DELAY, &sh->batch_head->state); | |
850 | sh->batch_head->bm_seq = seq; | |
851 | } | |
852 | ||
59fc630b | 853 | atomic_inc(&sh->count); |
854 | unlock_out: | |
855 | unlock_two_stripes(head, sh); | |
856 | out: | |
6d036f7d | 857 | raid5_release_stripe(head); |
59fc630b | 858 | } |
859 | ||
05616be5 N |
860 | /* Determine if 'data_offset' or 'new_data_offset' should be used |
861 | * in this stripe_head. | |
862 | */ | |
863 | static int use_new_offset(struct r5conf *conf, struct stripe_head *sh) | |
864 | { | |
865 | sector_t progress = conf->reshape_progress; | |
866 | /* Need a memory barrier to make sure we see the value | |
867 | * of conf->generation, or ->data_offset that was set before | |
868 | * reshape_progress was updated. | |
869 | */ | |
870 | smp_rmb(); | |
871 | if (progress == MaxSector) | |
872 | return 0; | |
873 | if (sh->generation == conf->generation - 1) | |
874 | return 0; | |
875 | /* We are in a reshape, and this is a new-generation stripe, | |
876 | * so use new_data_offset. | |
877 | */ | |
878 | return 1; | |
879 | } | |
880 | ||
765d704d SL |
881 | static void flush_deferred_bios(struct r5conf *conf) |
882 | { | |
883 | struct bio_list tmp; | |
884 | struct bio *bio; | |
885 | ||
886 | if (!conf->batch_bio_dispatch || !conf->group_cnt) | |
887 | return; | |
888 | ||
889 | bio_list_init(&tmp); | |
890 | spin_lock(&conf->pending_bios_lock); | |
891 | bio_list_merge(&tmp, &conf->pending_bios); | |
892 | bio_list_init(&conf->pending_bios); | |
893 | spin_unlock(&conf->pending_bios_lock); | |
894 | ||
895 | while ((bio = bio_list_pop(&tmp))) | |
896 | generic_make_request(bio); | |
897 | } | |
898 | ||
899 | static void defer_bio_issue(struct r5conf *conf, struct bio *bio) | |
900 | { | |
901 | /* | |
902 | * change group_cnt will drain all bios, so this is safe | |
903 | * | |
904 | * A read generally means a read-modify-write, which usually means a | |
905 | * randwrite, so we don't delay it | |
906 | */ | |
907 | if (!conf->batch_bio_dispatch || !conf->group_cnt || | |
908 | bio_op(bio) == REQ_OP_READ) { | |
909 | generic_make_request(bio); | |
910 | return; | |
911 | } | |
912 | spin_lock(&conf->pending_bios_lock); | |
913 | bio_list_add(&conf->pending_bios, bio); | |
914 | spin_unlock(&conf->pending_bios_lock); | |
915 | md_wakeup_thread(conf->mddev->thread); | |
916 | } | |
917 | ||
6712ecf8 | 918 | static void |
4246a0b6 | 919 | raid5_end_read_request(struct bio *bi); |
6712ecf8 | 920 | static void |
4246a0b6 | 921 | raid5_end_write_request(struct bio *bi); |
91c00924 | 922 | |
c4e5ac0a | 923 | static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s) |
91c00924 | 924 | { |
d1688a6d | 925 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 926 | int i, disks = sh->disks; |
59fc630b | 927 | struct stripe_head *head_sh = sh; |
91c00924 DW |
928 | |
929 | might_sleep(); | |
930 | ||
1e6d690b SL |
931 | if (!test_bit(STRIPE_R5C_CACHING, &sh->state)) { |
932 | /* writing out phase */ | |
d7bd398e SL |
933 | if (s->waiting_extra_page) |
934 | return; | |
1e6d690b SL |
935 | if (r5l_write_stripe(conf->log, sh) == 0) |
936 | return; | |
937 | } else { /* caching phase */ | |
938 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) { | |
939 | r5c_cache_data(conf->log, sh, s); | |
940 | return; | |
941 | } | |
942 | } | |
943 | ||
91c00924 | 944 | for (i = disks; i--; ) { |
796a5cf0 | 945 | int op, op_flags = 0; |
9a3e1101 | 946 | int replace_only = 0; |
977df362 N |
947 | struct bio *bi, *rbi; |
948 | struct md_rdev *rdev, *rrdev = NULL; | |
59fc630b | 949 | |
950 | sh = head_sh; | |
e9c7469b | 951 | if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) { |
796a5cf0 | 952 | op = REQ_OP_WRITE; |
e9c7469b | 953 | if (test_and_clear_bit(R5_WantFUA, &sh->dev[i].flags)) |
70fd7614 | 954 | op_flags = REQ_FUA; |
9e444768 | 955 | if (test_bit(R5_Discard, &sh->dev[i].flags)) |
796a5cf0 | 956 | op = REQ_OP_DISCARD; |
e9c7469b | 957 | } else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) |
796a5cf0 | 958 | op = REQ_OP_READ; |
9a3e1101 N |
959 | else if (test_and_clear_bit(R5_WantReplace, |
960 | &sh->dev[i].flags)) { | |
796a5cf0 | 961 | op = REQ_OP_WRITE; |
9a3e1101 N |
962 | replace_only = 1; |
963 | } else | |
91c00924 | 964 | continue; |
bc0934f0 | 965 | if (test_and_clear_bit(R5_SyncIO, &sh->dev[i].flags)) |
796a5cf0 | 966 | op_flags |= REQ_SYNC; |
91c00924 | 967 | |
59fc630b | 968 | again: |
91c00924 | 969 | bi = &sh->dev[i].req; |
977df362 | 970 | rbi = &sh->dev[i].rreq; /* For writing to replacement */ |
91c00924 | 971 | |
91c00924 | 972 | rcu_read_lock(); |
9a3e1101 | 973 | rrdev = rcu_dereference(conf->disks[i].replacement); |
dd054fce N |
974 | smp_mb(); /* Ensure that if rrdev is NULL, rdev won't be */ |
975 | rdev = rcu_dereference(conf->disks[i].rdev); | |
976 | if (!rdev) { | |
977 | rdev = rrdev; | |
978 | rrdev = NULL; | |
979 | } | |
796a5cf0 | 980 | if (op_is_write(op)) { |
9a3e1101 N |
981 | if (replace_only) |
982 | rdev = NULL; | |
dd054fce N |
983 | if (rdev == rrdev) |
984 | /* We raced and saw duplicates */ | |
985 | rrdev = NULL; | |
9a3e1101 | 986 | } else { |
59fc630b | 987 | if (test_bit(R5_ReadRepl, &head_sh->dev[i].flags) && rrdev) |
9a3e1101 N |
988 | rdev = rrdev; |
989 | rrdev = NULL; | |
990 | } | |
977df362 | 991 | |
91c00924 DW |
992 | if (rdev && test_bit(Faulty, &rdev->flags)) |
993 | rdev = NULL; | |
994 | if (rdev) | |
995 | atomic_inc(&rdev->nr_pending); | |
977df362 N |
996 | if (rrdev && test_bit(Faulty, &rrdev->flags)) |
997 | rrdev = NULL; | |
998 | if (rrdev) | |
999 | atomic_inc(&rrdev->nr_pending); | |
91c00924 DW |
1000 | rcu_read_unlock(); |
1001 | ||
73e92e51 | 1002 | /* We have already checked bad blocks for reads. Now |
977df362 N |
1003 | * need to check for writes. We never accept write errors |
1004 | * on the replacement, so we don't to check rrdev. | |
73e92e51 | 1005 | */ |
796a5cf0 | 1006 | while (op_is_write(op) && rdev && |
73e92e51 N |
1007 | test_bit(WriteErrorSeen, &rdev->flags)) { |
1008 | sector_t first_bad; | |
1009 | int bad_sectors; | |
1010 | int bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
1011 | &first_bad, &bad_sectors); | |
1012 | if (!bad) | |
1013 | break; | |
1014 | ||
1015 | if (bad < 0) { | |
1016 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1017 | if (!conf->mddev->external && | |
2953079c | 1018 | conf->mddev->sb_flags) { |
73e92e51 N |
1019 | /* It is very unlikely, but we might |
1020 | * still need to write out the | |
1021 | * bad block log - better give it | |
1022 | * a chance*/ | |
1023 | md_check_recovery(conf->mddev); | |
1024 | } | |
1850753d | 1025 | /* |
1026 | * Because md_wait_for_blocked_rdev | |
1027 | * will dec nr_pending, we must | |
1028 | * increment it first. | |
1029 | */ | |
1030 | atomic_inc(&rdev->nr_pending); | |
73e92e51 N |
1031 | md_wait_for_blocked_rdev(rdev, conf->mddev); |
1032 | } else { | |
1033 | /* Acknowledged bad block - skip the write */ | |
1034 | rdev_dec_pending(rdev, conf->mddev); | |
1035 | rdev = NULL; | |
1036 | } | |
1037 | } | |
1038 | ||
91c00924 | 1039 | if (rdev) { |
9a3e1101 N |
1040 | if (s->syncing || s->expanding || s->expanded |
1041 | || s->replacing) | |
91c00924 DW |
1042 | md_sync_acct(rdev->bdev, STRIPE_SECTORS); |
1043 | ||
2b7497f0 DW |
1044 | set_bit(STRIPE_IO_STARTED, &sh->state); |
1045 | ||
91c00924 | 1046 | bi->bi_bdev = rdev->bdev; |
796a5cf0 MC |
1047 | bio_set_op_attrs(bi, op, op_flags); |
1048 | bi->bi_end_io = op_is_write(op) | |
2f6db2a7 KO |
1049 | ? raid5_end_write_request |
1050 | : raid5_end_read_request; | |
1051 | bi->bi_private = sh; | |
1052 | ||
6296b960 | 1053 | pr_debug("%s: for %llu schedule op %d on disc %d\n", |
e46b272b | 1054 | __func__, (unsigned long long)sh->sector, |
1eff9d32 | 1055 | bi->bi_opf, i); |
91c00924 | 1056 | atomic_inc(&sh->count); |
59fc630b | 1057 | if (sh != head_sh) |
1058 | atomic_inc(&head_sh->count); | |
05616be5 | 1059 | if (use_new_offset(conf, sh)) |
4f024f37 | 1060 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1061 | + rdev->new_data_offset); |
1062 | else | |
4f024f37 | 1063 | bi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1064 | + rdev->data_offset); |
59fc630b | 1065 | if (test_bit(R5_ReadNoMerge, &head_sh->dev[i].flags)) |
1eff9d32 | 1066 | bi->bi_opf |= REQ_NOMERGE; |
3f9e7c14 | 1067 | |
d592a996 SL |
1068 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1069 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
86aa1397 SL |
1070 | |
1071 | if (!op_is_write(op) && | |
1072 | test_bit(R5_InJournal, &sh->dev[i].flags)) | |
1073 | /* | |
1074 | * issuing read for a page in journal, this | |
1075 | * must be preparing for prexor in rmw; read | |
1076 | * the data into orig_page | |
1077 | */ | |
1078 | sh->dev[i].vec.bv_page = sh->dev[i].orig_page; | |
1079 | else | |
1080 | sh->dev[i].vec.bv_page = sh->dev[i].page; | |
4997b72e | 1081 | bi->bi_vcnt = 1; |
91c00924 DW |
1082 | bi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1083 | bi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1084 | bi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1085 | /* |
1086 | * If this is discard request, set bi_vcnt 0. We don't | |
1087 | * want to confuse SCSI because SCSI will replace payload | |
1088 | */ | |
796a5cf0 | 1089 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1090 | bi->bi_vcnt = 0; |
977df362 N |
1091 | if (rrdev) |
1092 | set_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags); | |
e3620a3a JB |
1093 | |
1094 | if (conf->mddev->gendisk) | |
1095 | trace_block_bio_remap(bdev_get_queue(bi->bi_bdev), | |
1096 | bi, disk_devt(conf->mddev->gendisk), | |
1097 | sh->dev[i].sector); | |
765d704d | 1098 | defer_bio_issue(conf, bi); |
977df362 N |
1099 | } |
1100 | if (rrdev) { | |
9a3e1101 N |
1101 | if (s->syncing || s->expanding || s->expanded |
1102 | || s->replacing) | |
977df362 N |
1103 | md_sync_acct(rrdev->bdev, STRIPE_SECTORS); |
1104 | ||
1105 | set_bit(STRIPE_IO_STARTED, &sh->state); | |
1106 | ||
1107 | rbi->bi_bdev = rrdev->bdev; | |
796a5cf0 MC |
1108 | bio_set_op_attrs(rbi, op, op_flags); |
1109 | BUG_ON(!op_is_write(op)); | |
2f6db2a7 KO |
1110 | rbi->bi_end_io = raid5_end_write_request; |
1111 | rbi->bi_private = sh; | |
1112 | ||
6296b960 | 1113 | pr_debug("%s: for %llu schedule op %d on " |
977df362 N |
1114 | "replacement disc %d\n", |
1115 | __func__, (unsigned long long)sh->sector, | |
1eff9d32 | 1116 | rbi->bi_opf, i); |
977df362 | 1117 | atomic_inc(&sh->count); |
59fc630b | 1118 | if (sh != head_sh) |
1119 | atomic_inc(&head_sh->count); | |
05616be5 | 1120 | if (use_new_offset(conf, sh)) |
4f024f37 | 1121 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 N |
1122 | + rrdev->new_data_offset); |
1123 | else | |
4f024f37 | 1124 | rbi->bi_iter.bi_sector = (sh->sector |
05616be5 | 1125 | + rrdev->data_offset); |
d592a996 SL |
1126 | if (test_bit(R5_SkipCopy, &sh->dev[i].flags)) |
1127 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
1128 | sh->dev[i].rvec.bv_page = sh->dev[i].page; | |
4997b72e | 1129 | rbi->bi_vcnt = 1; |
977df362 N |
1130 | rbi->bi_io_vec[0].bv_len = STRIPE_SIZE; |
1131 | rbi->bi_io_vec[0].bv_offset = 0; | |
4f024f37 | 1132 | rbi->bi_iter.bi_size = STRIPE_SIZE; |
37c61ff3 SL |
1133 | /* |
1134 | * If this is discard request, set bi_vcnt 0. We don't | |
1135 | * want to confuse SCSI because SCSI will replace payload | |
1136 | */ | |
796a5cf0 | 1137 | if (op == REQ_OP_DISCARD) |
37c61ff3 | 1138 | rbi->bi_vcnt = 0; |
e3620a3a JB |
1139 | if (conf->mddev->gendisk) |
1140 | trace_block_bio_remap(bdev_get_queue(rbi->bi_bdev), | |
1141 | rbi, disk_devt(conf->mddev->gendisk), | |
1142 | sh->dev[i].sector); | |
765d704d | 1143 | defer_bio_issue(conf, rbi); |
977df362 N |
1144 | } |
1145 | if (!rdev && !rrdev) { | |
796a5cf0 | 1146 | if (op_is_write(op)) |
91c00924 | 1147 | set_bit(STRIPE_DEGRADED, &sh->state); |
6296b960 | 1148 | pr_debug("skip op %d on disc %d for sector %llu\n", |
1eff9d32 | 1149 | bi->bi_opf, i, (unsigned long long)sh->sector); |
91c00924 DW |
1150 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
1151 | set_bit(STRIPE_HANDLE, &sh->state); | |
1152 | } | |
59fc630b | 1153 | |
1154 | if (!head_sh->batch_head) | |
1155 | continue; | |
1156 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1157 | batch_list); | |
1158 | if (sh != head_sh) | |
1159 | goto again; | |
91c00924 DW |
1160 | } |
1161 | } | |
1162 | ||
1163 | static struct dma_async_tx_descriptor * | |
d592a996 SL |
1164 | async_copy_data(int frombio, struct bio *bio, struct page **page, |
1165 | sector_t sector, struct dma_async_tx_descriptor *tx, | |
1e6d690b | 1166 | struct stripe_head *sh, int no_skipcopy) |
91c00924 | 1167 | { |
7988613b KO |
1168 | struct bio_vec bvl; |
1169 | struct bvec_iter iter; | |
91c00924 | 1170 | struct page *bio_page; |
91c00924 | 1171 | int page_offset; |
a08abd8c | 1172 | struct async_submit_ctl submit; |
0403e382 | 1173 | enum async_tx_flags flags = 0; |
91c00924 | 1174 | |
4f024f37 KO |
1175 | if (bio->bi_iter.bi_sector >= sector) |
1176 | page_offset = (signed)(bio->bi_iter.bi_sector - sector) * 512; | |
91c00924 | 1177 | else |
4f024f37 | 1178 | page_offset = (signed)(sector - bio->bi_iter.bi_sector) * -512; |
a08abd8c | 1179 | |
0403e382 DW |
1180 | if (frombio) |
1181 | flags |= ASYNC_TX_FENCE; | |
1182 | init_async_submit(&submit, flags, tx, NULL, NULL, NULL); | |
1183 | ||
7988613b KO |
1184 | bio_for_each_segment(bvl, bio, iter) { |
1185 | int len = bvl.bv_len; | |
91c00924 DW |
1186 | int clen; |
1187 | int b_offset = 0; | |
1188 | ||
1189 | if (page_offset < 0) { | |
1190 | b_offset = -page_offset; | |
1191 | page_offset += b_offset; | |
1192 | len -= b_offset; | |
1193 | } | |
1194 | ||
1195 | if (len > 0 && page_offset + len > STRIPE_SIZE) | |
1196 | clen = STRIPE_SIZE - page_offset; | |
1197 | else | |
1198 | clen = len; | |
1199 | ||
1200 | if (clen > 0) { | |
7988613b KO |
1201 | b_offset += bvl.bv_offset; |
1202 | bio_page = bvl.bv_page; | |
d592a996 SL |
1203 | if (frombio) { |
1204 | if (sh->raid_conf->skip_copy && | |
1205 | b_offset == 0 && page_offset == 0 && | |
1e6d690b SL |
1206 | clen == STRIPE_SIZE && |
1207 | !no_skipcopy) | |
d592a996 SL |
1208 | *page = bio_page; |
1209 | else | |
1210 | tx = async_memcpy(*page, bio_page, page_offset, | |
a08abd8c | 1211 | b_offset, clen, &submit); |
d592a996 SL |
1212 | } else |
1213 | tx = async_memcpy(bio_page, *page, b_offset, | |
a08abd8c | 1214 | page_offset, clen, &submit); |
91c00924 | 1215 | } |
a08abd8c DW |
1216 | /* chain the operations */ |
1217 | submit.depend_tx = tx; | |
1218 | ||
91c00924 DW |
1219 | if (clen < len) /* hit end of page */ |
1220 | break; | |
1221 | page_offset += len; | |
1222 | } | |
1223 | ||
1224 | return tx; | |
1225 | } | |
1226 | ||
1227 | static void ops_complete_biofill(void *stripe_head_ref) | |
1228 | { | |
1229 | struct stripe_head *sh = stripe_head_ref; | |
34a6f80e | 1230 | struct bio_list return_bi = BIO_EMPTY_LIST; |
e4d84909 | 1231 | int i; |
91c00924 | 1232 | |
e46b272b | 1233 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1234 | (unsigned long long)sh->sector); |
1235 | ||
1236 | /* clear completed biofills */ | |
1237 | for (i = sh->disks; i--; ) { | |
1238 | struct r5dev *dev = &sh->dev[i]; | |
91c00924 DW |
1239 | |
1240 | /* acknowledge completion of a biofill operation */ | |
e4d84909 DW |
1241 | /* and check if we need to reply to a read request, |
1242 | * new R5_Wantfill requests are held off until | |
83de75cc | 1243 | * !STRIPE_BIOFILL_RUN |
e4d84909 DW |
1244 | */ |
1245 | if (test_and_clear_bit(R5_Wantfill, &dev->flags)) { | |
91c00924 | 1246 | struct bio *rbi, *rbi2; |
91c00924 | 1247 | |
91c00924 DW |
1248 | BUG_ON(!dev->read); |
1249 | rbi = dev->read; | |
1250 | dev->read = NULL; | |
4f024f37 | 1251 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 DW |
1252 | dev->sector + STRIPE_SECTORS) { |
1253 | rbi2 = r5_next_bio(rbi, dev->sector); | |
34a6f80e N |
1254 | if (!raid5_dec_bi_active_stripes(rbi)) |
1255 | bio_list_add(&return_bi, rbi); | |
91c00924 DW |
1256 | rbi = rbi2; |
1257 | } | |
1258 | } | |
1259 | } | |
83de75cc | 1260 | clear_bit(STRIPE_BIOFILL_RUN, &sh->state); |
91c00924 | 1261 | |
34a6f80e | 1262 | return_io(&return_bi); |
91c00924 | 1263 | |
e4d84909 | 1264 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1265 | raid5_release_stripe(sh); |
91c00924 DW |
1266 | } |
1267 | ||
1268 | static void ops_run_biofill(struct stripe_head *sh) | |
1269 | { | |
1270 | struct dma_async_tx_descriptor *tx = NULL; | |
a08abd8c | 1271 | struct async_submit_ctl submit; |
91c00924 DW |
1272 | int i; |
1273 | ||
59fc630b | 1274 | BUG_ON(sh->batch_head); |
e46b272b | 1275 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1276 | (unsigned long long)sh->sector); |
1277 | ||
1278 | for (i = sh->disks; i--; ) { | |
1279 | struct r5dev *dev = &sh->dev[i]; | |
1280 | if (test_bit(R5_Wantfill, &dev->flags)) { | |
1281 | struct bio *rbi; | |
b17459c0 | 1282 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1283 | dev->read = rbi = dev->toread; |
1284 | dev->toread = NULL; | |
b17459c0 | 1285 | spin_unlock_irq(&sh->stripe_lock); |
4f024f37 | 1286 | while (rbi && rbi->bi_iter.bi_sector < |
91c00924 | 1287 | dev->sector + STRIPE_SECTORS) { |
d592a996 | 1288 | tx = async_copy_data(0, rbi, &dev->page, |
1e6d690b | 1289 | dev->sector, tx, sh, 0); |
91c00924 DW |
1290 | rbi = r5_next_bio(rbi, dev->sector); |
1291 | } | |
1292 | } | |
1293 | } | |
1294 | ||
1295 | atomic_inc(&sh->count); | |
a08abd8c DW |
1296 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_biofill, sh, NULL); |
1297 | async_trigger_callback(&submit); | |
91c00924 DW |
1298 | } |
1299 | ||
4e7d2c0a | 1300 | static void mark_target_uptodate(struct stripe_head *sh, int target) |
91c00924 | 1301 | { |
4e7d2c0a | 1302 | struct r5dev *tgt; |
91c00924 | 1303 | |
4e7d2c0a DW |
1304 | if (target < 0) |
1305 | return; | |
91c00924 | 1306 | |
4e7d2c0a | 1307 | tgt = &sh->dev[target]; |
91c00924 DW |
1308 | set_bit(R5_UPTODATE, &tgt->flags); |
1309 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1310 | clear_bit(R5_Wantcompute, &tgt->flags); | |
4e7d2c0a DW |
1311 | } |
1312 | ||
ac6b53b6 | 1313 | static void ops_complete_compute(void *stripe_head_ref) |
91c00924 DW |
1314 | { |
1315 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1316 | |
e46b272b | 1317 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1318 | (unsigned long long)sh->sector); |
1319 | ||
ac6b53b6 | 1320 | /* mark the computed target(s) as uptodate */ |
4e7d2c0a | 1321 | mark_target_uptodate(sh, sh->ops.target); |
ac6b53b6 | 1322 | mark_target_uptodate(sh, sh->ops.target2); |
4e7d2c0a | 1323 | |
ecc65c9b DW |
1324 | clear_bit(STRIPE_COMPUTE_RUN, &sh->state); |
1325 | if (sh->check_state == check_state_compute_run) | |
1326 | sh->check_state = check_state_compute_result; | |
91c00924 | 1327 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1328 | raid5_release_stripe(sh); |
91c00924 DW |
1329 | } |
1330 | ||
d6f38f31 DW |
1331 | /* return a pointer to the address conversion region of the scribble buffer */ |
1332 | static addr_conv_t *to_addr_conv(struct stripe_head *sh, | |
46d5b785 | 1333 | struct raid5_percpu *percpu, int i) |
d6f38f31 | 1334 | { |
46d5b785 | 1335 | void *addr; |
1336 | ||
1337 | addr = flex_array_get(percpu->scribble, i); | |
1338 | return addr + sizeof(struct page *) * (sh->disks + 2); | |
1339 | } | |
1340 | ||
1341 | /* return a pointer to the address conversion region of the scribble buffer */ | |
1342 | static struct page **to_addr_page(struct raid5_percpu *percpu, int i) | |
1343 | { | |
1344 | void *addr; | |
1345 | ||
1346 | addr = flex_array_get(percpu->scribble, i); | |
1347 | return addr; | |
d6f38f31 DW |
1348 | } |
1349 | ||
1350 | static struct dma_async_tx_descriptor * | |
1351 | ops_run_compute5(struct stripe_head *sh, struct raid5_percpu *percpu) | |
91c00924 | 1352 | { |
91c00924 | 1353 | int disks = sh->disks; |
46d5b785 | 1354 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 DW |
1355 | int target = sh->ops.target; |
1356 | struct r5dev *tgt = &sh->dev[target]; | |
1357 | struct page *xor_dest = tgt->page; | |
1358 | int count = 0; | |
1359 | struct dma_async_tx_descriptor *tx; | |
a08abd8c | 1360 | struct async_submit_ctl submit; |
91c00924 DW |
1361 | int i; |
1362 | ||
59fc630b | 1363 | BUG_ON(sh->batch_head); |
1364 | ||
91c00924 | 1365 | pr_debug("%s: stripe %llu block: %d\n", |
e46b272b | 1366 | __func__, (unsigned long long)sh->sector, target); |
91c00924 DW |
1367 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); |
1368 | ||
1369 | for (i = disks; i--; ) | |
1370 | if (i != target) | |
1371 | xor_srcs[count++] = sh->dev[i].page; | |
1372 | ||
1373 | atomic_inc(&sh->count); | |
1374 | ||
0403e382 | 1375 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, NULL, |
46d5b785 | 1376 | ops_complete_compute, sh, to_addr_conv(sh, percpu, 0)); |
91c00924 | 1377 | if (unlikely(count == 1)) |
a08abd8c | 1378 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); |
91c00924 | 1379 | else |
a08abd8c | 1380 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 | 1381 | |
91c00924 DW |
1382 | return tx; |
1383 | } | |
1384 | ||
ac6b53b6 DW |
1385 | /* set_syndrome_sources - populate source buffers for gen_syndrome |
1386 | * @srcs - (struct page *) array of size sh->disks | |
1387 | * @sh - stripe_head to parse | |
1388 | * | |
1389 | * Populates srcs in proper layout order for the stripe and returns the | |
1390 | * 'count' of sources to be used in a call to async_gen_syndrome. The P | |
1391 | * destination buffer is recorded in srcs[count] and the Q destination | |
1392 | * is recorded in srcs[count+1]]. | |
1393 | */ | |
584acdd4 MS |
1394 | static int set_syndrome_sources(struct page **srcs, |
1395 | struct stripe_head *sh, | |
1396 | int srctype) | |
ac6b53b6 DW |
1397 | { |
1398 | int disks = sh->disks; | |
1399 | int syndrome_disks = sh->ddf_layout ? disks : (disks - 2); | |
1400 | int d0_idx = raid6_d0(sh); | |
1401 | int count; | |
1402 | int i; | |
1403 | ||
1404 | for (i = 0; i < disks; i++) | |
5dd33c9a | 1405 | srcs[i] = NULL; |
ac6b53b6 DW |
1406 | |
1407 | count = 0; | |
1408 | i = d0_idx; | |
1409 | do { | |
1410 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
584acdd4 | 1411 | struct r5dev *dev = &sh->dev[i]; |
ac6b53b6 | 1412 | |
584acdd4 MS |
1413 | if (i == sh->qd_idx || i == sh->pd_idx || |
1414 | (srctype == SYNDROME_SRC_ALL) || | |
1415 | (srctype == SYNDROME_SRC_WANT_DRAIN && | |
1e6d690b SL |
1416 | (test_bit(R5_Wantdrain, &dev->flags) || |
1417 | test_bit(R5_InJournal, &dev->flags))) || | |
584acdd4 | 1418 | (srctype == SYNDROME_SRC_WRITTEN && |
0977762f SL |
1419 | (dev->written || |
1420 | test_bit(R5_InJournal, &dev->flags)))) { | |
1e6d690b SL |
1421 | if (test_bit(R5_InJournal, &dev->flags)) |
1422 | srcs[slot] = sh->dev[i].orig_page; | |
1423 | else | |
1424 | srcs[slot] = sh->dev[i].page; | |
1425 | } | |
ac6b53b6 DW |
1426 | i = raid6_next_disk(i, disks); |
1427 | } while (i != d0_idx); | |
ac6b53b6 | 1428 | |
e4424fee | 1429 | return syndrome_disks; |
ac6b53b6 DW |
1430 | } |
1431 | ||
1432 | static struct dma_async_tx_descriptor * | |
1433 | ops_run_compute6_1(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1434 | { | |
1435 | int disks = sh->disks; | |
46d5b785 | 1436 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1437 | int target; |
1438 | int qd_idx = sh->qd_idx; | |
1439 | struct dma_async_tx_descriptor *tx; | |
1440 | struct async_submit_ctl submit; | |
1441 | struct r5dev *tgt; | |
1442 | struct page *dest; | |
1443 | int i; | |
1444 | int count; | |
1445 | ||
59fc630b | 1446 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1447 | if (sh->ops.target < 0) |
1448 | target = sh->ops.target2; | |
1449 | else if (sh->ops.target2 < 0) | |
1450 | target = sh->ops.target; | |
91c00924 | 1451 | else |
ac6b53b6 DW |
1452 | /* we should only have one valid target */ |
1453 | BUG(); | |
1454 | BUG_ON(target < 0); | |
1455 | pr_debug("%s: stripe %llu block: %d\n", | |
1456 | __func__, (unsigned long long)sh->sector, target); | |
1457 | ||
1458 | tgt = &sh->dev[target]; | |
1459 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1460 | dest = tgt->page; | |
1461 | ||
1462 | atomic_inc(&sh->count); | |
1463 | ||
1464 | if (target == qd_idx) { | |
584acdd4 | 1465 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1466 | blocks[count] = NULL; /* regenerating p is not necessary */ |
1467 | BUG_ON(blocks[count+1] != dest); /* q should already be set */ | |
0403e382 DW |
1468 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1469 | ops_complete_compute, sh, | |
46d5b785 | 1470 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1471 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
1472 | } else { | |
1473 | /* Compute any data- or p-drive using XOR */ | |
1474 | count = 0; | |
1475 | for (i = disks; i-- ; ) { | |
1476 | if (i == target || i == qd_idx) | |
1477 | continue; | |
1478 | blocks[count++] = sh->dev[i].page; | |
1479 | } | |
1480 | ||
0403e382 DW |
1481 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, |
1482 | NULL, ops_complete_compute, sh, | |
46d5b785 | 1483 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1484 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, &submit); |
1485 | } | |
91c00924 | 1486 | |
91c00924 DW |
1487 | return tx; |
1488 | } | |
1489 | ||
ac6b53b6 DW |
1490 | static struct dma_async_tx_descriptor * |
1491 | ops_run_compute6_2(struct stripe_head *sh, struct raid5_percpu *percpu) | |
1492 | { | |
1493 | int i, count, disks = sh->disks; | |
1494 | int syndrome_disks = sh->ddf_layout ? disks : disks-2; | |
1495 | int d0_idx = raid6_d0(sh); | |
1496 | int faila = -1, failb = -1; | |
1497 | int target = sh->ops.target; | |
1498 | int target2 = sh->ops.target2; | |
1499 | struct r5dev *tgt = &sh->dev[target]; | |
1500 | struct r5dev *tgt2 = &sh->dev[target2]; | |
1501 | struct dma_async_tx_descriptor *tx; | |
46d5b785 | 1502 | struct page **blocks = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1503 | struct async_submit_ctl submit; |
1504 | ||
59fc630b | 1505 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1506 | pr_debug("%s: stripe %llu block1: %d block2: %d\n", |
1507 | __func__, (unsigned long long)sh->sector, target, target2); | |
1508 | BUG_ON(target < 0 || target2 < 0); | |
1509 | BUG_ON(!test_bit(R5_Wantcompute, &tgt->flags)); | |
1510 | BUG_ON(!test_bit(R5_Wantcompute, &tgt2->flags)); | |
1511 | ||
6c910a78 | 1512 | /* we need to open-code set_syndrome_sources to handle the |
ac6b53b6 DW |
1513 | * slot number conversion for 'faila' and 'failb' |
1514 | */ | |
1515 | for (i = 0; i < disks ; i++) | |
5dd33c9a | 1516 | blocks[i] = NULL; |
ac6b53b6 DW |
1517 | count = 0; |
1518 | i = d0_idx; | |
1519 | do { | |
1520 | int slot = raid6_idx_to_slot(i, sh, &count, syndrome_disks); | |
1521 | ||
1522 | blocks[slot] = sh->dev[i].page; | |
1523 | ||
1524 | if (i == target) | |
1525 | faila = slot; | |
1526 | if (i == target2) | |
1527 | failb = slot; | |
1528 | i = raid6_next_disk(i, disks); | |
1529 | } while (i != d0_idx); | |
ac6b53b6 DW |
1530 | |
1531 | BUG_ON(faila == failb); | |
1532 | if (failb < faila) | |
1533 | swap(faila, failb); | |
1534 | pr_debug("%s: stripe: %llu faila: %d failb: %d\n", | |
1535 | __func__, (unsigned long long)sh->sector, faila, failb); | |
1536 | ||
1537 | atomic_inc(&sh->count); | |
1538 | ||
1539 | if (failb == syndrome_disks+1) { | |
1540 | /* Q disk is one of the missing disks */ | |
1541 | if (faila == syndrome_disks) { | |
1542 | /* Missing P+Q, just recompute */ | |
0403e382 DW |
1543 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1544 | ops_complete_compute, sh, | |
46d5b785 | 1545 | to_addr_conv(sh, percpu, 0)); |
e4424fee | 1546 | return async_gen_syndrome(blocks, 0, syndrome_disks+2, |
ac6b53b6 DW |
1547 | STRIPE_SIZE, &submit); |
1548 | } else { | |
1549 | struct page *dest; | |
1550 | int data_target; | |
1551 | int qd_idx = sh->qd_idx; | |
1552 | ||
1553 | /* Missing D+Q: recompute D from P, then recompute Q */ | |
1554 | if (target == qd_idx) | |
1555 | data_target = target2; | |
1556 | else | |
1557 | data_target = target; | |
1558 | ||
1559 | count = 0; | |
1560 | for (i = disks; i-- ; ) { | |
1561 | if (i == data_target || i == qd_idx) | |
1562 | continue; | |
1563 | blocks[count++] = sh->dev[i].page; | |
1564 | } | |
1565 | dest = sh->dev[data_target].page; | |
0403e382 DW |
1566 | init_async_submit(&submit, |
1567 | ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, | |
1568 | NULL, NULL, NULL, | |
46d5b785 | 1569 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1570 | tx = async_xor(dest, blocks, 0, count, STRIPE_SIZE, |
1571 | &submit); | |
1572 | ||
584acdd4 | 1573 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_ALL); |
0403e382 DW |
1574 | init_async_submit(&submit, ASYNC_TX_FENCE, tx, |
1575 | ops_complete_compute, sh, | |
46d5b785 | 1576 | to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1577 | return async_gen_syndrome(blocks, 0, count+2, |
1578 | STRIPE_SIZE, &submit); | |
1579 | } | |
ac6b53b6 | 1580 | } else { |
6c910a78 DW |
1581 | init_async_submit(&submit, ASYNC_TX_FENCE, NULL, |
1582 | ops_complete_compute, sh, | |
46d5b785 | 1583 | to_addr_conv(sh, percpu, 0)); |
6c910a78 DW |
1584 | if (failb == syndrome_disks) { |
1585 | /* We're missing D+P. */ | |
1586 | return async_raid6_datap_recov(syndrome_disks+2, | |
1587 | STRIPE_SIZE, faila, | |
1588 | blocks, &submit); | |
1589 | } else { | |
1590 | /* We're missing D+D. */ | |
1591 | return async_raid6_2data_recov(syndrome_disks+2, | |
1592 | STRIPE_SIZE, faila, failb, | |
1593 | blocks, &submit); | |
1594 | } | |
ac6b53b6 DW |
1595 | } |
1596 | } | |
1597 | ||
91c00924 DW |
1598 | static void ops_complete_prexor(void *stripe_head_ref) |
1599 | { | |
1600 | struct stripe_head *sh = stripe_head_ref; | |
1601 | ||
e46b272b | 1602 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 | 1603 | (unsigned long long)sh->sector); |
1e6d690b SL |
1604 | |
1605 | if (r5c_is_writeback(sh->raid_conf->log)) | |
1606 | /* | |
1607 | * raid5-cache write back uses orig_page during prexor. | |
1608 | * After prexor, it is time to free orig_page | |
1609 | */ | |
1610 | r5c_release_extra_page(sh); | |
91c00924 DW |
1611 | } |
1612 | ||
1613 | static struct dma_async_tx_descriptor * | |
584acdd4 MS |
1614 | ops_run_prexor5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1615 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1616 | { |
91c00924 | 1617 | int disks = sh->disks; |
46d5b785 | 1618 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1619 | int count = 0, pd_idx = sh->pd_idx, i; |
a08abd8c | 1620 | struct async_submit_ctl submit; |
91c00924 DW |
1621 | |
1622 | /* existing parity data subtracted */ | |
1623 | struct page *xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; | |
1624 | ||
59fc630b | 1625 | BUG_ON(sh->batch_head); |
e46b272b | 1626 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1627 | (unsigned long long)sh->sector); |
1628 | ||
1629 | for (i = disks; i--; ) { | |
1630 | struct r5dev *dev = &sh->dev[i]; | |
1631 | /* Only process blocks that are known to be uptodate */ | |
1e6d690b SL |
1632 | if (test_bit(R5_InJournal, &dev->flags)) |
1633 | xor_srcs[count++] = dev->orig_page; | |
1634 | else if (test_bit(R5_Wantdrain, &dev->flags)) | |
91c00924 DW |
1635 | xor_srcs[count++] = dev->page; |
1636 | } | |
1637 | ||
0403e382 | 1638 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_DROP_DST, tx, |
46d5b785 | 1639 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); |
a08abd8c | 1640 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); |
91c00924 DW |
1641 | |
1642 | return tx; | |
1643 | } | |
1644 | ||
584acdd4 MS |
1645 | static struct dma_async_tx_descriptor * |
1646 | ops_run_prexor6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1647 | struct dma_async_tx_descriptor *tx) | |
1648 | { | |
1649 | struct page **blocks = to_addr_page(percpu, 0); | |
1650 | int count; | |
1651 | struct async_submit_ctl submit; | |
1652 | ||
1653 | pr_debug("%s: stripe %llu\n", __func__, | |
1654 | (unsigned long long)sh->sector); | |
1655 | ||
1656 | count = set_syndrome_sources(blocks, sh, SYNDROME_SRC_WANT_DRAIN); | |
1657 | ||
1658 | init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_PQ_XOR_DST, tx, | |
1659 | ops_complete_prexor, sh, to_addr_conv(sh, percpu, 0)); | |
1660 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); | |
1661 | ||
1662 | return tx; | |
1663 | } | |
1664 | ||
91c00924 | 1665 | static struct dma_async_tx_descriptor * |
d8ee0728 | 1666 | ops_run_biodrain(struct stripe_head *sh, struct dma_async_tx_descriptor *tx) |
91c00924 | 1667 | { |
1e6d690b | 1668 | struct r5conf *conf = sh->raid_conf; |
91c00924 | 1669 | int disks = sh->disks; |
d8ee0728 | 1670 | int i; |
59fc630b | 1671 | struct stripe_head *head_sh = sh; |
91c00924 | 1672 | |
e46b272b | 1673 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1674 | (unsigned long long)sh->sector); |
1675 | ||
1676 | for (i = disks; i--; ) { | |
59fc630b | 1677 | struct r5dev *dev; |
91c00924 | 1678 | struct bio *chosen; |
91c00924 | 1679 | |
59fc630b | 1680 | sh = head_sh; |
1681 | if (test_and_clear_bit(R5_Wantdrain, &head_sh->dev[i].flags)) { | |
91c00924 DW |
1682 | struct bio *wbi; |
1683 | ||
59fc630b | 1684 | again: |
1685 | dev = &sh->dev[i]; | |
1e6d690b SL |
1686 | /* |
1687 | * clear R5_InJournal, so when rewriting a page in | |
1688 | * journal, it is not skipped by r5l_log_stripe() | |
1689 | */ | |
1690 | clear_bit(R5_InJournal, &dev->flags); | |
b17459c0 | 1691 | spin_lock_irq(&sh->stripe_lock); |
91c00924 DW |
1692 | chosen = dev->towrite; |
1693 | dev->towrite = NULL; | |
7a87f434 | 1694 | sh->overwrite_disks = 0; |
91c00924 DW |
1695 | BUG_ON(dev->written); |
1696 | wbi = dev->written = chosen; | |
b17459c0 | 1697 | spin_unlock_irq(&sh->stripe_lock); |
d592a996 | 1698 | WARN_ON(dev->page != dev->orig_page); |
91c00924 | 1699 | |
4f024f37 | 1700 | while (wbi && wbi->bi_iter.bi_sector < |
91c00924 | 1701 | dev->sector + STRIPE_SECTORS) { |
1eff9d32 | 1702 | if (wbi->bi_opf & REQ_FUA) |
e9c7469b | 1703 | set_bit(R5_WantFUA, &dev->flags); |
1eff9d32 | 1704 | if (wbi->bi_opf & REQ_SYNC) |
bc0934f0 | 1705 | set_bit(R5_SyncIO, &dev->flags); |
796a5cf0 | 1706 | if (bio_op(wbi) == REQ_OP_DISCARD) |
620125f2 | 1707 | set_bit(R5_Discard, &dev->flags); |
d592a996 SL |
1708 | else { |
1709 | tx = async_copy_data(1, wbi, &dev->page, | |
1e6d690b SL |
1710 | dev->sector, tx, sh, |
1711 | r5c_is_writeback(conf->log)); | |
1712 | if (dev->page != dev->orig_page && | |
1713 | !r5c_is_writeback(conf->log)) { | |
d592a996 SL |
1714 | set_bit(R5_SkipCopy, &dev->flags); |
1715 | clear_bit(R5_UPTODATE, &dev->flags); | |
1716 | clear_bit(R5_OVERWRITE, &dev->flags); | |
1717 | } | |
1718 | } | |
91c00924 DW |
1719 | wbi = r5_next_bio(wbi, dev->sector); |
1720 | } | |
59fc630b | 1721 | |
1722 | if (head_sh->batch_head) { | |
1723 | sh = list_first_entry(&sh->batch_list, | |
1724 | struct stripe_head, | |
1725 | batch_list); | |
1726 | if (sh == head_sh) | |
1727 | continue; | |
1728 | goto again; | |
1729 | } | |
91c00924 DW |
1730 | } |
1731 | } | |
1732 | ||
1733 | return tx; | |
1734 | } | |
1735 | ||
ac6b53b6 | 1736 | static void ops_complete_reconstruct(void *stripe_head_ref) |
91c00924 DW |
1737 | { |
1738 | struct stripe_head *sh = stripe_head_ref; | |
ac6b53b6 DW |
1739 | int disks = sh->disks; |
1740 | int pd_idx = sh->pd_idx; | |
1741 | int qd_idx = sh->qd_idx; | |
1742 | int i; | |
9e444768 | 1743 | bool fua = false, sync = false, discard = false; |
91c00924 | 1744 | |
e46b272b | 1745 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1746 | (unsigned long long)sh->sector); |
1747 | ||
bc0934f0 | 1748 | for (i = disks; i--; ) { |
e9c7469b | 1749 | fua |= test_bit(R5_WantFUA, &sh->dev[i].flags); |
bc0934f0 | 1750 | sync |= test_bit(R5_SyncIO, &sh->dev[i].flags); |
9e444768 | 1751 | discard |= test_bit(R5_Discard, &sh->dev[i].flags); |
bc0934f0 | 1752 | } |
e9c7469b | 1753 | |
91c00924 DW |
1754 | for (i = disks; i--; ) { |
1755 | struct r5dev *dev = &sh->dev[i]; | |
ac6b53b6 | 1756 | |
e9c7469b | 1757 | if (dev->written || i == pd_idx || i == qd_idx) { |
d592a996 | 1758 | if (!discard && !test_bit(R5_SkipCopy, &dev->flags)) |
9e444768 | 1759 | set_bit(R5_UPTODATE, &dev->flags); |
e9c7469b TH |
1760 | if (fua) |
1761 | set_bit(R5_WantFUA, &dev->flags); | |
bc0934f0 SL |
1762 | if (sync) |
1763 | set_bit(R5_SyncIO, &dev->flags); | |
e9c7469b | 1764 | } |
91c00924 DW |
1765 | } |
1766 | ||
d8ee0728 DW |
1767 | if (sh->reconstruct_state == reconstruct_state_drain_run) |
1768 | sh->reconstruct_state = reconstruct_state_drain_result; | |
1769 | else if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) | |
1770 | sh->reconstruct_state = reconstruct_state_prexor_drain_result; | |
1771 | else { | |
1772 | BUG_ON(sh->reconstruct_state != reconstruct_state_run); | |
1773 | sh->reconstruct_state = reconstruct_state_result; | |
1774 | } | |
91c00924 DW |
1775 | |
1776 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 1777 | raid5_release_stripe(sh); |
91c00924 DW |
1778 | } |
1779 | ||
1780 | static void | |
ac6b53b6 DW |
1781 | ops_run_reconstruct5(struct stripe_head *sh, struct raid5_percpu *percpu, |
1782 | struct dma_async_tx_descriptor *tx) | |
91c00924 | 1783 | { |
91c00924 | 1784 | int disks = sh->disks; |
59fc630b | 1785 | struct page **xor_srcs; |
a08abd8c | 1786 | struct async_submit_ctl submit; |
59fc630b | 1787 | int count, pd_idx = sh->pd_idx, i; |
91c00924 | 1788 | struct page *xor_dest; |
d8ee0728 | 1789 | int prexor = 0; |
91c00924 | 1790 | unsigned long flags; |
59fc630b | 1791 | int j = 0; |
1792 | struct stripe_head *head_sh = sh; | |
1793 | int last_stripe; | |
91c00924 | 1794 | |
e46b272b | 1795 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1796 | (unsigned long long)sh->sector); |
1797 | ||
620125f2 SL |
1798 | for (i = 0; i < sh->disks; i++) { |
1799 | if (pd_idx == i) | |
1800 | continue; | |
1801 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1802 | break; | |
1803 | } | |
1804 | if (i >= sh->disks) { | |
1805 | atomic_inc(&sh->count); | |
620125f2 SL |
1806 | set_bit(R5_Discard, &sh->dev[pd_idx].flags); |
1807 | ops_complete_reconstruct(sh); | |
1808 | return; | |
1809 | } | |
59fc630b | 1810 | again: |
1811 | count = 0; | |
1812 | xor_srcs = to_addr_page(percpu, j); | |
91c00924 DW |
1813 | /* check if prexor is active which means only process blocks |
1814 | * that are part of a read-modify-write (written) | |
1815 | */ | |
59fc630b | 1816 | if (head_sh->reconstruct_state == reconstruct_state_prexor_drain_run) { |
d8ee0728 | 1817 | prexor = 1; |
91c00924 DW |
1818 | xor_dest = xor_srcs[count++] = sh->dev[pd_idx].page; |
1819 | for (i = disks; i--; ) { | |
1820 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
1821 | if (head_sh->dev[i].written || |
1822 | test_bit(R5_InJournal, &head_sh->dev[i].flags)) | |
91c00924 DW |
1823 | xor_srcs[count++] = dev->page; |
1824 | } | |
1825 | } else { | |
1826 | xor_dest = sh->dev[pd_idx].page; | |
1827 | for (i = disks; i--; ) { | |
1828 | struct r5dev *dev = &sh->dev[i]; | |
1829 | if (i != pd_idx) | |
1830 | xor_srcs[count++] = dev->page; | |
1831 | } | |
1832 | } | |
1833 | ||
91c00924 DW |
1834 | /* 1/ if we prexor'd then the dest is reused as a source |
1835 | * 2/ if we did not prexor then we are redoing the parity | |
1836 | * set ASYNC_TX_XOR_DROP_DST and ASYNC_TX_XOR_ZERO_DST | |
1837 | * for the synchronous xor case | |
1838 | */ | |
59fc630b | 1839 | last_stripe = !head_sh->batch_head || |
1840 | list_first_entry(&sh->batch_list, | |
1841 | struct stripe_head, batch_list) == head_sh; | |
1842 | if (last_stripe) { | |
1843 | flags = ASYNC_TX_ACK | | |
1844 | (prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST); | |
1845 | ||
1846 | atomic_inc(&head_sh->count); | |
1847 | init_async_submit(&submit, flags, tx, ops_complete_reconstruct, head_sh, | |
1848 | to_addr_conv(sh, percpu, j)); | |
1849 | } else { | |
1850 | flags = prexor ? ASYNC_TX_XOR_DROP_DST : ASYNC_TX_XOR_ZERO_DST; | |
1851 | init_async_submit(&submit, flags, tx, NULL, NULL, | |
1852 | to_addr_conv(sh, percpu, j)); | |
1853 | } | |
91c00924 | 1854 | |
a08abd8c DW |
1855 | if (unlikely(count == 1)) |
1856 | tx = async_memcpy(xor_dest, xor_srcs[0], 0, 0, STRIPE_SIZE, &submit); | |
1857 | else | |
1858 | tx = async_xor(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, &submit); | |
59fc630b | 1859 | if (!last_stripe) { |
1860 | j++; | |
1861 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1862 | batch_list); | |
1863 | goto again; | |
1864 | } | |
91c00924 DW |
1865 | } |
1866 | ||
ac6b53b6 DW |
1867 | static void |
1868 | ops_run_reconstruct6(struct stripe_head *sh, struct raid5_percpu *percpu, | |
1869 | struct dma_async_tx_descriptor *tx) | |
1870 | { | |
1871 | struct async_submit_ctl submit; | |
59fc630b | 1872 | struct page **blocks; |
1873 | int count, i, j = 0; | |
1874 | struct stripe_head *head_sh = sh; | |
1875 | int last_stripe; | |
584acdd4 MS |
1876 | int synflags; |
1877 | unsigned long txflags; | |
ac6b53b6 DW |
1878 | |
1879 | pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); | |
1880 | ||
620125f2 SL |
1881 | for (i = 0; i < sh->disks; i++) { |
1882 | if (sh->pd_idx == i || sh->qd_idx == i) | |
1883 | continue; | |
1884 | if (!test_bit(R5_Discard, &sh->dev[i].flags)) | |
1885 | break; | |
1886 | } | |
1887 | if (i >= sh->disks) { | |
1888 | atomic_inc(&sh->count); | |
620125f2 SL |
1889 | set_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
1890 | set_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
1891 | ops_complete_reconstruct(sh); | |
1892 | return; | |
1893 | } | |
1894 | ||
59fc630b | 1895 | again: |
1896 | blocks = to_addr_page(percpu, j); | |
584acdd4 MS |
1897 | |
1898 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { | |
1899 | synflags = SYNDROME_SRC_WRITTEN; | |
1900 | txflags = ASYNC_TX_ACK | ASYNC_TX_PQ_XOR_DST; | |
1901 | } else { | |
1902 | synflags = SYNDROME_SRC_ALL; | |
1903 | txflags = ASYNC_TX_ACK; | |
1904 | } | |
1905 | ||
1906 | count = set_syndrome_sources(blocks, sh, synflags); | |
59fc630b | 1907 | last_stripe = !head_sh->batch_head || |
1908 | list_first_entry(&sh->batch_list, | |
1909 | struct stripe_head, batch_list) == head_sh; | |
1910 | ||
1911 | if (last_stripe) { | |
1912 | atomic_inc(&head_sh->count); | |
584acdd4 | 1913 | init_async_submit(&submit, txflags, tx, ops_complete_reconstruct, |
59fc630b | 1914 | head_sh, to_addr_conv(sh, percpu, j)); |
1915 | } else | |
1916 | init_async_submit(&submit, 0, tx, NULL, NULL, | |
1917 | to_addr_conv(sh, percpu, j)); | |
48769695 | 1918 | tx = async_gen_syndrome(blocks, 0, count+2, STRIPE_SIZE, &submit); |
59fc630b | 1919 | if (!last_stripe) { |
1920 | j++; | |
1921 | sh = list_first_entry(&sh->batch_list, struct stripe_head, | |
1922 | batch_list); | |
1923 | goto again; | |
1924 | } | |
91c00924 DW |
1925 | } |
1926 | ||
1927 | static void ops_complete_check(void *stripe_head_ref) | |
1928 | { | |
1929 | struct stripe_head *sh = stripe_head_ref; | |
91c00924 | 1930 | |
e46b272b | 1931 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1932 | (unsigned long long)sh->sector); |
1933 | ||
ecc65c9b | 1934 | sh->check_state = check_state_check_result; |
91c00924 | 1935 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 1936 | raid5_release_stripe(sh); |
91c00924 DW |
1937 | } |
1938 | ||
ac6b53b6 | 1939 | static void ops_run_check_p(struct stripe_head *sh, struct raid5_percpu *percpu) |
91c00924 | 1940 | { |
91c00924 | 1941 | int disks = sh->disks; |
ac6b53b6 DW |
1942 | int pd_idx = sh->pd_idx; |
1943 | int qd_idx = sh->qd_idx; | |
1944 | struct page *xor_dest; | |
46d5b785 | 1945 | struct page **xor_srcs = to_addr_page(percpu, 0); |
91c00924 | 1946 | struct dma_async_tx_descriptor *tx; |
a08abd8c | 1947 | struct async_submit_ctl submit; |
ac6b53b6 DW |
1948 | int count; |
1949 | int i; | |
91c00924 | 1950 | |
e46b272b | 1951 | pr_debug("%s: stripe %llu\n", __func__, |
91c00924 DW |
1952 | (unsigned long long)sh->sector); |
1953 | ||
59fc630b | 1954 | BUG_ON(sh->batch_head); |
ac6b53b6 DW |
1955 | count = 0; |
1956 | xor_dest = sh->dev[pd_idx].page; | |
1957 | xor_srcs[count++] = xor_dest; | |
91c00924 | 1958 | for (i = disks; i--; ) { |
ac6b53b6 DW |
1959 | if (i == pd_idx || i == qd_idx) |
1960 | continue; | |
1961 | xor_srcs[count++] = sh->dev[i].page; | |
91c00924 DW |
1962 | } |
1963 | ||
d6f38f31 | 1964 | init_async_submit(&submit, 0, NULL, NULL, NULL, |
46d5b785 | 1965 | to_addr_conv(sh, percpu, 0)); |
099f53cb | 1966 | tx = async_xor_val(xor_dest, xor_srcs, 0, count, STRIPE_SIZE, |
a08abd8c | 1967 | &sh->ops.zero_sum_result, &submit); |
91c00924 | 1968 | |
91c00924 | 1969 | atomic_inc(&sh->count); |
a08abd8c DW |
1970 | init_async_submit(&submit, ASYNC_TX_ACK, tx, ops_complete_check, sh, NULL); |
1971 | tx = async_trigger_callback(&submit); | |
91c00924 DW |
1972 | } |
1973 | ||
ac6b53b6 DW |
1974 | static void ops_run_check_pq(struct stripe_head *sh, struct raid5_percpu *percpu, int checkp) |
1975 | { | |
46d5b785 | 1976 | struct page **srcs = to_addr_page(percpu, 0); |
ac6b53b6 DW |
1977 | struct async_submit_ctl submit; |
1978 | int count; | |
1979 | ||
1980 | pr_debug("%s: stripe %llu checkp: %d\n", __func__, | |
1981 | (unsigned long long)sh->sector, checkp); | |
1982 | ||
59fc630b | 1983 | BUG_ON(sh->batch_head); |
584acdd4 | 1984 | count = set_syndrome_sources(srcs, sh, SYNDROME_SRC_ALL); |
ac6b53b6 DW |
1985 | if (!checkp) |
1986 | srcs[count] = NULL; | |
91c00924 | 1987 | |
91c00924 | 1988 | atomic_inc(&sh->count); |
ac6b53b6 | 1989 | init_async_submit(&submit, ASYNC_TX_ACK, NULL, ops_complete_check, |
46d5b785 | 1990 | sh, to_addr_conv(sh, percpu, 0)); |
ac6b53b6 DW |
1991 | async_syndrome_val(srcs, 0, count+2, STRIPE_SIZE, |
1992 | &sh->ops.zero_sum_result, percpu->spare_page, &submit); | |
91c00924 DW |
1993 | } |
1994 | ||
51acbcec | 1995 | static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request) |
91c00924 DW |
1996 | { |
1997 | int overlap_clear = 0, i, disks = sh->disks; | |
1998 | struct dma_async_tx_descriptor *tx = NULL; | |
d1688a6d | 1999 | struct r5conf *conf = sh->raid_conf; |
ac6b53b6 | 2000 | int level = conf->level; |
d6f38f31 DW |
2001 | struct raid5_percpu *percpu; |
2002 | unsigned long cpu; | |
91c00924 | 2003 | |
d6f38f31 DW |
2004 | cpu = get_cpu(); |
2005 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
83de75cc | 2006 | if (test_bit(STRIPE_OP_BIOFILL, &ops_request)) { |
91c00924 DW |
2007 | ops_run_biofill(sh); |
2008 | overlap_clear++; | |
2009 | } | |
2010 | ||
7b3a871e | 2011 | if (test_bit(STRIPE_OP_COMPUTE_BLK, &ops_request)) { |
ac6b53b6 DW |
2012 | if (level < 6) |
2013 | tx = ops_run_compute5(sh, percpu); | |
2014 | else { | |
2015 | if (sh->ops.target2 < 0 || sh->ops.target < 0) | |
2016 | tx = ops_run_compute6_1(sh, percpu); | |
2017 | else | |
2018 | tx = ops_run_compute6_2(sh, percpu); | |
2019 | } | |
2020 | /* terminate the chain if reconstruct is not set to be run */ | |
2021 | if (tx && !test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) | |
7b3a871e DW |
2022 | async_tx_ack(tx); |
2023 | } | |
91c00924 | 2024 | |
584acdd4 MS |
2025 | if (test_bit(STRIPE_OP_PREXOR, &ops_request)) { |
2026 | if (level < 6) | |
2027 | tx = ops_run_prexor5(sh, percpu, tx); | |
2028 | else | |
2029 | tx = ops_run_prexor6(sh, percpu, tx); | |
2030 | } | |
91c00924 | 2031 | |
600aa109 | 2032 | if (test_bit(STRIPE_OP_BIODRAIN, &ops_request)) { |
d8ee0728 | 2033 | tx = ops_run_biodrain(sh, tx); |
91c00924 DW |
2034 | overlap_clear++; |
2035 | } | |
2036 | ||
ac6b53b6 DW |
2037 | if (test_bit(STRIPE_OP_RECONSTRUCT, &ops_request)) { |
2038 | if (level < 6) | |
2039 | ops_run_reconstruct5(sh, percpu, tx); | |
2040 | else | |
2041 | ops_run_reconstruct6(sh, percpu, tx); | |
2042 | } | |
91c00924 | 2043 | |
ac6b53b6 DW |
2044 | if (test_bit(STRIPE_OP_CHECK, &ops_request)) { |
2045 | if (sh->check_state == check_state_run) | |
2046 | ops_run_check_p(sh, percpu); | |
2047 | else if (sh->check_state == check_state_run_q) | |
2048 | ops_run_check_pq(sh, percpu, 0); | |
2049 | else if (sh->check_state == check_state_run_pq) | |
2050 | ops_run_check_pq(sh, percpu, 1); | |
2051 | else | |
2052 | BUG(); | |
2053 | } | |
91c00924 | 2054 | |
59fc630b | 2055 | if (overlap_clear && !sh->batch_head) |
91c00924 DW |
2056 | for (i = disks; i--; ) { |
2057 | struct r5dev *dev = &sh->dev[i]; | |
2058 | if (test_and_clear_bit(R5_Overlap, &dev->flags)) | |
2059 | wake_up(&sh->raid_conf->wait_for_overlap); | |
2060 | } | |
d6f38f31 | 2061 | put_cpu(); |
91c00924 DW |
2062 | } |
2063 | ||
5f9d1fde SL |
2064 | static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp, |
2065 | int disks) | |
f18c1a35 N |
2066 | { |
2067 | struct stripe_head *sh; | |
5f9d1fde | 2068 | int i; |
f18c1a35 N |
2069 | |
2070 | sh = kmem_cache_zalloc(sc, gfp); | |
2071 | if (sh) { | |
2072 | spin_lock_init(&sh->stripe_lock); | |
2073 | spin_lock_init(&sh->batch_lock); | |
2074 | INIT_LIST_HEAD(&sh->batch_list); | |
2075 | INIT_LIST_HEAD(&sh->lru); | |
a39f7afd | 2076 | INIT_LIST_HEAD(&sh->r5c); |
d7bd398e | 2077 | INIT_LIST_HEAD(&sh->log_list); |
f18c1a35 | 2078 | atomic_set(&sh->count, 1); |
a39f7afd | 2079 | sh->log_start = MaxSector; |
5f9d1fde SL |
2080 | for (i = 0; i < disks; i++) { |
2081 | struct r5dev *dev = &sh->dev[i]; | |
2082 | ||
3a83f467 ML |
2083 | bio_init(&dev->req, &dev->vec, 1); |
2084 | bio_init(&dev->rreq, &dev->rvec, 1); | |
5f9d1fde | 2085 | } |
f18c1a35 N |
2086 | } |
2087 | return sh; | |
2088 | } | |
486f0644 | 2089 | static int grow_one_stripe(struct r5conf *conf, gfp_t gfp) |
1da177e4 LT |
2090 | { |
2091 | struct stripe_head *sh; | |
f18c1a35 | 2092 | |
5f9d1fde | 2093 | sh = alloc_stripe(conf->slab_cache, gfp, conf->pool_size); |
3f294f4f N |
2094 | if (!sh) |
2095 | return 0; | |
6ce32846 | 2096 | |
3f294f4f | 2097 | sh->raid_conf = conf; |
3f294f4f | 2098 | |
a9683a79 | 2099 | if (grow_buffers(sh, gfp)) { |
e4e11e38 | 2100 | shrink_buffers(sh); |
3f294f4f N |
2101 | kmem_cache_free(conf->slab_cache, sh); |
2102 | return 0; | |
2103 | } | |
486f0644 N |
2104 | sh->hash_lock_index = |
2105 | conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS; | |
3f294f4f | 2106 | /* we just created an active stripe so... */ |
3f294f4f | 2107 | atomic_inc(&conf->active_stripes); |
59fc630b | 2108 | |
6d036f7d | 2109 | raid5_release_stripe(sh); |
486f0644 | 2110 | conf->max_nr_stripes++; |
3f294f4f N |
2111 | return 1; |
2112 | } | |
2113 | ||
d1688a6d | 2114 | static int grow_stripes(struct r5conf *conf, int num) |
3f294f4f | 2115 | { |
e18b890b | 2116 | struct kmem_cache *sc; |
5e5e3e78 | 2117 | int devs = max(conf->raid_disks, conf->previous_raid_disks); |
1da177e4 | 2118 | |
f4be6b43 N |
2119 | if (conf->mddev->gendisk) |
2120 | sprintf(conf->cache_name[0], | |
2121 | "raid%d-%s", conf->level, mdname(conf->mddev)); | |
2122 | else | |
2123 | sprintf(conf->cache_name[0], | |
2124 | "raid%d-%p", conf->level, conf->mddev); | |
2125 | sprintf(conf->cache_name[1], "%s-alt", conf->cache_name[0]); | |
2126 | ||
ad01c9e3 N |
2127 | conf->active_name = 0; |
2128 | sc = kmem_cache_create(conf->cache_name[conf->active_name], | |
1da177e4 | 2129 | sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), |
20c2df83 | 2130 | 0, 0, NULL); |
1da177e4 LT |
2131 | if (!sc) |
2132 | return 1; | |
2133 | conf->slab_cache = sc; | |
ad01c9e3 | 2134 | conf->pool_size = devs; |
486f0644 N |
2135 | while (num--) |
2136 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
1da177e4 | 2137 | return 1; |
486f0644 | 2138 | |
1da177e4 LT |
2139 | return 0; |
2140 | } | |
29269553 | 2141 | |
d6f38f31 DW |
2142 | /** |
2143 | * scribble_len - return the required size of the scribble region | |
2144 | * @num - total number of disks in the array | |
2145 | * | |
2146 | * The size must be enough to contain: | |
2147 | * 1/ a struct page pointer for each device in the array +2 | |
2148 | * 2/ room to convert each entry in (1) to its corresponding dma | |
2149 | * (dma_map_page()) or page (page_address()) address. | |
2150 | * | |
2151 | * Note: the +2 is for the destination buffers of the ddf/raid6 case where we | |
2152 | * calculate over all devices (not just the data blocks), using zeros in place | |
2153 | * of the P and Q blocks. | |
2154 | */ | |
46d5b785 | 2155 | static struct flex_array *scribble_alloc(int num, int cnt, gfp_t flags) |
d6f38f31 | 2156 | { |
46d5b785 | 2157 | struct flex_array *ret; |
d6f38f31 DW |
2158 | size_t len; |
2159 | ||
2160 | len = sizeof(struct page *) * (num+2) + sizeof(addr_conv_t) * (num+2); | |
46d5b785 | 2161 | ret = flex_array_alloc(len, cnt, flags); |
2162 | if (!ret) | |
2163 | return NULL; | |
2164 | /* always prealloc all elements, so no locking is required */ | |
2165 | if (flex_array_prealloc(ret, 0, cnt, flags)) { | |
2166 | flex_array_free(ret); | |
2167 | return NULL; | |
2168 | } | |
2169 | return ret; | |
d6f38f31 DW |
2170 | } |
2171 | ||
738a2738 N |
2172 | static int resize_chunks(struct r5conf *conf, int new_disks, int new_sectors) |
2173 | { | |
2174 | unsigned long cpu; | |
2175 | int err = 0; | |
2176 | ||
27a353c0 SL |
2177 | /* |
2178 | * Never shrink. And mddev_suspend() could deadlock if this is called | |
2179 | * from raid5d. In that case, scribble_disks and scribble_sectors | |
2180 | * should equal to new_disks and new_sectors | |
2181 | */ | |
2182 | if (conf->scribble_disks >= new_disks && | |
2183 | conf->scribble_sectors >= new_sectors) | |
2184 | return 0; | |
738a2738 N |
2185 | mddev_suspend(conf->mddev); |
2186 | get_online_cpus(); | |
2187 | for_each_present_cpu(cpu) { | |
2188 | struct raid5_percpu *percpu; | |
2189 | struct flex_array *scribble; | |
2190 | ||
2191 | percpu = per_cpu_ptr(conf->percpu, cpu); | |
2192 | scribble = scribble_alloc(new_disks, | |
2193 | new_sectors / STRIPE_SECTORS, | |
2194 | GFP_NOIO); | |
2195 | ||
2196 | if (scribble) { | |
2197 | flex_array_free(percpu->scribble); | |
2198 | percpu->scribble = scribble; | |
2199 | } else { | |
2200 | err = -ENOMEM; | |
2201 | break; | |
2202 | } | |
2203 | } | |
2204 | put_online_cpus(); | |
2205 | mddev_resume(conf->mddev); | |
27a353c0 SL |
2206 | if (!err) { |
2207 | conf->scribble_disks = new_disks; | |
2208 | conf->scribble_sectors = new_sectors; | |
2209 | } | |
738a2738 N |
2210 | return err; |
2211 | } | |
2212 | ||
d1688a6d | 2213 | static int resize_stripes(struct r5conf *conf, int newsize) |
ad01c9e3 N |
2214 | { |
2215 | /* Make all the stripes able to hold 'newsize' devices. | |
2216 | * New slots in each stripe get 'page' set to a new page. | |
2217 | * | |
2218 | * This happens in stages: | |
2219 | * 1/ create a new kmem_cache and allocate the required number of | |
2220 | * stripe_heads. | |
83f0d77a | 2221 | * 2/ gather all the old stripe_heads and transfer the pages across |
ad01c9e3 N |
2222 | * to the new stripe_heads. This will have the side effect of |
2223 | * freezing the array as once all stripe_heads have been collected, | |
2224 | * no IO will be possible. Old stripe heads are freed once their | |
2225 | * pages have been transferred over, and the old kmem_cache is | |
2226 | * freed when all stripes are done. | |
2227 | * 3/ reallocate conf->disks to be suitable bigger. If this fails, | |
2228 | * we simple return a failre status - no need to clean anything up. | |
2229 | * 4/ allocate new pages for the new slots in the new stripe_heads. | |
2230 | * If this fails, we don't bother trying the shrink the | |
2231 | * stripe_heads down again, we just leave them as they are. | |
2232 | * As each stripe_head is processed the new one is released into | |
2233 | * active service. | |
2234 | * | |
2235 | * Once step2 is started, we cannot afford to wait for a write, | |
2236 | * so we use GFP_NOIO allocations. | |
2237 | */ | |
2238 | struct stripe_head *osh, *nsh; | |
2239 | LIST_HEAD(newstripes); | |
2240 | struct disk_info *ndisks; | |
b5470dc5 | 2241 | int err; |
e18b890b | 2242 | struct kmem_cache *sc; |
ad01c9e3 | 2243 | int i; |
566c09c5 | 2244 | int hash, cnt; |
ad01c9e3 N |
2245 | |
2246 | if (newsize <= conf->pool_size) | |
2247 | return 0; /* never bother to shrink */ | |
2248 | ||
b5470dc5 DW |
2249 | err = md_allow_write(conf->mddev); |
2250 | if (err) | |
2251 | return err; | |
2a2275d6 | 2252 | |
ad01c9e3 N |
2253 | /* Step 1 */ |
2254 | sc = kmem_cache_create(conf->cache_name[1-conf->active_name], | |
2255 | sizeof(struct stripe_head)+(newsize-1)*sizeof(struct r5dev), | |
20c2df83 | 2256 | 0, 0, NULL); |
ad01c9e3 N |
2257 | if (!sc) |
2258 | return -ENOMEM; | |
2259 | ||
2d5b569b N |
2260 | /* Need to ensure auto-resizing doesn't interfere */ |
2261 | mutex_lock(&conf->cache_size_mutex); | |
2262 | ||
ad01c9e3 | 2263 | for (i = conf->max_nr_stripes; i; i--) { |
5f9d1fde | 2264 | nsh = alloc_stripe(sc, GFP_KERNEL, newsize); |
ad01c9e3 N |
2265 | if (!nsh) |
2266 | break; | |
2267 | ||
ad01c9e3 | 2268 | nsh->raid_conf = conf; |
ad01c9e3 N |
2269 | list_add(&nsh->lru, &newstripes); |
2270 | } | |
2271 | if (i) { | |
2272 | /* didn't get enough, give up */ | |
2273 | while (!list_empty(&newstripes)) { | |
2274 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2275 | list_del(&nsh->lru); | |
2276 | kmem_cache_free(sc, nsh); | |
2277 | } | |
2278 | kmem_cache_destroy(sc); | |
2d5b569b | 2279 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2280 | return -ENOMEM; |
2281 | } | |
2282 | /* Step 2 - Must use GFP_NOIO now. | |
2283 | * OK, we have enough stripes, start collecting inactive | |
2284 | * stripes and copying them over | |
2285 | */ | |
566c09c5 SL |
2286 | hash = 0; |
2287 | cnt = 0; | |
ad01c9e3 | 2288 | list_for_each_entry(nsh, &newstripes, lru) { |
566c09c5 | 2289 | lock_device_hash_lock(conf, hash); |
6ab2a4b8 | 2290 | wait_event_cmd(conf->wait_for_stripe, |
566c09c5 SL |
2291 | !list_empty(conf->inactive_list + hash), |
2292 | unlock_device_hash_lock(conf, hash), | |
2293 | lock_device_hash_lock(conf, hash)); | |
2294 | osh = get_free_stripe(conf, hash); | |
2295 | unlock_device_hash_lock(conf, hash); | |
f18c1a35 | 2296 | |
d592a996 | 2297 | for(i=0; i<conf->pool_size; i++) { |
ad01c9e3 | 2298 | nsh->dev[i].page = osh->dev[i].page; |
d592a996 SL |
2299 | nsh->dev[i].orig_page = osh->dev[i].page; |
2300 | } | |
566c09c5 | 2301 | nsh->hash_lock_index = hash; |
ad01c9e3 | 2302 | kmem_cache_free(conf->slab_cache, osh); |
566c09c5 SL |
2303 | cnt++; |
2304 | if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS + | |
2305 | !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) { | |
2306 | hash++; | |
2307 | cnt = 0; | |
2308 | } | |
ad01c9e3 N |
2309 | } |
2310 | kmem_cache_destroy(conf->slab_cache); | |
2311 | ||
2312 | /* Step 3. | |
2313 | * At this point, we are holding all the stripes so the array | |
2314 | * is completely stalled, so now is a good time to resize | |
d6f38f31 | 2315 | * conf->disks and the scribble region |
ad01c9e3 N |
2316 | */ |
2317 | ndisks = kzalloc(newsize * sizeof(struct disk_info), GFP_NOIO); | |
2318 | if (ndisks) { | |
d7bd398e | 2319 | for (i = 0; i < conf->pool_size; i++) |
ad01c9e3 | 2320 | ndisks[i] = conf->disks[i]; |
d7bd398e SL |
2321 | |
2322 | for (i = conf->pool_size; i < newsize; i++) { | |
2323 | ndisks[i].extra_page = alloc_page(GFP_NOIO); | |
2324 | if (!ndisks[i].extra_page) | |
2325 | err = -ENOMEM; | |
2326 | } | |
2327 | ||
2328 | if (err) { | |
2329 | for (i = conf->pool_size; i < newsize; i++) | |
2330 | if (ndisks[i].extra_page) | |
2331 | put_page(ndisks[i].extra_page); | |
2332 | kfree(ndisks); | |
2333 | } else { | |
2334 | kfree(conf->disks); | |
2335 | conf->disks = ndisks; | |
2336 | } | |
ad01c9e3 N |
2337 | } else |
2338 | err = -ENOMEM; | |
2339 | ||
2d5b569b | 2340 | mutex_unlock(&conf->cache_size_mutex); |
ad01c9e3 N |
2341 | /* Step 4, return new stripes to service */ |
2342 | while(!list_empty(&newstripes)) { | |
2343 | nsh = list_entry(newstripes.next, struct stripe_head, lru); | |
2344 | list_del_init(&nsh->lru); | |
d6f38f31 | 2345 | |
ad01c9e3 N |
2346 | for (i=conf->raid_disks; i < newsize; i++) |
2347 | if (nsh->dev[i].page == NULL) { | |
2348 | struct page *p = alloc_page(GFP_NOIO); | |
2349 | nsh->dev[i].page = p; | |
d592a996 | 2350 | nsh->dev[i].orig_page = p; |
ad01c9e3 N |
2351 | if (!p) |
2352 | err = -ENOMEM; | |
2353 | } | |
6d036f7d | 2354 | raid5_release_stripe(nsh); |
ad01c9e3 N |
2355 | } |
2356 | /* critical section pass, GFP_NOIO no longer needed */ | |
2357 | ||
2358 | conf->slab_cache = sc; | |
2359 | conf->active_name = 1-conf->active_name; | |
6e9eac2d N |
2360 | if (!err) |
2361 | conf->pool_size = newsize; | |
ad01c9e3 N |
2362 | return err; |
2363 | } | |
1da177e4 | 2364 | |
486f0644 | 2365 | static int drop_one_stripe(struct r5conf *conf) |
1da177e4 LT |
2366 | { |
2367 | struct stripe_head *sh; | |
49895bcc | 2368 | int hash = (conf->max_nr_stripes - 1) & STRIPE_HASH_LOCKS_MASK; |
1da177e4 | 2369 | |
566c09c5 SL |
2370 | spin_lock_irq(conf->hash_locks + hash); |
2371 | sh = get_free_stripe(conf, hash); | |
2372 | spin_unlock_irq(conf->hash_locks + hash); | |
3f294f4f N |
2373 | if (!sh) |
2374 | return 0; | |
78bafebd | 2375 | BUG_ON(atomic_read(&sh->count)); |
e4e11e38 | 2376 | shrink_buffers(sh); |
3f294f4f N |
2377 | kmem_cache_free(conf->slab_cache, sh); |
2378 | atomic_dec(&conf->active_stripes); | |
486f0644 | 2379 | conf->max_nr_stripes--; |
3f294f4f N |
2380 | return 1; |
2381 | } | |
2382 | ||
d1688a6d | 2383 | static void shrink_stripes(struct r5conf *conf) |
3f294f4f | 2384 | { |
486f0644 N |
2385 | while (conf->max_nr_stripes && |
2386 | drop_one_stripe(conf)) | |
2387 | ; | |
3f294f4f | 2388 | |
644df1a8 | 2389 | kmem_cache_destroy(conf->slab_cache); |
1da177e4 LT |
2390 | conf->slab_cache = NULL; |
2391 | } | |
2392 | ||
4246a0b6 | 2393 | static void raid5_end_read_request(struct bio * bi) |
1da177e4 | 2394 | { |
99c0fb5f | 2395 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2396 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2397 | int disks = sh->disks, i; |
d6950432 | 2398 | char b[BDEVNAME_SIZE]; |
dd054fce | 2399 | struct md_rdev *rdev = NULL; |
05616be5 | 2400 | sector_t s; |
1da177e4 LT |
2401 | |
2402 | for (i=0 ; i<disks; i++) | |
2403 | if (bi == &sh->dev[i].req) | |
2404 | break; | |
2405 | ||
4246a0b6 | 2406 | pr_debug("end_read_request %llu/%d, count: %d, error %d.\n", |
45b4233c | 2407 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2408 | bi->bi_error); |
1da177e4 | 2409 | if (i == disks) { |
5f9d1fde | 2410 | bio_reset(bi); |
1da177e4 | 2411 | BUG(); |
6712ecf8 | 2412 | return; |
1da177e4 | 2413 | } |
14a75d3e | 2414 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
dd054fce N |
2415 | /* If replacement finished while this request was outstanding, |
2416 | * 'replacement' might be NULL already. | |
2417 | * In that case it moved down to 'rdev'. | |
2418 | * rdev is not removed until all requests are finished. | |
2419 | */ | |
14a75d3e | 2420 | rdev = conf->disks[i].replacement; |
dd054fce | 2421 | if (!rdev) |
14a75d3e | 2422 | rdev = conf->disks[i].rdev; |
1da177e4 | 2423 | |
05616be5 N |
2424 | if (use_new_offset(conf, sh)) |
2425 | s = sh->sector + rdev->new_data_offset; | |
2426 | else | |
2427 | s = sh->sector + rdev->data_offset; | |
4246a0b6 | 2428 | if (!bi->bi_error) { |
1da177e4 | 2429 | set_bit(R5_UPTODATE, &sh->dev[i].flags); |
4e5314b5 | 2430 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { |
14a75d3e N |
2431 | /* Note that this cannot happen on a |
2432 | * replacement device. We just fail those on | |
2433 | * any error | |
2434 | */ | |
cc6167b4 N |
2435 | pr_info_ratelimited( |
2436 | "md/raid:%s: read error corrected (%lu sectors at %llu on %s)\n", | |
8bda470e | 2437 | mdname(conf->mddev), STRIPE_SECTORS, |
05616be5 | 2438 | (unsigned long long)s, |
8bda470e | 2439 | bdevname(rdev->bdev, b)); |
ddd5115f | 2440 | atomic_add(STRIPE_SECTORS, &rdev->corrected_errors); |
4e5314b5 N |
2441 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2442 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
3f9e7c14 | 2443 | } else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) |
2444 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2445 | ||
86aa1397 SL |
2446 | if (test_bit(R5_InJournal, &sh->dev[i].flags)) |
2447 | /* | |
2448 | * end read for a page in journal, this | |
2449 | * must be preparing for prexor in rmw | |
2450 | */ | |
2451 | set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags); | |
2452 | ||
14a75d3e N |
2453 | if (atomic_read(&rdev->read_errors)) |
2454 | atomic_set(&rdev->read_errors, 0); | |
1da177e4 | 2455 | } else { |
14a75d3e | 2456 | const char *bdn = bdevname(rdev->bdev, b); |
ba22dcbf | 2457 | int retry = 0; |
2e8ac303 | 2458 | int set_bad = 0; |
d6950432 | 2459 | |
1da177e4 | 2460 | clear_bit(R5_UPTODATE, &sh->dev[i].flags); |
d6950432 | 2461 | atomic_inc(&rdev->read_errors); |
14a75d3e | 2462 | if (test_bit(R5_ReadRepl, &sh->dev[i].flags)) |
cc6167b4 N |
2463 | pr_warn_ratelimited( |
2464 | "md/raid:%s: read error on replacement device (sector %llu on %s).\n", | |
14a75d3e | 2465 | mdname(conf->mddev), |
05616be5 | 2466 | (unsigned long long)s, |
14a75d3e | 2467 | bdn); |
2e8ac303 | 2468 | else if (conf->mddev->degraded >= conf->max_degraded) { |
2469 | set_bad = 1; | |
cc6167b4 N |
2470 | pr_warn_ratelimited( |
2471 | "md/raid:%s: read error not correctable (sector %llu on %s).\n", | |
8bda470e | 2472 | mdname(conf->mddev), |
05616be5 | 2473 | (unsigned long long)s, |
8bda470e | 2474 | bdn); |
2e8ac303 | 2475 | } else if (test_bit(R5_ReWrite, &sh->dev[i].flags)) { |
4e5314b5 | 2476 | /* Oh, no!!! */ |
2e8ac303 | 2477 | set_bad = 1; |
cc6167b4 N |
2478 | pr_warn_ratelimited( |
2479 | "md/raid:%s: read error NOT corrected!! (sector %llu on %s).\n", | |
8bda470e | 2480 | mdname(conf->mddev), |
05616be5 | 2481 | (unsigned long long)s, |
8bda470e | 2482 | bdn); |
2e8ac303 | 2483 | } else if (atomic_read(&rdev->read_errors) |
ba22dcbf | 2484 | > conf->max_nr_stripes) |
cc6167b4 | 2485 | pr_warn("md/raid:%s: Too many read errors, failing device %s.\n", |
d6950432 | 2486 | mdname(conf->mddev), bdn); |
ba22dcbf N |
2487 | else |
2488 | retry = 1; | |
edfa1f65 BY |
2489 | if (set_bad && test_bit(In_sync, &rdev->flags) |
2490 | && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) | |
2491 | retry = 1; | |
ba22dcbf | 2492 | if (retry) |
3f9e7c14 | 2493 | if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) { |
2494 | set_bit(R5_ReadError, &sh->dev[i].flags); | |
2495 | clear_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
2496 | } else | |
2497 | set_bit(R5_ReadNoMerge, &sh->dev[i].flags); | |
ba22dcbf | 2498 | else { |
4e5314b5 N |
2499 | clear_bit(R5_ReadError, &sh->dev[i].flags); |
2500 | clear_bit(R5_ReWrite, &sh->dev[i].flags); | |
2e8ac303 | 2501 | if (!(set_bad |
2502 | && test_bit(In_sync, &rdev->flags) | |
2503 | && rdev_set_badblocks( | |
2504 | rdev, sh->sector, STRIPE_SECTORS, 0))) | |
2505 | md_error(conf->mddev, rdev); | |
ba22dcbf | 2506 | } |
1da177e4 | 2507 | } |
14a75d3e | 2508 | rdev_dec_pending(rdev, conf->mddev); |
c9445555 | 2509 | bio_reset(bi); |
1da177e4 LT |
2510 | clear_bit(R5_LOCKED, &sh->dev[i].flags); |
2511 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 2512 | raid5_release_stripe(sh); |
1da177e4 LT |
2513 | } |
2514 | ||
4246a0b6 | 2515 | static void raid5_end_write_request(struct bio *bi) |
1da177e4 | 2516 | { |
99c0fb5f | 2517 | struct stripe_head *sh = bi->bi_private; |
d1688a6d | 2518 | struct r5conf *conf = sh->raid_conf; |
7ecaa1e6 | 2519 | int disks = sh->disks, i; |
977df362 | 2520 | struct md_rdev *uninitialized_var(rdev); |
b84db560 N |
2521 | sector_t first_bad; |
2522 | int bad_sectors; | |
977df362 | 2523 | int replacement = 0; |
1da177e4 | 2524 | |
977df362 N |
2525 | for (i = 0 ; i < disks; i++) { |
2526 | if (bi == &sh->dev[i].req) { | |
2527 | rdev = conf->disks[i].rdev; | |
1da177e4 | 2528 | break; |
977df362 N |
2529 | } |
2530 | if (bi == &sh->dev[i].rreq) { | |
2531 | rdev = conf->disks[i].replacement; | |
dd054fce N |
2532 | if (rdev) |
2533 | replacement = 1; | |
2534 | else | |
2535 | /* rdev was removed and 'replacement' | |
2536 | * replaced it. rdev is not removed | |
2537 | * until all requests are finished. | |
2538 | */ | |
2539 | rdev = conf->disks[i].rdev; | |
977df362 N |
2540 | break; |
2541 | } | |
2542 | } | |
4246a0b6 | 2543 | pr_debug("end_write_request %llu/%d, count %d, error: %d.\n", |
1da177e4 | 2544 | (unsigned long long)sh->sector, i, atomic_read(&sh->count), |
4246a0b6 | 2545 | bi->bi_error); |
1da177e4 | 2546 | if (i == disks) { |
5f9d1fde | 2547 | bio_reset(bi); |
1da177e4 | 2548 | BUG(); |
6712ecf8 | 2549 | return; |
1da177e4 LT |
2550 | } |
2551 | ||
977df362 | 2552 | if (replacement) { |
4246a0b6 | 2553 | if (bi->bi_error) |
977df362 N |
2554 | md_error(conf->mddev, rdev); |
2555 | else if (is_badblock(rdev, sh->sector, | |
2556 | STRIPE_SECTORS, | |
2557 | &first_bad, &bad_sectors)) | |
2558 | set_bit(R5_MadeGoodRepl, &sh->dev[i].flags); | |
2559 | } else { | |
4246a0b6 | 2560 | if (bi->bi_error) { |
9f97e4b1 | 2561 | set_bit(STRIPE_DEGRADED, &sh->state); |
977df362 N |
2562 | set_bit(WriteErrorSeen, &rdev->flags); |
2563 | set_bit(R5_WriteError, &sh->dev[i].flags); | |
3a6de292 N |
2564 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2565 | set_bit(MD_RECOVERY_NEEDED, | |
2566 | &rdev->mddev->recovery); | |
977df362 N |
2567 | } else if (is_badblock(rdev, sh->sector, |
2568 | STRIPE_SECTORS, | |
c0b32972 | 2569 | &first_bad, &bad_sectors)) { |
977df362 | 2570 | set_bit(R5_MadeGood, &sh->dev[i].flags); |
c0b32972 N |
2571 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) |
2572 | /* That was a successful write so make | |
2573 | * sure it looks like we already did | |
2574 | * a re-write. | |
2575 | */ | |
2576 | set_bit(R5_ReWrite, &sh->dev[i].flags); | |
2577 | } | |
977df362 N |
2578 | } |
2579 | rdev_dec_pending(rdev, conf->mddev); | |
1da177e4 | 2580 | |
4246a0b6 | 2581 | if (sh->batch_head && bi->bi_error && !replacement) |
72ac7330 | 2582 | set_bit(STRIPE_BATCH_ERR, &sh->batch_head->state); |
2583 | ||
c9445555 | 2584 | bio_reset(bi); |
977df362 N |
2585 | if (!test_and_clear_bit(R5_DOUBLE_LOCKED, &sh->dev[i].flags)) |
2586 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
1da177e4 | 2587 | set_bit(STRIPE_HANDLE, &sh->state); |
6d036f7d | 2588 | raid5_release_stripe(sh); |
59fc630b | 2589 | |
2590 | if (sh->batch_head && sh != sh->batch_head) | |
6d036f7d | 2591 | raid5_release_stripe(sh->batch_head); |
1da177e4 LT |
2592 | } |
2593 | ||
784052ec | 2594 | static void raid5_build_block(struct stripe_head *sh, int i, int previous) |
1da177e4 LT |
2595 | { |
2596 | struct r5dev *dev = &sh->dev[i]; | |
2597 | ||
1da177e4 | 2598 | dev->flags = 0; |
6d036f7d | 2599 | dev->sector = raid5_compute_blocknr(sh, i, previous); |
1da177e4 LT |
2600 | } |
2601 | ||
849674e4 | 2602 | static void raid5_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
2603 | { |
2604 | char b[BDEVNAME_SIZE]; | |
d1688a6d | 2605 | struct r5conf *conf = mddev->private; |
908f4fbd | 2606 | unsigned long flags; |
0c55e022 | 2607 | pr_debug("raid456: error called\n"); |
1da177e4 | 2608 | |
908f4fbd N |
2609 | spin_lock_irqsave(&conf->device_lock, flags); |
2610 | clear_bit(In_sync, &rdev->flags); | |
2e38a37f | 2611 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd N |
2612 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2613 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
2614 | ||
de393cde | 2615 | set_bit(Blocked, &rdev->flags); |
6f8d0c77 | 2616 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
2617 | set_mask_bits(&mddev->sb_flags, 0, |
2618 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
cc6167b4 N |
2619 | pr_crit("md/raid:%s: Disk failure on %s, disabling device.\n" |
2620 | "md/raid:%s: Operation continuing on %d devices.\n", | |
2621 | mdname(mddev), | |
2622 | bdevname(rdev->bdev, b), | |
2623 | mdname(mddev), | |
2624 | conf->raid_disks - mddev->degraded); | |
2e38a37f | 2625 | r5c_update_on_rdev_error(mddev); |
16a53ecc | 2626 | } |
1da177e4 LT |
2627 | |
2628 | /* | |
2629 | * Input: a 'big' sector number, | |
2630 | * Output: index of the data and parity disk, and the sector # in them. | |
2631 | */ | |
6d036f7d SL |
2632 | sector_t raid5_compute_sector(struct r5conf *conf, sector_t r_sector, |
2633 | int previous, int *dd_idx, | |
2634 | struct stripe_head *sh) | |
1da177e4 | 2635 | { |
6e3b96ed | 2636 | sector_t stripe, stripe2; |
35f2a591 | 2637 | sector_t chunk_number; |
1da177e4 | 2638 | unsigned int chunk_offset; |
911d4ee8 | 2639 | int pd_idx, qd_idx; |
67cc2b81 | 2640 | int ddf_layout = 0; |
1da177e4 | 2641 | sector_t new_sector; |
e183eaed N |
2642 | int algorithm = previous ? conf->prev_algo |
2643 | : conf->algorithm; | |
09c9e5fa AN |
2644 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2645 | : conf->chunk_sectors; | |
112bf897 N |
2646 | int raid_disks = previous ? conf->previous_raid_disks |
2647 | : conf->raid_disks; | |
2648 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 LT |
2649 | |
2650 | /* First compute the information on this sector */ | |
2651 | ||
2652 | /* | |
2653 | * Compute the chunk number and the sector offset inside the chunk | |
2654 | */ | |
2655 | chunk_offset = sector_div(r_sector, sectors_per_chunk); | |
2656 | chunk_number = r_sector; | |
1da177e4 LT |
2657 | |
2658 | /* | |
2659 | * Compute the stripe number | |
2660 | */ | |
35f2a591 N |
2661 | stripe = chunk_number; |
2662 | *dd_idx = sector_div(stripe, data_disks); | |
6e3b96ed | 2663 | stripe2 = stripe; |
1da177e4 LT |
2664 | /* |
2665 | * Select the parity disk based on the user selected algorithm. | |
2666 | */ | |
84789554 | 2667 | pd_idx = qd_idx = -1; |
16a53ecc N |
2668 | switch(conf->level) { |
2669 | case 4: | |
911d4ee8 | 2670 | pd_idx = data_disks; |
16a53ecc N |
2671 | break; |
2672 | case 5: | |
e183eaed | 2673 | switch (algorithm) { |
1da177e4 | 2674 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2675 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2676 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2677 | (*dd_idx)++; |
2678 | break; | |
2679 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2680 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2681 | if (*dd_idx >= pd_idx) |
1da177e4 LT |
2682 | (*dd_idx)++; |
2683 | break; | |
2684 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2685 | pd_idx = data_disks - sector_div(stripe2, raid_disks); |
911d4ee8 | 2686 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 LT |
2687 | break; |
2688 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2689 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 | 2690 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; |
1da177e4 | 2691 | break; |
99c0fb5f N |
2692 | case ALGORITHM_PARITY_0: |
2693 | pd_idx = 0; | |
2694 | (*dd_idx)++; | |
2695 | break; | |
2696 | case ALGORITHM_PARITY_N: | |
2697 | pd_idx = data_disks; | |
2698 | break; | |
1da177e4 | 2699 | default: |
99c0fb5f | 2700 | BUG(); |
16a53ecc N |
2701 | } |
2702 | break; | |
2703 | case 6: | |
2704 | ||
e183eaed | 2705 | switch (algorithm) { |
16a53ecc | 2706 | case ALGORITHM_LEFT_ASYMMETRIC: |
6e3b96ed | 2707 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2708 | qd_idx = pd_idx + 1; |
2709 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2710 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2711 | qd_idx = 0; |
2712 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2713 | (*dd_idx) += 2; /* D D P Q D */ |
2714 | break; | |
2715 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
6e3b96ed | 2716 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2717 | qd_idx = pd_idx + 1; |
2718 | if (pd_idx == raid_disks-1) { | |
99c0fb5f | 2719 | (*dd_idx)++; /* Q D D D P */ |
911d4ee8 N |
2720 | qd_idx = 0; |
2721 | } else if (*dd_idx >= pd_idx) | |
16a53ecc N |
2722 | (*dd_idx) += 2; /* D D P Q D */ |
2723 | break; | |
2724 | case ALGORITHM_LEFT_SYMMETRIC: | |
6e3b96ed | 2725 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
911d4ee8 N |
2726 | qd_idx = (pd_idx + 1) % raid_disks; |
2727 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc N |
2728 | break; |
2729 | case ALGORITHM_RIGHT_SYMMETRIC: | |
6e3b96ed | 2730 | pd_idx = sector_div(stripe2, raid_disks); |
911d4ee8 N |
2731 | qd_idx = (pd_idx + 1) % raid_disks; |
2732 | *dd_idx = (pd_idx + 2 + *dd_idx) % raid_disks; | |
16a53ecc | 2733 | break; |
99c0fb5f N |
2734 | |
2735 | case ALGORITHM_PARITY_0: | |
2736 | pd_idx = 0; | |
2737 | qd_idx = 1; | |
2738 | (*dd_idx) += 2; | |
2739 | break; | |
2740 | case ALGORITHM_PARITY_N: | |
2741 | pd_idx = data_disks; | |
2742 | qd_idx = data_disks + 1; | |
2743 | break; | |
2744 | ||
2745 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
2746 | /* Exactly the same as RIGHT_ASYMMETRIC, but or | |
2747 | * of blocks for computing Q is different. | |
2748 | */ | |
6e3b96ed | 2749 | pd_idx = sector_div(stripe2, raid_disks); |
99c0fb5f N |
2750 | qd_idx = pd_idx + 1; |
2751 | if (pd_idx == raid_disks-1) { | |
2752 | (*dd_idx)++; /* Q D D D P */ | |
2753 | qd_idx = 0; | |
2754 | } else if (*dd_idx >= pd_idx) | |
2755 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2756 | ddf_layout = 1; |
99c0fb5f N |
2757 | break; |
2758 | ||
2759 | case ALGORITHM_ROTATING_N_RESTART: | |
2760 | /* Same a left_asymmetric, by first stripe is | |
2761 | * D D D P Q rather than | |
2762 | * Q D D D P | |
2763 | */ | |
6e3b96ed N |
2764 | stripe2 += 1; |
2765 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); | |
99c0fb5f N |
2766 | qd_idx = pd_idx + 1; |
2767 | if (pd_idx == raid_disks-1) { | |
2768 | (*dd_idx)++; /* Q D D D P */ | |
2769 | qd_idx = 0; | |
2770 | } else if (*dd_idx >= pd_idx) | |
2771 | (*dd_idx) += 2; /* D D P Q D */ | |
67cc2b81 | 2772 | ddf_layout = 1; |
99c0fb5f N |
2773 | break; |
2774 | ||
2775 | case ALGORITHM_ROTATING_N_CONTINUE: | |
2776 | /* Same as left_symmetric but Q is before P */ | |
6e3b96ed | 2777 | pd_idx = raid_disks - 1 - sector_div(stripe2, raid_disks); |
99c0fb5f N |
2778 | qd_idx = (pd_idx + raid_disks - 1) % raid_disks; |
2779 | *dd_idx = (pd_idx + 1 + *dd_idx) % raid_disks; | |
67cc2b81 | 2780 | ddf_layout = 1; |
99c0fb5f N |
2781 | break; |
2782 | ||
2783 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2784 | /* RAID5 left_asymmetric, with Q on last device */ | |
6e3b96ed | 2785 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2786 | if (*dd_idx >= pd_idx) |
2787 | (*dd_idx)++; | |
2788 | qd_idx = raid_disks - 1; | |
2789 | break; | |
2790 | ||
2791 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
6e3b96ed | 2792 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2793 | if (*dd_idx >= pd_idx) |
2794 | (*dd_idx)++; | |
2795 | qd_idx = raid_disks - 1; | |
2796 | break; | |
2797 | ||
2798 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
6e3b96ed | 2799 | pd_idx = data_disks - sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2800 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2801 | qd_idx = raid_disks - 1; | |
2802 | break; | |
2803 | ||
2804 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
6e3b96ed | 2805 | pd_idx = sector_div(stripe2, raid_disks-1); |
99c0fb5f N |
2806 | *dd_idx = (pd_idx + 1 + *dd_idx) % (raid_disks-1); |
2807 | qd_idx = raid_disks - 1; | |
2808 | break; | |
2809 | ||
2810 | case ALGORITHM_PARITY_0_6: | |
2811 | pd_idx = 0; | |
2812 | (*dd_idx)++; | |
2813 | qd_idx = raid_disks - 1; | |
2814 | break; | |
2815 | ||
16a53ecc | 2816 | default: |
99c0fb5f | 2817 | BUG(); |
16a53ecc N |
2818 | } |
2819 | break; | |
1da177e4 LT |
2820 | } |
2821 | ||
911d4ee8 N |
2822 | if (sh) { |
2823 | sh->pd_idx = pd_idx; | |
2824 | sh->qd_idx = qd_idx; | |
67cc2b81 | 2825 | sh->ddf_layout = ddf_layout; |
911d4ee8 | 2826 | } |
1da177e4 LT |
2827 | /* |
2828 | * Finally, compute the new sector number | |
2829 | */ | |
2830 | new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; | |
2831 | return new_sector; | |
2832 | } | |
2833 | ||
6d036f7d | 2834 | sector_t raid5_compute_blocknr(struct stripe_head *sh, int i, int previous) |
1da177e4 | 2835 | { |
d1688a6d | 2836 | struct r5conf *conf = sh->raid_conf; |
b875e531 N |
2837 | int raid_disks = sh->disks; |
2838 | int data_disks = raid_disks - conf->max_degraded; | |
1da177e4 | 2839 | sector_t new_sector = sh->sector, check; |
09c9e5fa AN |
2840 | int sectors_per_chunk = previous ? conf->prev_chunk_sectors |
2841 | : conf->chunk_sectors; | |
e183eaed N |
2842 | int algorithm = previous ? conf->prev_algo |
2843 | : conf->algorithm; | |
1da177e4 LT |
2844 | sector_t stripe; |
2845 | int chunk_offset; | |
35f2a591 N |
2846 | sector_t chunk_number; |
2847 | int dummy1, dd_idx = i; | |
1da177e4 | 2848 | sector_t r_sector; |
911d4ee8 | 2849 | struct stripe_head sh2; |
1da177e4 LT |
2850 | |
2851 | chunk_offset = sector_div(new_sector, sectors_per_chunk); | |
2852 | stripe = new_sector; | |
1da177e4 | 2853 | |
16a53ecc N |
2854 | if (i == sh->pd_idx) |
2855 | return 0; | |
2856 | switch(conf->level) { | |
2857 | case 4: break; | |
2858 | case 5: | |
e183eaed | 2859 | switch (algorithm) { |
1da177e4 LT |
2860 | case ALGORITHM_LEFT_ASYMMETRIC: |
2861 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
2862 | if (i > sh->pd_idx) | |
2863 | i--; | |
2864 | break; | |
2865 | case ALGORITHM_LEFT_SYMMETRIC: | |
2866 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2867 | if (i < sh->pd_idx) | |
2868 | i += raid_disks; | |
2869 | i -= (sh->pd_idx + 1); | |
2870 | break; | |
99c0fb5f N |
2871 | case ALGORITHM_PARITY_0: |
2872 | i -= 1; | |
2873 | break; | |
2874 | case ALGORITHM_PARITY_N: | |
2875 | break; | |
1da177e4 | 2876 | default: |
99c0fb5f | 2877 | BUG(); |
16a53ecc N |
2878 | } |
2879 | break; | |
2880 | case 6: | |
d0dabf7e | 2881 | if (i == sh->qd_idx) |
16a53ecc | 2882 | return 0; /* It is the Q disk */ |
e183eaed | 2883 | switch (algorithm) { |
16a53ecc N |
2884 | case ALGORITHM_LEFT_ASYMMETRIC: |
2885 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
99c0fb5f N |
2886 | case ALGORITHM_ROTATING_ZERO_RESTART: |
2887 | case ALGORITHM_ROTATING_N_RESTART: | |
2888 | if (sh->pd_idx == raid_disks-1) | |
2889 | i--; /* Q D D D P */ | |
16a53ecc N |
2890 | else if (i > sh->pd_idx) |
2891 | i -= 2; /* D D P Q D */ | |
2892 | break; | |
2893 | case ALGORITHM_LEFT_SYMMETRIC: | |
2894 | case ALGORITHM_RIGHT_SYMMETRIC: | |
2895 | if (sh->pd_idx == raid_disks-1) | |
2896 | i--; /* Q D D D P */ | |
2897 | else { | |
2898 | /* D D P Q D */ | |
2899 | if (i < sh->pd_idx) | |
2900 | i += raid_disks; | |
2901 | i -= (sh->pd_idx + 2); | |
2902 | } | |
2903 | break; | |
99c0fb5f N |
2904 | case ALGORITHM_PARITY_0: |
2905 | i -= 2; | |
2906 | break; | |
2907 | case ALGORITHM_PARITY_N: | |
2908 | break; | |
2909 | case ALGORITHM_ROTATING_N_CONTINUE: | |
e4424fee | 2910 | /* Like left_symmetric, but P is before Q */ |
99c0fb5f N |
2911 | if (sh->pd_idx == 0) |
2912 | i--; /* P D D D Q */ | |
e4424fee N |
2913 | else { |
2914 | /* D D Q P D */ | |
2915 | if (i < sh->pd_idx) | |
2916 | i += raid_disks; | |
2917 | i -= (sh->pd_idx + 1); | |
2918 | } | |
99c0fb5f N |
2919 | break; |
2920 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
2921 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
2922 | if (i > sh->pd_idx) | |
2923 | i--; | |
2924 | break; | |
2925 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
2926 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
2927 | if (i < sh->pd_idx) | |
2928 | i += data_disks + 1; | |
2929 | i -= (sh->pd_idx + 1); | |
2930 | break; | |
2931 | case ALGORITHM_PARITY_0_6: | |
2932 | i -= 1; | |
2933 | break; | |
16a53ecc | 2934 | default: |
99c0fb5f | 2935 | BUG(); |
16a53ecc N |
2936 | } |
2937 | break; | |
1da177e4 LT |
2938 | } |
2939 | ||
2940 | chunk_number = stripe * data_disks + i; | |
35f2a591 | 2941 | r_sector = chunk_number * sectors_per_chunk + chunk_offset; |
1da177e4 | 2942 | |
112bf897 | 2943 | check = raid5_compute_sector(conf, r_sector, |
784052ec | 2944 | previous, &dummy1, &sh2); |
911d4ee8 N |
2945 | if (check != sh->sector || dummy1 != dd_idx || sh2.pd_idx != sh->pd_idx |
2946 | || sh2.qd_idx != sh->qd_idx) { | |
cc6167b4 N |
2947 | pr_warn("md/raid:%s: compute_blocknr: map not correct\n", |
2948 | mdname(conf->mddev)); | |
1da177e4 LT |
2949 | return 0; |
2950 | } | |
2951 | return r_sector; | |
2952 | } | |
2953 | ||
07e83364 SL |
2954 | /* |
2955 | * There are cases where we want handle_stripe_dirtying() and | |
2956 | * schedule_reconstruction() to delay towrite to some dev of a stripe. | |
2957 | * | |
2958 | * This function checks whether we want to delay the towrite. Specifically, | |
2959 | * we delay the towrite when: | |
2960 | * | |
2961 | * 1. degraded stripe has a non-overwrite to the missing dev, AND this | |
2962 | * stripe has data in journal (for other devices). | |
2963 | * | |
2964 | * In this case, when reading data for the non-overwrite dev, it is | |
2965 | * necessary to handle complex rmw of write back cache (prexor with | |
2966 | * orig_page, and xor with page). To keep read path simple, we would | |
2967 | * like to flush data in journal to RAID disks first, so complex rmw | |
2968 | * is handled in the write patch (handle_stripe_dirtying). | |
2969 | * | |
39b99586 SL |
2970 | * 2. when journal space is critical (R5C_LOG_CRITICAL=1) |
2971 | * | |
2972 | * It is important to be able to flush all stripes in raid5-cache. | |
2973 | * Therefore, we need reserve some space on the journal device for | |
2974 | * these flushes. If flush operation includes pending writes to the | |
2975 | * stripe, we need to reserve (conf->raid_disk + 1) pages per stripe | |
2976 | * for the flush out. If we exclude these pending writes from flush | |
2977 | * operation, we only need (conf->max_degraded + 1) pages per stripe. | |
2978 | * Therefore, excluding pending writes in these cases enables more | |
2979 | * efficient use of the journal device. | |
2980 | * | |
2981 | * Note: To make sure the stripe makes progress, we only delay | |
2982 | * towrite for stripes with data already in journal (injournal > 0). | |
2983 | * When LOG_CRITICAL, stripes with injournal == 0 will be sent to | |
2984 | * no_space_stripes list. | |
2985 | * | |
07e83364 | 2986 | */ |
39b99586 SL |
2987 | static inline bool delay_towrite(struct r5conf *conf, |
2988 | struct r5dev *dev, | |
2989 | struct stripe_head_state *s) | |
07e83364 | 2990 | { |
39b99586 SL |
2991 | /* case 1 above */ |
2992 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
2993 | !test_bit(R5_Insync, &dev->flags) && s->injournal) | |
2994 | return true; | |
2995 | /* case 2 above */ | |
2996 | if (test_bit(R5C_LOG_CRITICAL, &conf->cache_state) && | |
2997 | s->injournal > 0) | |
2998 | return true; | |
2999 | return false; | |
07e83364 SL |
3000 | } |
3001 | ||
600aa109 | 3002 | static void |
c0f7bddb | 3003 | schedule_reconstruction(struct stripe_head *sh, struct stripe_head_state *s, |
600aa109 | 3004 | int rcw, int expand) |
e33129d8 | 3005 | { |
584acdd4 | 3006 | int i, pd_idx = sh->pd_idx, qd_idx = sh->qd_idx, disks = sh->disks; |
d1688a6d | 3007 | struct r5conf *conf = sh->raid_conf; |
c0f7bddb | 3008 | int level = conf->level; |
e33129d8 DW |
3009 | |
3010 | if (rcw) { | |
1e6d690b SL |
3011 | /* |
3012 | * In some cases, handle_stripe_dirtying initially decided to | |
3013 | * run rmw and allocates extra page for prexor. However, rcw is | |
3014 | * cheaper later on. We need to free the extra page now, | |
3015 | * because we won't be able to do that in ops_complete_prexor(). | |
3016 | */ | |
3017 | r5c_release_extra_page(sh); | |
e33129d8 DW |
3018 | |
3019 | for (i = disks; i--; ) { | |
3020 | struct r5dev *dev = &sh->dev[i]; | |
3021 | ||
39b99586 | 3022 | if (dev->towrite && !delay_towrite(conf, dev, s)) { |
e33129d8 | 3023 | set_bit(R5_LOCKED, &dev->flags); |
d8ee0728 | 3024 | set_bit(R5_Wantdrain, &dev->flags); |
e33129d8 DW |
3025 | if (!expand) |
3026 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3027 | s->locked++; |
1e6d690b SL |
3028 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3029 | set_bit(R5_LOCKED, &dev->flags); | |
3030 | s->locked++; | |
e33129d8 DW |
3031 | } |
3032 | } | |
ce7d363a N |
3033 | /* if we are not expanding this is a proper write request, and |
3034 | * there will be bios with new data to be drained into the | |
3035 | * stripe cache | |
3036 | */ | |
3037 | if (!expand) { | |
3038 | if (!s->locked) | |
3039 | /* False alarm, nothing to do */ | |
3040 | return; | |
3041 | sh->reconstruct_state = reconstruct_state_drain_run; | |
3042 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3043 | } else | |
3044 | sh->reconstruct_state = reconstruct_state_run; | |
3045 | ||
3046 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
3047 | ||
c0f7bddb | 3048 | if (s->locked + conf->max_degraded == disks) |
8b3e6cdc | 3049 | if (!test_and_set_bit(STRIPE_FULL_WRITE, &sh->state)) |
c0f7bddb | 3050 | atomic_inc(&conf->pending_full_writes); |
e33129d8 DW |
3051 | } else { |
3052 | BUG_ON(!(test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags) || | |
3053 | test_bit(R5_Wantcompute, &sh->dev[pd_idx].flags))); | |
584acdd4 MS |
3054 | BUG_ON(level == 6 && |
3055 | (!(test_bit(R5_UPTODATE, &sh->dev[qd_idx].flags) || | |
3056 | test_bit(R5_Wantcompute, &sh->dev[qd_idx].flags)))); | |
e33129d8 | 3057 | |
e33129d8 DW |
3058 | for (i = disks; i--; ) { |
3059 | struct r5dev *dev = &sh->dev[i]; | |
584acdd4 | 3060 | if (i == pd_idx || i == qd_idx) |
e33129d8 DW |
3061 | continue; |
3062 | ||
e33129d8 DW |
3063 | if (dev->towrite && |
3064 | (test_bit(R5_UPTODATE, &dev->flags) || | |
d8ee0728 DW |
3065 | test_bit(R5_Wantcompute, &dev->flags))) { |
3066 | set_bit(R5_Wantdrain, &dev->flags); | |
e33129d8 DW |
3067 | set_bit(R5_LOCKED, &dev->flags); |
3068 | clear_bit(R5_UPTODATE, &dev->flags); | |
600aa109 | 3069 | s->locked++; |
1e6d690b SL |
3070 | } else if (test_bit(R5_InJournal, &dev->flags)) { |
3071 | set_bit(R5_LOCKED, &dev->flags); | |
3072 | s->locked++; | |
e33129d8 DW |
3073 | } |
3074 | } | |
ce7d363a N |
3075 | if (!s->locked) |
3076 | /* False alarm - nothing to do */ | |
3077 | return; | |
3078 | sh->reconstruct_state = reconstruct_state_prexor_drain_run; | |
3079 | set_bit(STRIPE_OP_PREXOR, &s->ops_request); | |
3080 | set_bit(STRIPE_OP_BIODRAIN, &s->ops_request); | |
3081 | set_bit(STRIPE_OP_RECONSTRUCT, &s->ops_request); | |
e33129d8 DW |
3082 | } |
3083 | ||
c0f7bddb | 3084 | /* keep the parity disk(s) locked while asynchronous operations |
e33129d8 DW |
3085 | * are in flight |
3086 | */ | |
3087 | set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); | |
3088 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
600aa109 | 3089 | s->locked++; |
e33129d8 | 3090 | |
c0f7bddb YT |
3091 | if (level == 6) { |
3092 | int qd_idx = sh->qd_idx; | |
3093 | struct r5dev *dev = &sh->dev[qd_idx]; | |
3094 | ||
3095 | set_bit(R5_LOCKED, &dev->flags); | |
3096 | clear_bit(R5_UPTODATE, &dev->flags); | |
3097 | s->locked++; | |
3098 | } | |
3099 | ||
600aa109 | 3100 | pr_debug("%s: stripe %llu locked: %d ops_request: %lx\n", |
e46b272b | 3101 | __func__, (unsigned long long)sh->sector, |
600aa109 | 3102 | s->locked, s->ops_request); |
e33129d8 | 3103 | } |
16a53ecc | 3104 | |
1da177e4 LT |
3105 | /* |
3106 | * Each stripe/dev can have one or more bion attached. | |
16a53ecc | 3107 | * toread/towrite point to the first in a chain. |
1da177e4 LT |
3108 | * The bi_next chain must be in order. |
3109 | */ | |
da41ba65 | 3110 | static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, |
3111 | int forwrite, int previous) | |
1da177e4 LT |
3112 | { |
3113 | struct bio **bip; | |
d1688a6d | 3114 | struct r5conf *conf = sh->raid_conf; |
72626685 | 3115 | int firstwrite=0; |
1da177e4 | 3116 | |
cbe47ec5 | 3117 | pr_debug("adding bi b#%llu to stripe s#%llu\n", |
4f024f37 | 3118 | (unsigned long long)bi->bi_iter.bi_sector, |
1da177e4 LT |
3119 | (unsigned long long)sh->sector); |
3120 | ||
b17459c0 SL |
3121 | /* |
3122 | * If several bio share a stripe. The bio bi_phys_segments acts as a | |
3123 | * reference count to avoid race. The reference count should already be | |
3124 | * increased before this function is called (for example, in | |
849674e4 | 3125 | * raid5_make_request()), so other bio sharing this stripe will not free the |
b17459c0 SL |
3126 | * stripe. If a stripe is owned by one stripe, the stripe lock will |
3127 | * protect it. | |
3128 | */ | |
3129 | spin_lock_irq(&sh->stripe_lock); | |
59fc630b | 3130 | /* Don't allow new IO added to stripes in batch list */ |
3131 | if (sh->batch_head) | |
3132 | goto overlap; | |
72626685 | 3133 | if (forwrite) { |
1da177e4 | 3134 | bip = &sh->dev[dd_idx].towrite; |
7eaf7e8e | 3135 | if (*bip == NULL) |
72626685 N |
3136 | firstwrite = 1; |
3137 | } else | |
1da177e4 | 3138 | bip = &sh->dev[dd_idx].toread; |
4f024f37 KO |
3139 | while (*bip && (*bip)->bi_iter.bi_sector < bi->bi_iter.bi_sector) { |
3140 | if (bio_end_sector(*bip) > bi->bi_iter.bi_sector) | |
1da177e4 LT |
3141 | goto overlap; |
3142 | bip = & (*bip)->bi_next; | |
3143 | } | |
4f024f37 | 3144 | if (*bip && (*bip)->bi_iter.bi_sector < bio_end_sector(bi)) |
1da177e4 LT |
3145 | goto overlap; |
3146 | ||
da41ba65 | 3147 | if (!forwrite || previous) |
3148 | clear_bit(STRIPE_BATCH_READY, &sh->state); | |
3149 | ||
78bafebd | 3150 | BUG_ON(*bip && bi->bi_next && (*bip) != bi->bi_next); |
1da177e4 LT |
3151 | if (*bip) |
3152 | bi->bi_next = *bip; | |
3153 | *bip = bi; | |
e7836bd6 | 3154 | raid5_inc_bi_active_stripes(bi); |
72626685 | 3155 | |
1da177e4 LT |
3156 | if (forwrite) { |
3157 | /* check if page is covered */ | |
3158 | sector_t sector = sh->dev[dd_idx].sector; | |
3159 | for (bi=sh->dev[dd_idx].towrite; | |
3160 | sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && | |
4f024f37 | 3161 | bi && bi->bi_iter.bi_sector <= sector; |
1da177e4 | 3162 | bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { |
f73a1c7d KO |
3163 | if (bio_end_sector(bi) >= sector) |
3164 | sector = bio_end_sector(bi); | |
1da177e4 LT |
3165 | } |
3166 | if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) | |
7a87f434 | 3167 | if (!test_and_set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags)) |
3168 | sh->overwrite_disks++; | |
1da177e4 | 3169 | } |
cbe47ec5 N |
3170 | |
3171 | pr_debug("added bi b#%llu to stripe s#%llu, disk %d.\n", | |
4f024f37 | 3172 | (unsigned long long)(*bip)->bi_iter.bi_sector, |
cbe47ec5 N |
3173 | (unsigned long long)sh->sector, dd_idx); |
3174 | ||
3175 | if (conf->mddev->bitmap && firstwrite) { | |
d0852df5 N |
3176 | /* Cannot hold spinlock over bitmap_startwrite, |
3177 | * but must ensure this isn't added to a batch until | |
3178 | * we have added to the bitmap and set bm_seq. | |
3179 | * So set STRIPE_BITMAP_PENDING to prevent | |
3180 | * batching. | |
3181 | * If multiple add_stripe_bio() calls race here they | |
3182 | * much all set STRIPE_BITMAP_PENDING. So only the first one | |
3183 | * to complete "bitmap_startwrite" gets to set | |
3184 | * STRIPE_BIT_DELAY. This is important as once a stripe | |
3185 | * is added to a batch, STRIPE_BIT_DELAY cannot be changed | |
3186 | * any more. | |
3187 | */ | |
3188 | set_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3189 | spin_unlock_irq(&sh->stripe_lock); | |
cbe47ec5 N |
3190 | bitmap_startwrite(conf->mddev->bitmap, sh->sector, |
3191 | STRIPE_SECTORS, 0); | |
d0852df5 N |
3192 | spin_lock_irq(&sh->stripe_lock); |
3193 | clear_bit(STRIPE_BITMAP_PENDING, &sh->state); | |
3194 | if (!sh->batch_head) { | |
3195 | sh->bm_seq = conf->seq_flush+1; | |
3196 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
3197 | } | |
cbe47ec5 | 3198 | } |
d0852df5 | 3199 | spin_unlock_irq(&sh->stripe_lock); |
59fc630b | 3200 | |
3201 | if (stripe_can_batch(sh)) | |
3202 | stripe_add_to_batch_list(conf, sh); | |
1da177e4 LT |
3203 | return 1; |
3204 | ||
3205 | overlap: | |
3206 | set_bit(R5_Overlap, &sh->dev[dd_idx].flags); | |
b17459c0 | 3207 | spin_unlock_irq(&sh->stripe_lock); |
1da177e4 LT |
3208 | return 0; |
3209 | } | |
3210 | ||
d1688a6d | 3211 | static void end_reshape(struct r5conf *conf); |
29269553 | 3212 | |
d1688a6d | 3213 | static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous, |
911d4ee8 | 3214 | struct stripe_head *sh) |
ccfcc3c1 | 3215 | { |
784052ec | 3216 | int sectors_per_chunk = |
09c9e5fa | 3217 | previous ? conf->prev_chunk_sectors : conf->chunk_sectors; |
911d4ee8 | 3218 | int dd_idx; |
2d2063ce | 3219 | int chunk_offset = sector_div(stripe, sectors_per_chunk); |
112bf897 | 3220 | int disks = previous ? conf->previous_raid_disks : conf->raid_disks; |
2d2063ce | 3221 | |
112bf897 N |
3222 | raid5_compute_sector(conf, |
3223 | stripe * (disks - conf->max_degraded) | |
b875e531 | 3224 | *sectors_per_chunk + chunk_offset, |
112bf897 | 3225 | previous, |
911d4ee8 | 3226 | &dd_idx, sh); |
ccfcc3c1 N |
3227 | } |
3228 | ||
a4456856 | 3229 | static void |
d1688a6d | 3230 | handle_failed_stripe(struct r5conf *conf, struct stripe_head *sh, |
a4456856 | 3231 | struct stripe_head_state *s, int disks, |
34a6f80e | 3232 | struct bio_list *return_bi) |
a4456856 DW |
3233 | { |
3234 | int i; | |
59fc630b | 3235 | BUG_ON(sh->batch_head); |
a4456856 DW |
3236 | for (i = disks; i--; ) { |
3237 | struct bio *bi; | |
3238 | int bitmap_end = 0; | |
3239 | ||
3240 | if (test_bit(R5_ReadError, &sh->dev[i].flags)) { | |
3cb03002 | 3241 | struct md_rdev *rdev; |
a4456856 DW |
3242 | rcu_read_lock(); |
3243 | rdev = rcu_dereference(conf->disks[i].rdev); | |
f5b67ae8 N |
3244 | if (rdev && test_bit(In_sync, &rdev->flags) && |
3245 | !test_bit(Faulty, &rdev->flags)) | |
7f0da59b N |
3246 | atomic_inc(&rdev->nr_pending); |
3247 | else | |
3248 | rdev = NULL; | |
a4456856 | 3249 | rcu_read_unlock(); |
7f0da59b N |
3250 | if (rdev) { |
3251 | if (!rdev_set_badblocks( | |
3252 | rdev, | |
3253 | sh->sector, | |
3254 | STRIPE_SECTORS, 0)) | |
3255 | md_error(conf->mddev, rdev); | |
3256 | rdev_dec_pending(rdev, conf->mddev); | |
3257 | } | |
a4456856 | 3258 | } |
b17459c0 | 3259 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3260 | /* fail all writes first */ |
3261 | bi = sh->dev[i].towrite; | |
3262 | sh->dev[i].towrite = NULL; | |
7a87f434 | 3263 | sh->overwrite_disks = 0; |
b17459c0 | 3264 | spin_unlock_irq(&sh->stripe_lock); |
1ed850f3 | 3265 | if (bi) |
a4456856 | 3266 | bitmap_end = 1; |
a4456856 | 3267 | |
0576b1c6 SL |
3268 | r5l_stripe_write_finished(sh); |
3269 | ||
a4456856 DW |
3270 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3271 | wake_up(&conf->wait_for_overlap); | |
3272 | ||
4f024f37 | 3273 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3274 | sh->dev[i].sector + STRIPE_SECTORS) { |
3275 | struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3276 | |
3277 | bi->bi_error = -EIO; | |
e7836bd6 | 3278 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3279 | md_write_end(conf->mddev); |
34a6f80e | 3280 | bio_list_add(return_bi, bi); |
a4456856 DW |
3281 | } |
3282 | bi = nextbi; | |
3283 | } | |
7eaf7e8e SL |
3284 | if (bitmap_end) |
3285 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3286 | STRIPE_SECTORS, 0, 0); | |
3287 | bitmap_end = 0; | |
a4456856 DW |
3288 | /* and fail all 'written' */ |
3289 | bi = sh->dev[i].written; | |
3290 | sh->dev[i].written = NULL; | |
d592a996 SL |
3291 | if (test_and_clear_bit(R5_SkipCopy, &sh->dev[i].flags)) { |
3292 | WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
3293 | sh->dev[i].page = sh->dev[i].orig_page; | |
3294 | } | |
3295 | ||
a4456856 | 3296 | if (bi) bitmap_end = 1; |
4f024f37 | 3297 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3298 | sh->dev[i].sector + STRIPE_SECTORS) { |
3299 | struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3300 | |
3301 | bi->bi_error = -EIO; | |
e7836bd6 | 3302 | if (!raid5_dec_bi_active_stripes(bi)) { |
a4456856 | 3303 | md_write_end(conf->mddev); |
34a6f80e | 3304 | bio_list_add(return_bi, bi); |
a4456856 DW |
3305 | } |
3306 | bi = bi2; | |
3307 | } | |
3308 | ||
b5e98d65 DW |
3309 | /* fail any reads if this device is non-operational and |
3310 | * the data has not reached the cache yet. | |
3311 | */ | |
3312 | if (!test_bit(R5_Wantfill, &sh->dev[i].flags) && | |
6e74a9cf | 3313 | s->failed > conf->max_degraded && |
b5e98d65 DW |
3314 | (!test_bit(R5_Insync, &sh->dev[i].flags) || |
3315 | test_bit(R5_ReadError, &sh->dev[i].flags))) { | |
143c4d05 | 3316 | spin_lock_irq(&sh->stripe_lock); |
a4456856 DW |
3317 | bi = sh->dev[i].toread; |
3318 | sh->dev[i].toread = NULL; | |
143c4d05 | 3319 | spin_unlock_irq(&sh->stripe_lock); |
a4456856 DW |
3320 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) |
3321 | wake_up(&conf->wait_for_overlap); | |
ebda780b SL |
3322 | if (bi) |
3323 | s->to_read--; | |
4f024f37 | 3324 | while (bi && bi->bi_iter.bi_sector < |
a4456856 DW |
3325 | sh->dev[i].sector + STRIPE_SECTORS) { |
3326 | struct bio *nextbi = | |
3327 | r5_next_bio(bi, sh->dev[i].sector); | |
4246a0b6 CH |
3328 | |
3329 | bi->bi_error = -EIO; | |
34a6f80e N |
3330 | if (!raid5_dec_bi_active_stripes(bi)) |
3331 | bio_list_add(return_bi, bi); | |
a4456856 DW |
3332 | bi = nextbi; |
3333 | } | |
3334 | } | |
a4456856 DW |
3335 | if (bitmap_end) |
3336 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, | |
3337 | STRIPE_SECTORS, 0, 0); | |
8cfa7b0f N |
3338 | /* If we were in the middle of a write the parity block might |
3339 | * still be locked - so just clear all R5_LOCKED flags | |
3340 | */ | |
3341 | clear_bit(R5_LOCKED, &sh->dev[i].flags); | |
a4456856 | 3342 | } |
ebda780b SL |
3343 | s->to_write = 0; |
3344 | s->written = 0; | |
a4456856 | 3345 | |
8b3e6cdc DW |
3346 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) |
3347 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3348 | md_wakeup_thread(conf->mddev->thread); | |
a4456856 DW |
3349 | } |
3350 | ||
7f0da59b | 3351 | static void |
d1688a6d | 3352 | handle_failed_sync(struct r5conf *conf, struct stripe_head *sh, |
7f0da59b N |
3353 | struct stripe_head_state *s) |
3354 | { | |
3355 | int abort = 0; | |
3356 | int i; | |
3357 | ||
59fc630b | 3358 | BUG_ON(sh->batch_head); |
7f0da59b | 3359 | clear_bit(STRIPE_SYNCING, &sh->state); |
f8dfcffd N |
3360 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
3361 | wake_up(&conf->wait_for_overlap); | |
7f0da59b | 3362 | s->syncing = 0; |
9a3e1101 | 3363 | s->replacing = 0; |
7f0da59b | 3364 | /* There is nothing more to do for sync/check/repair. |
18b9837e N |
3365 | * Don't even need to abort as that is handled elsewhere |
3366 | * if needed, and not always wanted e.g. if there is a known | |
3367 | * bad block here. | |
9a3e1101 | 3368 | * For recover/replace we need to record a bad block on all |
7f0da59b N |
3369 | * non-sync devices, or abort the recovery |
3370 | */ | |
18b9837e N |
3371 | if (test_bit(MD_RECOVERY_RECOVER, &conf->mddev->recovery)) { |
3372 | /* During recovery devices cannot be removed, so | |
3373 | * locking and refcounting of rdevs is not needed | |
3374 | */ | |
e50d3992 | 3375 | rcu_read_lock(); |
18b9837e | 3376 | for (i = 0; i < conf->raid_disks; i++) { |
e50d3992 | 3377 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); |
18b9837e N |
3378 | if (rdev |
3379 | && !test_bit(Faulty, &rdev->flags) | |
3380 | && !test_bit(In_sync, &rdev->flags) | |
3381 | && !rdev_set_badblocks(rdev, sh->sector, | |
3382 | STRIPE_SECTORS, 0)) | |
3383 | abort = 1; | |
e50d3992 | 3384 | rdev = rcu_dereference(conf->disks[i].replacement); |
18b9837e N |
3385 | if (rdev |
3386 | && !test_bit(Faulty, &rdev->flags) | |
3387 | && !test_bit(In_sync, &rdev->flags) | |
3388 | && !rdev_set_badblocks(rdev, sh->sector, | |
3389 | STRIPE_SECTORS, 0)) | |
3390 | abort = 1; | |
3391 | } | |
e50d3992 | 3392 | rcu_read_unlock(); |
18b9837e N |
3393 | if (abort) |
3394 | conf->recovery_disabled = | |
3395 | conf->mddev->recovery_disabled; | |
7f0da59b | 3396 | } |
18b9837e | 3397 | md_done_sync(conf->mddev, STRIPE_SECTORS, !abort); |
7f0da59b N |
3398 | } |
3399 | ||
9a3e1101 N |
3400 | static int want_replace(struct stripe_head *sh, int disk_idx) |
3401 | { | |
3402 | struct md_rdev *rdev; | |
3403 | int rv = 0; | |
3f232d6a N |
3404 | |
3405 | rcu_read_lock(); | |
3406 | rdev = rcu_dereference(sh->raid_conf->disks[disk_idx].replacement); | |
9a3e1101 N |
3407 | if (rdev |
3408 | && !test_bit(Faulty, &rdev->flags) | |
3409 | && !test_bit(In_sync, &rdev->flags) | |
3410 | && (rdev->recovery_offset <= sh->sector | |
3411 | || rdev->mddev->recovery_cp <= sh->sector)) | |
3412 | rv = 1; | |
3f232d6a | 3413 | rcu_read_unlock(); |
9a3e1101 N |
3414 | return rv; |
3415 | } | |
3416 | ||
2c58f06e N |
3417 | static int need_this_block(struct stripe_head *sh, struct stripe_head_state *s, |
3418 | int disk_idx, int disks) | |
a4456856 | 3419 | { |
5599becc | 3420 | struct r5dev *dev = &sh->dev[disk_idx]; |
f2b3b44d N |
3421 | struct r5dev *fdev[2] = { &sh->dev[s->failed_num[0]], |
3422 | &sh->dev[s->failed_num[1]] }; | |
ea664c82 | 3423 | int i; |
5599becc | 3424 | |
a79cfe12 N |
3425 | |
3426 | if (test_bit(R5_LOCKED, &dev->flags) || | |
3427 | test_bit(R5_UPTODATE, &dev->flags)) | |
3428 | /* No point reading this as we already have it or have | |
3429 | * decided to get it. | |
3430 | */ | |
3431 | return 0; | |
3432 | ||
3433 | if (dev->toread || | |
3434 | (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags))) | |
3435 | /* We need this block to directly satisfy a request */ | |
3436 | return 1; | |
3437 | ||
3438 | if (s->syncing || s->expanding || | |
3439 | (s->replacing && want_replace(sh, disk_idx))) | |
3440 | /* When syncing, or expanding we read everything. | |
3441 | * When replacing, we need the replaced block. | |
3442 | */ | |
3443 | return 1; | |
3444 | ||
3445 | if ((s->failed >= 1 && fdev[0]->toread) || | |
3446 | (s->failed >= 2 && fdev[1]->toread)) | |
3447 | /* If we want to read from a failed device, then | |
3448 | * we need to actually read every other device. | |
3449 | */ | |
3450 | return 1; | |
3451 | ||
a9d56950 N |
3452 | /* Sometimes neither read-modify-write nor reconstruct-write |
3453 | * cycles can work. In those cases we read every block we | |
3454 | * can. Then the parity-update is certain to have enough to | |
3455 | * work with. | |
3456 | * This can only be a problem when we need to write something, | |
3457 | * and some device has failed. If either of those tests | |
3458 | * fail we need look no further. | |
3459 | */ | |
3460 | if (!s->failed || !s->to_write) | |
3461 | return 0; | |
3462 | ||
3463 | if (test_bit(R5_Insync, &dev->flags) && | |
3464 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3465 | /* Pre-reads at not permitted until after short delay | |
3466 | * to gather multiple requests. However if this | |
3467 | * device is no Insync, the block could only be be computed | |
3468 | * and there is no need to delay that. | |
3469 | */ | |
3470 | return 0; | |
ea664c82 | 3471 | |
36707bb2 | 3472 | for (i = 0; i < s->failed && i < 2; i++) { |
ea664c82 N |
3473 | if (fdev[i]->towrite && |
3474 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
3475 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) | |
3476 | /* If we have a partial write to a failed | |
3477 | * device, then we will need to reconstruct | |
3478 | * the content of that device, so all other | |
3479 | * devices must be read. | |
3480 | */ | |
3481 | return 1; | |
3482 | } | |
3483 | ||
3484 | /* If we are forced to do a reconstruct-write, either because | |
3485 | * the current RAID6 implementation only supports that, or | |
3486 | * or because parity cannot be trusted and we are currently | |
3487 | * recovering it, there is extra need to be careful. | |
3488 | * If one of the devices that we would need to read, because | |
3489 | * it is not being overwritten (and maybe not written at all) | |
3490 | * is missing/faulty, then we need to read everything we can. | |
3491 | */ | |
3492 | if (sh->raid_conf->level != 6 && | |
3493 | sh->sector < sh->raid_conf->mddev->recovery_cp) | |
3494 | /* reconstruct-write isn't being forced */ | |
3495 | return 0; | |
36707bb2 | 3496 | for (i = 0; i < s->failed && i < 2; i++) { |
10d82c5f N |
3497 | if (s->failed_num[i] != sh->pd_idx && |
3498 | s->failed_num[i] != sh->qd_idx && | |
3499 | !test_bit(R5_UPTODATE, &fdev[i]->flags) && | |
ea664c82 N |
3500 | !test_bit(R5_OVERWRITE, &fdev[i]->flags)) |
3501 | return 1; | |
3502 | } | |
3503 | ||
2c58f06e N |
3504 | return 0; |
3505 | } | |
3506 | ||
ba02684d SL |
3507 | /* fetch_block - checks the given member device to see if its data needs |
3508 | * to be read or computed to satisfy a request. | |
3509 | * | |
3510 | * Returns 1 when no more member devices need to be checked, otherwise returns | |
3511 | * 0 to tell the loop in handle_stripe_fill to continue | |
3512 | */ | |
2c58f06e N |
3513 | static int fetch_block(struct stripe_head *sh, struct stripe_head_state *s, |
3514 | int disk_idx, int disks) | |
3515 | { | |
3516 | struct r5dev *dev = &sh->dev[disk_idx]; | |
3517 | ||
3518 | /* is the data in this block needed, and can we get it? */ | |
3519 | if (need_this_block(sh, s, disk_idx, disks)) { | |
5599becc YT |
3520 | /* we would like to get this block, possibly by computing it, |
3521 | * otherwise read it if the backing disk is insync | |
3522 | */ | |
3523 | BUG_ON(test_bit(R5_Wantcompute, &dev->flags)); | |
3524 | BUG_ON(test_bit(R5_Wantread, &dev->flags)); | |
b0c783b3 | 3525 | BUG_ON(sh->batch_head); |
5599becc | 3526 | if ((s->uptodate == disks - 1) && |
f2b3b44d N |
3527 | (s->failed && (disk_idx == s->failed_num[0] || |
3528 | disk_idx == s->failed_num[1]))) { | |
5599becc YT |
3529 | /* have disk failed, and we're requested to fetch it; |
3530 | * do compute it | |
a4456856 | 3531 | */ |
5599becc YT |
3532 | pr_debug("Computing stripe %llu block %d\n", |
3533 | (unsigned long long)sh->sector, disk_idx); | |
3534 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3535 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3536 | set_bit(R5_Wantcompute, &dev->flags); | |
3537 | sh->ops.target = disk_idx; | |
3538 | sh->ops.target2 = -1; /* no 2nd target */ | |
3539 | s->req_compute = 1; | |
93b3dbce N |
3540 | /* Careful: from this point on 'uptodate' is in the eye |
3541 | * of raid_run_ops which services 'compute' operations | |
3542 | * before writes. R5_Wantcompute flags a block that will | |
3543 | * be R5_UPTODATE by the time it is needed for a | |
3544 | * subsequent operation. | |
3545 | */ | |
5599becc YT |
3546 | s->uptodate++; |
3547 | return 1; | |
3548 | } else if (s->uptodate == disks-2 && s->failed >= 2) { | |
3549 | /* Computing 2-failure is *very* expensive; only | |
3550 | * do it if failed >= 2 | |
3551 | */ | |
3552 | int other; | |
3553 | for (other = disks; other--; ) { | |
3554 | if (other == disk_idx) | |
3555 | continue; | |
3556 | if (!test_bit(R5_UPTODATE, | |
3557 | &sh->dev[other].flags)) | |
3558 | break; | |
a4456856 | 3559 | } |
5599becc YT |
3560 | BUG_ON(other < 0); |
3561 | pr_debug("Computing stripe %llu blocks %d,%d\n", | |
3562 | (unsigned long long)sh->sector, | |
3563 | disk_idx, other); | |
3564 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
3565 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
3566 | set_bit(R5_Wantcompute, &sh->dev[disk_idx].flags); | |
3567 | set_bit(R5_Wantcompute, &sh->dev[other].flags); | |
3568 | sh->ops.target = disk_idx; | |
3569 | sh->ops.target2 = other; | |
3570 | s->uptodate += 2; | |
3571 | s->req_compute = 1; | |
3572 | return 1; | |
3573 | } else if (test_bit(R5_Insync, &dev->flags)) { | |
3574 | set_bit(R5_LOCKED, &dev->flags); | |
3575 | set_bit(R5_Wantread, &dev->flags); | |
3576 | s->locked++; | |
3577 | pr_debug("Reading block %d (sync=%d)\n", | |
3578 | disk_idx, s->syncing); | |
a4456856 DW |
3579 | } |
3580 | } | |
5599becc YT |
3581 | |
3582 | return 0; | |
3583 | } | |
3584 | ||
3585 | /** | |
93b3dbce | 3586 | * handle_stripe_fill - read or compute data to satisfy pending requests. |
5599becc | 3587 | */ |
93b3dbce N |
3588 | static void handle_stripe_fill(struct stripe_head *sh, |
3589 | struct stripe_head_state *s, | |
3590 | int disks) | |
5599becc YT |
3591 | { |
3592 | int i; | |
3593 | ||
3594 | /* look for blocks to read/compute, skip this if a compute | |
3595 | * is already in flight, or if the stripe contents are in the | |
3596 | * midst of changing due to a write | |
3597 | */ | |
3598 | if (!test_bit(STRIPE_COMPUTE_RUN, &sh->state) && !sh->check_state && | |
07e83364 SL |
3599 | !sh->reconstruct_state) { |
3600 | ||
3601 | /* | |
3602 | * For degraded stripe with data in journal, do not handle | |
3603 | * read requests yet, instead, flush the stripe to raid | |
3604 | * disks first, this avoids handling complex rmw of write | |
3605 | * back cache (prexor with orig_page, and then xor with | |
3606 | * page) in the read path | |
3607 | */ | |
3608 | if (s->injournal && s->failed) { | |
3609 | if (test_bit(STRIPE_R5C_CACHING, &sh->state)) | |
3610 | r5c_make_stripe_write_out(sh); | |
3611 | goto out; | |
3612 | } | |
3613 | ||
5599becc | 3614 | for (i = disks; i--; ) |
93b3dbce | 3615 | if (fetch_block(sh, s, i, disks)) |
5599becc | 3616 | break; |
07e83364 SL |
3617 | } |
3618 | out: | |
a4456856 DW |
3619 | set_bit(STRIPE_HANDLE, &sh->state); |
3620 | } | |
3621 | ||
787b76fa N |
3622 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
3623 | unsigned long handle_flags); | |
1fe797e6 | 3624 | /* handle_stripe_clean_event |
a4456856 DW |
3625 | * any written block on an uptodate or failed drive can be returned. |
3626 | * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but | |
3627 | * never LOCKED, so we don't need to test 'failed' directly. | |
3628 | */ | |
d1688a6d | 3629 | static void handle_stripe_clean_event(struct r5conf *conf, |
34a6f80e | 3630 | struct stripe_head *sh, int disks, struct bio_list *return_bi) |
a4456856 DW |
3631 | { |
3632 | int i; | |
3633 | struct r5dev *dev; | |
f8dfcffd | 3634 | int discard_pending = 0; |
59fc630b | 3635 | struct stripe_head *head_sh = sh; |
3636 | bool do_endio = false; | |
a4456856 DW |
3637 | |
3638 | for (i = disks; i--; ) | |
3639 | if (sh->dev[i].written) { | |
3640 | dev = &sh->dev[i]; | |
3641 | if (!test_bit(R5_LOCKED, &dev->flags) && | |
9e444768 | 3642 | (test_bit(R5_UPTODATE, &dev->flags) || |
d592a996 SL |
3643 | test_bit(R5_Discard, &dev->flags) || |
3644 | test_bit(R5_SkipCopy, &dev->flags))) { | |
a4456856 DW |
3645 | /* We can return any write requests */ |
3646 | struct bio *wbi, *wbi2; | |
45b4233c | 3647 | pr_debug("Return write for disc %d\n", i); |
ca64cae9 N |
3648 | if (test_and_clear_bit(R5_Discard, &dev->flags)) |
3649 | clear_bit(R5_UPTODATE, &dev->flags); | |
d592a996 SL |
3650 | if (test_and_clear_bit(R5_SkipCopy, &dev->flags)) { |
3651 | WARN_ON(test_bit(R5_UPTODATE, &dev->flags)); | |
d592a996 | 3652 | } |
59fc630b | 3653 | do_endio = true; |
3654 | ||
3655 | returnbi: | |
3656 | dev->page = dev->orig_page; | |
a4456856 DW |
3657 | wbi = dev->written; |
3658 | dev->written = NULL; | |
4f024f37 | 3659 | while (wbi && wbi->bi_iter.bi_sector < |
a4456856 DW |
3660 | dev->sector + STRIPE_SECTORS) { |
3661 | wbi2 = r5_next_bio(wbi, dev->sector); | |
e7836bd6 | 3662 | if (!raid5_dec_bi_active_stripes(wbi)) { |
a4456856 | 3663 | md_write_end(conf->mddev); |
34a6f80e | 3664 | bio_list_add(return_bi, wbi); |
a4456856 DW |
3665 | } |
3666 | wbi = wbi2; | |
3667 | } | |
7eaf7e8e SL |
3668 | bitmap_endwrite(conf->mddev->bitmap, sh->sector, |
3669 | STRIPE_SECTORS, | |
a4456856 | 3670 | !test_bit(STRIPE_DEGRADED, &sh->state), |
7eaf7e8e | 3671 | 0); |
59fc630b | 3672 | if (head_sh->batch_head) { |
3673 | sh = list_first_entry(&sh->batch_list, | |
3674 | struct stripe_head, | |
3675 | batch_list); | |
3676 | if (sh != head_sh) { | |
3677 | dev = &sh->dev[i]; | |
3678 | goto returnbi; | |
3679 | } | |
3680 | } | |
3681 | sh = head_sh; | |
3682 | dev = &sh->dev[i]; | |
f8dfcffd N |
3683 | } else if (test_bit(R5_Discard, &dev->flags)) |
3684 | discard_pending = 1; | |
3685 | } | |
f6bed0ef | 3686 | |
0576b1c6 SL |
3687 | r5l_stripe_write_finished(sh); |
3688 | ||
f8dfcffd N |
3689 | if (!discard_pending && |
3690 | test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)) { | |
b8a9d66d | 3691 | int hash; |
f8dfcffd N |
3692 | clear_bit(R5_Discard, &sh->dev[sh->pd_idx].flags); |
3693 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); | |
3694 | if (sh->qd_idx >= 0) { | |
3695 | clear_bit(R5_Discard, &sh->dev[sh->qd_idx].flags); | |
3696 | clear_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags); | |
3697 | } | |
3698 | /* now that discard is done we can proceed with any sync */ | |
3699 | clear_bit(STRIPE_DISCARD, &sh->state); | |
d47648fc SL |
3700 | /* |
3701 | * SCSI discard will change some bio fields and the stripe has | |
3702 | * no updated data, so remove it from hash list and the stripe | |
3703 | * will be reinitialized | |
3704 | */ | |
59fc630b | 3705 | unhash: |
b8a9d66d RG |
3706 | hash = sh->hash_lock_index; |
3707 | spin_lock_irq(conf->hash_locks + hash); | |
d47648fc | 3708 | remove_hash(sh); |
b8a9d66d | 3709 | spin_unlock_irq(conf->hash_locks + hash); |
59fc630b | 3710 | if (head_sh->batch_head) { |
3711 | sh = list_first_entry(&sh->batch_list, | |
3712 | struct stripe_head, batch_list); | |
3713 | if (sh != head_sh) | |
3714 | goto unhash; | |
3715 | } | |
59fc630b | 3716 | sh = head_sh; |
3717 | ||
f8dfcffd N |
3718 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state)) |
3719 | set_bit(STRIPE_HANDLE, &sh->state); | |
3720 | ||
3721 | } | |
8b3e6cdc DW |
3722 | |
3723 | if (test_and_clear_bit(STRIPE_FULL_WRITE, &sh->state)) | |
3724 | if (atomic_dec_and_test(&conf->pending_full_writes)) | |
3725 | md_wakeup_thread(conf->mddev->thread); | |
59fc630b | 3726 | |
787b76fa N |
3727 | if (head_sh->batch_head && do_endio) |
3728 | break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS); | |
a4456856 DW |
3729 | } |
3730 | ||
86aa1397 SL |
3731 | /* |
3732 | * For RMW in write back cache, we need extra page in prexor to store the | |
3733 | * old data. This page is stored in dev->orig_page. | |
3734 | * | |
3735 | * This function checks whether we have data for prexor. The exact logic | |
3736 | * is: | |
3737 | * R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE) | |
3738 | */ | |
3739 | static inline bool uptodate_for_rmw(struct r5dev *dev) | |
3740 | { | |
3741 | return (test_bit(R5_UPTODATE, &dev->flags)) && | |
3742 | (!test_bit(R5_InJournal, &dev->flags) || | |
3743 | test_bit(R5_OrigPageUPTDODATE, &dev->flags)); | |
3744 | } | |
3745 | ||
d7bd398e SL |
3746 | static int handle_stripe_dirtying(struct r5conf *conf, |
3747 | struct stripe_head *sh, | |
3748 | struct stripe_head_state *s, | |
3749 | int disks) | |
a4456856 DW |
3750 | { |
3751 | int rmw = 0, rcw = 0, i; | |
a7854487 AL |
3752 | sector_t recovery_cp = conf->mddev->recovery_cp; |
3753 | ||
584acdd4 | 3754 | /* Check whether resync is now happening or should start. |
a7854487 AL |
3755 | * If yes, then the array is dirty (after unclean shutdown or |
3756 | * initial creation), so parity in some stripes might be inconsistent. | |
3757 | * In this case, we need to always do reconstruct-write, to ensure | |
3758 | * that in case of drive failure or read-error correction, we | |
3759 | * generate correct data from the parity. | |
3760 | */ | |
584acdd4 | 3761 | if (conf->rmw_level == PARITY_DISABLE_RMW || |
26ac1073 N |
3762 | (recovery_cp < MaxSector && sh->sector >= recovery_cp && |
3763 | s->failed == 0)) { | |
a7854487 | 3764 | /* Calculate the real rcw later - for now make it |
c8ac1803 N |
3765 | * look like rcw is cheaper |
3766 | */ | |
3767 | rcw = 1; rmw = 2; | |
584acdd4 MS |
3768 | pr_debug("force RCW rmw_level=%u, recovery_cp=%llu sh->sector=%llu\n", |
3769 | conf->rmw_level, (unsigned long long)recovery_cp, | |
a7854487 | 3770 | (unsigned long long)sh->sector); |
c8ac1803 | 3771 | } else for (i = disks; i--; ) { |
a4456856 DW |
3772 | /* would I have to read this buffer for read_modify_write */ |
3773 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3774 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
07e83364 | 3775 | i == sh->pd_idx || i == sh->qd_idx || |
1e6d690b | 3776 | test_bit(R5_InJournal, &dev->flags)) && |
a4456856 | 3777 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3778 | !(uptodate_for_rmw(dev) || |
f38e1219 | 3779 | test_bit(R5_Wantcompute, &dev->flags))) { |
a4456856 DW |
3780 | if (test_bit(R5_Insync, &dev->flags)) |
3781 | rmw++; | |
3782 | else | |
3783 | rmw += 2*disks; /* cannot read it */ | |
3784 | } | |
3785 | /* Would I have to read this buffer for reconstruct_write */ | |
584acdd4 MS |
3786 | if (!test_bit(R5_OVERWRITE, &dev->flags) && |
3787 | i != sh->pd_idx && i != sh->qd_idx && | |
a4456856 | 3788 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3789 | !(test_bit(R5_UPTODATE, &dev->flags) || |
1e6d690b | 3790 | test_bit(R5_Wantcompute, &dev->flags))) { |
67f45548 N |
3791 | if (test_bit(R5_Insync, &dev->flags)) |
3792 | rcw++; | |
a4456856 DW |
3793 | else |
3794 | rcw += 2*disks; | |
3795 | } | |
3796 | } | |
1e6d690b | 3797 | |
39b99586 SL |
3798 | pr_debug("for sector %llu state 0x%lx, rmw=%d rcw=%d\n", |
3799 | (unsigned long long)sh->sector, sh->state, rmw, rcw); | |
a4456856 | 3800 | set_bit(STRIPE_HANDLE, &sh->state); |
41257580 | 3801 | if ((rmw < rcw || (rmw == rcw && conf->rmw_level == PARITY_PREFER_RMW)) && rmw > 0) { |
a4456856 | 3802 | /* prefer read-modify-write, but need to get some data */ |
e3620a3a JB |
3803 | if (conf->mddev->queue) |
3804 | blk_add_trace_msg(conf->mddev->queue, | |
3805 | "raid5 rmw %llu %d", | |
3806 | (unsigned long long)sh->sector, rmw); | |
a4456856 DW |
3807 | for (i = disks; i--; ) { |
3808 | struct r5dev *dev = &sh->dev[i]; | |
1e6d690b SL |
3809 | if (test_bit(R5_InJournal, &dev->flags) && |
3810 | dev->page == dev->orig_page && | |
3811 | !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) { | |
3812 | /* alloc page for prexor */ | |
d7bd398e SL |
3813 | struct page *p = alloc_page(GFP_NOIO); |
3814 | ||
3815 | if (p) { | |
3816 | dev->orig_page = p; | |
3817 | continue; | |
3818 | } | |
3819 | ||
3820 | /* | |
3821 | * alloc_page() failed, try use | |
3822 | * disk_info->extra_page | |
3823 | */ | |
3824 | if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE, | |
3825 | &conf->cache_state)) { | |
3826 | r5c_use_extra_page(sh); | |
3827 | break; | |
3828 | } | |
1e6d690b | 3829 | |
d7bd398e SL |
3830 | /* extra_page in use, add to delayed_list */ |
3831 | set_bit(STRIPE_DELAYED, &sh->state); | |
3832 | s->waiting_extra_page = 1; | |
3833 | return -EAGAIN; | |
1e6d690b | 3834 | } |
d7bd398e | 3835 | } |
1e6d690b | 3836 | |
d7bd398e SL |
3837 | for (i = disks; i--; ) { |
3838 | struct r5dev *dev = &sh->dev[i]; | |
39b99586 | 3839 | if (((dev->towrite && !delay_towrite(conf, dev, s)) || |
1e6d690b SL |
3840 | i == sh->pd_idx || i == sh->qd_idx || |
3841 | test_bit(R5_InJournal, &dev->flags)) && | |
a4456856 | 3842 | !test_bit(R5_LOCKED, &dev->flags) && |
86aa1397 | 3843 | !(uptodate_for_rmw(dev) || |
1e6d690b | 3844 | test_bit(R5_Wantcompute, &dev->flags)) && |
a4456856 | 3845 | test_bit(R5_Insync, &dev->flags)) { |
67f45548 N |
3846 | if (test_bit(STRIPE_PREREAD_ACTIVE, |
3847 | &sh->state)) { | |
3848 | pr_debug("Read_old block %d for r-m-w\n", | |
3849 | i); | |
a4456856 DW |
3850 | set_bit(R5_LOCKED, &dev->flags); |
3851 | set_bit(R5_Wantread, &dev->flags); | |
3852 | s->locked++; | |
3853 | } else { | |
3854 | set_bit(STRIPE_DELAYED, &sh->state); | |
3855 | set_bit(STRIPE_HANDLE, &sh->state); | |
3856 | } | |
3857 | } | |
3858 | } | |
a9add5d9 | 3859 | } |
41257580 | 3860 | if ((rcw < rmw || (rcw == rmw && conf->rmw_level != PARITY_PREFER_RMW)) && rcw > 0) { |
a4456856 | 3861 | /* want reconstruct write, but need to get some data */ |
a9add5d9 | 3862 | int qread =0; |
c8ac1803 | 3863 | rcw = 0; |
a4456856 DW |
3864 | for (i = disks; i--; ) { |
3865 | struct r5dev *dev = &sh->dev[i]; | |
3866 | if (!test_bit(R5_OVERWRITE, &dev->flags) && | |
c8ac1803 | 3867 | i != sh->pd_idx && i != sh->qd_idx && |
a4456856 | 3868 | !test_bit(R5_LOCKED, &dev->flags) && |
f38e1219 | 3869 | !(test_bit(R5_UPTODATE, &dev->flags) || |
c8ac1803 N |
3870 | test_bit(R5_Wantcompute, &dev->flags))) { |
3871 | rcw++; | |
67f45548 N |
3872 | if (test_bit(R5_Insync, &dev->flags) && |
3873 | test_bit(STRIPE_PREREAD_ACTIVE, | |
3874 | &sh->state)) { | |
45b4233c | 3875 | pr_debug("Read_old block " |
a4456856 DW |
3876 | "%d for Reconstruct\n", i); |
3877 | set_bit(R5_LOCKED, &dev->flags); | |
3878 | set_bit(R5_Wantread, &dev->flags); | |
3879 | s->locked++; | |
a9add5d9 | 3880 | qread++; |
a4456856 DW |
3881 | } else { |
3882 | set_bit(STRIPE_DELAYED, &sh->state); | |
3883 | set_bit(STRIPE_HANDLE, &sh->state); | |
3884 | } | |
3885 | } | |
3886 | } | |
e3620a3a | 3887 | if (rcw && conf->mddev->queue) |
a9add5d9 N |
3888 | blk_add_trace_msg(conf->mddev->queue, "raid5 rcw %llu %d %d %d", |
3889 | (unsigned long long)sh->sector, | |
3890 | rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); | |
c8ac1803 | 3891 | } |
b1b02fe9 N |
3892 | |
3893 | if (rcw > disks && rmw > disks && | |
3894 | !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
3895 | set_bit(STRIPE_DELAYED, &sh->state); | |
3896 | ||
a4456856 DW |
3897 | /* now if nothing is locked, and if we have enough data, |
3898 | * we can start a write request | |
3899 | */ | |
f38e1219 DW |
3900 | /* since handle_stripe can be called at any time we need to handle the |
3901 | * case where a compute block operation has been submitted and then a | |
ac6b53b6 DW |
3902 | * subsequent call wants to start a write request. raid_run_ops only |
3903 | * handles the case where compute block and reconstruct are requested | |
f38e1219 DW |
3904 | * simultaneously. If this is not the case then new writes need to be |
3905 | * held off until the compute completes. | |
3906 | */ | |
976ea8d4 DW |
3907 | if ((s->req_compute || !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) && |
3908 | (s->locked == 0 && (rcw == 0 || rmw == 0) && | |
1e6d690b | 3909 | !test_bit(STRIPE_BIT_DELAY, &sh->state))) |
c0f7bddb | 3910 | schedule_reconstruction(sh, s, rcw == 0, 0); |
d7bd398e | 3911 | return 0; |
a4456856 DW |
3912 | } |
3913 | ||
d1688a6d | 3914 | static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh, |
a4456856 DW |
3915 | struct stripe_head_state *s, int disks) |
3916 | { | |
ecc65c9b | 3917 | struct r5dev *dev = NULL; |
bd2ab670 | 3918 | |
59fc630b | 3919 | BUG_ON(sh->batch_head); |
a4456856 | 3920 | set_bit(STRIPE_HANDLE, &sh->state); |
e89f8962 | 3921 | |
ecc65c9b DW |
3922 | switch (sh->check_state) { |
3923 | case check_state_idle: | |
3924 | /* start a new check operation if there are no failures */ | |
bd2ab670 | 3925 | if (s->failed == 0) { |
bd2ab670 | 3926 | BUG_ON(s->uptodate != disks); |
ecc65c9b DW |
3927 | sh->check_state = check_state_run; |
3928 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
bd2ab670 | 3929 | clear_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags); |
bd2ab670 | 3930 | s->uptodate--; |
ecc65c9b | 3931 | break; |
bd2ab670 | 3932 | } |
f2b3b44d | 3933 | dev = &sh->dev[s->failed_num[0]]; |
ecc65c9b DW |
3934 | /* fall through */ |
3935 | case check_state_compute_result: | |
3936 | sh->check_state = check_state_idle; | |
3937 | if (!dev) | |
3938 | dev = &sh->dev[sh->pd_idx]; | |
3939 | ||
3940 | /* check that a write has not made the stripe insync */ | |
3941 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
3942 | break; | |
c8894419 | 3943 | |
a4456856 | 3944 | /* either failed parity check, or recovery is happening */ |
a4456856 DW |
3945 | BUG_ON(!test_bit(R5_UPTODATE, &dev->flags)); |
3946 | BUG_ON(s->uptodate != disks); | |
3947 | ||
3948 | set_bit(R5_LOCKED, &dev->flags); | |
ecc65c9b | 3949 | s->locked++; |
a4456856 | 3950 | set_bit(R5_Wantwrite, &dev->flags); |
830ea016 | 3951 | |
a4456856 | 3952 | clear_bit(STRIPE_DEGRADED, &sh->state); |
a4456856 | 3953 | set_bit(STRIPE_INSYNC, &sh->state); |
ecc65c9b DW |
3954 | break; |
3955 | case check_state_run: | |
3956 | break; /* we will be called again upon completion */ | |
3957 | case check_state_check_result: | |
3958 | sh->check_state = check_state_idle; | |
3959 | ||
3960 | /* if a failure occurred during the check operation, leave | |
3961 | * STRIPE_INSYNC not set and let the stripe be handled again | |
3962 | */ | |
3963 | if (s->failed) | |
3964 | break; | |
3965 | ||
3966 | /* handle a successful check operation, if parity is correct | |
3967 | * we are done. Otherwise update the mismatch count and repair | |
3968 | * parity if !MD_RECOVERY_CHECK | |
3969 | */ | |
ad283ea4 | 3970 | if ((sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) == 0) |
ecc65c9b DW |
3971 | /* parity is correct (on disc, |
3972 | * not in buffer any more) | |
3973 | */ | |
3974 | set_bit(STRIPE_INSYNC, &sh->state); | |
3975 | else { | |
7f7583d4 | 3976 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
ecc65c9b DW |
3977 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
3978 | /* don't try to repair!! */ | |
3979 | set_bit(STRIPE_INSYNC, &sh->state); | |
3980 | else { | |
3981 | sh->check_state = check_state_compute_run; | |
976ea8d4 | 3982 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); |
ecc65c9b DW |
3983 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); |
3984 | set_bit(R5_Wantcompute, | |
3985 | &sh->dev[sh->pd_idx].flags); | |
3986 | sh->ops.target = sh->pd_idx; | |
ac6b53b6 | 3987 | sh->ops.target2 = -1; |
ecc65c9b DW |
3988 | s->uptodate++; |
3989 | } | |
3990 | } | |
3991 | break; | |
3992 | case check_state_compute_run: | |
3993 | break; | |
3994 | default: | |
cc6167b4 | 3995 | pr_err("%s: unknown check_state: %d sector: %llu\n", |
ecc65c9b DW |
3996 | __func__, sh->check_state, |
3997 | (unsigned long long) sh->sector); | |
3998 | BUG(); | |
a4456856 DW |
3999 | } |
4000 | } | |
4001 | ||
d1688a6d | 4002 | static void handle_parity_checks6(struct r5conf *conf, struct stripe_head *sh, |
36d1c647 | 4003 | struct stripe_head_state *s, |
f2b3b44d | 4004 | int disks) |
a4456856 | 4005 | { |
a4456856 | 4006 | int pd_idx = sh->pd_idx; |
34e04e87 | 4007 | int qd_idx = sh->qd_idx; |
d82dfee0 | 4008 | struct r5dev *dev; |
a4456856 | 4009 | |
59fc630b | 4010 | BUG_ON(sh->batch_head); |
a4456856 DW |
4011 | set_bit(STRIPE_HANDLE, &sh->state); |
4012 | ||
4013 | BUG_ON(s->failed > 2); | |
d82dfee0 | 4014 | |
a4456856 DW |
4015 | /* Want to check and possibly repair P and Q. |
4016 | * However there could be one 'failed' device, in which | |
4017 | * case we can only check one of them, possibly using the | |
4018 | * other to generate missing data | |
4019 | */ | |
4020 | ||
d82dfee0 DW |
4021 | switch (sh->check_state) { |
4022 | case check_state_idle: | |
4023 | /* start a new check operation if there are < 2 failures */ | |
f2b3b44d | 4024 | if (s->failed == s->q_failed) { |
d82dfee0 | 4025 | /* The only possible failed device holds Q, so it |
a4456856 DW |
4026 | * makes sense to check P (If anything else were failed, |
4027 | * we would have used P to recreate it). | |
4028 | */ | |
d82dfee0 | 4029 | sh->check_state = check_state_run; |
a4456856 | 4030 | } |
f2b3b44d | 4031 | if (!s->q_failed && s->failed < 2) { |
d82dfee0 | 4032 | /* Q is not failed, and we didn't use it to generate |
a4456856 DW |
4033 | * anything, so it makes sense to check it |
4034 | */ | |
d82dfee0 DW |
4035 | if (sh->check_state == check_state_run) |
4036 | sh->check_state = check_state_run_pq; | |
4037 | else | |
4038 | sh->check_state = check_state_run_q; | |
a4456856 | 4039 | } |
a4456856 | 4040 | |
d82dfee0 DW |
4041 | /* discard potentially stale zero_sum_result */ |
4042 | sh->ops.zero_sum_result = 0; | |
a4456856 | 4043 | |
d82dfee0 DW |
4044 | if (sh->check_state == check_state_run) { |
4045 | /* async_xor_zero_sum destroys the contents of P */ | |
4046 | clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); | |
4047 | s->uptodate--; | |
a4456856 | 4048 | } |
d82dfee0 DW |
4049 | if (sh->check_state >= check_state_run && |
4050 | sh->check_state <= check_state_run_pq) { | |
4051 | /* async_syndrome_zero_sum preserves P and Q, so | |
4052 | * no need to mark them !uptodate here | |
4053 | */ | |
4054 | set_bit(STRIPE_OP_CHECK, &s->ops_request); | |
4055 | break; | |
a4456856 DW |
4056 | } |
4057 | ||
d82dfee0 DW |
4058 | /* we have 2-disk failure */ |
4059 | BUG_ON(s->failed != 2); | |
4060 | /* fall through */ | |
4061 | case check_state_compute_result: | |
4062 | sh->check_state = check_state_idle; | |
a4456856 | 4063 | |
d82dfee0 DW |
4064 | /* check that a write has not made the stripe insync */ |
4065 | if (test_bit(STRIPE_INSYNC, &sh->state)) | |
4066 | break; | |
a4456856 DW |
4067 | |
4068 | /* now write out any block on a failed drive, | |
d82dfee0 | 4069 | * or P or Q if they were recomputed |
a4456856 | 4070 | */ |
d82dfee0 | 4071 | BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ |
a4456856 | 4072 | if (s->failed == 2) { |
f2b3b44d | 4073 | dev = &sh->dev[s->failed_num[1]]; |
a4456856 DW |
4074 | s->locked++; |
4075 | set_bit(R5_LOCKED, &dev->flags); | |
4076 | set_bit(R5_Wantwrite, &dev->flags); | |
4077 | } | |
4078 | if (s->failed >= 1) { | |
f2b3b44d | 4079 | dev = &sh->dev[s->failed_num[0]]; |
a4456856 DW |
4080 | s->locked++; |
4081 | set_bit(R5_LOCKED, &dev->flags); | |
4082 | set_bit(R5_Wantwrite, &dev->flags); | |
4083 | } | |
d82dfee0 | 4084 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { |
a4456856 DW |
4085 | dev = &sh->dev[pd_idx]; |
4086 | s->locked++; | |
4087 | set_bit(R5_LOCKED, &dev->flags); | |
4088 | set_bit(R5_Wantwrite, &dev->flags); | |
4089 | } | |
d82dfee0 | 4090 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { |
a4456856 DW |
4091 | dev = &sh->dev[qd_idx]; |
4092 | s->locked++; | |
4093 | set_bit(R5_LOCKED, &dev->flags); | |
4094 | set_bit(R5_Wantwrite, &dev->flags); | |
4095 | } | |
4096 | clear_bit(STRIPE_DEGRADED, &sh->state); | |
4097 | ||
4098 | set_bit(STRIPE_INSYNC, &sh->state); | |
d82dfee0 DW |
4099 | break; |
4100 | case check_state_run: | |
4101 | case check_state_run_q: | |
4102 | case check_state_run_pq: | |
4103 | break; /* we will be called again upon completion */ | |
4104 | case check_state_check_result: | |
4105 | sh->check_state = check_state_idle; | |
4106 | ||
4107 | /* handle a successful check operation, if parity is correct | |
4108 | * we are done. Otherwise update the mismatch count and repair | |
4109 | * parity if !MD_RECOVERY_CHECK | |
4110 | */ | |
4111 | if (sh->ops.zero_sum_result == 0) { | |
4112 | /* both parities are correct */ | |
4113 | if (!s->failed) | |
4114 | set_bit(STRIPE_INSYNC, &sh->state); | |
4115 | else { | |
4116 | /* in contrast to the raid5 case we can validate | |
4117 | * parity, but still have a failure to write | |
4118 | * back | |
4119 | */ | |
4120 | sh->check_state = check_state_compute_result; | |
4121 | /* Returning at this point means that we may go | |
4122 | * off and bring p and/or q uptodate again so | |
4123 | * we make sure to check zero_sum_result again | |
4124 | * to verify if p or q need writeback | |
4125 | */ | |
4126 | } | |
4127 | } else { | |
7f7583d4 | 4128 | atomic64_add(STRIPE_SECTORS, &conf->mddev->resync_mismatches); |
d82dfee0 DW |
4129 | if (test_bit(MD_RECOVERY_CHECK, &conf->mddev->recovery)) |
4130 | /* don't try to repair!! */ | |
4131 | set_bit(STRIPE_INSYNC, &sh->state); | |
4132 | else { | |
4133 | int *target = &sh->ops.target; | |
4134 | ||
4135 | sh->ops.target = -1; | |
4136 | sh->ops.target2 = -1; | |
4137 | sh->check_state = check_state_compute_run; | |
4138 | set_bit(STRIPE_COMPUTE_RUN, &sh->state); | |
4139 | set_bit(STRIPE_OP_COMPUTE_BLK, &s->ops_request); | |
4140 | if (sh->ops.zero_sum_result & SUM_CHECK_P_RESULT) { | |
4141 | set_bit(R5_Wantcompute, | |
4142 | &sh->dev[pd_idx].flags); | |
4143 | *target = pd_idx; | |
4144 | target = &sh->ops.target2; | |
4145 | s->uptodate++; | |
4146 | } | |
4147 | if (sh->ops.zero_sum_result & SUM_CHECK_Q_RESULT) { | |
4148 | set_bit(R5_Wantcompute, | |
4149 | &sh->dev[qd_idx].flags); | |
4150 | *target = qd_idx; | |
4151 | s->uptodate++; | |
4152 | } | |
4153 | } | |
4154 | } | |
4155 | break; | |
4156 | case check_state_compute_run: | |
4157 | break; | |
4158 | default: | |
cc6167b4 N |
4159 | pr_warn("%s: unknown check_state: %d sector: %llu\n", |
4160 | __func__, sh->check_state, | |
4161 | (unsigned long long) sh->sector); | |
d82dfee0 | 4162 | BUG(); |
a4456856 DW |
4163 | } |
4164 | } | |
4165 | ||
d1688a6d | 4166 | static void handle_stripe_expansion(struct r5conf *conf, struct stripe_head *sh) |
a4456856 DW |
4167 | { |
4168 | int i; | |
4169 | ||
4170 | /* We have read all the blocks in this stripe and now we need to | |
4171 | * copy some of them into a target stripe for expand. | |
4172 | */ | |
f0a50d37 | 4173 | struct dma_async_tx_descriptor *tx = NULL; |
59fc630b | 4174 | BUG_ON(sh->batch_head); |
a4456856 DW |
4175 | clear_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
4176 | for (i = 0; i < sh->disks; i++) | |
34e04e87 | 4177 | if (i != sh->pd_idx && i != sh->qd_idx) { |
911d4ee8 | 4178 | int dd_idx, j; |
a4456856 | 4179 | struct stripe_head *sh2; |
a08abd8c | 4180 | struct async_submit_ctl submit; |
a4456856 | 4181 | |
6d036f7d | 4182 | sector_t bn = raid5_compute_blocknr(sh, i, 1); |
911d4ee8 N |
4183 | sector_t s = raid5_compute_sector(conf, bn, 0, |
4184 | &dd_idx, NULL); | |
6d036f7d | 4185 | sh2 = raid5_get_active_stripe(conf, s, 0, 1, 1); |
a4456856 DW |
4186 | if (sh2 == NULL) |
4187 | /* so far only the early blocks of this stripe | |
4188 | * have been requested. When later blocks | |
4189 | * get requested, we will try again | |
4190 | */ | |
4191 | continue; | |
4192 | if (!test_bit(STRIPE_EXPANDING, &sh2->state) || | |
4193 | test_bit(R5_Expanded, &sh2->dev[dd_idx].flags)) { | |
4194 | /* must have already done this block */ | |
6d036f7d | 4195 | raid5_release_stripe(sh2); |
a4456856 DW |
4196 | continue; |
4197 | } | |
f0a50d37 DW |
4198 | |
4199 | /* place all the copies on one channel */ | |
a08abd8c | 4200 | init_async_submit(&submit, 0, tx, NULL, NULL, NULL); |
f0a50d37 | 4201 | tx = async_memcpy(sh2->dev[dd_idx].page, |
88ba2aa5 | 4202 | sh->dev[i].page, 0, 0, STRIPE_SIZE, |
a08abd8c | 4203 | &submit); |
f0a50d37 | 4204 | |
a4456856 DW |
4205 | set_bit(R5_Expanded, &sh2->dev[dd_idx].flags); |
4206 | set_bit(R5_UPTODATE, &sh2->dev[dd_idx].flags); | |
4207 | for (j = 0; j < conf->raid_disks; j++) | |
4208 | if (j != sh2->pd_idx && | |
86c374ba | 4209 | j != sh2->qd_idx && |
a4456856 DW |
4210 | !test_bit(R5_Expanded, &sh2->dev[j].flags)) |
4211 | break; | |
4212 | if (j == conf->raid_disks) { | |
4213 | set_bit(STRIPE_EXPAND_READY, &sh2->state); | |
4214 | set_bit(STRIPE_HANDLE, &sh2->state); | |
4215 | } | |
6d036f7d | 4216 | raid5_release_stripe(sh2); |
f0a50d37 | 4217 | |
a4456856 | 4218 | } |
a2e08551 | 4219 | /* done submitting copies, wait for them to complete */ |
749586b7 | 4220 | async_tx_quiesce(&tx); |
a4456856 | 4221 | } |
1da177e4 LT |
4222 | |
4223 | /* | |
4224 | * handle_stripe - do things to a stripe. | |
4225 | * | |
9a3e1101 N |
4226 | * We lock the stripe by setting STRIPE_ACTIVE and then examine the |
4227 | * state of various bits to see what needs to be done. | |
1da177e4 | 4228 | * Possible results: |
9a3e1101 N |
4229 | * return some read requests which now have data |
4230 | * return some write requests which are safely on storage | |
1da177e4 LT |
4231 | * schedule a read on some buffers |
4232 | * schedule a write of some buffers | |
4233 | * return confirmation of parity correctness | |
4234 | * | |
1da177e4 | 4235 | */ |
a4456856 | 4236 | |
acfe726b | 4237 | static void analyse_stripe(struct stripe_head *sh, struct stripe_head_state *s) |
1da177e4 | 4238 | { |
d1688a6d | 4239 | struct r5conf *conf = sh->raid_conf; |
f416885e | 4240 | int disks = sh->disks; |
474af965 N |
4241 | struct r5dev *dev; |
4242 | int i; | |
9a3e1101 | 4243 | int do_recovery = 0; |
1da177e4 | 4244 | |
acfe726b N |
4245 | memset(s, 0, sizeof(*s)); |
4246 | ||
dabc4ec6 | 4247 | s->expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state) && !sh->batch_head; |
4248 | s->expanded = test_bit(STRIPE_EXPAND_READY, &sh->state) && !sh->batch_head; | |
acfe726b N |
4249 | s->failed_num[0] = -1; |
4250 | s->failed_num[1] = -1; | |
6e74a9cf | 4251 | s->log_failed = r5l_log_disk_error(conf); |
1da177e4 | 4252 | |
acfe726b | 4253 | /* Now to look around and see what can be done */ |
1da177e4 | 4254 | rcu_read_lock(); |
16a53ecc | 4255 | for (i=disks; i--; ) { |
3cb03002 | 4256 | struct md_rdev *rdev; |
31c176ec N |
4257 | sector_t first_bad; |
4258 | int bad_sectors; | |
4259 | int is_bad = 0; | |
acfe726b | 4260 | |
16a53ecc | 4261 | dev = &sh->dev[i]; |
1da177e4 | 4262 | |
45b4233c | 4263 | pr_debug("check %d: state 0x%lx read %p write %p written %p\n", |
9a3e1101 N |
4264 | i, dev->flags, |
4265 | dev->toread, dev->towrite, dev->written); | |
6c0069c0 YT |
4266 | /* maybe we can reply to a read |
4267 | * | |
4268 | * new wantfill requests are only permitted while | |
4269 | * ops_complete_biofill is guaranteed to be inactive | |
4270 | */ | |
4271 | if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread && | |
4272 | !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) | |
4273 | set_bit(R5_Wantfill, &dev->flags); | |
1da177e4 | 4274 | |
16a53ecc | 4275 | /* now count some things */ |
cc94015a N |
4276 | if (test_bit(R5_LOCKED, &dev->flags)) |
4277 | s->locked++; | |
4278 | if (test_bit(R5_UPTODATE, &dev->flags)) | |
4279 | s->uptodate++; | |
2d6e4ecc | 4280 | if (test_bit(R5_Wantcompute, &dev->flags)) { |
cc94015a N |
4281 | s->compute++; |
4282 | BUG_ON(s->compute > 2); | |
2d6e4ecc | 4283 | } |
1da177e4 | 4284 | |
acfe726b | 4285 | if (test_bit(R5_Wantfill, &dev->flags)) |
cc94015a | 4286 | s->to_fill++; |
acfe726b | 4287 | else if (dev->toread) |
cc94015a | 4288 | s->to_read++; |
16a53ecc | 4289 | if (dev->towrite) { |
cc94015a | 4290 | s->to_write++; |
16a53ecc | 4291 | if (!test_bit(R5_OVERWRITE, &dev->flags)) |
cc94015a | 4292 | s->non_overwrite++; |
16a53ecc | 4293 | } |
a4456856 | 4294 | if (dev->written) |
cc94015a | 4295 | s->written++; |
14a75d3e N |
4296 | /* Prefer to use the replacement for reads, but only |
4297 | * if it is recovered enough and has no bad blocks. | |
4298 | */ | |
4299 | rdev = rcu_dereference(conf->disks[i].replacement); | |
4300 | if (rdev && !test_bit(Faulty, &rdev->flags) && | |
4301 | rdev->recovery_offset >= sh->sector + STRIPE_SECTORS && | |
4302 | !is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4303 | &first_bad, &bad_sectors)) | |
4304 | set_bit(R5_ReadRepl, &dev->flags); | |
4305 | else { | |
e6030cb0 | 4306 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
9a3e1101 | 4307 | set_bit(R5_NeedReplace, &dev->flags); |
e6030cb0 N |
4308 | else |
4309 | clear_bit(R5_NeedReplace, &dev->flags); | |
14a75d3e N |
4310 | rdev = rcu_dereference(conf->disks[i].rdev); |
4311 | clear_bit(R5_ReadRepl, &dev->flags); | |
4312 | } | |
9283d8c5 N |
4313 | if (rdev && test_bit(Faulty, &rdev->flags)) |
4314 | rdev = NULL; | |
31c176ec N |
4315 | if (rdev) { |
4316 | is_bad = is_badblock(rdev, sh->sector, STRIPE_SECTORS, | |
4317 | &first_bad, &bad_sectors); | |
4318 | if (s->blocked_rdev == NULL | |
4319 | && (test_bit(Blocked, &rdev->flags) | |
4320 | || is_bad < 0)) { | |
4321 | if (is_bad < 0) | |
4322 | set_bit(BlockedBadBlocks, | |
4323 | &rdev->flags); | |
4324 | s->blocked_rdev = rdev; | |
4325 | atomic_inc(&rdev->nr_pending); | |
4326 | } | |
6bfe0b49 | 4327 | } |
415e72d0 N |
4328 | clear_bit(R5_Insync, &dev->flags); |
4329 | if (!rdev) | |
4330 | /* Not in-sync */; | |
31c176ec N |
4331 | else if (is_bad) { |
4332 | /* also not in-sync */ | |
18b9837e N |
4333 | if (!test_bit(WriteErrorSeen, &rdev->flags) && |
4334 | test_bit(R5_UPTODATE, &dev->flags)) { | |
31c176ec N |
4335 | /* treat as in-sync, but with a read error |
4336 | * which we can now try to correct | |
4337 | */ | |
4338 | set_bit(R5_Insync, &dev->flags); | |
4339 | set_bit(R5_ReadError, &dev->flags); | |
4340 | } | |
4341 | } else if (test_bit(In_sync, &rdev->flags)) | |
415e72d0 | 4342 | set_bit(R5_Insync, &dev->flags); |
30d7a483 | 4343 | else if (sh->sector + STRIPE_SECTORS <= rdev->recovery_offset) |
415e72d0 | 4344 | /* in sync if before recovery_offset */ |
30d7a483 N |
4345 | set_bit(R5_Insync, &dev->flags); |
4346 | else if (test_bit(R5_UPTODATE, &dev->flags) && | |
4347 | test_bit(R5_Expanded, &dev->flags)) | |
4348 | /* If we've reshaped into here, we assume it is Insync. | |
4349 | * We will shortly update recovery_offset to make | |
4350 | * it official. | |
4351 | */ | |
4352 | set_bit(R5_Insync, &dev->flags); | |
4353 | ||
1cc03eb9 | 4354 | if (test_bit(R5_WriteError, &dev->flags)) { |
14a75d3e N |
4355 | /* This flag does not apply to '.replacement' |
4356 | * only to .rdev, so make sure to check that*/ | |
4357 | struct md_rdev *rdev2 = rcu_dereference( | |
4358 | conf->disks[i].rdev); | |
4359 | if (rdev2 == rdev) | |
4360 | clear_bit(R5_Insync, &dev->flags); | |
4361 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
bc2607f3 | 4362 | s->handle_bad_blocks = 1; |
14a75d3e | 4363 | atomic_inc(&rdev2->nr_pending); |
bc2607f3 N |
4364 | } else |
4365 | clear_bit(R5_WriteError, &dev->flags); | |
4366 | } | |
1cc03eb9 | 4367 | if (test_bit(R5_MadeGood, &dev->flags)) { |
14a75d3e N |
4368 | /* This flag does not apply to '.replacement' |
4369 | * only to .rdev, so make sure to check that*/ | |
4370 | struct md_rdev *rdev2 = rcu_dereference( | |
4371 | conf->disks[i].rdev); | |
4372 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
b84db560 | 4373 | s->handle_bad_blocks = 1; |
14a75d3e | 4374 | atomic_inc(&rdev2->nr_pending); |
b84db560 N |
4375 | } else |
4376 | clear_bit(R5_MadeGood, &dev->flags); | |
4377 | } | |
977df362 N |
4378 | if (test_bit(R5_MadeGoodRepl, &dev->flags)) { |
4379 | struct md_rdev *rdev2 = rcu_dereference( | |
4380 | conf->disks[i].replacement); | |
4381 | if (rdev2 && !test_bit(Faulty, &rdev2->flags)) { | |
4382 | s->handle_bad_blocks = 1; | |
4383 | atomic_inc(&rdev2->nr_pending); | |
4384 | } else | |
4385 | clear_bit(R5_MadeGoodRepl, &dev->flags); | |
4386 | } | |
415e72d0 | 4387 | if (!test_bit(R5_Insync, &dev->flags)) { |
16a53ecc N |
4388 | /* The ReadError flag will just be confusing now */ |
4389 | clear_bit(R5_ReadError, &dev->flags); | |
4390 | clear_bit(R5_ReWrite, &dev->flags); | |
1da177e4 | 4391 | } |
415e72d0 N |
4392 | if (test_bit(R5_ReadError, &dev->flags)) |
4393 | clear_bit(R5_Insync, &dev->flags); | |
4394 | if (!test_bit(R5_Insync, &dev->flags)) { | |
cc94015a N |
4395 | if (s->failed < 2) |
4396 | s->failed_num[s->failed] = i; | |
4397 | s->failed++; | |
9a3e1101 N |
4398 | if (rdev && !test_bit(Faulty, &rdev->flags)) |
4399 | do_recovery = 1; | |
415e72d0 | 4400 | } |
2ded3703 SL |
4401 | |
4402 | if (test_bit(R5_InJournal, &dev->flags)) | |
4403 | s->injournal++; | |
1e6d690b SL |
4404 | if (test_bit(R5_InJournal, &dev->flags) && dev->written) |
4405 | s->just_cached++; | |
1da177e4 | 4406 | } |
9a3e1101 N |
4407 | if (test_bit(STRIPE_SYNCING, &sh->state)) { |
4408 | /* If there is a failed device being replaced, | |
4409 | * we must be recovering. | |
4410 | * else if we are after recovery_cp, we must be syncing | |
c6d2e084 | 4411 | * else if MD_RECOVERY_REQUESTED is set, we also are syncing. |
9a3e1101 N |
4412 | * else we can only be replacing |
4413 | * sync and recovery both need to read all devices, and so | |
4414 | * use the same flag. | |
4415 | */ | |
4416 | if (do_recovery || | |
c6d2e084 | 4417 | sh->sector >= conf->mddev->recovery_cp || |
4418 | test_bit(MD_RECOVERY_REQUESTED, &(conf->mddev->recovery))) | |
9a3e1101 N |
4419 | s->syncing = 1; |
4420 | else | |
4421 | s->replacing = 1; | |
4422 | } | |
1da177e4 | 4423 | rcu_read_unlock(); |
cc94015a N |
4424 | } |
4425 | ||
59fc630b | 4426 | static int clear_batch_ready(struct stripe_head *sh) |
4427 | { | |
b15a9dbd N |
4428 | /* Return '1' if this is a member of batch, or |
4429 | * '0' if it is a lone stripe or a head which can now be | |
4430 | * handled. | |
4431 | */ | |
59fc630b | 4432 | struct stripe_head *tmp; |
4433 | if (!test_and_clear_bit(STRIPE_BATCH_READY, &sh->state)) | |
b15a9dbd | 4434 | return (sh->batch_head && sh->batch_head != sh); |
59fc630b | 4435 | spin_lock(&sh->stripe_lock); |
4436 | if (!sh->batch_head) { | |
4437 | spin_unlock(&sh->stripe_lock); | |
4438 | return 0; | |
4439 | } | |
4440 | ||
4441 | /* | |
4442 | * this stripe could be added to a batch list before we check | |
4443 | * BATCH_READY, skips it | |
4444 | */ | |
4445 | if (sh->batch_head != sh) { | |
4446 | spin_unlock(&sh->stripe_lock); | |
4447 | return 1; | |
4448 | } | |
4449 | spin_lock(&sh->batch_lock); | |
4450 | list_for_each_entry(tmp, &sh->batch_list, batch_list) | |
4451 | clear_bit(STRIPE_BATCH_READY, &tmp->state); | |
4452 | spin_unlock(&sh->batch_lock); | |
4453 | spin_unlock(&sh->stripe_lock); | |
4454 | ||
4455 | /* | |
4456 | * BATCH_READY is cleared, no new stripes can be added. | |
4457 | * batch_list can be accessed without lock | |
4458 | */ | |
4459 | return 0; | |
4460 | } | |
4461 | ||
3960ce79 N |
4462 | static void break_stripe_batch_list(struct stripe_head *head_sh, |
4463 | unsigned long handle_flags) | |
72ac7330 | 4464 | { |
4e3d62ff | 4465 | struct stripe_head *sh, *next; |
72ac7330 | 4466 | int i; |
fb642b92 | 4467 | int do_wakeup = 0; |
72ac7330 | 4468 | |
bb27051f N |
4469 | list_for_each_entry_safe(sh, next, &head_sh->batch_list, batch_list) { |
4470 | ||
72ac7330 | 4471 | list_del_init(&sh->batch_list); |
4472 | ||
fb3229d5 | 4473 | WARN_ONCE(sh->state & ((1 << STRIPE_ACTIVE) | |
1b956f7a N |
4474 | (1 << STRIPE_SYNCING) | |
4475 | (1 << STRIPE_REPLACED) | | |
1b956f7a N |
4476 | (1 << STRIPE_DELAYED) | |
4477 | (1 << STRIPE_BIT_DELAY) | | |
4478 | (1 << STRIPE_FULL_WRITE) | | |
4479 | (1 << STRIPE_BIOFILL_RUN) | | |
4480 | (1 << STRIPE_COMPUTE_RUN) | | |
4481 | (1 << STRIPE_OPS_REQ_PENDING) | | |
4482 | (1 << STRIPE_DISCARD) | | |
4483 | (1 << STRIPE_BATCH_READY) | | |
4484 | (1 << STRIPE_BATCH_ERR) | | |
fb3229d5 SL |
4485 | (1 << STRIPE_BITMAP_PENDING)), |
4486 | "stripe state: %lx\n", sh->state); | |
4487 | WARN_ONCE(head_sh->state & ((1 << STRIPE_DISCARD) | | |
4488 | (1 << STRIPE_REPLACED)), | |
4489 | "head stripe state: %lx\n", head_sh->state); | |
1b956f7a N |
4490 | |
4491 | set_mask_bits(&sh->state, ~(STRIPE_EXPAND_SYNC_FLAGS | | |
550da24f | 4492 | (1 << STRIPE_PREREAD_ACTIVE) | |
1b956f7a N |
4493 | (1 << STRIPE_DEGRADED)), |
4494 | head_sh->state & (1 << STRIPE_INSYNC)); | |
4495 | ||
72ac7330 | 4496 | sh->check_state = head_sh->check_state; |
4497 | sh->reconstruct_state = head_sh->reconstruct_state; | |
fb642b92 N |
4498 | for (i = 0; i < sh->disks; i++) { |
4499 | if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) | |
4500 | do_wakeup = 1; | |
72ac7330 | 4501 | sh->dev[i].flags = head_sh->dev[i].flags & |
4502 | (~((1 << R5_WriteError) | (1 << R5_Overlap))); | |
fb642b92 | 4503 | } |
72ac7330 | 4504 | spin_lock_irq(&sh->stripe_lock); |
4505 | sh->batch_head = NULL; | |
4506 | spin_unlock_irq(&sh->stripe_lock); | |
3960ce79 N |
4507 | if (handle_flags == 0 || |
4508 | sh->state & handle_flags) | |
4509 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 4510 | raid5_release_stripe(sh); |
72ac7330 | 4511 | } |
fb642b92 N |
4512 | spin_lock_irq(&head_sh->stripe_lock); |
4513 | head_sh->batch_head = NULL; | |
4514 | spin_unlock_irq(&head_sh->stripe_lock); | |
4515 | for (i = 0; i < head_sh->disks; i++) | |
4516 | if (test_and_clear_bit(R5_Overlap, &head_sh->dev[i].flags)) | |
4517 | do_wakeup = 1; | |
3960ce79 N |
4518 | if (head_sh->state & handle_flags) |
4519 | set_bit(STRIPE_HANDLE, &head_sh->state); | |
fb642b92 N |
4520 | |
4521 | if (do_wakeup) | |
4522 | wake_up(&head_sh->raid_conf->wait_for_overlap); | |
72ac7330 | 4523 | } |
4524 | ||
cc94015a N |
4525 | static void handle_stripe(struct stripe_head *sh) |
4526 | { | |
4527 | struct stripe_head_state s; | |
d1688a6d | 4528 | struct r5conf *conf = sh->raid_conf; |
3687c061 | 4529 | int i; |
84789554 N |
4530 | int prexor; |
4531 | int disks = sh->disks; | |
474af965 | 4532 | struct r5dev *pdev, *qdev; |
cc94015a N |
4533 | |
4534 | clear_bit(STRIPE_HANDLE, &sh->state); | |
257a4b42 | 4535 | if (test_and_set_bit_lock(STRIPE_ACTIVE, &sh->state)) { |
cc94015a N |
4536 | /* already being handled, ensure it gets handled |
4537 | * again when current action finishes */ | |
4538 | set_bit(STRIPE_HANDLE, &sh->state); | |
4539 | return; | |
4540 | } | |
4541 | ||
59fc630b | 4542 | if (clear_batch_ready(sh) ) { |
4543 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); | |
4544 | return; | |
4545 | } | |
4546 | ||
4e3d62ff | 4547 | if (test_and_clear_bit(STRIPE_BATCH_ERR, &sh->state)) |
3960ce79 | 4548 | break_stripe_batch_list(sh, 0); |
72ac7330 | 4549 | |
dabc4ec6 | 4550 | if (test_bit(STRIPE_SYNC_REQUESTED, &sh->state) && !sh->batch_head) { |
f8dfcffd N |
4551 | spin_lock(&sh->stripe_lock); |
4552 | /* Cannot process 'sync' concurrently with 'discard' */ | |
4553 | if (!test_bit(STRIPE_DISCARD, &sh->state) && | |
4554 | test_and_clear_bit(STRIPE_SYNC_REQUESTED, &sh->state)) { | |
4555 | set_bit(STRIPE_SYNCING, &sh->state); | |
4556 | clear_bit(STRIPE_INSYNC, &sh->state); | |
f94c0b66 | 4557 | clear_bit(STRIPE_REPLACED, &sh->state); |
f8dfcffd N |
4558 | } |
4559 | spin_unlock(&sh->stripe_lock); | |
cc94015a N |
4560 | } |
4561 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4562 | ||
4563 | pr_debug("handling stripe %llu, state=%#lx cnt=%d, " | |
4564 | "pd_idx=%d, qd_idx=%d\n, check:%d, reconstruct:%d\n", | |
4565 | (unsigned long long)sh->sector, sh->state, | |
4566 | atomic_read(&sh->count), sh->pd_idx, sh->qd_idx, | |
4567 | sh->check_state, sh->reconstruct_state); | |
3687c061 | 4568 | |
acfe726b | 4569 | analyse_stripe(sh, &s); |
c5a31000 | 4570 | |
b70abcb2 SL |
4571 | if (test_bit(STRIPE_LOG_TRAPPED, &sh->state)) |
4572 | goto finish; | |
4573 | ||
bc2607f3 N |
4574 | if (s.handle_bad_blocks) { |
4575 | set_bit(STRIPE_HANDLE, &sh->state); | |
4576 | goto finish; | |
4577 | } | |
4578 | ||
474af965 N |
4579 | if (unlikely(s.blocked_rdev)) { |
4580 | if (s.syncing || s.expanding || s.expanded || | |
9a3e1101 | 4581 | s.replacing || s.to_write || s.written) { |
474af965 N |
4582 | set_bit(STRIPE_HANDLE, &sh->state); |
4583 | goto finish; | |
4584 | } | |
4585 | /* There is nothing for the blocked_rdev to block */ | |
4586 | rdev_dec_pending(s.blocked_rdev, conf->mddev); | |
4587 | s.blocked_rdev = NULL; | |
4588 | } | |
4589 | ||
4590 | if (s.to_fill && !test_bit(STRIPE_BIOFILL_RUN, &sh->state)) { | |
4591 | set_bit(STRIPE_OP_BIOFILL, &s.ops_request); | |
4592 | set_bit(STRIPE_BIOFILL_RUN, &sh->state); | |
4593 | } | |
4594 | ||
4595 | pr_debug("locked=%d uptodate=%d to_read=%d" | |
4596 | " to_write=%d failed=%d failed_num=%d,%d\n", | |
4597 | s.locked, s.uptodate, s.to_read, s.to_write, s.failed, | |
4598 | s.failed_num[0], s.failed_num[1]); | |
4599 | /* check if the array has lost more than max_degraded devices and, | |
4600 | * if so, some requests might need to be failed. | |
4601 | */ | |
6e74a9cf | 4602 | if (s.failed > conf->max_degraded || s.log_failed) { |
9a3f530f N |
4603 | sh->check_state = 0; |
4604 | sh->reconstruct_state = 0; | |
626f2092 | 4605 | break_stripe_batch_list(sh, 0); |
9a3f530f N |
4606 | if (s.to_read+s.to_write+s.written) |
4607 | handle_failed_stripe(conf, sh, &s, disks, &s.return_bi); | |
9a3e1101 | 4608 | if (s.syncing + s.replacing) |
9a3f530f N |
4609 | handle_failed_sync(conf, sh, &s); |
4610 | } | |
474af965 | 4611 | |
84789554 N |
4612 | /* Now we check to see if any write operations have recently |
4613 | * completed | |
4614 | */ | |
4615 | prexor = 0; | |
4616 | if (sh->reconstruct_state == reconstruct_state_prexor_drain_result) | |
4617 | prexor = 1; | |
4618 | if (sh->reconstruct_state == reconstruct_state_drain_result || | |
4619 | sh->reconstruct_state == reconstruct_state_prexor_drain_result) { | |
4620 | sh->reconstruct_state = reconstruct_state_idle; | |
4621 | ||
4622 | /* All the 'written' buffers and the parity block are ready to | |
4623 | * be written back to disk | |
4624 | */ | |
9e444768 SL |
4625 | BUG_ON(!test_bit(R5_UPTODATE, &sh->dev[sh->pd_idx].flags) && |
4626 | !test_bit(R5_Discard, &sh->dev[sh->pd_idx].flags)); | |
84789554 | 4627 | BUG_ON(sh->qd_idx >= 0 && |
9e444768 SL |
4628 | !test_bit(R5_UPTODATE, &sh->dev[sh->qd_idx].flags) && |
4629 | !test_bit(R5_Discard, &sh->dev[sh->qd_idx].flags)); | |
84789554 N |
4630 | for (i = disks; i--; ) { |
4631 | struct r5dev *dev = &sh->dev[i]; | |
4632 | if (test_bit(R5_LOCKED, &dev->flags) && | |
4633 | (i == sh->pd_idx || i == sh->qd_idx || | |
1e6d690b SL |
4634 | dev->written || test_bit(R5_InJournal, |
4635 | &dev->flags))) { | |
84789554 N |
4636 | pr_debug("Writing block %d\n", i); |
4637 | set_bit(R5_Wantwrite, &dev->flags); | |
4638 | if (prexor) | |
4639 | continue; | |
9c4bdf69 N |
4640 | if (s.failed > 1) |
4641 | continue; | |
84789554 N |
4642 | if (!test_bit(R5_Insync, &dev->flags) || |
4643 | ((i == sh->pd_idx || i == sh->qd_idx) && | |
4644 | s.failed == 0)) | |
4645 | set_bit(STRIPE_INSYNC, &sh->state); | |
4646 | } | |
4647 | } | |
4648 | if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4649 | s.dec_preread_active = 1; | |
4650 | } | |
4651 | ||
ef5b7c69 N |
4652 | /* |
4653 | * might be able to return some write requests if the parity blocks | |
4654 | * are safe, or on a failed drive | |
4655 | */ | |
4656 | pdev = &sh->dev[sh->pd_idx]; | |
4657 | s.p_failed = (s.failed >= 1 && s.failed_num[0] == sh->pd_idx) | |
4658 | || (s.failed >= 2 && s.failed_num[1] == sh->pd_idx); | |
4659 | qdev = &sh->dev[sh->qd_idx]; | |
4660 | s.q_failed = (s.failed >= 1 && s.failed_num[0] == sh->qd_idx) | |
4661 | || (s.failed >= 2 && s.failed_num[1] == sh->qd_idx) | |
4662 | || conf->level < 6; | |
4663 | ||
4664 | if (s.written && | |
4665 | (s.p_failed || ((test_bit(R5_Insync, &pdev->flags) | |
4666 | && !test_bit(R5_LOCKED, &pdev->flags) | |
4667 | && (test_bit(R5_UPTODATE, &pdev->flags) || | |
4668 | test_bit(R5_Discard, &pdev->flags))))) && | |
4669 | (s.q_failed || ((test_bit(R5_Insync, &qdev->flags) | |
4670 | && !test_bit(R5_LOCKED, &qdev->flags) | |
4671 | && (test_bit(R5_UPTODATE, &qdev->flags) || | |
4672 | test_bit(R5_Discard, &qdev->flags)))))) | |
4673 | handle_stripe_clean_event(conf, sh, disks, &s.return_bi); | |
4674 | ||
1e6d690b SL |
4675 | if (s.just_cached) |
4676 | r5c_handle_cached_data_endio(conf, sh, disks, &s.return_bi); | |
4677 | r5l_stripe_write_finished(sh); | |
4678 | ||
ef5b7c69 N |
4679 | /* Now we might consider reading some blocks, either to check/generate |
4680 | * parity, or to satisfy requests | |
4681 | * or to load a block that is being partially written. | |
4682 | */ | |
4683 | if (s.to_read || s.non_overwrite | |
4684 | || (conf->level == 6 && s.to_write && s.failed) | |
4685 | || (s.syncing && (s.uptodate + s.compute < disks)) | |
4686 | || s.replacing | |
4687 | || s.expanding) | |
4688 | handle_stripe_fill(sh, &s, disks); | |
4689 | ||
2ded3703 SL |
4690 | /* |
4691 | * When the stripe finishes full journal write cycle (write to journal | |
4692 | * and raid disk), this is the clean up procedure so it is ready for | |
4693 | * next operation. | |
4694 | */ | |
4695 | r5c_finish_stripe_write_out(conf, sh, &s); | |
4696 | ||
4697 | /* | |
4698 | * Now to consider new write requests, cache write back and what else, | |
4699 | * if anything should be read. We do not handle new writes when: | |
84789554 N |
4700 | * 1/ A 'write' operation (copy+xor) is already in flight. |
4701 | * 2/ A 'check' operation is in flight, as it may clobber the parity | |
4702 | * block. | |
2ded3703 | 4703 | * 3/ A r5c cache log write is in flight. |
84789554 | 4704 | */ |
2ded3703 SL |
4705 | |
4706 | if (!sh->reconstruct_state && !sh->check_state && !sh->log_io) { | |
4707 | if (!r5c_is_writeback(conf->log)) { | |
4708 | if (s.to_write) | |
4709 | handle_stripe_dirtying(conf, sh, &s, disks); | |
4710 | } else { /* write back cache */ | |
4711 | int ret = 0; | |
4712 | ||
4713 | /* First, try handle writes in caching phase */ | |
4714 | if (s.to_write) | |
4715 | ret = r5c_try_caching_write(conf, sh, &s, | |
4716 | disks); | |
4717 | /* | |
4718 | * If caching phase failed: ret == -EAGAIN | |
4719 | * OR | |
4720 | * stripe under reclaim: !caching && injournal | |
4721 | * | |
4722 | * fall back to handle_stripe_dirtying() | |
4723 | */ | |
4724 | if (ret == -EAGAIN || | |
4725 | /* stripe under reclaim: !caching && injournal */ | |
4726 | (!test_bit(STRIPE_R5C_CACHING, &sh->state) && | |
d7bd398e SL |
4727 | s.injournal > 0)) { |
4728 | ret = handle_stripe_dirtying(conf, sh, &s, | |
4729 | disks); | |
4730 | if (ret == -EAGAIN) | |
4731 | goto finish; | |
4732 | } | |
2ded3703 SL |
4733 | } |
4734 | } | |
84789554 N |
4735 | |
4736 | /* maybe we need to check and possibly fix the parity for this stripe | |
4737 | * Any reads will already have been scheduled, so we just see if enough | |
4738 | * data is available. The parity check is held off while parity | |
4739 | * dependent operations are in flight. | |
4740 | */ | |
4741 | if (sh->check_state || | |
4742 | (s.syncing && s.locked == 0 && | |
4743 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && | |
4744 | !test_bit(STRIPE_INSYNC, &sh->state))) { | |
4745 | if (conf->level == 6) | |
4746 | handle_parity_checks6(conf, sh, &s, disks); | |
4747 | else | |
4748 | handle_parity_checks5(conf, sh, &s, disks); | |
4749 | } | |
c5a31000 | 4750 | |
f94c0b66 N |
4751 | if ((s.replacing || s.syncing) && s.locked == 0 |
4752 | && !test_bit(STRIPE_COMPUTE_RUN, &sh->state) | |
4753 | && !test_bit(STRIPE_REPLACED, &sh->state)) { | |
9a3e1101 N |
4754 | /* Write out to replacement devices where possible */ |
4755 | for (i = 0; i < conf->raid_disks; i++) | |
f94c0b66 N |
4756 | if (test_bit(R5_NeedReplace, &sh->dev[i].flags)) { |
4757 | WARN_ON(!test_bit(R5_UPTODATE, &sh->dev[i].flags)); | |
9a3e1101 N |
4758 | set_bit(R5_WantReplace, &sh->dev[i].flags); |
4759 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4760 | s.locked++; | |
4761 | } | |
f94c0b66 N |
4762 | if (s.replacing) |
4763 | set_bit(STRIPE_INSYNC, &sh->state); | |
4764 | set_bit(STRIPE_REPLACED, &sh->state); | |
9a3e1101 N |
4765 | } |
4766 | if ((s.syncing || s.replacing) && s.locked == 0 && | |
f94c0b66 | 4767 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state) && |
9a3e1101 | 4768 | test_bit(STRIPE_INSYNC, &sh->state)) { |
c5a31000 N |
4769 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); |
4770 | clear_bit(STRIPE_SYNCING, &sh->state); | |
f8dfcffd N |
4771 | if (test_and_clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags)) |
4772 | wake_up(&conf->wait_for_overlap); | |
c5a31000 N |
4773 | } |
4774 | ||
4775 | /* If the failed drives are just a ReadError, then we might need | |
4776 | * to progress the repair/check process | |
4777 | */ | |
4778 | if (s.failed <= conf->max_degraded && !conf->mddev->ro) | |
4779 | for (i = 0; i < s.failed; i++) { | |
4780 | struct r5dev *dev = &sh->dev[s.failed_num[i]]; | |
4781 | if (test_bit(R5_ReadError, &dev->flags) | |
4782 | && !test_bit(R5_LOCKED, &dev->flags) | |
4783 | && test_bit(R5_UPTODATE, &dev->flags) | |
4784 | ) { | |
4785 | if (!test_bit(R5_ReWrite, &dev->flags)) { | |
4786 | set_bit(R5_Wantwrite, &dev->flags); | |
4787 | set_bit(R5_ReWrite, &dev->flags); | |
4788 | set_bit(R5_LOCKED, &dev->flags); | |
4789 | s.locked++; | |
4790 | } else { | |
4791 | /* let's read it back */ | |
4792 | set_bit(R5_Wantread, &dev->flags); | |
4793 | set_bit(R5_LOCKED, &dev->flags); | |
4794 | s.locked++; | |
4795 | } | |
4796 | } | |
4797 | } | |
4798 | ||
3687c061 N |
4799 | /* Finish reconstruct operations initiated by the expansion process */ |
4800 | if (sh->reconstruct_state == reconstruct_state_result) { | |
4801 | struct stripe_head *sh_src | |
6d036f7d | 4802 | = raid5_get_active_stripe(conf, sh->sector, 1, 1, 1); |
3687c061 N |
4803 | if (sh_src && test_bit(STRIPE_EXPAND_SOURCE, &sh_src->state)) { |
4804 | /* sh cannot be written until sh_src has been read. | |
4805 | * so arrange for sh to be delayed a little | |
4806 | */ | |
4807 | set_bit(STRIPE_DELAYED, &sh->state); | |
4808 | set_bit(STRIPE_HANDLE, &sh->state); | |
4809 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, | |
4810 | &sh_src->state)) | |
4811 | atomic_inc(&conf->preread_active_stripes); | |
6d036f7d | 4812 | raid5_release_stripe(sh_src); |
3687c061 N |
4813 | goto finish; |
4814 | } | |
4815 | if (sh_src) | |
6d036f7d | 4816 | raid5_release_stripe(sh_src); |
3687c061 N |
4817 | |
4818 | sh->reconstruct_state = reconstruct_state_idle; | |
4819 | clear_bit(STRIPE_EXPANDING, &sh->state); | |
4820 | for (i = conf->raid_disks; i--; ) { | |
4821 | set_bit(R5_Wantwrite, &sh->dev[i].flags); | |
4822 | set_bit(R5_LOCKED, &sh->dev[i].flags); | |
4823 | s.locked++; | |
4824 | } | |
4825 | } | |
f416885e | 4826 | |
3687c061 N |
4827 | if (s.expanded && test_bit(STRIPE_EXPANDING, &sh->state) && |
4828 | !sh->reconstruct_state) { | |
4829 | /* Need to write out all blocks after computing parity */ | |
4830 | sh->disks = conf->raid_disks; | |
4831 | stripe_set_idx(sh->sector, conf, 0, sh); | |
4832 | schedule_reconstruction(sh, &s, 1, 1); | |
4833 | } else if (s.expanded && !sh->reconstruct_state && s.locked == 0) { | |
4834 | clear_bit(STRIPE_EXPAND_READY, &sh->state); | |
4835 | atomic_dec(&conf->reshape_stripes); | |
4836 | wake_up(&conf->wait_for_overlap); | |
4837 | md_done_sync(conf->mddev, STRIPE_SECTORS, 1); | |
4838 | } | |
4839 | ||
4840 | if (s.expanding && s.locked == 0 && | |
4841 | !test_bit(STRIPE_COMPUTE_RUN, &sh->state)) | |
4842 | handle_stripe_expansion(conf, sh); | |
16a53ecc | 4843 | |
3687c061 | 4844 | finish: |
6bfe0b49 | 4845 | /* wait for this device to become unblocked */ |
5f066c63 N |
4846 | if (unlikely(s.blocked_rdev)) { |
4847 | if (conf->mddev->external) | |
4848 | md_wait_for_blocked_rdev(s.blocked_rdev, | |
4849 | conf->mddev); | |
4850 | else | |
4851 | /* Internal metadata will immediately | |
4852 | * be written by raid5d, so we don't | |
4853 | * need to wait here. | |
4854 | */ | |
4855 | rdev_dec_pending(s.blocked_rdev, | |
4856 | conf->mddev); | |
4857 | } | |
6bfe0b49 | 4858 | |
bc2607f3 N |
4859 | if (s.handle_bad_blocks) |
4860 | for (i = disks; i--; ) { | |
3cb03002 | 4861 | struct md_rdev *rdev; |
bc2607f3 N |
4862 | struct r5dev *dev = &sh->dev[i]; |
4863 | if (test_and_clear_bit(R5_WriteError, &dev->flags)) { | |
4864 | /* We own a safe reference to the rdev */ | |
4865 | rdev = conf->disks[i].rdev; | |
4866 | if (!rdev_set_badblocks(rdev, sh->sector, | |
4867 | STRIPE_SECTORS, 0)) | |
4868 | md_error(conf->mddev, rdev); | |
4869 | rdev_dec_pending(rdev, conf->mddev); | |
4870 | } | |
b84db560 N |
4871 | if (test_and_clear_bit(R5_MadeGood, &dev->flags)) { |
4872 | rdev = conf->disks[i].rdev; | |
4873 | rdev_clear_badblocks(rdev, sh->sector, | |
c6563a8c | 4874 | STRIPE_SECTORS, 0); |
b84db560 N |
4875 | rdev_dec_pending(rdev, conf->mddev); |
4876 | } | |
977df362 N |
4877 | if (test_and_clear_bit(R5_MadeGoodRepl, &dev->flags)) { |
4878 | rdev = conf->disks[i].replacement; | |
dd054fce N |
4879 | if (!rdev) |
4880 | /* rdev have been moved down */ | |
4881 | rdev = conf->disks[i].rdev; | |
977df362 | 4882 | rdev_clear_badblocks(rdev, sh->sector, |
c6563a8c | 4883 | STRIPE_SECTORS, 0); |
977df362 N |
4884 | rdev_dec_pending(rdev, conf->mddev); |
4885 | } | |
bc2607f3 N |
4886 | } |
4887 | ||
6c0069c0 YT |
4888 | if (s.ops_request) |
4889 | raid_run_ops(sh, s.ops_request); | |
4890 | ||
f0e43bcd | 4891 | ops_run_io(sh, &s); |
16a53ecc | 4892 | |
c5709ef6 | 4893 | if (s.dec_preread_active) { |
729a1866 | 4894 | /* We delay this until after ops_run_io so that if make_request |
e9c7469b | 4895 | * is waiting on a flush, it won't continue until the writes |
729a1866 N |
4896 | * have actually been submitted. |
4897 | */ | |
4898 | atomic_dec(&conf->preread_active_stripes); | |
4899 | if (atomic_read(&conf->preread_active_stripes) < | |
4900 | IO_THRESHOLD) | |
4901 | md_wakeup_thread(conf->mddev->thread); | |
4902 | } | |
4903 | ||
c3cce6cd | 4904 | if (!bio_list_empty(&s.return_bi)) { |
2953079c | 4905 | if (test_bit(MD_SB_CHANGE_PENDING, &conf->mddev->sb_flags)) { |
c3cce6cd N |
4906 | spin_lock_irq(&conf->device_lock); |
4907 | bio_list_merge(&conf->return_bi, &s.return_bi); | |
4908 | spin_unlock_irq(&conf->device_lock); | |
4909 | md_wakeup_thread(conf->mddev->thread); | |
4910 | } else | |
4911 | return_io(&s.return_bi); | |
4912 | } | |
16a53ecc | 4913 | |
257a4b42 | 4914 | clear_bit_unlock(STRIPE_ACTIVE, &sh->state); |
16a53ecc N |
4915 | } |
4916 | ||
d1688a6d | 4917 | static void raid5_activate_delayed(struct r5conf *conf) |
16a53ecc N |
4918 | { |
4919 | if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { | |
4920 | while (!list_empty(&conf->delayed_list)) { | |
4921 | struct list_head *l = conf->delayed_list.next; | |
4922 | struct stripe_head *sh; | |
4923 | sh = list_entry(l, struct stripe_head, lru); | |
4924 | list_del_init(l); | |
4925 | clear_bit(STRIPE_DELAYED, &sh->state); | |
4926 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
4927 | atomic_inc(&conf->preread_active_stripes); | |
8b3e6cdc | 4928 | list_add_tail(&sh->lru, &conf->hold_list); |
851c30c9 | 4929 | raid5_wakeup_stripe_thread(sh); |
16a53ecc | 4930 | } |
482c0834 | 4931 | } |
16a53ecc N |
4932 | } |
4933 | ||
566c09c5 SL |
4934 | static void activate_bit_delay(struct r5conf *conf, |
4935 | struct list_head *temp_inactive_list) | |
16a53ecc N |
4936 | { |
4937 | /* device_lock is held */ | |
4938 | struct list_head head; | |
4939 | list_add(&head, &conf->bitmap_list); | |
4940 | list_del_init(&conf->bitmap_list); | |
4941 | while (!list_empty(&head)) { | |
4942 | struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru); | |
566c09c5 | 4943 | int hash; |
16a53ecc N |
4944 | list_del_init(&sh->lru); |
4945 | atomic_inc(&sh->count); | |
566c09c5 SL |
4946 | hash = sh->hash_lock_index; |
4947 | __release_stripe(conf, sh, &temp_inactive_list[hash]); | |
16a53ecc N |
4948 | } |
4949 | } | |
4950 | ||
5c675f83 | 4951 | static int raid5_congested(struct mddev *mddev, int bits) |
f022b2fd | 4952 | { |
d1688a6d | 4953 | struct r5conf *conf = mddev->private; |
f022b2fd N |
4954 | |
4955 | /* No difference between reads and writes. Just check | |
4956 | * how busy the stripe_cache is | |
4957 | */ | |
3fa841d7 | 4958 | |
5423399a | 4959 | if (test_bit(R5_INACTIVE_BLOCKED, &conf->cache_state)) |
f022b2fd | 4960 | return 1; |
a39f7afd SL |
4961 | |
4962 | /* Also checks whether there is pressure on r5cache log space */ | |
4963 | if (test_bit(R5C_LOG_TIGHT, &conf->cache_state)) | |
4964 | return 1; | |
f022b2fd N |
4965 | if (conf->quiesce) |
4966 | return 1; | |
4bda556a | 4967 | if (atomic_read(&conf->empty_inactive_list_nr)) |
f022b2fd N |
4968 | return 1; |
4969 | ||
4970 | return 0; | |
4971 | } | |
4972 | ||
fd01b88c | 4973 | static int in_chunk_boundary(struct mddev *mddev, struct bio *bio) |
f679623f | 4974 | { |
3cb5edf4 | 4975 | struct r5conf *conf = mddev->private; |
4f024f37 | 4976 | sector_t sector = bio->bi_iter.bi_sector + get_start_sect(bio->bi_bdev); |
3cb5edf4 | 4977 | unsigned int chunk_sectors; |
aa8b57aa | 4978 | unsigned int bio_sectors = bio_sectors(bio); |
f679623f | 4979 | |
3cb5edf4 | 4980 | chunk_sectors = min(conf->chunk_sectors, conf->prev_chunk_sectors); |
f679623f RBJ |
4981 | return chunk_sectors >= |
4982 | ((sector & (chunk_sectors - 1)) + bio_sectors); | |
4983 | } | |
4984 | ||
46031f9a RBJ |
4985 | /* |
4986 | * add bio to the retry LIFO ( in O(1) ... we are in interrupt ) | |
4987 | * later sampled by raid5d. | |
4988 | */ | |
d1688a6d | 4989 | static void add_bio_to_retry(struct bio *bi,struct r5conf *conf) |
46031f9a RBJ |
4990 | { |
4991 | unsigned long flags; | |
4992 | ||
4993 | spin_lock_irqsave(&conf->device_lock, flags); | |
4994 | ||
4995 | bi->bi_next = conf->retry_read_aligned_list; | |
4996 | conf->retry_read_aligned_list = bi; | |
4997 | ||
4998 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
4999 | md_wakeup_thread(conf->mddev->thread); | |
5000 | } | |
5001 | ||
d1688a6d | 5002 | static struct bio *remove_bio_from_retry(struct r5conf *conf) |
46031f9a RBJ |
5003 | { |
5004 | struct bio *bi; | |
5005 | ||
5006 | bi = conf->retry_read_aligned; | |
5007 | if (bi) { | |
5008 | conf->retry_read_aligned = NULL; | |
5009 | return bi; | |
5010 | } | |
5011 | bi = conf->retry_read_aligned_list; | |
5012 | if(bi) { | |
387bb173 | 5013 | conf->retry_read_aligned_list = bi->bi_next; |
46031f9a | 5014 | bi->bi_next = NULL; |
960e739d JA |
5015 | /* |
5016 | * this sets the active strip count to 1 and the processed | |
5017 | * strip count to zero (upper 8 bits) | |
5018 | */ | |
e7836bd6 | 5019 | raid5_set_bi_stripes(bi, 1); /* biased count of active stripes */ |
46031f9a RBJ |
5020 | } |
5021 | ||
5022 | return bi; | |
5023 | } | |
5024 | ||
f679623f RBJ |
5025 | /* |
5026 | * The "raid5_align_endio" should check if the read succeeded and if it | |
5027 | * did, call bio_endio on the original bio (having bio_put the new bio | |
5028 | * first). | |
5029 | * If the read failed.. | |
5030 | */ | |
4246a0b6 | 5031 | static void raid5_align_endio(struct bio *bi) |
f679623f RBJ |
5032 | { |
5033 | struct bio* raid_bi = bi->bi_private; | |
fd01b88c | 5034 | struct mddev *mddev; |
d1688a6d | 5035 | struct r5conf *conf; |
3cb03002 | 5036 | struct md_rdev *rdev; |
9b81c842 | 5037 | int error = bi->bi_error; |
46031f9a | 5038 | |
f679623f | 5039 | bio_put(bi); |
46031f9a | 5040 | |
46031f9a RBJ |
5041 | rdev = (void*)raid_bi->bi_next; |
5042 | raid_bi->bi_next = NULL; | |
2b7f2228 N |
5043 | mddev = rdev->mddev; |
5044 | conf = mddev->private; | |
46031f9a RBJ |
5045 | |
5046 | rdev_dec_pending(rdev, conf->mddev); | |
5047 | ||
9b81c842 | 5048 | if (!error) { |
0a82a8d1 LT |
5049 | trace_block_bio_complete(bdev_get_queue(raid_bi->bi_bdev), |
5050 | raid_bi, 0); | |
4246a0b6 | 5051 | bio_endio(raid_bi); |
46031f9a | 5052 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 5053 | wake_up(&conf->wait_for_quiescent); |
6712ecf8 | 5054 | return; |
46031f9a RBJ |
5055 | } |
5056 | ||
45b4233c | 5057 | pr_debug("raid5_align_endio : io error...handing IO for a retry\n"); |
46031f9a RBJ |
5058 | |
5059 | add_bio_to_retry(raid_bi, conf); | |
f679623f RBJ |
5060 | } |
5061 | ||
7ef6b12a | 5062 | static int raid5_read_one_chunk(struct mddev *mddev, struct bio *raid_bio) |
f679623f | 5063 | { |
d1688a6d | 5064 | struct r5conf *conf = mddev->private; |
8553fe7e | 5065 | int dd_idx; |
f679623f | 5066 | struct bio* align_bi; |
3cb03002 | 5067 | struct md_rdev *rdev; |
671488cc | 5068 | sector_t end_sector; |
f679623f RBJ |
5069 | |
5070 | if (!in_chunk_boundary(mddev, raid_bio)) { | |
7ef6b12a | 5071 | pr_debug("%s: non aligned\n", __func__); |
f679623f RBJ |
5072 | return 0; |
5073 | } | |
5074 | /* | |
d7a10308 | 5075 | * use bio_clone_fast to make a copy of the bio |
f679623f | 5076 | */ |
d7a10308 | 5077 | align_bi = bio_clone_fast(raid_bio, GFP_NOIO, mddev->bio_set); |
f679623f RBJ |
5078 | if (!align_bi) |
5079 | return 0; | |
5080 | /* | |
5081 | * set bi_end_io to a new function, and set bi_private to the | |
5082 | * original bio. | |
5083 | */ | |
5084 | align_bi->bi_end_io = raid5_align_endio; | |
5085 | align_bi->bi_private = raid_bio; | |
5086 | /* | |
5087 | * compute position | |
5088 | */ | |
4f024f37 KO |
5089 | align_bi->bi_iter.bi_sector = |
5090 | raid5_compute_sector(conf, raid_bio->bi_iter.bi_sector, | |
5091 | 0, &dd_idx, NULL); | |
f679623f | 5092 | |
f73a1c7d | 5093 | end_sector = bio_end_sector(align_bi); |
f679623f | 5094 | rcu_read_lock(); |
671488cc N |
5095 | rdev = rcu_dereference(conf->disks[dd_idx].replacement); |
5096 | if (!rdev || test_bit(Faulty, &rdev->flags) || | |
5097 | rdev->recovery_offset < end_sector) { | |
5098 | rdev = rcu_dereference(conf->disks[dd_idx].rdev); | |
5099 | if (rdev && | |
5100 | (test_bit(Faulty, &rdev->flags) || | |
5101 | !(test_bit(In_sync, &rdev->flags) || | |
5102 | rdev->recovery_offset >= end_sector))) | |
5103 | rdev = NULL; | |
5104 | } | |
03b047f4 SL |
5105 | |
5106 | if (r5c_big_stripe_cached(conf, align_bi->bi_iter.bi_sector)) { | |
5107 | rcu_read_unlock(); | |
5108 | bio_put(align_bi); | |
5109 | return 0; | |
5110 | } | |
5111 | ||
671488cc | 5112 | if (rdev) { |
31c176ec N |
5113 | sector_t first_bad; |
5114 | int bad_sectors; | |
5115 | ||
f679623f RBJ |
5116 | atomic_inc(&rdev->nr_pending); |
5117 | rcu_read_unlock(); | |
46031f9a RBJ |
5118 | raid_bio->bi_next = (void*)rdev; |
5119 | align_bi->bi_bdev = rdev->bdev; | |
b7c44ed9 | 5120 | bio_clear_flag(align_bi, BIO_SEG_VALID); |
46031f9a | 5121 | |
7140aafc | 5122 | if (is_badblock(rdev, align_bi->bi_iter.bi_sector, |
4f024f37 | 5123 | bio_sectors(align_bi), |
31c176ec | 5124 | &first_bad, &bad_sectors)) { |
387bb173 NB |
5125 | bio_put(align_bi); |
5126 | rdev_dec_pending(rdev, mddev); | |
5127 | return 0; | |
5128 | } | |
5129 | ||
6c0544e2 | 5130 | /* No reshape active, so we can trust rdev->data_offset */ |
4f024f37 | 5131 | align_bi->bi_iter.bi_sector += rdev->data_offset; |
6c0544e2 | 5132 | |
46031f9a | 5133 | spin_lock_irq(&conf->device_lock); |
b1b46486 | 5134 | wait_event_lock_irq(conf->wait_for_quiescent, |
46031f9a | 5135 | conf->quiesce == 0, |
eed8c02e | 5136 | conf->device_lock); |
46031f9a RBJ |
5137 | atomic_inc(&conf->active_aligned_reads); |
5138 | spin_unlock_irq(&conf->device_lock); | |
5139 | ||
e3620a3a JB |
5140 | if (mddev->gendisk) |
5141 | trace_block_bio_remap(bdev_get_queue(align_bi->bi_bdev), | |
5142 | align_bi, disk_devt(mddev->gendisk), | |
4f024f37 | 5143 | raid_bio->bi_iter.bi_sector); |
f679623f RBJ |
5144 | generic_make_request(align_bi); |
5145 | return 1; | |
5146 | } else { | |
5147 | rcu_read_unlock(); | |
46031f9a | 5148 | bio_put(align_bi); |
f679623f RBJ |
5149 | return 0; |
5150 | } | |
5151 | } | |
5152 | ||
7ef6b12a ML |
5153 | static struct bio *chunk_aligned_read(struct mddev *mddev, struct bio *raid_bio) |
5154 | { | |
5155 | struct bio *split; | |
5156 | ||
5157 | do { | |
5158 | sector_t sector = raid_bio->bi_iter.bi_sector; | |
5159 | unsigned chunk_sects = mddev->chunk_sectors; | |
5160 | unsigned sectors = chunk_sects - (sector & (chunk_sects-1)); | |
5161 | ||
5162 | if (sectors < bio_sectors(raid_bio)) { | |
5163 | split = bio_split(raid_bio, sectors, GFP_NOIO, fs_bio_set); | |
5164 | bio_chain(split, raid_bio); | |
5165 | } else | |
5166 | split = raid_bio; | |
5167 | ||
5168 | if (!raid5_read_one_chunk(mddev, split)) { | |
5169 | if (split != raid_bio) | |
5170 | generic_make_request(raid_bio); | |
5171 | return split; | |
5172 | } | |
5173 | } while (split != raid_bio); | |
5174 | ||
5175 | return NULL; | |
5176 | } | |
5177 | ||
8b3e6cdc DW |
5178 | /* __get_priority_stripe - get the next stripe to process |
5179 | * | |
5180 | * Full stripe writes are allowed to pass preread active stripes up until | |
5181 | * the bypass_threshold is exceeded. In general the bypass_count | |
5182 | * increments when the handle_list is handled before the hold_list; however, it | |
5183 | * will not be incremented when STRIPE_IO_STARTED is sampled set signifying a | |
5184 | * stripe with in flight i/o. The bypass_count will be reset when the | |
5185 | * head of the hold_list has changed, i.e. the head was promoted to the | |
5186 | * handle_list. | |
5187 | */ | |
851c30c9 | 5188 | static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group) |
8b3e6cdc | 5189 | { |
535ae4eb | 5190 | struct stripe_head *sh, *tmp; |
851c30c9 | 5191 | struct list_head *handle_list = NULL; |
535ae4eb SL |
5192 | struct r5worker_group *wg; |
5193 | bool second_try = !r5c_is_writeback(conf->log); | |
5194 | bool try_loprio = test_bit(R5C_LOG_TIGHT, &conf->cache_state); | |
851c30c9 | 5195 | |
535ae4eb SL |
5196 | again: |
5197 | wg = NULL; | |
5198 | sh = NULL; | |
851c30c9 | 5199 | if (conf->worker_cnt_per_group == 0) { |
535ae4eb SL |
5200 | handle_list = try_loprio ? &conf->loprio_list : |
5201 | &conf->handle_list; | |
851c30c9 | 5202 | } else if (group != ANY_GROUP) { |
535ae4eb SL |
5203 | handle_list = try_loprio ? &conf->worker_groups[group].loprio_list : |
5204 | &conf->worker_groups[group].handle_list; | |
bfc90cb0 | 5205 | wg = &conf->worker_groups[group]; |
851c30c9 SL |
5206 | } else { |
5207 | int i; | |
5208 | for (i = 0; i < conf->group_cnt; i++) { | |
535ae4eb SL |
5209 | handle_list = try_loprio ? &conf->worker_groups[i].loprio_list : |
5210 | &conf->worker_groups[i].handle_list; | |
bfc90cb0 | 5211 | wg = &conf->worker_groups[i]; |
851c30c9 SL |
5212 | if (!list_empty(handle_list)) |
5213 | break; | |
5214 | } | |
5215 | } | |
8b3e6cdc DW |
5216 | |
5217 | pr_debug("%s: handle: %s hold: %s full_writes: %d bypass_count: %d\n", | |
5218 | __func__, | |
851c30c9 | 5219 | list_empty(handle_list) ? "empty" : "busy", |
8b3e6cdc DW |
5220 | list_empty(&conf->hold_list) ? "empty" : "busy", |
5221 | atomic_read(&conf->pending_full_writes), conf->bypass_count); | |
5222 | ||
851c30c9 SL |
5223 | if (!list_empty(handle_list)) { |
5224 | sh = list_entry(handle_list->next, typeof(*sh), lru); | |
8b3e6cdc DW |
5225 | |
5226 | if (list_empty(&conf->hold_list)) | |
5227 | conf->bypass_count = 0; | |
5228 | else if (!test_bit(STRIPE_IO_STARTED, &sh->state)) { | |
5229 | if (conf->hold_list.next == conf->last_hold) | |
5230 | conf->bypass_count++; | |
5231 | else { | |
5232 | conf->last_hold = conf->hold_list.next; | |
5233 | conf->bypass_count -= conf->bypass_threshold; | |
5234 | if (conf->bypass_count < 0) | |
5235 | conf->bypass_count = 0; | |
5236 | } | |
5237 | } | |
5238 | } else if (!list_empty(&conf->hold_list) && | |
5239 | ((conf->bypass_threshold && | |
5240 | conf->bypass_count > conf->bypass_threshold) || | |
5241 | atomic_read(&conf->pending_full_writes) == 0)) { | |
851c30c9 SL |
5242 | |
5243 | list_for_each_entry(tmp, &conf->hold_list, lru) { | |
5244 | if (conf->worker_cnt_per_group == 0 || | |
5245 | group == ANY_GROUP || | |
5246 | !cpu_online(tmp->cpu) || | |
5247 | cpu_to_group(tmp->cpu) == group) { | |
5248 | sh = tmp; | |
5249 | break; | |
5250 | } | |
5251 | } | |
5252 | ||
5253 | if (sh) { | |
5254 | conf->bypass_count -= conf->bypass_threshold; | |
5255 | if (conf->bypass_count < 0) | |
5256 | conf->bypass_count = 0; | |
5257 | } | |
bfc90cb0 | 5258 | wg = NULL; |
851c30c9 SL |
5259 | } |
5260 | ||
535ae4eb SL |
5261 | if (!sh) { |
5262 | if (second_try) | |
5263 | return NULL; | |
5264 | second_try = true; | |
5265 | try_loprio = !try_loprio; | |
5266 | goto again; | |
5267 | } | |
8b3e6cdc | 5268 | |
bfc90cb0 SL |
5269 | if (wg) { |
5270 | wg->stripes_cnt--; | |
5271 | sh->group = NULL; | |
5272 | } | |
8b3e6cdc | 5273 | list_del_init(&sh->lru); |
c7a6d35e | 5274 | BUG_ON(atomic_inc_return(&sh->count) != 1); |
8b3e6cdc DW |
5275 | return sh; |
5276 | } | |
f679623f | 5277 | |
8811b596 SL |
5278 | struct raid5_plug_cb { |
5279 | struct blk_plug_cb cb; | |
5280 | struct list_head list; | |
566c09c5 | 5281 | struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS]; |
8811b596 SL |
5282 | }; |
5283 | ||
5284 | static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) | |
5285 | { | |
5286 | struct raid5_plug_cb *cb = container_of( | |
5287 | blk_cb, struct raid5_plug_cb, cb); | |
5288 | struct stripe_head *sh; | |
5289 | struct mddev *mddev = cb->cb.data; | |
5290 | struct r5conf *conf = mddev->private; | |
a9add5d9 | 5291 | int cnt = 0; |
566c09c5 | 5292 | int hash; |
8811b596 SL |
5293 | |
5294 | if (cb->list.next && !list_empty(&cb->list)) { | |
5295 | spin_lock_irq(&conf->device_lock); | |
5296 | while (!list_empty(&cb->list)) { | |
5297 | sh = list_first_entry(&cb->list, struct stripe_head, lru); | |
5298 | list_del_init(&sh->lru); | |
5299 | /* | |
5300 | * avoid race release_stripe_plug() sees | |
5301 | * STRIPE_ON_UNPLUG_LIST clear but the stripe | |
5302 | * is still in our list | |
5303 | */ | |
4e857c58 | 5304 | smp_mb__before_atomic(); |
8811b596 | 5305 | clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); |
773ca82f SL |
5306 | /* |
5307 | * STRIPE_ON_RELEASE_LIST could be set here. In that | |
5308 | * case, the count is always > 1 here | |
5309 | */ | |
566c09c5 SL |
5310 | hash = sh->hash_lock_index; |
5311 | __release_stripe(conf, sh, &cb->temp_inactive_list[hash]); | |
a9add5d9 | 5312 | cnt++; |
8811b596 SL |
5313 | } |
5314 | spin_unlock_irq(&conf->device_lock); | |
5315 | } | |
566c09c5 SL |
5316 | release_inactive_stripe_list(conf, cb->temp_inactive_list, |
5317 | NR_STRIPE_HASH_LOCKS); | |
e3620a3a JB |
5318 | if (mddev->queue) |
5319 | trace_block_unplug(mddev->queue, cnt, !from_schedule); | |
8811b596 SL |
5320 | kfree(cb); |
5321 | } | |
5322 | ||
5323 | static void release_stripe_plug(struct mddev *mddev, | |
5324 | struct stripe_head *sh) | |
5325 | { | |
5326 | struct blk_plug_cb *blk_cb = blk_check_plugged( | |
5327 | raid5_unplug, mddev, | |
5328 | sizeof(struct raid5_plug_cb)); | |
5329 | struct raid5_plug_cb *cb; | |
5330 | ||
5331 | if (!blk_cb) { | |
6d036f7d | 5332 | raid5_release_stripe(sh); |
8811b596 SL |
5333 | return; |
5334 | } | |
5335 | ||
5336 | cb = container_of(blk_cb, struct raid5_plug_cb, cb); | |
5337 | ||
566c09c5 SL |
5338 | if (cb->list.next == NULL) { |
5339 | int i; | |
8811b596 | 5340 | INIT_LIST_HEAD(&cb->list); |
566c09c5 SL |
5341 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) |
5342 | INIT_LIST_HEAD(cb->temp_inactive_list + i); | |
5343 | } | |
8811b596 SL |
5344 | |
5345 | if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)) | |
5346 | list_add_tail(&sh->lru, &cb->list); | |
5347 | else | |
6d036f7d | 5348 | raid5_release_stripe(sh); |
8811b596 SL |
5349 | } |
5350 | ||
620125f2 SL |
5351 | static void make_discard_request(struct mddev *mddev, struct bio *bi) |
5352 | { | |
5353 | struct r5conf *conf = mddev->private; | |
5354 | sector_t logical_sector, last_sector; | |
5355 | struct stripe_head *sh; | |
5356 | int remaining; | |
5357 | int stripe_sectors; | |
5358 | ||
5359 | if (mddev->reshape_position != MaxSector) | |
5360 | /* Skip discard while reshape is happening */ | |
5361 | return; | |
5362 | ||
4f024f37 KO |
5363 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
5364 | last_sector = bi->bi_iter.bi_sector + (bi->bi_iter.bi_size>>9); | |
620125f2 SL |
5365 | |
5366 | bi->bi_next = NULL; | |
5367 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
5368 | ||
5369 | stripe_sectors = conf->chunk_sectors * | |
5370 | (conf->raid_disks - conf->max_degraded); | |
5371 | logical_sector = DIV_ROUND_UP_SECTOR_T(logical_sector, | |
5372 | stripe_sectors); | |
5373 | sector_div(last_sector, stripe_sectors); | |
5374 | ||
5375 | logical_sector *= conf->chunk_sectors; | |
5376 | last_sector *= conf->chunk_sectors; | |
5377 | ||
5378 | for (; logical_sector < last_sector; | |
5379 | logical_sector += STRIPE_SECTORS) { | |
5380 | DEFINE_WAIT(w); | |
5381 | int d; | |
5382 | again: | |
6d036f7d | 5383 | sh = raid5_get_active_stripe(conf, logical_sector, 0, 0, 0); |
620125f2 SL |
5384 | prepare_to_wait(&conf->wait_for_overlap, &w, |
5385 | TASK_UNINTERRUPTIBLE); | |
f8dfcffd N |
5386 | set_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); |
5387 | if (test_bit(STRIPE_SYNCING, &sh->state)) { | |
6d036f7d | 5388 | raid5_release_stripe(sh); |
f8dfcffd N |
5389 | schedule(); |
5390 | goto again; | |
5391 | } | |
5392 | clear_bit(R5_Overlap, &sh->dev[sh->pd_idx].flags); | |
620125f2 SL |
5393 | spin_lock_irq(&sh->stripe_lock); |
5394 | for (d = 0; d < conf->raid_disks; d++) { | |
5395 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5396 | continue; | |
5397 | if (sh->dev[d].towrite || sh->dev[d].toread) { | |
5398 | set_bit(R5_Overlap, &sh->dev[d].flags); | |
5399 | spin_unlock_irq(&sh->stripe_lock); | |
6d036f7d | 5400 | raid5_release_stripe(sh); |
620125f2 SL |
5401 | schedule(); |
5402 | goto again; | |
5403 | } | |
5404 | } | |
f8dfcffd | 5405 | set_bit(STRIPE_DISCARD, &sh->state); |
620125f2 | 5406 | finish_wait(&conf->wait_for_overlap, &w); |
7a87f434 | 5407 | sh->overwrite_disks = 0; |
620125f2 SL |
5408 | for (d = 0; d < conf->raid_disks; d++) { |
5409 | if (d == sh->pd_idx || d == sh->qd_idx) | |
5410 | continue; | |
5411 | sh->dev[d].towrite = bi; | |
5412 | set_bit(R5_OVERWRITE, &sh->dev[d].flags); | |
5413 | raid5_inc_bi_active_stripes(bi); | |
7a87f434 | 5414 | sh->overwrite_disks++; |
620125f2 SL |
5415 | } |
5416 | spin_unlock_irq(&sh->stripe_lock); | |
5417 | if (conf->mddev->bitmap) { | |
5418 | for (d = 0; | |
5419 | d < conf->raid_disks - conf->max_degraded; | |
5420 | d++) | |
5421 | bitmap_startwrite(mddev->bitmap, | |
5422 | sh->sector, | |
5423 | STRIPE_SECTORS, | |
5424 | 0); | |
5425 | sh->bm_seq = conf->seq_flush + 1; | |
5426 | set_bit(STRIPE_BIT_DELAY, &sh->state); | |
5427 | } | |
5428 | ||
5429 | set_bit(STRIPE_HANDLE, &sh->state); | |
5430 | clear_bit(STRIPE_DELAYED, &sh->state); | |
5431 | if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) | |
5432 | atomic_inc(&conf->preread_active_stripes); | |
5433 | release_stripe_plug(mddev, sh); | |
5434 | } | |
5435 | ||
5436 | remaining = raid5_dec_bi_active_stripes(bi); | |
5437 | if (remaining == 0) { | |
5438 | md_write_end(mddev); | |
4246a0b6 | 5439 | bio_endio(bi); |
620125f2 SL |
5440 | } |
5441 | } | |
5442 | ||
849674e4 | 5443 | static void raid5_make_request(struct mddev *mddev, struct bio * bi) |
1da177e4 | 5444 | { |
d1688a6d | 5445 | struct r5conf *conf = mddev->private; |
911d4ee8 | 5446 | int dd_idx; |
1da177e4 LT |
5447 | sector_t new_sector; |
5448 | sector_t logical_sector, last_sector; | |
5449 | struct stripe_head *sh; | |
a362357b | 5450 | const int rw = bio_data_dir(bi); |
49077326 | 5451 | int remaining; |
27c0f68f SL |
5452 | DEFINE_WAIT(w); |
5453 | bool do_prepare; | |
3bddb7f8 | 5454 | bool do_flush = false; |
1da177e4 | 5455 | |
1eff9d32 | 5456 | if (unlikely(bi->bi_opf & REQ_PREFLUSH)) { |
828cbe98 SL |
5457 | int ret = r5l_handle_flush_request(conf->log, bi); |
5458 | ||
5459 | if (ret == 0) | |
5460 | return; | |
5461 | if (ret == -ENODEV) { | |
5462 | md_flush_request(mddev, bi); | |
5463 | return; | |
5464 | } | |
5465 | /* ret == -EAGAIN, fallback */ | |
3bddb7f8 SL |
5466 | /* |
5467 | * if r5l_handle_flush_request() didn't clear REQ_PREFLUSH, | |
5468 | * we need to flush journal device | |
5469 | */ | |
5470 | do_flush = bi->bi_opf & REQ_PREFLUSH; | |
e5dcdd80 N |
5471 | } |
5472 | ||
3d310eb7 | 5473 | md_write_start(mddev, bi); |
06d91a5f | 5474 | |
9ffc8f7c EM |
5475 | /* |
5476 | * If array is degraded, better not do chunk aligned read because | |
5477 | * later we might have to read it again in order to reconstruct | |
5478 | * data on failed drives. | |
5479 | */ | |
5480 | if (rw == READ && mddev->degraded == 0 && | |
7ef6b12a ML |
5481 | mddev->reshape_position == MaxSector) { |
5482 | bi = chunk_aligned_read(mddev, bi); | |
5483 | if (!bi) | |
5484 | return; | |
5485 | } | |
52488615 | 5486 | |
796a5cf0 | 5487 | if (unlikely(bio_op(bi) == REQ_OP_DISCARD)) { |
620125f2 SL |
5488 | make_discard_request(mddev, bi); |
5489 | return; | |
5490 | } | |
5491 | ||
4f024f37 | 5492 | logical_sector = bi->bi_iter.bi_sector & ~((sector_t)STRIPE_SECTORS-1); |
f73a1c7d | 5493 | last_sector = bio_end_sector(bi); |
1da177e4 LT |
5494 | bi->bi_next = NULL; |
5495 | bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ | |
06d91a5f | 5496 | |
27c0f68f | 5497 | prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5498 | for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { |
b5663ba4 | 5499 | int previous; |
c46501b2 | 5500 | int seq; |
b578d55f | 5501 | |
27c0f68f | 5502 | do_prepare = false; |
7ecaa1e6 | 5503 | retry: |
c46501b2 | 5504 | seq = read_seqcount_begin(&conf->gen_lock); |
b5663ba4 | 5505 | previous = 0; |
27c0f68f SL |
5506 | if (do_prepare) |
5507 | prepare_to_wait(&conf->wait_for_overlap, &w, | |
5508 | TASK_UNINTERRUPTIBLE); | |
b0f9ec04 | 5509 | if (unlikely(conf->reshape_progress != MaxSector)) { |
fef9c61f | 5510 | /* spinlock is needed as reshape_progress may be |
df8e7f76 N |
5511 | * 64bit on a 32bit platform, and so it might be |
5512 | * possible to see a half-updated value | |
aeb878b0 | 5513 | * Of course reshape_progress could change after |
df8e7f76 N |
5514 | * the lock is dropped, so once we get a reference |
5515 | * to the stripe that we think it is, we will have | |
5516 | * to check again. | |
5517 | */ | |
7ecaa1e6 | 5518 | spin_lock_irq(&conf->device_lock); |
2c810cdd | 5519 | if (mddev->reshape_backwards |
fef9c61f N |
5520 | ? logical_sector < conf->reshape_progress |
5521 | : logical_sector >= conf->reshape_progress) { | |
b5663ba4 N |
5522 | previous = 1; |
5523 | } else { | |
2c810cdd | 5524 | if (mddev->reshape_backwards |
fef9c61f N |
5525 | ? logical_sector < conf->reshape_safe |
5526 | : logical_sector >= conf->reshape_safe) { | |
b578d55f N |
5527 | spin_unlock_irq(&conf->device_lock); |
5528 | schedule(); | |
27c0f68f | 5529 | do_prepare = true; |
b578d55f N |
5530 | goto retry; |
5531 | } | |
5532 | } | |
7ecaa1e6 N |
5533 | spin_unlock_irq(&conf->device_lock); |
5534 | } | |
16a53ecc | 5535 | |
112bf897 N |
5536 | new_sector = raid5_compute_sector(conf, logical_sector, |
5537 | previous, | |
911d4ee8 | 5538 | &dd_idx, NULL); |
849674e4 | 5539 | pr_debug("raid456: raid5_make_request, sector %llu logical %llu\n", |
c46501b2 | 5540 | (unsigned long long)new_sector, |
1da177e4 LT |
5541 | (unsigned long long)logical_sector); |
5542 | ||
6d036f7d | 5543 | sh = raid5_get_active_stripe(conf, new_sector, previous, |
1eff9d32 | 5544 | (bi->bi_opf & REQ_RAHEAD), 0); |
1da177e4 | 5545 | if (sh) { |
b0f9ec04 | 5546 | if (unlikely(previous)) { |
7ecaa1e6 | 5547 | /* expansion might have moved on while waiting for a |
df8e7f76 N |
5548 | * stripe, so we must do the range check again. |
5549 | * Expansion could still move past after this | |
5550 | * test, but as we are holding a reference to | |
5551 | * 'sh', we know that if that happens, | |
5552 | * STRIPE_EXPANDING will get set and the expansion | |
5553 | * won't proceed until we finish with the stripe. | |
7ecaa1e6 N |
5554 | */ |
5555 | int must_retry = 0; | |
5556 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5557 | if (mddev->reshape_backwards |
b0f9ec04 N |
5558 | ? logical_sector >= conf->reshape_progress |
5559 | : logical_sector < conf->reshape_progress) | |
7ecaa1e6 N |
5560 | /* mismatch, need to try again */ |
5561 | must_retry = 1; | |
5562 | spin_unlock_irq(&conf->device_lock); | |
5563 | if (must_retry) { | |
6d036f7d | 5564 | raid5_release_stripe(sh); |
7a3ab908 | 5565 | schedule(); |
27c0f68f | 5566 | do_prepare = true; |
7ecaa1e6 N |
5567 | goto retry; |
5568 | } | |
5569 | } | |
c46501b2 N |
5570 | if (read_seqcount_retry(&conf->gen_lock, seq)) { |
5571 | /* Might have got the wrong stripe_head | |
5572 | * by accident | |
5573 | */ | |
6d036f7d | 5574 | raid5_release_stripe(sh); |
c46501b2 N |
5575 | goto retry; |
5576 | } | |
e62e58a5 | 5577 | |
ffd96e35 | 5578 | if (rw == WRITE && |
a5c308d4 | 5579 | logical_sector >= mddev->suspend_lo && |
e464eafd | 5580 | logical_sector < mddev->suspend_hi) { |
6d036f7d | 5581 | raid5_release_stripe(sh); |
e62e58a5 N |
5582 | /* As the suspend_* range is controlled by |
5583 | * userspace, we want an interruptible | |
5584 | * wait. | |
5585 | */ | |
5586 | flush_signals(current); | |
5587 | prepare_to_wait(&conf->wait_for_overlap, | |
5588 | &w, TASK_INTERRUPTIBLE); | |
5589 | if (logical_sector >= mddev->suspend_lo && | |
27c0f68f | 5590 | logical_sector < mddev->suspend_hi) { |
e62e58a5 | 5591 | schedule(); |
27c0f68f SL |
5592 | do_prepare = true; |
5593 | } | |
e464eafd N |
5594 | goto retry; |
5595 | } | |
7ecaa1e6 N |
5596 | |
5597 | if (test_bit(STRIPE_EXPANDING, &sh->state) || | |
da41ba65 | 5598 | !add_stripe_bio(sh, bi, dd_idx, rw, previous)) { |
7ecaa1e6 N |
5599 | /* Stripe is busy expanding or |
5600 | * add failed due to overlap. Flush everything | |
1da177e4 LT |
5601 | * and wait a while |
5602 | */ | |
482c0834 | 5603 | md_wakeup_thread(mddev->thread); |
6d036f7d | 5604 | raid5_release_stripe(sh); |
1da177e4 | 5605 | schedule(); |
27c0f68f | 5606 | do_prepare = true; |
1da177e4 LT |
5607 | goto retry; |
5608 | } | |
3bddb7f8 SL |
5609 | if (do_flush) { |
5610 | set_bit(STRIPE_R5C_PREFLUSH, &sh->state); | |
5611 | /* we only need flush for one stripe */ | |
5612 | do_flush = false; | |
5613 | } | |
5614 | ||
6ed3003c N |
5615 | set_bit(STRIPE_HANDLE, &sh->state); |
5616 | clear_bit(STRIPE_DELAYED, &sh->state); | |
59fc630b | 5617 | if ((!sh->batch_head || sh == sh->batch_head) && |
1eff9d32 | 5618 | (bi->bi_opf & REQ_SYNC) && |
729a1866 N |
5619 | !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) |
5620 | atomic_inc(&conf->preread_active_stripes); | |
8811b596 | 5621 | release_stripe_plug(mddev, sh); |
1da177e4 LT |
5622 | } else { |
5623 | /* cannot get stripe for read-ahead, just give-up */ | |
4246a0b6 | 5624 | bi->bi_error = -EIO; |
1da177e4 LT |
5625 | break; |
5626 | } | |
1da177e4 | 5627 | } |
27c0f68f | 5628 | finish_wait(&conf->wait_for_overlap, &w); |
7c13edc8 | 5629 | |
e7836bd6 | 5630 | remaining = raid5_dec_bi_active_stripes(bi); |
f6344757 | 5631 | if (remaining == 0) { |
1da177e4 | 5632 | |
16a53ecc | 5633 | if ( rw == WRITE ) |
1da177e4 | 5634 | md_write_end(mddev); |
6712ecf8 | 5635 | |
0a82a8d1 LT |
5636 | trace_block_bio_complete(bdev_get_queue(bi->bi_bdev), |
5637 | bi, 0); | |
4246a0b6 | 5638 | bio_endio(bi); |
1da177e4 | 5639 | } |
1da177e4 LT |
5640 | } |
5641 | ||
fd01b88c | 5642 | static sector_t raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks); |
b522adcd | 5643 | |
fd01b88c | 5644 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *skipped) |
1da177e4 | 5645 | { |
52c03291 N |
5646 | /* reshaping is quite different to recovery/resync so it is |
5647 | * handled quite separately ... here. | |
5648 | * | |
5649 | * On each call to sync_request, we gather one chunk worth of | |
5650 | * destination stripes and flag them as expanding. | |
5651 | * Then we find all the source stripes and request reads. | |
5652 | * As the reads complete, handle_stripe will copy the data | |
5653 | * into the destination stripe and release that stripe. | |
5654 | */ | |
d1688a6d | 5655 | struct r5conf *conf = mddev->private; |
1da177e4 | 5656 | struct stripe_head *sh; |
ccfcc3c1 | 5657 | sector_t first_sector, last_sector; |
f416885e N |
5658 | int raid_disks = conf->previous_raid_disks; |
5659 | int data_disks = raid_disks - conf->max_degraded; | |
5660 | int new_data_disks = conf->raid_disks - conf->max_degraded; | |
52c03291 N |
5661 | int i; |
5662 | int dd_idx; | |
c8f517c4 | 5663 | sector_t writepos, readpos, safepos; |
ec32a2bd | 5664 | sector_t stripe_addr; |
7a661381 | 5665 | int reshape_sectors; |
ab69ae12 | 5666 | struct list_head stripes; |
92140480 | 5667 | sector_t retn; |
52c03291 | 5668 | |
fef9c61f N |
5669 | if (sector_nr == 0) { |
5670 | /* If restarting in the middle, skip the initial sectors */ | |
2c810cdd | 5671 | if (mddev->reshape_backwards && |
fef9c61f N |
5672 | conf->reshape_progress < raid5_size(mddev, 0, 0)) { |
5673 | sector_nr = raid5_size(mddev, 0, 0) | |
5674 | - conf->reshape_progress; | |
6cbd8148 N |
5675 | } else if (mddev->reshape_backwards && |
5676 | conf->reshape_progress == MaxSector) { | |
5677 | /* shouldn't happen, but just in case, finish up.*/ | |
5678 | sector_nr = MaxSector; | |
2c810cdd | 5679 | } else if (!mddev->reshape_backwards && |
fef9c61f N |
5680 | conf->reshape_progress > 0) |
5681 | sector_nr = conf->reshape_progress; | |
f416885e | 5682 | sector_div(sector_nr, new_data_disks); |
fef9c61f | 5683 | if (sector_nr) { |
8dee7211 N |
5684 | mddev->curr_resync_completed = sector_nr; |
5685 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
fef9c61f | 5686 | *skipped = 1; |
92140480 N |
5687 | retn = sector_nr; |
5688 | goto finish; | |
fef9c61f | 5689 | } |
52c03291 N |
5690 | } |
5691 | ||
7a661381 N |
5692 | /* We need to process a full chunk at a time. |
5693 | * If old and new chunk sizes differ, we need to process the | |
5694 | * largest of these | |
5695 | */ | |
3cb5edf4 N |
5696 | |
5697 | reshape_sectors = max(conf->chunk_sectors, conf->prev_chunk_sectors); | |
7a661381 | 5698 | |
b5254dd5 N |
5699 | /* We update the metadata at least every 10 seconds, or when |
5700 | * the data about to be copied would over-write the source of | |
5701 | * the data at the front of the range. i.e. one new_stripe | |
5702 | * along from reshape_progress new_maps to after where | |
5703 | * reshape_safe old_maps to | |
52c03291 | 5704 | */ |
fef9c61f | 5705 | writepos = conf->reshape_progress; |
f416885e | 5706 | sector_div(writepos, new_data_disks); |
c8f517c4 N |
5707 | readpos = conf->reshape_progress; |
5708 | sector_div(readpos, data_disks); | |
fef9c61f | 5709 | safepos = conf->reshape_safe; |
f416885e | 5710 | sector_div(safepos, data_disks); |
2c810cdd | 5711 | if (mddev->reshape_backwards) { |
c74c0d76 N |
5712 | BUG_ON(writepos < reshape_sectors); |
5713 | writepos -= reshape_sectors; | |
c8f517c4 | 5714 | readpos += reshape_sectors; |
7a661381 | 5715 | safepos += reshape_sectors; |
fef9c61f | 5716 | } else { |
7a661381 | 5717 | writepos += reshape_sectors; |
c74c0d76 N |
5718 | /* readpos and safepos are worst-case calculations. |
5719 | * A negative number is overly pessimistic, and causes | |
5720 | * obvious problems for unsigned storage. So clip to 0. | |
5721 | */ | |
ed37d83e N |
5722 | readpos -= min_t(sector_t, reshape_sectors, readpos); |
5723 | safepos -= min_t(sector_t, reshape_sectors, safepos); | |
fef9c61f | 5724 | } |
52c03291 | 5725 | |
b5254dd5 N |
5726 | /* Having calculated the 'writepos' possibly use it |
5727 | * to set 'stripe_addr' which is where we will write to. | |
5728 | */ | |
5729 | if (mddev->reshape_backwards) { | |
5730 | BUG_ON(conf->reshape_progress == 0); | |
5731 | stripe_addr = writepos; | |
5732 | BUG_ON((mddev->dev_sectors & | |
5733 | ~((sector_t)reshape_sectors - 1)) | |
5734 | - reshape_sectors - stripe_addr | |
5735 | != sector_nr); | |
5736 | } else { | |
5737 | BUG_ON(writepos != sector_nr + reshape_sectors); | |
5738 | stripe_addr = sector_nr; | |
5739 | } | |
5740 | ||
c8f517c4 N |
5741 | /* 'writepos' is the most advanced device address we might write. |
5742 | * 'readpos' is the least advanced device address we might read. | |
5743 | * 'safepos' is the least address recorded in the metadata as having | |
5744 | * been reshaped. | |
b5254dd5 N |
5745 | * If there is a min_offset_diff, these are adjusted either by |
5746 | * increasing the safepos/readpos if diff is negative, or | |
5747 | * increasing writepos if diff is positive. | |
5748 | * If 'readpos' is then behind 'writepos', there is no way that we can | |
c8f517c4 N |
5749 | * ensure safety in the face of a crash - that must be done by userspace |
5750 | * making a backup of the data. So in that case there is no particular | |
5751 | * rush to update metadata. | |
5752 | * Otherwise if 'safepos' is behind 'writepos', then we really need to | |
5753 | * update the metadata to advance 'safepos' to match 'readpos' so that | |
5754 | * we can be safe in the event of a crash. | |
5755 | * So we insist on updating metadata if safepos is behind writepos and | |
5756 | * readpos is beyond writepos. | |
5757 | * In any case, update the metadata every 10 seconds. | |
5758 | * Maybe that number should be configurable, but I'm not sure it is | |
5759 | * worth it.... maybe it could be a multiple of safemode_delay??? | |
5760 | */ | |
b5254dd5 N |
5761 | if (conf->min_offset_diff < 0) { |
5762 | safepos += -conf->min_offset_diff; | |
5763 | readpos += -conf->min_offset_diff; | |
5764 | } else | |
5765 | writepos += conf->min_offset_diff; | |
5766 | ||
2c810cdd | 5767 | if ((mddev->reshape_backwards |
c8f517c4 N |
5768 | ? (safepos > writepos && readpos < writepos) |
5769 | : (safepos < writepos && readpos > writepos)) || | |
5770 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
52c03291 N |
5771 | /* Cannot proceed until we've updated the superblock... */ |
5772 | wait_event(conf->wait_for_overlap, | |
c91abf5a N |
5773 | atomic_read(&conf->reshape_stripes)==0 |
5774 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); | |
5775 | if (atomic_read(&conf->reshape_stripes) != 0) | |
5776 | return 0; | |
fef9c61f | 5777 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5778 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5779 | conf->reshape_checkpoint = jiffies; |
2953079c | 5780 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
52c03291 | 5781 | md_wakeup_thread(mddev->thread); |
2953079c | 5782 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
5783 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5784 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5785 | return 0; | |
52c03291 | 5786 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5787 | conf->reshape_safe = mddev->reshape_position; |
52c03291 N |
5788 | spin_unlock_irq(&conf->device_lock); |
5789 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5790 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
52c03291 N |
5791 | } |
5792 | ||
ab69ae12 | 5793 | INIT_LIST_HEAD(&stripes); |
7a661381 | 5794 | for (i = 0; i < reshape_sectors; i += STRIPE_SECTORS) { |
52c03291 | 5795 | int j; |
a9f326eb | 5796 | int skipped_disk = 0; |
6d036f7d | 5797 | sh = raid5_get_active_stripe(conf, stripe_addr+i, 0, 0, 1); |
52c03291 N |
5798 | set_bit(STRIPE_EXPANDING, &sh->state); |
5799 | atomic_inc(&conf->reshape_stripes); | |
5800 | /* If any of this stripe is beyond the end of the old | |
5801 | * array, then we need to zero those blocks | |
5802 | */ | |
5803 | for (j=sh->disks; j--;) { | |
5804 | sector_t s; | |
5805 | if (j == sh->pd_idx) | |
5806 | continue; | |
f416885e | 5807 | if (conf->level == 6 && |
d0dabf7e | 5808 | j == sh->qd_idx) |
f416885e | 5809 | continue; |
6d036f7d | 5810 | s = raid5_compute_blocknr(sh, j, 0); |
b522adcd | 5811 | if (s < raid5_size(mddev, 0, 0)) { |
a9f326eb | 5812 | skipped_disk = 1; |
52c03291 N |
5813 | continue; |
5814 | } | |
5815 | memset(page_address(sh->dev[j].page), 0, STRIPE_SIZE); | |
5816 | set_bit(R5_Expanded, &sh->dev[j].flags); | |
5817 | set_bit(R5_UPTODATE, &sh->dev[j].flags); | |
5818 | } | |
a9f326eb | 5819 | if (!skipped_disk) { |
52c03291 N |
5820 | set_bit(STRIPE_EXPAND_READY, &sh->state); |
5821 | set_bit(STRIPE_HANDLE, &sh->state); | |
5822 | } | |
ab69ae12 | 5823 | list_add(&sh->lru, &stripes); |
52c03291 N |
5824 | } |
5825 | spin_lock_irq(&conf->device_lock); | |
2c810cdd | 5826 | if (mddev->reshape_backwards) |
7a661381 | 5827 | conf->reshape_progress -= reshape_sectors * new_data_disks; |
fef9c61f | 5828 | else |
7a661381 | 5829 | conf->reshape_progress += reshape_sectors * new_data_disks; |
52c03291 N |
5830 | spin_unlock_irq(&conf->device_lock); |
5831 | /* Ok, those stripe are ready. We can start scheduling | |
5832 | * reads on the source stripes. | |
5833 | * The source stripes are determined by mapping the first and last | |
5834 | * block on the destination stripes. | |
5835 | */ | |
52c03291 | 5836 | first_sector = |
ec32a2bd | 5837 | raid5_compute_sector(conf, stripe_addr*(new_data_disks), |
911d4ee8 | 5838 | 1, &dd_idx, NULL); |
52c03291 | 5839 | last_sector = |
0e6e0271 | 5840 | raid5_compute_sector(conf, ((stripe_addr+reshape_sectors) |
09c9e5fa | 5841 | * new_data_disks - 1), |
911d4ee8 | 5842 | 1, &dd_idx, NULL); |
58c0fed4 AN |
5843 | if (last_sector >= mddev->dev_sectors) |
5844 | last_sector = mddev->dev_sectors - 1; | |
52c03291 | 5845 | while (first_sector <= last_sector) { |
6d036f7d | 5846 | sh = raid5_get_active_stripe(conf, first_sector, 1, 0, 1); |
52c03291 N |
5847 | set_bit(STRIPE_EXPAND_SOURCE, &sh->state); |
5848 | set_bit(STRIPE_HANDLE, &sh->state); | |
6d036f7d | 5849 | raid5_release_stripe(sh); |
52c03291 N |
5850 | first_sector += STRIPE_SECTORS; |
5851 | } | |
ab69ae12 N |
5852 | /* Now that the sources are clearly marked, we can release |
5853 | * the destination stripes | |
5854 | */ | |
5855 | while (!list_empty(&stripes)) { | |
5856 | sh = list_entry(stripes.next, struct stripe_head, lru); | |
5857 | list_del_init(&sh->lru); | |
6d036f7d | 5858 | raid5_release_stripe(sh); |
ab69ae12 | 5859 | } |
c6207277 N |
5860 | /* If this takes us to the resync_max point where we have to pause, |
5861 | * then we need to write out the superblock. | |
5862 | */ | |
7a661381 | 5863 | sector_nr += reshape_sectors; |
92140480 N |
5864 | retn = reshape_sectors; |
5865 | finish: | |
c5e19d90 N |
5866 | if (mddev->curr_resync_completed > mddev->resync_max || |
5867 | (sector_nr - mddev->curr_resync_completed) * 2 | |
c03f6a19 | 5868 | >= mddev->resync_max - mddev->curr_resync_completed) { |
c6207277 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 | goto ret; | |
fef9c61f | 5875 | mddev->reshape_position = conf->reshape_progress; |
75d3da43 | 5876 | mddev->curr_resync_completed = sector_nr; |
c8f517c4 | 5877 | conf->reshape_checkpoint = jiffies; |
2953079c | 5878 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
c6207277 N |
5879 | md_wakeup_thread(mddev->thread); |
5880 | wait_event(mddev->sb_wait, | |
2953079c | 5881 | !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) |
c91abf5a N |
5882 | || test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
5883 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5884 | goto ret; | |
c6207277 | 5885 | spin_lock_irq(&conf->device_lock); |
fef9c61f | 5886 | conf->reshape_safe = mddev->reshape_position; |
c6207277 N |
5887 | spin_unlock_irq(&conf->device_lock); |
5888 | wake_up(&conf->wait_for_overlap); | |
acb180b0 | 5889 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); |
c6207277 | 5890 | } |
c91abf5a | 5891 | ret: |
92140480 | 5892 | return retn; |
52c03291 N |
5893 | } |
5894 | ||
849674e4 SL |
5895 | static inline sector_t raid5_sync_request(struct mddev *mddev, sector_t sector_nr, |
5896 | int *skipped) | |
52c03291 | 5897 | { |
d1688a6d | 5898 | struct r5conf *conf = mddev->private; |
52c03291 | 5899 | struct stripe_head *sh; |
58c0fed4 | 5900 | sector_t max_sector = mddev->dev_sectors; |
57dab0bd | 5901 | sector_t sync_blocks; |
16a53ecc N |
5902 | int still_degraded = 0; |
5903 | int i; | |
1da177e4 | 5904 | |
72626685 | 5905 | if (sector_nr >= max_sector) { |
1da177e4 | 5906 | /* just being told to finish up .. nothing much to do */ |
cea9c228 | 5907 | |
29269553 N |
5908 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
5909 | end_reshape(conf); | |
5910 | return 0; | |
5911 | } | |
72626685 N |
5912 | |
5913 | if (mddev->curr_resync < max_sector) /* aborted */ | |
5914 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
5915 | &sync_blocks, 1); | |
16a53ecc | 5916 | else /* completed sync */ |
72626685 N |
5917 | conf->fullsync = 0; |
5918 | bitmap_close_sync(mddev->bitmap); | |
5919 | ||
1da177e4 LT |
5920 | return 0; |
5921 | } | |
ccfcc3c1 | 5922 | |
64bd660b N |
5923 | /* Allow raid5_quiesce to complete */ |
5924 | wait_event(conf->wait_for_overlap, conf->quiesce != 2); | |
5925 | ||
52c03291 N |
5926 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) |
5927 | return reshape_request(mddev, sector_nr, skipped); | |
f6705578 | 5928 | |
c6207277 N |
5929 | /* No need to check resync_max as we never do more than one |
5930 | * stripe, and as resync_max will always be on a chunk boundary, | |
5931 | * if the check in md_do_sync didn't fire, there is no chance | |
5932 | * of overstepping resync_max here | |
5933 | */ | |
5934 | ||
16a53ecc | 5935 | /* if there is too many failed drives and we are trying |
1da177e4 LT |
5936 | * to resync, then assert that we are finished, because there is |
5937 | * nothing we can do. | |
5938 | */ | |
3285edf1 | 5939 | if (mddev->degraded >= conf->max_degraded && |
16a53ecc | 5940 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
58c0fed4 | 5941 | sector_t rv = mddev->dev_sectors - sector_nr; |
57afd89f | 5942 | *skipped = 1; |
1da177e4 LT |
5943 | return rv; |
5944 | } | |
6f608040 | 5945 | if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
5946 | !conf->fullsync && | |
5947 | !bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && | |
5948 | sync_blocks >= STRIPE_SECTORS) { | |
72626685 N |
5949 | /* we can skip this block, and probably more */ |
5950 | sync_blocks /= STRIPE_SECTORS; | |
5951 | *skipped = 1; | |
5952 | return sync_blocks * STRIPE_SECTORS; /* keep things rounded to whole stripes */ | |
5953 | } | |
1da177e4 | 5954 | |
c40f341f | 5955 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, false); |
b47490c9 | 5956 | |
6d036f7d | 5957 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 1, 0); |
1da177e4 | 5958 | if (sh == NULL) { |
6d036f7d | 5959 | sh = raid5_get_active_stripe(conf, sector_nr, 0, 0, 0); |
1da177e4 | 5960 | /* make sure we don't swamp the stripe cache if someone else |
16a53ecc | 5961 | * is trying to get access |
1da177e4 | 5962 | */ |
66c006a5 | 5963 | schedule_timeout_uninterruptible(1); |
1da177e4 | 5964 | } |
16a53ecc | 5965 | /* Need to check if array will still be degraded after recovery/resync |
16d9cfab EM |
5966 | * Note in case of > 1 drive failures it's possible we're rebuilding |
5967 | * one drive while leaving another faulty drive in array. | |
16a53ecc | 5968 | */ |
16d9cfab EM |
5969 | rcu_read_lock(); |
5970 | for (i = 0; i < conf->raid_disks; i++) { | |
5971 | struct md_rdev *rdev = ACCESS_ONCE(conf->disks[i].rdev); | |
5972 | ||
5973 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
16a53ecc | 5974 | still_degraded = 1; |
16d9cfab EM |
5975 | } |
5976 | rcu_read_unlock(); | |
16a53ecc N |
5977 | |
5978 | bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, still_degraded); | |
5979 | ||
83206d66 | 5980 | set_bit(STRIPE_SYNC_REQUESTED, &sh->state); |
053f5b65 | 5981 | set_bit(STRIPE_HANDLE, &sh->state); |
1da177e4 | 5982 | |
6d036f7d | 5983 | raid5_release_stripe(sh); |
1da177e4 LT |
5984 | |
5985 | return STRIPE_SECTORS; | |
5986 | } | |
5987 | ||
d1688a6d | 5988 | static int retry_aligned_read(struct r5conf *conf, struct bio *raid_bio) |
46031f9a RBJ |
5989 | { |
5990 | /* We may not be able to submit a whole bio at once as there | |
5991 | * may not be enough stripe_heads available. | |
5992 | * We cannot pre-allocate enough stripe_heads as we may need | |
5993 | * more than exist in the cache (if we allow ever large chunks). | |
5994 | * So we do one stripe head at a time and record in | |
5995 | * ->bi_hw_segments how many have been done. | |
5996 | * | |
5997 | * We *know* that this entire raid_bio is in one chunk, so | |
5998 | * it will be only one 'dd_idx' and only need one call to raid5_compute_sector. | |
5999 | */ | |
6000 | struct stripe_head *sh; | |
911d4ee8 | 6001 | int dd_idx; |
46031f9a RBJ |
6002 | sector_t sector, logical_sector, last_sector; |
6003 | int scnt = 0; | |
6004 | int remaining; | |
6005 | int handled = 0; | |
6006 | ||
4f024f37 KO |
6007 | logical_sector = raid_bio->bi_iter.bi_sector & |
6008 | ~((sector_t)STRIPE_SECTORS-1); | |
112bf897 | 6009 | sector = raid5_compute_sector(conf, logical_sector, |
911d4ee8 | 6010 | 0, &dd_idx, NULL); |
f73a1c7d | 6011 | last_sector = bio_end_sector(raid_bio); |
46031f9a RBJ |
6012 | |
6013 | for (; logical_sector < last_sector; | |
387bb173 NB |
6014 | logical_sector += STRIPE_SECTORS, |
6015 | sector += STRIPE_SECTORS, | |
6016 | scnt++) { | |
46031f9a | 6017 | |
e7836bd6 | 6018 | if (scnt < raid5_bi_processed_stripes(raid_bio)) |
46031f9a RBJ |
6019 | /* already done this stripe */ |
6020 | continue; | |
6021 | ||
6d036f7d | 6022 | sh = raid5_get_active_stripe(conf, sector, 0, 1, 1); |
46031f9a RBJ |
6023 | |
6024 | if (!sh) { | |
6025 | /* failed to get a stripe - must wait */ | |
e7836bd6 | 6026 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
46031f9a RBJ |
6027 | conf->retry_read_aligned = raid_bio; |
6028 | return handled; | |
6029 | } | |
6030 | ||
da41ba65 | 6031 | if (!add_stripe_bio(sh, raid_bio, dd_idx, 0, 0)) { |
6d036f7d | 6032 | raid5_release_stripe(sh); |
e7836bd6 | 6033 | raid5_set_bi_processed_stripes(raid_bio, scnt); |
387bb173 NB |
6034 | conf->retry_read_aligned = raid_bio; |
6035 | return handled; | |
6036 | } | |
6037 | ||
3f9e7c14 | 6038 | set_bit(R5_ReadNoMerge, &sh->dev[dd_idx].flags); |
36d1c647 | 6039 | handle_stripe(sh); |
6d036f7d | 6040 | raid5_release_stripe(sh); |
46031f9a RBJ |
6041 | handled++; |
6042 | } | |
e7836bd6 | 6043 | remaining = raid5_dec_bi_active_stripes(raid_bio); |
0a82a8d1 LT |
6044 | if (remaining == 0) { |
6045 | trace_block_bio_complete(bdev_get_queue(raid_bio->bi_bdev), | |
6046 | raid_bio, 0); | |
4246a0b6 | 6047 | bio_endio(raid_bio); |
0a82a8d1 | 6048 | } |
46031f9a | 6049 | if (atomic_dec_and_test(&conf->active_aligned_reads)) |
b1b46486 | 6050 | wake_up(&conf->wait_for_quiescent); |
46031f9a RBJ |
6051 | return handled; |
6052 | } | |
6053 | ||
bfc90cb0 | 6054 | static int handle_active_stripes(struct r5conf *conf, int group, |
566c09c5 SL |
6055 | struct r5worker *worker, |
6056 | struct list_head *temp_inactive_list) | |
46a06401 SL |
6057 | { |
6058 | struct stripe_head *batch[MAX_STRIPE_BATCH], *sh; | |
566c09c5 SL |
6059 | int i, batch_size = 0, hash; |
6060 | bool release_inactive = false; | |
46a06401 SL |
6061 | |
6062 | while (batch_size < MAX_STRIPE_BATCH && | |
851c30c9 | 6063 | (sh = __get_priority_stripe(conf, group)) != NULL) |
46a06401 SL |
6064 | batch[batch_size++] = sh; |
6065 | ||
566c09c5 SL |
6066 | if (batch_size == 0) { |
6067 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6068 | if (!list_empty(temp_inactive_list + i)) | |
6069 | break; | |
a8c34f91 SL |
6070 | if (i == NR_STRIPE_HASH_LOCKS) { |
6071 | spin_unlock_irq(&conf->device_lock); | |
6072 | r5l_flush_stripe_to_raid(conf->log); | |
6073 | spin_lock_irq(&conf->device_lock); | |
566c09c5 | 6074 | return batch_size; |
a8c34f91 | 6075 | } |
566c09c5 SL |
6076 | release_inactive = true; |
6077 | } | |
46a06401 SL |
6078 | spin_unlock_irq(&conf->device_lock); |
6079 | ||
566c09c5 SL |
6080 | release_inactive_stripe_list(conf, temp_inactive_list, |
6081 | NR_STRIPE_HASH_LOCKS); | |
6082 | ||
a8c34f91 | 6083 | r5l_flush_stripe_to_raid(conf->log); |
566c09c5 SL |
6084 | if (release_inactive) { |
6085 | spin_lock_irq(&conf->device_lock); | |
6086 | return 0; | |
6087 | } | |
6088 | ||
46a06401 SL |
6089 | for (i = 0; i < batch_size; i++) |
6090 | handle_stripe(batch[i]); | |
f6bed0ef | 6091 | r5l_write_stripe_run(conf->log); |
46a06401 SL |
6092 | |
6093 | cond_resched(); | |
6094 | ||
6095 | spin_lock_irq(&conf->device_lock); | |
566c09c5 SL |
6096 | for (i = 0; i < batch_size; i++) { |
6097 | hash = batch[i]->hash_lock_index; | |
6098 | __release_stripe(conf, batch[i], &temp_inactive_list[hash]); | |
6099 | } | |
46a06401 SL |
6100 | return batch_size; |
6101 | } | |
46031f9a | 6102 | |
851c30c9 SL |
6103 | static void raid5_do_work(struct work_struct *work) |
6104 | { | |
6105 | struct r5worker *worker = container_of(work, struct r5worker, work); | |
6106 | struct r5worker_group *group = worker->group; | |
6107 | struct r5conf *conf = group->conf; | |
6108 | int group_id = group - conf->worker_groups; | |
6109 | int handled; | |
6110 | struct blk_plug plug; | |
6111 | ||
6112 | pr_debug("+++ raid5worker active\n"); | |
6113 | ||
6114 | blk_start_plug(&plug); | |
6115 | handled = 0; | |
6116 | spin_lock_irq(&conf->device_lock); | |
6117 | while (1) { | |
6118 | int batch_size, released; | |
6119 | ||
566c09c5 | 6120 | released = release_stripe_list(conf, worker->temp_inactive_list); |
851c30c9 | 6121 | |
566c09c5 SL |
6122 | batch_size = handle_active_stripes(conf, group_id, worker, |
6123 | worker->temp_inactive_list); | |
bfc90cb0 | 6124 | worker->working = false; |
851c30c9 SL |
6125 | if (!batch_size && !released) |
6126 | break; | |
6127 | handled += batch_size; | |
6128 | } | |
6129 | pr_debug("%d stripes handled\n", handled); | |
6130 | ||
6131 | spin_unlock_irq(&conf->device_lock); | |
6132 | blk_finish_plug(&plug); | |
6133 | ||
6134 | pr_debug("--- raid5worker inactive\n"); | |
6135 | } | |
6136 | ||
1da177e4 LT |
6137 | /* |
6138 | * This is our raid5 kernel thread. | |
6139 | * | |
6140 | * We scan the hash table for stripes which can be handled now. | |
6141 | * During the scan, completed stripes are saved for us by the interrupt | |
6142 | * handler, so that they will not have to wait for our next wakeup. | |
6143 | */ | |
4ed8731d | 6144 | static void raid5d(struct md_thread *thread) |
1da177e4 | 6145 | { |
4ed8731d | 6146 | struct mddev *mddev = thread->mddev; |
d1688a6d | 6147 | struct r5conf *conf = mddev->private; |
1da177e4 | 6148 | int handled; |
e1dfa0a2 | 6149 | struct blk_plug plug; |
1da177e4 | 6150 | |
45b4233c | 6151 | pr_debug("+++ raid5d active\n"); |
1da177e4 LT |
6152 | |
6153 | md_check_recovery(mddev); | |
1da177e4 | 6154 | |
c3cce6cd | 6155 | if (!bio_list_empty(&conf->return_bi) && |
2953079c | 6156 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6157 | struct bio_list tmp = BIO_EMPTY_LIST; |
6158 | spin_lock_irq(&conf->device_lock); | |
2953079c | 6159 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
c3cce6cd N |
6160 | bio_list_merge(&tmp, &conf->return_bi); |
6161 | bio_list_init(&conf->return_bi); | |
6162 | } | |
6163 | spin_unlock_irq(&conf->device_lock); | |
6164 | return_io(&tmp); | |
6165 | } | |
6166 | ||
e1dfa0a2 | 6167 | blk_start_plug(&plug); |
1da177e4 LT |
6168 | handled = 0; |
6169 | spin_lock_irq(&conf->device_lock); | |
6170 | while (1) { | |
46031f9a | 6171 | struct bio *bio; |
773ca82f SL |
6172 | int batch_size, released; |
6173 | ||
566c09c5 | 6174 | released = release_stripe_list(conf, conf->temp_inactive_list); |
edbe83ab N |
6175 | if (released) |
6176 | clear_bit(R5_DID_ALLOC, &conf->cache_state); | |
1da177e4 | 6177 | |
0021b7bc | 6178 | if ( |
7c13edc8 N |
6179 | !list_empty(&conf->bitmap_list)) { |
6180 | /* Now is a good time to flush some bitmap updates */ | |
6181 | conf->seq_flush++; | |
700e432d | 6182 | spin_unlock_irq(&conf->device_lock); |
72626685 | 6183 | bitmap_unplug(mddev->bitmap); |
700e432d | 6184 | spin_lock_irq(&conf->device_lock); |
7c13edc8 | 6185 | conf->seq_write = conf->seq_flush; |
566c09c5 | 6186 | activate_bit_delay(conf, conf->temp_inactive_list); |
72626685 | 6187 | } |
0021b7bc | 6188 | raid5_activate_delayed(conf); |
72626685 | 6189 | |
46031f9a RBJ |
6190 | while ((bio = remove_bio_from_retry(conf))) { |
6191 | int ok; | |
6192 | spin_unlock_irq(&conf->device_lock); | |
6193 | ok = retry_aligned_read(conf, bio); | |
6194 | spin_lock_irq(&conf->device_lock); | |
6195 | if (!ok) | |
6196 | break; | |
6197 | handled++; | |
6198 | } | |
6199 | ||
566c09c5 SL |
6200 | batch_size = handle_active_stripes(conf, ANY_GROUP, NULL, |
6201 | conf->temp_inactive_list); | |
773ca82f | 6202 | if (!batch_size && !released) |
1da177e4 | 6203 | break; |
46a06401 | 6204 | handled += batch_size; |
1da177e4 | 6205 | |
2953079c | 6206 | if (mddev->sb_flags & ~(1 << MD_SB_CHANGE_PENDING)) { |
46a06401 | 6207 | spin_unlock_irq(&conf->device_lock); |
de393cde | 6208 | md_check_recovery(mddev); |
46a06401 SL |
6209 | spin_lock_irq(&conf->device_lock); |
6210 | } | |
1da177e4 | 6211 | } |
45b4233c | 6212 | pr_debug("%d stripes handled\n", handled); |
1da177e4 LT |
6213 | |
6214 | spin_unlock_irq(&conf->device_lock); | |
2d5b569b N |
6215 | if (test_and_clear_bit(R5_ALLOC_MORE, &conf->cache_state) && |
6216 | mutex_trylock(&conf->cache_size_mutex)) { | |
edbe83ab N |
6217 | grow_one_stripe(conf, __GFP_NOWARN); |
6218 | /* Set flag even if allocation failed. This helps | |
6219 | * slow down allocation requests when mem is short | |
6220 | */ | |
6221 | set_bit(R5_DID_ALLOC, &conf->cache_state); | |
2d5b569b | 6222 | mutex_unlock(&conf->cache_size_mutex); |
edbe83ab | 6223 | } |
1da177e4 | 6224 | |
765d704d SL |
6225 | flush_deferred_bios(conf); |
6226 | ||
0576b1c6 SL |
6227 | r5l_flush_stripe_to_raid(conf->log); |
6228 | ||
c9f21aaf | 6229 | async_tx_issue_pending_all(); |
e1dfa0a2 | 6230 | blk_finish_plug(&plug); |
1da177e4 | 6231 | |
45b4233c | 6232 | pr_debug("--- raid5d inactive\n"); |
1da177e4 LT |
6233 | } |
6234 | ||
3f294f4f | 6235 | static ssize_t |
fd01b88c | 6236 | raid5_show_stripe_cache_size(struct mddev *mddev, char *page) |
3f294f4f | 6237 | { |
7b1485ba N |
6238 | struct r5conf *conf; |
6239 | int ret = 0; | |
6240 | spin_lock(&mddev->lock); | |
6241 | conf = mddev->private; | |
96de1e66 | 6242 | if (conf) |
edbe83ab | 6243 | ret = sprintf(page, "%d\n", conf->min_nr_stripes); |
7b1485ba N |
6244 | spin_unlock(&mddev->lock); |
6245 | return ret; | |
3f294f4f N |
6246 | } |
6247 | ||
c41d4ac4 | 6248 | int |
fd01b88c | 6249 | raid5_set_cache_size(struct mddev *mddev, int size) |
3f294f4f | 6250 | { |
d1688a6d | 6251 | struct r5conf *conf = mddev->private; |
b5470dc5 DW |
6252 | int err; |
6253 | ||
c41d4ac4 | 6254 | if (size <= 16 || size > 32768) |
3f294f4f | 6255 | return -EINVAL; |
486f0644 | 6256 | |
edbe83ab | 6257 | conf->min_nr_stripes = size; |
2d5b569b | 6258 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6259 | while (size < conf->max_nr_stripes && |
6260 | drop_one_stripe(conf)) | |
6261 | ; | |
2d5b569b | 6262 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6263 | |
edbe83ab | 6264 | |
b5470dc5 DW |
6265 | err = md_allow_write(mddev); |
6266 | if (err) | |
6267 | return err; | |
486f0644 | 6268 | |
2d5b569b | 6269 | mutex_lock(&conf->cache_size_mutex); |
486f0644 N |
6270 | while (size > conf->max_nr_stripes) |
6271 | if (!grow_one_stripe(conf, GFP_KERNEL)) | |
6272 | break; | |
2d5b569b | 6273 | mutex_unlock(&conf->cache_size_mutex); |
486f0644 | 6274 | |
c41d4ac4 N |
6275 | return 0; |
6276 | } | |
6277 | EXPORT_SYMBOL(raid5_set_cache_size); | |
6278 | ||
6279 | static ssize_t | |
fd01b88c | 6280 | raid5_store_stripe_cache_size(struct mddev *mddev, const char *page, size_t len) |
c41d4ac4 | 6281 | { |
6791875e | 6282 | struct r5conf *conf; |
c41d4ac4 N |
6283 | unsigned long new; |
6284 | int err; | |
6285 | ||
6286 | if (len >= PAGE_SIZE) | |
6287 | return -EINVAL; | |
b29bebd6 | 6288 | if (kstrtoul(page, 10, &new)) |
c41d4ac4 | 6289 | return -EINVAL; |
6791875e | 6290 | err = mddev_lock(mddev); |
c41d4ac4 N |
6291 | if (err) |
6292 | return err; | |
6791875e N |
6293 | conf = mddev->private; |
6294 | if (!conf) | |
6295 | err = -ENODEV; | |
6296 | else | |
6297 | err = raid5_set_cache_size(mddev, new); | |
6298 | mddev_unlock(mddev); | |
6299 | ||
6300 | return err ?: len; | |
3f294f4f | 6301 | } |
007583c9 | 6302 | |
96de1e66 N |
6303 | static struct md_sysfs_entry |
6304 | raid5_stripecache_size = __ATTR(stripe_cache_size, S_IRUGO | S_IWUSR, | |
6305 | raid5_show_stripe_cache_size, | |
6306 | raid5_store_stripe_cache_size); | |
3f294f4f | 6307 | |
d06f191f MS |
6308 | static ssize_t |
6309 | raid5_show_rmw_level(struct mddev *mddev, char *page) | |
6310 | { | |
6311 | struct r5conf *conf = mddev->private; | |
6312 | if (conf) | |
6313 | return sprintf(page, "%d\n", conf->rmw_level); | |
6314 | else | |
6315 | return 0; | |
6316 | } | |
6317 | ||
6318 | static ssize_t | |
6319 | raid5_store_rmw_level(struct mddev *mddev, const char *page, size_t len) | |
6320 | { | |
6321 | struct r5conf *conf = mddev->private; | |
6322 | unsigned long new; | |
6323 | ||
6324 | if (!conf) | |
6325 | return -ENODEV; | |
6326 | ||
6327 | if (len >= PAGE_SIZE) | |
6328 | return -EINVAL; | |
6329 | ||
6330 | if (kstrtoul(page, 10, &new)) | |
6331 | return -EINVAL; | |
6332 | ||
6333 | if (new != PARITY_DISABLE_RMW && !raid6_call.xor_syndrome) | |
6334 | return -EINVAL; | |
6335 | ||
6336 | if (new != PARITY_DISABLE_RMW && | |
6337 | new != PARITY_ENABLE_RMW && | |
6338 | new != PARITY_PREFER_RMW) | |
6339 | return -EINVAL; | |
6340 | ||
6341 | conf->rmw_level = new; | |
6342 | return len; | |
6343 | } | |
6344 | ||
6345 | static struct md_sysfs_entry | |
6346 | raid5_rmw_level = __ATTR(rmw_level, S_IRUGO | S_IWUSR, | |
6347 | raid5_show_rmw_level, | |
6348 | raid5_store_rmw_level); | |
6349 | ||
6350 | ||
8b3e6cdc | 6351 | static ssize_t |
fd01b88c | 6352 | raid5_show_preread_threshold(struct mddev *mddev, char *page) |
8b3e6cdc | 6353 | { |
7b1485ba N |
6354 | struct r5conf *conf; |
6355 | int ret = 0; | |
6356 | spin_lock(&mddev->lock); | |
6357 | conf = mddev->private; | |
8b3e6cdc | 6358 | if (conf) |
7b1485ba N |
6359 | ret = sprintf(page, "%d\n", conf->bypass_threshold); |
6360 | spin_unlock(&mddev->lock); | |
6361 | return ret; | |
8b3e6cdc DW |
6362 | } |
6363 | ||
6364 | static ssize_t | |
fd01b88c | 6365 | raid5_store_preread_threshold(struct mddev *mddev, const char *page, size_t len) |
8b3e6cdc | 6366 | { |
6791875e | 6367 | struct r5conf *conf; |
4ef197d8 | 6368 | unsigned long new; |
6791875e N |
6369 | int err; |
6370 | ||
8b3e6cdc DW |
6371 | if (len >= PAGE_SIZE) |
6372 | return -EINVAL; | |
b29bebd6 | 6373 | if (kstrtoul(page, 10, &new)) |
8b3e6cdc | 6374 | return -EINVAL; |
6791875e N |
6375 | |
6376 | err = mddev_lock(mddev); | |
6377 | if (err) | |
6378 | return err; | |
6379 | conf = mddev->private; | |
6380 | if (!conf) | |
6381 | err = -ENODEV; | |
edbe83ab | 6382 | else if (new > conf->min_nr_stripes) |
6791875e N |
6383 | err = -EINVAL; |
6384 | else | |
6385 | conf->bypass_threshold = new; | |
6386 | mddev_unlock(mddev); | |
6387 | return err ?: len; | |
8b3e6cdc DW |
6388 | } |
6389 | ||
6390 | static struct md_sysfs_entry | |
6391 | raid5_preread_bypass_threshold = __ATTR(preread_bypass_threshold, | |
6392 | S_IRUGO | S_IWUSR, | |
6393 | raid5_show_preread_threshold, | |
6394 | raid5_store_preread_threshold); | |
6395 | ||
d592a996 SL |
6396 | static ssize_t |
6397 | raid5_show_skip_copy(struct mddev *mddev, char *page) | |
6398 | { | |
7b1485ba N |
6399 | struct r5conf *conf; |
6400 | int ret = 0; | |
6401 | spin_lock(&mddev->lock); | |
6402 | conf = mddev->private; | |
d592a996 | 6403 | if (conf) |
7b1485ba N |
6404 | ret = sprintf(page, "%d\n", conf->skip_copy); |
6405 | spin_unlock(&mddev->lock); | |
6406 | return ret; | |
d592a996 SL |
6407 | } |
6408 | ||
6409 | static ssize_t | |
6410 | raid5_store_skip_copy(struct mddev *mddev, const char *page, size_t len) | |
6411 | { | |
6791875e | 6412 | struct r5conf *conf; |
d592a996 | 6413 | unsigned long new; |
6791875e N |
6414 | int err; |
6415 | ||
d592a996 SL |
6416 | if (len >= PAGE_SIZE) |
6417 | return -EINVAL; | |
d592a996 SL |
6418 | if (kstrtoul(page, 10, &new)) |
6419 | return -EINVAL; | |
6420 | new = !!new; | |
6791875e N |
6421 | |
6422 | err = mddev_lock(mddev); | |
6423 | if (err) | |
6424 | return err; | |
6425 | conf = mddev->private; | |
6426 | if (!conf) | |
6427 | err = -ENODEV; | |
6428 | else if (new != conf->skip_copy) { | |
6429 | mddev_suspend(mddev); | |
6430 | conf->skip_copy = new; | |
6431 | if (new) | |
dc3b17cc | 6432 | mddev->queue->backing_dev_info->capabilities |= |
6791875e N |
6433 | BDI_CAP_STABLE_WRITES; |
6434 | else | |
dc3b17cc | 6435 | mddev->queue->backing_dev_info->capabilities &= |
6791875e N |
6436 | ~BDI_CAP_STABLE_WRITES; |
6437 | mddev_resume(mddev); | |
6438 | } | |
6439 | mddev_unlock(mddev); | |
6440 | return err ?: len; | |
d592a996 SL |
6441 | } |
6442 | ||
6443 | static struct md_sysfs_entry | |
6444 | raid5_skip_copy = __ATTR(skip_copy, S_IRUGO | S_IWUSR, | |
6445 | raid5_show_skip_copy, | |
6446 | raid5_store_skip_copy); | |
6447 | ||
3f294f4f | 6448 | static ssize_t |
fd01b88c | 6449 | stripe_cache_active_show(struct mddev *mddev, char *page) |
3f294f4f | 6450 | { |
d1688a6d | 6451 | struct r5conf *conf = mddev->private; |
96de1e66 N |
6452 | if (conf) |
6453 | return sprintf(page, "%d\n", atomic_read(&conf->active_stripes)); | |
6454 | else | |
6455 | return 0; | |
3f294f4f N |
6456 | } |
6457 | ||
96de1e66 N |
6458 | static struct md_sysfs_entry |
6459 | raid5_stripecache_active = __ATTR_RO(stripe_cache_active); | |
3f294f4f | 6460 | |
b721420e SL |
6461 | static ssize_t |
6462 | raid5_show_group_thread_cnt(struct mddev *mddev, char *page) | |
6463 | { | |
7b1485ba N |
6464 | struct r5conf *conf; |
6465 | int ret = 0; | |
6466 | spin_lock(&mddev->lock); | |
6467 | conf = mddev->private; | |
b721420e | 6468 | if (conf) |
7b1485ba N |
6469 | ret = sprintf(page, "%d\n", conf->worker_cnt_per_group); |
6470 | spin_unlock(&mddev->lock); | |
6471 | return ret; | |
b721420e SL |
6472 | } |
6473 | ||
60aaf933 | 6474 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6475 | int *group_cnt, | |
6476 | int *worker_cnt_per_group, | |
6477 | struct r5worker_group **worker_groups); | |
b721420e SL |
6478 | static ssize_t |
6479 | raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len) | |
6480 | { | |
6791875e | 6481 | struct r5conf *conf; |
b721420e SL |
6482 | unsigned long new; |
6483 | int err; | |
60aaf933 | 6484 | struct r5worker_group *new_groups, *old_groups; |
6485 | int group_cnt, worker_cnt_per_group; | |
b721420e SL |
6486 | |
6487 | if (len >= PAGE_SIZE) | |
6488 | return -EINVAL; | |
b721420e SL |
6489 | if (kstrtoul(page, 10, &new)) |
6490 | return -EINVAL; | |
6491 | ||
6791875e N |
6492 | err = mddev_lock(mddev); |
6493 | if (err) | |
6494 | return err; | |
6495 | conf = mddev->private; | |
6496 | if (!conf) | |
6497 | err = -ENODEV; | |
6498 | else if (new != conf->worker_cnt_per_group) { | |
6499 | mddev_suspend(mddev); | |
b721420e | 6500 | |
6791875e N |
6501 | old_groups = conf->worker_groups; |
6502 | if (old_groups) | |
6503 | flush_workqueue(raid5_wq); | |
d206dcfa | 6504 | |
6791875e N |
6505 | err = alloc_thread_groups(conf, new, |
6506 | &group_cnt, &worker_cnt_per_group, | |
6507 | &new_groups); | |
6508 | if (!err) { | |
6509 | spin_lock_irq(&conf->device_lock); | |
6510 | conf->group_cnt = group_cnt; | |
6511 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6512 | conf->worker_groups = new_groups; | |
6513 | spin_unlock_irq(&conf->device_lock); | |
b721420e | 6514 | |
6791875e N |
6515 | if (old_groups) |
6516 | kfree(old_groups[0].workers); | |
6517 | kfree(old_groups); | |
6518 | } | |
6519 | mddev_resume(mddev); | |
b721420e | 6520 | } |
6791875e | 6521 | mddev_unlock(mddev); |
b721420e | 6522 | |
6791875e | 6523 | return err ?: len; |
b721420e SL |
6524 | } |
6525 | ||
6526 | static struct md_sysfs_entry | |
6527 | raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR, | |
6528 | raid5_show_group_thread_cnt, | |
6529 | raid5_store_group_thread_cnt); | |
6530 | ||
007583c9 | 6531 | static struct attribute *raid5_attrs[] = { |
3f294f4f N |
6532 | &raid5_stripecache_size.attr, |
6533 | &raid5_stripecache_active.attr, | |
8b3e6cdc | 6534 | &raid5_preread_bypass_threshold.attr, |
b721420e | 6535 | &raid5_group_thread_cnt.attr, |
d592a996 | 6536 | &raid5_skip_copy.attr, |
d06f191f | 6537 | &raid5_rmw_level.attr, |
2c7da14b | 6538 | &r5c_journal_mode.attr, |
3f294f4f N |
6539 | NULL, |
6540 | }; | |
007583c9 N |
6541 | static struct attribute_group raid5_attrs_group = { |
6542 | .name = NULL, | |
6543 | .attrs = raid5_attrs, | |
3f294f4f N |
6544 | }; |
6545 | ||
60aaf933 | 6546 | static int alloc_thread_groups(struct r5conf *conf, int cnt, |
6547 | int *group_cnt, | |
6548 | int *worker_cnt_per_group, | |
6549 | struct r5worker_group **worker_groups) | |
851c30c9 | 6550 | { |
566c09c5 | 6551 | int i, j, k; |
851c30c9 SL |
6552 | ssize_t size; |
6553 | struct r5worker *workers; | |
6554 | ||
60aaf933 | 6555 | *worker_cnt_per_group = cnt; |
851c30c9 | 6556 | if (cnt == 0) { |
60aaf933 | 6557 | *group_cnt = 0; |
6558 | *worker_groups = NULL; | |
851c30c9 SL |
6559 | return 0; |
6560 | } | |
60aaf933 | 6561 | *group_cnt = num_possible_nodes(); |
851c30c9 | 6562 | size = sizeof(struct r5worker) * cnt; |
60aaf933 | 6563 | workers = kzalloc(size * *group_cnt, GFP_NOIO); |
6564 | *worker_groups = kzalloc(sizeof(struct r5worker_group) * | |
6565 | *group_cnt, GFP_NOIO); | |
6566 | if (!*worker_groups || !workers) { | |
851c30c9 | 6567 | kfree(workers); |
60aaf933 | 6568 | kfree(*worker_groups); |
851c30c9 SL |
6569 | return -ENOMEM; |
6570 | } | |
6571 | ||
60aaf933 | 6572 | for (i = 0; i < *group_cnt; i++) { |
851c30c9 SL |
6573 | struct r5worker_group *group; |
6574 | ||
0c775d52 | 6575 | group = &(*worker_groups)[i]; |
851c30c9 | 6576 | INIT_LIST_HEAD(&group->handle_list); |
535ae4eb | 6577 | INIT_LIST_HEAD(&group->loprio_list); |
851c30c9 SL |
6578 | group->conf = conf; |
6579 | group->workers = workers + i * cnt; | |
6580 | ||
6581 | for (j = 0; j < cnt; j++) { | |
566c09c5 SL |
6582 | struct r5worker *worker = group->workers + j; |
6583 | worker->group = group; | |
6584 | INIT_WORK(&worker->work, raid5_do_work); | |
6585 | ||
6586 | for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++) | |
6587 | INIT_LIST_HEAD(worker->temp_inactive_list + k); | |
851c30c9 SL |
6588 | } |
6589 | } | |
6590 | ||
6591 | return 0; | |
6592 | } | |
6593 | ||
6594 | static void free_thread_groups(struct r5conf *conf) | |
6595 | { | |
6596 | if (conf->worker_groups) | |
6597 | kfree(conf->worker_groups[0].workers); | |
6598 | kfree(conf->worker_groups); | |
6599 | conf->worker_groups = NULL; | |
6600 | } | |
6601 | ||
80c3a6ce | 6602 | static sector_t |
fd01b88c | 6603 | raid5_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 6604 | { |
d1688a6d | 6605 | struct r5conf *conf = mddev->private; |
80c3a6ce DW |
6606 | |
6607 | if (!sectors) | |
6608 | sectors = mddev->dev_sectors; | |
5e5e3e78 | 6609 | if (!raid_disks) |
7ec05478 | 6610 | /* size is defined by the smallest of previous and new size */ |
5e5e3e78 | 6611 | raid_disks = min(conf->raid_disks, conf->previous_raid_disks); |
80c3a6ce | 6612 | |
3cb5edf4 N |
6613 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
6614 | sectors &= ~((sector_t)conf->prev_chunk_sectors - 1); | |
80c3a6ce DW |
6615 | return sectors * (raid_disks - conf->max_degraded); |
6616 | } | |
6617 | ||
789b5e03 ON |
6618 | static void free_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) |
6619 | { | |
6620 | safe_put_page(percpu->spare_page); | |
46d5b785 | 6621 | if (percpu->scribble) |
6622 | flex_array_free(percpu->scribble); | |
789b5e03 ON |
6623 | percpu->spare_page = NULL; |
6624 | percpu->scribble = NULL; | |
6625 | } | |
6626 | ||
6627 | static int alloc_scratch_buffer(struct r5conf *conf, struct raid5_percpu *percpu) | |
6628 | { | |
6629 | if (conf->level == 6 && !percpu->spare_page) | |
6630 | percpu->spare_page = alloc_page(GFP_KERNEL); | |
6631 | if (!percpu->scribble) | |
46d5b785 | 6632 | percpu->scribble = scribble_alloc(max(conf->raid_disks, |
738a2738 N |
6633 | conf->previous_raid_disks), |
6634 | max(conf->chunk_sectors, | |
6635 | conf->prev_chunk_sectors) | |
6636 | / STRIPE_SECTORS, | |
6637 | GFP_KERNEL); | |
789b5e03 ON |
6638 | |
6639 | if (!percpu->scribble || (conf->level == 6 && !percpu->spare_page)) { | |
6640 | free_scratch_buffer(conf, percpu); | |
6641 | return -ENOMEM; | |
6642 | } | |
6643 | ||
6644 | return 0; | |
6645 | } | |
6646 | ||
29c6d1bb | 6647 | static int raid456_cpu_dead(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6648 | { |
29c6d1bb SAS |
6649 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
6650 | ||
6651 | free_scratch_buffer(conf, per_cpu_ptr(conf->percpu, cpu)); | |
6652 | return 0; | |
6653 | } | |
36d1c647 | 6654 | |
29c6d1bb SAS |
6655 | static void raid5_free_percpu(struct r5conf *conf) |
6656 | { | |
36d1c647 DW |
6657 | if (!conf->percpu) |
6658 | return; | |
6659 | ||
29c6d1bb | 6660 | cpuhp_state_remove_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
36d1c647 DW |
6661 | free_percpu(conf->percpu); |
6662 | } | |
6663 | ||
d1688a6d | 6664 | static void free_conf(struct r5conf *conf) |
95fc17aa | 6665 | { |
d7bd398e SL |
6666 | int i; |
6667 | ||
5c7e81c3 SL |
6668 | if (conf->log) |
6669 | r5l_exit_log(conf->log); | |
30c89465 | 6670 | if (conf->shrinker.nr_deferred) |
edbe83ab | 6671 | unregister_shrinker(&conf->shrinker); |
5c7e81c3 | 6672 | |
851c30c9 | 6673 | free_thread_groups(conf); |
95fc17aa | 6674 | shrink_stripes(conf); |
36d1c647 | 6675 | raid5_free_percpu(conf); |
d7bd398e SL |
6676 | for (i = 0; i < conf->pool_size; i++) |
6677 | if (conf->disks[i].extra_page) | |
6678 | put_page(conf->disks[i].extra_page); | |
95fc17aa DW |
6679 | kfree(conf->disks); |
6680 | kfree(conf->stripe_hashtbl); | |
6681 | kfree(conf); | |
6682 | } | |
6683 | ||
29c6d1bb | 6684 | static int raid456_cpu_up_prepare(unsigned int cpu, struct hlist_node *node) |
36d1c647 | 6685 | { |
29c6d1bb | 6686 | struct r5conf *conf = hlist_entry_safe(node, struct r5conf, node); |
36d1c647 DW |
6687 | struct raid5_percpu *percpu = per_cpu_ptr(conf->percpu, cpu); |
6688 | ||
29c6d1bb | 6689 | if (alloc_scratch_buffer(conf, percpu)) { |
cc6167b4 N |
6690 | pr_warn("%s: failed memory allocation for cpu%u\n", |
6691 | __func__, cpu); | |
29c6d1bb | 6692 | return -ENOMEM; |
36d1c647 | 6693 | } |
29c6d1bb | 6694 | return 0; |
36d1c647 | 6695 | } |
36d1c647 | 6696 | |
d1688a6d | 6697 | static int raid5_alloc_percpu(struct r5conf *conf) |
36d1c647 | 6698 | { |
789b5e03 | 6699 | int err = 0; |
36d1c647 | 6700 | |
789b5e03 ON |
6701 | conf->percpu = alloc_percpu(struct raid5_percpu); |
6702 | if (!conf->percpu) | |
36d1c647 | 6703 | return -ENOMEM; |
789b5e03 | 6704 | |
29c6d1bb | 6705 | err = cpuhp_state_add_instance(CPUHP_MD_RAID5_PREPARE, &conf->node); |
27a353c0 SL |
6706 | if (!err) { |
6707 | conf->scribble_disks = max(conf->raid_disks, | |
6708 | conf->previous_raid_disks); | |
6709 | conf->scribble_sectors = max(conf->chunk_sectors, | |
6710 | conf->prev_chunk_sectors); | |
6711 | } | |
36d1c647 DW |
6712 | return err; |
6713 | } | |
6714 | ||
edbe83ab N |
6715 | static unsigned long raid5_cache_scan(struct shrinker *shrink, |
6716 | struct shrink_control *sc) | |
6717 | { | |
6718 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
2d5b569b N |
6719 | unsigned long ret = SHRINK_STOP; |
6720 | ||
6721 | if (mutex_trylock(&conf->cache_size_mutex)) { | |
6722 | ret= 0; | |
49895bcc N |
6723 | while (ret < sc->nr_to_scan && |
6724 | conf->max_nr_stripes > conf->min_nr_stripes) { | |
2d5b569b N |
6725 | if (drop_one_stripe(conf) == 0) { |
6726 | ret = SHRINK_STOP; | |
6727 | break; | |
6728 | } | |
6729 | ret++; | |
6730 | } | |
6731 | mutex_unlock(&conf->cache_size_mutex); | |
edbe83ab N |
6732 | } |
6733 | return ret; | |
6734 | } | |
6735 | ||
6736 | static unsigned long raid5_cache_count(struct shrinker *shrink, | |
6737 | struct shrink_control *sc) | |
6738 | { | |
6739 | struct r5conf *conf = container_of(shrink, struct r5conf, shrinker); | |
6740 | ||
6741 | if (conf->max_nr_stripes < conf->min_nr_stripes) | |
6742 | /* unlikely, but not impossible */ | |
6743 | return 0; | |
6744 | return conf->max_nr_stripes - conf->min_nr_stripes; | |
6745 | } | |
6746 | ||
d1688a6d | 6747 | static struct r5conf *setup_conf(struct mddev *mddev) |
1da177e4 | 6748 | { |
d1688a6d | 6749 | struct r5conf *conf; |
5e5e3e78 | 6750 | int raid_disk, memory, max_disks; |
3cb03002 | 6751 | struct md_rdev *rdev; |
1da177e4 | 6752 | struct disk_info *disk; |
0232605d | 6753 | char pers_name[6]; |
566c09c5 | 6754 | int i; |
60aaf933 | 6755 | int group_cnt, worker_cnt_per_group; |
6756 | struct r5worker_group *new_group; | |
1da177e4 | 6757 | |
91adb564 N |
6758 | if (mddev->new_level != 5 |
6759 | && mddev->new_level != 4 | |
6760 | && mddev->new_level != 6) { | |
cc6167b4 N |
6761 | pr_warn("md/raid:%s: raid level not set to 4/5/6 (%d)\n", |
6762 | mdname(mddev), mddev->new_level); | |
91adb564 | 6763 | return ERR_PTR(-EIO); |
1da177e4 | 6764 | } |
91adb564 N |
6765 | if ((mddev->new_level == 5 |
6766 | && !algorithm_valid_raid5(mddev->new_layout)) || | |
6767 | (mddev->new_level == 6 | |
6768 | && !algorithm_valid_raid6(mddev->new_layout))) { | |
cc6167b4 N |
6769 | pr_warn("md/raid:%s: layout %d not supported\n", |
6770 | mdname(mddev), mddev->new_layout); | |
91adb564 | 6771 | return ERR_PTR(-EIO); |
99c0fb5f | 6772 | } |
91adb564 | 6773 | if (mddev->new_level == 6 && mddev->raid_disks < 4) { |
cc6167b4 N |
6774 | pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n", |
6775 | mdname(mddev), mddev->raid_disks); | |
91adb564 | 6776 | return ERR_PTR(-EINVAL); |
4bbf3771 N |
6777 | } |
6778 | ||
664e7c41 AN |
6779 | if (!mddev->new_chunk_sectors || |
6780 | (mddev->new_chunk_sectors << 9) % PAGE_SIZE || | |
6781 | !is_power_of_2(mddev->new_chunk_sectors)) { | |
cc6167b4 N |
6782 | pr_warn("md/raid:%s: invalid chunk size %d\n", |
6783 | mdname(mddev), mddev->new_chunk_sectors << 9); | |
91adb564 | 6784 | return ERR_PTR(-EINVAL); |
f6705578 N |
6785 | } |
6786 | ||
d1688a6d | 6787 | conf = kzalloc(sizeof(struct r5conf), GFP_KERNEL); |
91adb564 | 6788 | if (conf == NULL) |
1da177e4 | 6789 | goto abort; |
851c30c9 | 6790 | /* Don't enable multi-threading by default*/ |
60aaf933 | 6791 | if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group, |
6792 | &new_group)) { | |
6793 | conf->group_cnt = group_cnt; | |
6794 | conf->worker_cnt_per_group = worker_cnt_per_group; | |
6795 | conf->worker_groups = new_group; | |
6796 | } else | |
851c30c9 | 6797 | goto abort; |
f5efd45a | 6798 | spin_lock_init(&conf->device_lock); |
c46501b2 | 6799 | seqcount_init(&conf->gen_lock); |
2d5b569b | 6800 | mutex_init(&conf->cache_size_mutex); |
b1b46486 | 6801 | init_waitqueue_head(&conf->wait_for_quiescent); |
6ab2a4b8 | 6802 | init_waitqueue_head(&conf->wait_for_stripe); |
f5efd45a DW |
6803 | init_waitqueue_head(&conf->wait_for_overlap); |
6804 | INIT_LIST_HEAD(&conf->handle_list); | |
535ae4eb | 6805 | INIT_LIST_HEAD(&conf->loprio_list); |
f5efd45a DW |
6806 | INIT_LIST_HEAD(&conf->hold_list); |
6807 | INIT_LIST_HEAD(&conf->delayed_list); | |
6808 | INIT_LIST_HEAD(&conf->bitmap_list); | |
c3cce6cd | 6809 | bio_list_init(&conf->return_bi); |
773ca82f | 6810 | init_llist_head(&conf->released_stripes); |
f5efd45a DW |
6811 | atomic_set(&conf->active_stripes, 0); |
6812 | atomic_set(&conf->preread_active_stripes, 0); | |
6813 | atomic_set(&conf->active_aligned_reads, 0); | |
765d704d SL |
6814 | bio_list_init(&conf->pending_bios); |
6815 | spin_lock_init(&conf->pending_bios_lock); | |
6816 | conf->batch_bio_dispatch = true; | |
6817 | rdev_for_each(rdev, mddev) { | |
6818 | if (test_bit(Journal, &rdev->flags)) | |
6819 | continue; | |
6820 | if (blk_queue_nonrot(bdev_get_queue(rdev->bdev))) { | |
6821 | conf->batch_bio_dispatch = false; | |
6822 | break; | |
6823 | } | |
6824 | } | |
6825 | ||
f5efd45a | 6826 | conf->bypass_threshold = BYPASS_THRESHOLD; |
d890fa2b | 6827 | conf->recovery_disabled = mddev->recovery_disabled - 1; |
91adb564 N |
6828 | |
6829 | conf->raid_disks = mddev->raid_disks; | |
6830 | if (mddev->reshape_position == MaxSector) | |
6831 | conf->previous_raid_disks = mddev->raid_disks; | |
6832 | else | |
f6705578 | 6833 | conf->previous_raid_disks = mddev->raid_disks - mddev->delta_disks; |
5e5e3e78 | 6834 | max_disks = max(conf->raid_disks, conf->previous_raid_disks); |
f6705578 | 6835 | |
5e5e3e78 | 6836 | conf->disks = kzalloc(max_disks * sizeof(struct disk_info), |
b55e6bfc | 6837 | GFP_KERNEL); |
d7bd398e | 6838 | |
b55e6bfc N |
6839 | if (!conf->disks) |
6840 | goto abort; | |
9ffae0cf | 6841 | |
d7bd398e SL |
6842 | for (i = 0; i < max_disks; i++) { |
6843 | conf->disks[i].extra_page = alloc_page(GFP_KERNEL); | |
6844 | if (!conf->disks[i].extra_page) | |
6845 | goto abort; | |
6846 | } | |
6847 | ||
1da177e4 LT |
6848 | conf->mddev = mddev; |
6849 | ||
fccddba0 | 6850 | if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL) |
1da177e4 | 6851 | goto abort; |
1da177e4 | 6852 | |
566c09c5 SL |
6853 | /* We init hash_locks[0] separately to that it can be used |
6854 | * as the reference lock in the spin_lock_nest_lock() call | |
6855 | * in lock_all_device_hash_locks_irq in order to convince | |
6856 | * lockdep that we know what we are doing. | |
6857 | */ | |
6858 | spin_lock_init(conf->hash_locks); | |
6859 | for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++) | |
6860 | spin_lock_init(conf->hash_locks + i); | |
6861 | ||
6862 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6863 | INIT_LIST_HEAD(conf->inactive_list + i); | |
6864 | ||
6865 | for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) | |
6866 | INIT_LIST_HEAD(conf->temp_inactive_list + i); | |
6867 | ||
1e6d690b SL |
6868 | atomic_set(&conf->r5c_cached_full_stripes, 0); |
6869 | INIT_LIST_HEAD(&conf->r5c_full_stripe_list); | |
6870 | atomic_set(&conf->r5c_cached_partial_stripes, 0); | |
6871 | INIT_LIST_HEAD(&conf->r5c_partial_stripe_list); | |
e33fbb9c SL |
6872 | atomic_set(&conf->r5c_flushing_full_stripes, 0); |
6873 | atomic_set(&conf->r5c_flushing_partial_stripes, 0); | |
1e6d690b | 6874 | |
36d1c647 | 6875 | conf->level = mddev->new_level; |
46d5b785 | 6876 | conf->chunk_sectors = mddev->new_chunk_sectors; |
36d1c647 DW |
6877 | if (raid5_alloc_percpu(conf) != 0) |
6878 | goto abort; | |
6879 | ||
0c55e022 | 6880 | pr_debug("raid456: run(%s) called.\n", mdname(mddev)); |
1da177e4 | 6881 | |
dafb20fa | 6882 | rdev_for_each(rdev, mddev) { |
1da177e4 | 6883 | raid_disk = rdev->raid_disk; |
5e5e3e78 | 6884 | if (raid_disk >= max_disks |
f2076e7d | 6885 | || raid_disk < 0 || test_bit(Journal, &rdev->flags)) |
1da177e4 LT |
6886 | continue; |
6887 | disk = conf->disks + raid_disk; | |
6888 | ||
17045f52 N |
6889 | if (test_bit(Replacement, &rdev->flags)) { |
6890 | if (disk->replacement) | |
6891 | goto abort; | |
6892 | disk->replacement = rdev; | |
6893 | } else { | |
6894 | if (disk->rdev) | |
6895 | goto abort; | |
6896 | disk->rdev = rdev; | |
6897 | } | |
1da177e4 | 6898 | |
b2d444d7 | 6899 | if (test_bit(In_sync, &rdev->flags)) { |
1da177e4 | 6900 | char b[BDEVNAME_SIZE]; |
cc6167b4 N |
6901 | pr_info("md/raid:%s: device %s operational as raid disk %d\n", |
6902 | mdname(mddev), bdevname(rdev->bdev, b), raid_disk); | |
d6b212f4 | 6903 | } else if (rdev->saved_raid_disk != raid_disk) |
8c2e870a NB |
6904 | /* Cannot rely on bitmap to complete recovery */ |
6905 | conf->fullsync = 1; | |
1da177e4 LT |
6906 | } |
6907 | ||
91adb564 | 6908 | conf->level = mddev->new_level; |
584acdd4 | 6909 | if (conf->level == 6) { |
16a53ecc | 6910 | conf->max_degraded = 2; |
584acdd4 MS |
6911 | if (raid6_call.xor_syndrome) |
6912 | conf->rmw_level = PARITY_ENABLE_RMW; | |
6913 | else | |
6914 | conf->rmw_level = PARITY_DISABLE_RMW; | |
6915 | } else { | |
16a53ecc | 6916 | conf->max_degraded = 1; |
584acdd4 MS |
6917 | conf->rmw_level = PARITY_ENABLE_RMW; |
6918 | } | |
91adb564 | 6919 | conf->algorithm = mddev->new_layout; |
fef9c61f | 6920 | conf->reshape_progress = mddev->reshape_position; |
e183eaed | 6921 | if (conf->reshape_progress != MaxSector) { |
09c9e5fa | 6922 | conf->prev_chunk_sectors = mddev->chunk_sectors; |
e183eaed | 6923 | conf->prev_algo = mddev->layout; |
5cac6bcb N |
6924 | } else { |
6925 | conf->prev_chunk_sectors = conf->chunk_sectors; | |
6926 | conf->prev_algo = conf->algorithm; | |
e183eaed | 6927 | } |
1da177e4 | 6928 | |
edbe83ab | 6929 | conf->min_nr_stripes = NR_STRIPES; |
ad5b0f76 SL |
6930 | if (mddev->reshape_position != MaxSector) { |
6931 | int stripes = max_t(int, | |
6932 | ((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4, | |
6933 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4); | |
6934 | conf->min_nr_stripes = max(NR_STRIPES, stripes); | |
6935 | if (conf->min_nr_stripes != NR_STRIPES) | |
cc6167b4 | 6936 | pr_info("md/raid:%s: force stripe size %d for reshape\n", |
ad5b0f76 SL |
6937 | mdname(mddev), conf->min_nr_stripes); |
6938 | } | |
edbe83ab | 6939 | memory = conf->min_nr_stripes * (sizeof(struct stripe_head) + |
5e5e3e78 | 6940 | max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; |
4bda556a | 6941 | atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS); |
edbe83ab | 6942 | if (grow_stripes(conf, conf->min_nr_stripes)) { |
cc6167b4 N |
6943 | pr_warn("md/raid:%s: couldn't allocate %dkB for buffers\n", |
6944 | mdname(mddev), memory); | |
91adb564 N |
6945 | goto abort; |
6946 | } else | |
cc6167b4 | 6947 | pr_debug("md/raid:%s: allocated %dkB\n", mdname(mddev), memory); |
edbe83ab N |
6948 | /* |
6949 | * Losing a stripe head costs more than the time to refill it, | |
6950 | * it reduces the queue depth and so can hurt throughput. | |
6951 | * So set it rather large, scaled by number of devices. | |
6952 | */ | |
6953 | conf->shrinker.seeks = DEFAULT_SEEKS * conf->raid_disks * 4; | |
6954 | conf->shrinker.scan_objects = raid5_cache_scan; | |
6955 | conf->shrinker.count_objects = raid5_cache_count; | |
6956 | conf->shrinker.batch = 128; | |
6957 | conf->shrinker.flags = 0; | |
6a0f53ff | 6958 | if (register_shrinker(&conf->shrinker)) { |
cc6167b4 N |
6959 | pr_warn("md/raid:%s: couldn't register shrinker.\n", |
6960 | mdname(mddev)); | |
6a0f53ff CY |
6961 | goto abort; |
6962 | } | |
1da177e4 | 6963 | |
0232605d N |
6964 | sprintf(pers_name, "raid%d", mddev->new_level); |
6965 | conf->thread = md_register_thread(raid5d, mddev, pers_name); | |
91adb564 | 6966 | if (!conf->thread) { |
cc6167b4 N |
6967 | pr_warn("md/raid:%s: couldn't allocate thread.\n", |
6968 | mdname(mddev)); | |
16a53ecc N |
6969 | goto abort; |
6970 | } | |
91adb564 N |
6971 | |
6972 | return conf; | |
6973 | ||
6974 | abort: | |
6975 | if (conf) { | |
95fc17aa | 6976 | free_conf(conf); |
91adb564 N |
6977 | return ERR_PTR(-EIO); |
6978 | } else | |
6979 | return ERR_PTR(-ENOMEM); | |
6980 | } | |
6981 | ||
c148ffdc N |
6982 | static int only_parity(int raid_disk, int algo, int raid_disks, int max_degraded) |
6983 | { | |
6984 | switch (algo) { | |
6985 | case ALGORITHM_PARITY_0: | |
6986 | if (raid_disk < max_degraded) | |
6987 | return 1; | |
6988 | break; | |
6989 | case ALGORITHM_PARITY_N: | |
6990 | if (raid_disk >= raid_disks - max_degraded) | |
6991 | return 1; | |
6992 | break; | |
6993 | case ALGORITHM_PARITY_0_6: | |
f72ffdd6 | 6994 | if (raid_disk == 0 || |
c148ffdc N |
6995 | raid_disk == raid_disks - 1) |
6996 | return 1; | |
6997 | break; | |
6998 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
6999 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
7000 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
7001 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
7002 | if (raid_disk == raid_disks - 1) | |
7003 | return 1; | |
7004 | } | |
7005 | return 0; | |
7006 | } | |
7007 | ||
849674e4 | 7008 | static int raid5_run(struct mddev *mddev) |
91adb564 | 7009 | { |
d1688a6d | 7010 | struct r5conf *conf; |
9f7c2220 | 7011 | int working_disks = 0; |
c148ffdc | 7012 | int dirty_parity_disks = 0; |
3cb03002 | 7013 | struct md_rdev *rdev; |
713cf5a6 | 7014 | struct md_rdev *journal_dev = NULL; |
c148ffdc | 7015 | sector_t reshape_offset = 0; |
17045f52 | 7016 | int i; |
b5254dd5 N |
7017 | long long min_offset_diff = 0; |
7018 | int first = 1; | |
91adb564 | 7019 | |
8c6ac868 | 7020 | if (mddev->recovery_cp != MaxSector) |
cc6167b4 N |
7021 | pr_notice("md/raid:%s: not clean -- starting background reconstruction\n", |
7022 | mdname(mddev)); | |
b5254dd5 N |
7023 | |
7024 | rdev_for_each(rdev, mddev) { | |
7025 | long long diff; | |
713cf5a6 | 7026 | |
f2076e7d | 7027 | if (test_bit(Journal, &rdev->flags)) { |
713cf5a6 | 7028 | journal_dev = rdev; |
f2076e7d SL |
7029 | continue; |
7030 | } | |
b5254dd5 N |
7031 | if (rdev->raid_disk < 0) |
7032 | continue; | |
7033 | diff = (rdev->new_data_offset - rdev->data_offset); | |
7034 | if (first) { | |
7035 | min_offset_diff = diff; | |
7036 | first = 0; | |
7037 | } else if (mddev->reshape_backwards && | |
7038 | diff < min_offset_diff) | |
7039 | min_offset_diff = diff; | |
7040 | else if (!mddev->reshape_backwards && | |
7041 | diff > min_offset_diff) | |
7042 | min_offset_diff = diff; | |
7043 | } | |
7044 | ||
91adb564 N |
7045 | if (mddev->reshape_position != MaxSector) { |
7046 | /* Check that we can continue the reshape. | |
b5254dd5 N |
7047 | * Difficulties arise if the stripe we would write to |
7048 | * next is at or after the stripe we would read from next. | |
7049 | * For a reshape that changes the number of devices, this | |
7050 | * is only possible for a very short time, and mdadm makes | |
7051 | * sure that time appears to have past before assembling | |
7052 | * the array. So we fail if that time hasn't passed. | |
7053 | * For a reshape that keeps the number of devices the same | |
7054 | * mdadm must be monitoring the reshape can keeping the | |
7055 | * critical areas read-only and backed up. It will start | |
7056 | * the array in read-only mode, so we check for that. | |
91adb564 N |
7057 | */ |
7058 | sector_t here_new, here_old; | |
7059 | int old_disks; | |
18b00334 | 7060 | int max_degraded = (mddev->level == 6 ? 2 : 1); |
05256d98 N |
7061 | int chunk_sectors; |
7062 | int new_data_disks; | |
91adb564 | 7063 | |
713cf5a6 | 7064 | if (journal_dev) { |
cc6167b4 N |
7065 | pr_warn("md/raid:%s: don't support reshape with journal - aborting.\n", |
7066 | mdname(mddev)); | |
713cf5a6 SL |
7067 | return -EINVAL; |
7068 | } | |
7069 | ||
88ce4930 | 7070 | if (mddev->new_level != mddev->level) { |
cc6167b4 N |
7071 | pr_warn("md/raid:%s: unsupported reshape required - aborting.\n", |
7072 | mdname(mddev)); | |
91adb564 N |
7073 | return -EINVAL; |
7074 | } | |
91adb564 N |
7075 | old_disks = mddev->raid_disks - mddev->delta_disks; |
7076 | /* reshape_position must be on a new-stripe boundary, and one | |
7077 | * further up in new geometry must map after here in old | |
7078 | * geometry. | |
05256d98 N |
7079 | * If the chunk sizes are different, then as we perform reshape |
7080 | * in units of the largest of the two, reshape_position needs | |
7081 | * be a multiple of the largest chunk size times new data disks. | |
91adb564 N |
7082 | */ |
7083 | here_new = mddev->reshape_position; | |
05256d98 N |
7084 | chunk_sectors = max(mddev->chunk_sectors, mddev->new_chunk_sectors); |
7085 | new_data_disks = mddev->raid_disks - max_degraded; | |
7086 | if (sector_div(here_new, chunk_sectors * new_data_disks)) { | |
cc6167b4 N |
7087 | pr_warn("md/raid:%s: reshape_position not on a stripe boundary\n", |
7088 | mdname(mddev)); | |
91adb564 N |
7089 | return -EINVAL; |
7090 | } | |
05256d98 | 7091 | reshape_offset = here_new * chunk_sectors; |
91adb564 N |
7092 | /* here_new is the stripe we will write to */ |
7093 | here_old = mddev->reshape_position; | |
05256d98 | 7094 | sector_div(here_old, chunk_sectors * (old_disks-max_degraded)); |
91adb564 N |
7095 | /* here_old is the first stripe that we might need to read |
7096 | * from */ | |
67ac6011 N |
7097 | if (mddev->delta_disks == 0) { |
7098 | /* We cannot be sure it is safe to start an in-place | |
b5254dd5 | 7099 | * reshape. It is only safe if user-space is monitoring |
67ac6011 N |
7100 | * and taking constant backups. |
7101 | * mdadm always starts a situation like this in | |
7102 | * readonly mode so it can take control before | |
7103 | * allowing any writes. So just check for that. | |
7104 | */ | |
b5254dd5 N |
7105 | if (abs(min_offset_diff) >= mddev->chunk_sectors && |
7106 | abs(min_offset_diff) >= mddev->new_chunk_sectors) | |
7107 | /* not really in-place - so OK */; | |
7108 | else if (mddev->ro == 0) { | |
cc6167b4 N |
7109 | pr_warn("md/raid:%s: in-place reshape must be started in read-only mode - aborting\n", |
7110 | mdname(mddev)); | |
67ac6011 N |
7111 | return -EINVAL; |
7112 | } | |
2c810cdd | 7113 | } else if (mddev->reshape_backwards |
05256d98 N |
7114 | ? (here_new * chunk_sectors + min_offset_diff <= |
7115 | here_old * chunk_sectors) | |
7116 | : (here_new * chunk_sectors >= | |
7117 | here_old * chunk_sectors + (-min_offset_diff))) { | |
91adb564 | 7118 | /* Reading from the same stripe as writing to - bad */ |
cc6167b4 N |
7119 | pr_warn("md/raid:%s: reshape_position too early for auto-recovery - aborting.\n", |
7120 | mdname(mddev)); | |
91adb564 N |
7121 | return -EINVAL; |
7122 | } | |
cc6167b4 | 7123 | pr_debug("md/raid:%s: reshape will continue\n", mdname(mddev)); |
91adb564 N |
7124 | /* OK, we should be able to continue; */ |
7125 | } else { | |
7126 | BUG_ON(mddev->level != mddev->new_level); | |
7127 | BUG_ON(mddev->layout != mddev->new_layout); | |
664e7c41 | 7128 | BUG_ON(mddev->chunk_sectors != mddev->new_chunk_sectors); |
91adb564 | 7129 | BUG_ON(mddev->delta_disks != 0); |
1da177e4 | 7130 | } |
91adb564 | 7131 | |
245f46c2 N |
7132 | if (mddev->private == NULL) |
7133 | conf = setup_conf(mddev); | |
7134 | else | |
7135 | conf = mddev->private; | |
7136 | ||
91adb564 N |
7137 | if (IS_ERR(conf)) |
7138 | return PTR_ERR(conf); | |
7139 | ||
486b0f7b SL |
7140 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
7141 | if (!journal_dev) { | |
cc6167b4 N |
7142 | pr_warn("md/raid:%s: journal disk is missing, force array readonly\n", |
7143 | mdname(mddev)); | |
486b0f7b SL |
7144 | mddev->ro = 1; |
7145 | set_disk_ro(mddev->gendisk, 1); | |
7146 | } else if (mddev->recovery_cp == MaxSector) | |
7147 | set_bit(MD_JOURNAL_CLEAN, &mddev->flags); | |
7dde2ad3 SL |
7148 | } |
7149 | ||
b5254dd5 | 7150 | conf->min_offset_diff = min_offset_diff; |
91adb564 N |
7151 | mddev->thread = conf->thread; |
7152 | conf->thread = NULL; | |
7153 | mddev->private = conf; | |
7154 | ||
17045f52 N |
7155 | for (i = 0; i < conf->raid_disks && conf->previous_raid_disks; |
7156 | i++) { | |
7157 | rdev = conf->disks[i].rdev; | |
7158 | if (!rdev && conf->disks[i].replacement) { | |
7159 | /* The replacement is all we have yet */ | |
7160 | rdev = conf->disks[i].replacement; | |
7161 | conf->disks[i].replacement = NULL; | |
7162 | clear_bit(Replacement, &rdev->flags); | |
7163 | conf->disks[i].rdev = rdev; | |
7164 | } | |
7165 | if (!rdev) | |
c148ffdc | 7166 | continue; |
17045f52 N |
7167 | if (conf->disks[i].replacement && |
7168 | conf->reshape_progress != MaxSector) { | |
7169 | /* replacements and reshape simply do not mix. */ | |
cc6167b4 | 7170 | pr_warn("md: cannot handle concurrent replacement and reshape.\n"); |
17045f52 N |
7171 | goto abort; |
7172 | } | |
2f115882 | 7173 | if (test_bit(In_sync, &rdev->flags)) { |
91adb564 | 7174 | working_disks++; |
2f115882 N |
7175 | continue; |
7176 | } | |
c148ffdc N |
7177 | /* This disc is not fully in-sync. However if it |
7178 | * just stored parity (beyond the recovery_offset), | |
7179 | * when we don't need to be concerned about the | |
7180 | * array being dirty. | |
7181 | * When reshape goes 'backwards', we never have | |
7182 | * partially completed devices, so we only need | |
7183 | * to worry about reshape going forwards. | |
7184 | */ | |
7185 | /* Hack because v0.91 doesn't store recovery_offset properly. */ | |
7186 | if (mddev->major_version == 0 && | |
7187 | mddev->minor_version > 90) | |
7188 | rdev->recovery_offset = reshape_offset; | |
5026d7a9 | 7189 | |
c148ffdc N |
7190 | if (rdev->recovery_offset < reshape_offset) { |
7191 | /* We need to check old and new layout */ | |
7192 | if (!only_parity(rdev->raid_disk, | |
7193 | conf->algorithm, | |
7194 | conf->raid_disks, | |
7195 | conf->max_degraded)) | |
7196 | continue; | |
7197 | } | |
7198 | if (!only_parity(rdev->raid_disk, | |
7199 | conf->prev_algo, | |
7200 | conf->previous_raid_disks, | |
7201 | conf->max_degraded)) | |
7202 | continue; | |
7203 | dirty_parity_disks++; | |
7204 | } | |
91adb564 | 7205 | |
17045f52 N |
7206 | /* |
7207 | * 0 for a fully functional array, 1 or 2 for a degraded array. | |
7208 | */ | |
2e38a37f | 7209 | mddev->degraded = raid5_calc_degraded(conf); |
91adb564 | 7210 | |
674806d6 | 7211 | if (has_failed(conf)) { |
cc6167b4 | 7212 | pr_crit("md/raid:%s: not enough operational devices (%d/%d failed)\n", |
02c2de8c | 7213 | mdname(mddev), mddev->degraded, conf->raid_disks); |
1da177e4 LT |
7214 | goto abort; |
7215 | } | |
7216 | ||
91adb564 | 7217 | /* device size must be a multiple of chunk size */ |
9d8f0363 | 7218 | mddev->dev_sectors &= ~(mddev->chunk_sectors - 1); |
91adb564 N |
7219 | mddev->resync_max_sectors = mddev->dev_sectors; |
7220 | ||
c148ffdc | 7221 | if (mddev->degraded > dirty_parity_disks && |
1da177e4 | 7222 | mddev->recovery_cp != MaxSector) { |
6ff8d8ec | 7223 | if (mddev->ok_start_degraded) |
cc6167b4 N |
7224 | pr_crit("md/raid:%s: starting dirty degraded array - data corruption possible.\n", |
7225 | mdname(mddev)); | |
6ff8d8ec | 7226 | else { |
cc6167b4 N |
7227 | pr_crit("md/raid:%s: cannot start dirty degraded array.\n", |
7228 | mdname(mddev)); | |
6ff8d8ec N |
7229 | goto abort; |
7230 | } | |
1da177e4 LT |
7231 | } |
7232 | ||
cc6167b4 N |
7233 | pr_info("md/raid:%s: raid level %d active with %d out of %d devices, algorithm %d\n", |
7234 | mdname(mddev), conf->level, | |
7235 | mddev->raid_disks-mddev->degraded, mddev->raid_disks, | |
7236 | mddev->new_layout); | |
1da177e4 LT |
7237 | |
7238 | print_raid5_conf(conf); | |
7239 | ||
fef9c61f | 7240 | if (conf->reshape_progress != MaxSector) { |
fef9c61f | 7241 | conf->reshape_safe = conf->reshape_progress; |
f6705578 N |
7242 | atomic_set(&conf->reshape_stripes, 0); |
7243 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
7244 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
7245 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
7246 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7247 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7248 | "reshape"); |
f6705578 N |
7249 | } |
7250 | ||
1da177e4 | 7251 | /* Ok, everything is just fine now */ |
a64c876f N |
7252 | if (mddev->to_remove == &raid5_attrs_group) |
7253 | mddev->to_remove = NULL; | |
00bcb4ac N |
7254 | else if (mddev->kobj.sd && |
7255 | sysfs_create_group(&mddev->kobj, &raid5_attrs_group)) | |
cc6167b4 N |
7256 | pr_warn("raid5: failed to create sysfs attributes for %s\n", |
7257 | mdname(mddev)); | |
4a5add49 | 7258 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); |
7a5febe9 | 7259 | |
4a5add49 | 7260 | if (mddev->queue) { |
9f7c2220 | 7261 | int chunk_size; |
620125f2 | 7262 | bool discard_supported = true; |
4a5add49 N |
7263 | /* read-ahead size must cover two whole stripes, which |
7264 | * is 2 * (datadisks) * chunksize where 'n' is the | |
7265 | * number of raid devices | |
7266 | */ | |
7267 | int data_disks = conf->previous_raid_disks - conf->max_degraded; | |
7268 | int stripe = data_disks * | |
7269 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); | |
dc3b17cc JK |
7270 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7271 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
91adb564 | 7272 | |
9f7c2220 N |
7273 | chunk_size = mddev->chunk_sectors << 9; |
7274 | blk_queue_io_min(mddev->queue, chunk_size); | |
7275 | blk_queue_io_opt(mddev->queue, chunk_size * | |
7276 | (conf->raid_disks - conf->max_degraded)); | |
c78afc62 | 7277 | mddev->queue->limits.raid_partial_stripes_expensive = 1; |
620125f2 SL |
7278 | /* |
7279 | * We can only discard a whole stripe. It doesn't make sense to | |
7280 | * discard data disk but write parity disk | |
7281 | */ | |
7282 | stripe = stripe * PAGE_SIZE; | |
4ac6875e N |
7283 | /* Round up to power of 2, as discard handling |
7284 | * currently assumes that */ | |
7285 | while ((stripe-1) & stripe) | |
7286 | stripe = (stripe | (stripe-1)) + 1; | |
620125f2 SL |
7287 | mddev->queue->limits.discard_alignment = stripe; |
7288 | mddev->queue->limits.discard_granularity = stripe; | |
e8d7c332 KK |
7289 | |
7290 | /* | |
7291 | * We use 16-bit counter of active stripes in bi_phys_segments | |
7292 | * (minus one for over-loaded initialization) | |
7293 | */ | |
7294 | blk_queue_max_hw_sectors(mddev->queue, 0xfffe * STRIPE_SECTORS); | |
7295 | blk_queue_max_discard_sectors(mddev->queue, | |
7296 | 0xfffe * STRIPE_SECTORS); | |
7297 | ||
620125f2 SL |
7298 | /* |
7299 | * unaligned part of discard request will be ignored, so can't | |
8e0e99ba | 7300 | * guarantee discard_zeroes_data |
620125f2 SL |
7301 | */ |
7302 | mddev->queue->limits.discard_zeroes_data = 0; | |
8f6c2e4b | 7303 | |
5026d7a9 PA |
7304 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
7305 | ||
05616be5 | 7306 | rdev_for_each(rdev, mddev) { |
9f7c2220 N |
7307 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7308 | rdev->data_offset << 9); | |
05616be5 N |
7309 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
7310 | rdev->new_data_offset << 9); | |
620125f2 SL |
7311 | /* |
7312 | * discard_zeroes_data is required, otherwise data | |
7313 | * could be lost. Consider a scenario: discard a stripe | |
7314 | * (the stripe could be inconsistent if | |
7315 | * discard_zeroes_data is 0); write one disk of the | |
7316 | * stripe (the stripe could be inconsistent again | |
7317 | * depending on which disks are used to calculate | |
7318 | * parity); the disk is broken; The stripe data of this | |
7319 | * disk is lost. | |
7320 | */ | |
7321 | if (!blk_queue_discard(bdev_get_queue(rdev->bdev)) || | |
7322 | !bdev_get_queue(rdev->bdev)-> | |
7323 | limits.discard_zeroes_data) | |
7324 | discard_supported = false; | |
8e0e99ba N |
7325 | /* Unfortunately, discard_zeroes_data is not currently |
7326 | * a guarantee - just a hint. So we only allow DISCARD | |
7327 | * if the sysadmin has confirmed that only safe devices | |
7328 | * are in use by setting a module parameter. | |
7329 | */ | |
7330 | if (!devices_handle_discard_safely) { | |
7331 | if (discard_supported) { | |
7332 | pr_info("md/raid456: discard support disabled due to uncertainty.\n"); | |
7333 | pr_info("Set raid456.devices_handle_discard_safely=Y to override.\n"); | |
7334 | } | |
7335 | discard_supported = false; | |
7336 | } | |
05616be5 | 7337 | } |
620125f2 SL |
7338 | |
7339 | if (discard_supported && | |
e7597e69 JS |
7340 | mddev->queue->limits.max_discard_sectors >= (stripe >> 9) && |
7341 | mddev->queue->limits.discard_granularity >= stripe) | |
620125f2 SL |
7342 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, |
7343 | mddev->queue); | |
7344 | else | |
7345 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
7346 | mddev->queue); | |
1dffdddd SL |
7347 | |
7348 | blk_queue_max_hw_sectors(mddev->queue, UINT_MAX); | |
9f7c2220 | 7349 | } |
23032a0e | 7350 | |
5c7e81c3 SL |
7351 | if (journal_dev) { |
7352 | char b[BDEVNAME_SIZE]; | |
7353 | ||
cc6167b4 N |
7354 | pr_debug("md/raid:%s: using device %s as journal\n", |
7355 | mdname(mddev), bdevname(journal_dev->bdev, b)); | |
5aabf7c4 SL |
7356 | if (r5l_init_log(conf, journal_dev)) |
7357 | goto abort; | |
5c7e81c3 SL |
7358 | } |
7359 | ||
1da177e4 LT |
7360 | return 0; |
7361 | abort: | |
01f96c0a | 7362 | md_unregister_thread(&mddev->thread); |
e4f869d9 N |
7363 | print_raid5_conf(conf); |
7364 | free_conf(conf); | |
1da177e4 | 7365 | mddev->private = NULL; |
cc6167b4 | 7366 | pr_warn("md/raid:%s: failed to run raid set.\n", mdname(mddev)); |
1da177e4 LT |
7367 | return -EIO; |
7368 | } | |
7369 | ||
afa0f557 | 7370 | static void raid5_free(struct mddev *mddev, void *priv) |
1da177e4 | 7371 | { |
afa0f557 | 7372 | struct r5conf *conf = priv; |
1da177e4 | 7373 | |
95fc17aa | 7374 | free_conf(conf); |
a64c876f | 7375 | mddev->to_remove = &raid5_attrs_group; |
1da177e4 LT |
7376 | } |
7377 | ||
849674e4 | 7378 | static void raid5_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 7379 | { |
d1688a6d | 7380 | struct r5conf *conf = mddev->private; |
1da177e4 LT |
7381 | int i; |
7382 | ||
9d8f0363 | 7383 | seq_printf(seq, " level %d, %dk chunk, algorithm %d", mddev->level, |
3cb5edf4 | 7384 | conf->chunk_sectors / 2, mddev->layout); |
02c2de8c | 7385 | seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->raid_disks - mddev->degraded); |
5fd13351 N |
7386 | rcu_read_lock(); |
7387 | for (i = 0; i < conf->raid_disks; i++) { | |
7388 | struct md_rdev *rdev = rcu_dereference(conf->disks[i].rdev); | |
7389 | seq_printf (seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
7390 | } | |
7391 | rcu_read_unlock(); | |
1da177e4 | 7392 | seq_printf (seq, "]"); |
1da177e4 LT |
7393 | } |
7394 | ||
d1688a6d | 7395 | static void print_raid5_conf (struct r5conf *conf) |
1da177e4 LT |
7396 | { |
7397 | int i; | |
7398 | struct disk_info *tmp; | |
7399 | ||
cc6167b4 | 7400 | pr_debug("RAID conf printout:\n"); |
1da177e4 | 7401 | if (!conf) { |
cc6167b4 | 7402 | pr_debug("(conf==NULL)\n"); |
1da177e4 LT |
7403 | return; |
7404 | } | |
cc6167b4 | 7405 | pr_debug(" --- level:%d rd:%d wd:%d\n", conf->level, |
0c55e022 N |
7406 | conf->raid_disks, |
7407 | conf->raid_disks - conf->mddev->degraded); | |
1da177e4 LT |
7408 | |
7409 | for (i = 0; i < conf->raid_disks; i++) { | |
7410 | char b[BDEVNAME_SIZE]; | |
7411 | tmp = conf->disks + i; | |
7412 | if (tmp->rdev) | |
cc6167b4 | 7413 | pr_debug(" disk %d, o:%d, dev:%s\n", |
0c55e022 N |
7414 | i, !test_bit(Faulty, &tmp->rdev->flags), |
7415 | bdevname(tmp->rdev->bdev, b)); | |
1da177e4 LT |
7416 | } |
7417 | } | |
7418 | ||
fd01b88c | 7419 | static int raid5_spare_active(struct mddev *mddev) |
1da177e4 LT |
7420 | { |
7421 | int i; | |
d1688a6d | 7422 | struct r5conf *conf = mddev->private; |
1da177e4 | 7423 | struct disk_info *tmp; |
6b965620 N |
7424 | int count = 0; |
7425 | unsigned long flags; | |
1da177e4 LT |
7426 | |
7427 | for (i = 0; i < conf->raid_disks; i++) { | |
7428 | tmp = conf->disks + i; | |
dd054fce N |
7429 | if (tmp->replacement |
7430 | && tmp->replacement->recovery_offset == MaxSector | |
7431 | && !test_bit(Faulty, &tmp->replacement->flags) | |
7432 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
7433 | /* Replacement has just become active. */ | |
7434 | if (!tmp->rdev | |
7435 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
7436 | count++; | |
7437 | if (tmp->rdev) { | |
7438 | /* Replaced device not technically faulty, | |
7439 | * but we need to be sure it gets removed | |
7440 | * and never re-added. | |
7441 | */ | |
7442 | set_bit(Faulty, &tmp->rdev->flags); | |
7443 | sysfs_notify_dirent_safe( | |
7444 | tmp->rdev->sysfs_state); | |
7445 | } | |
7446 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
7447 | } else if (tmp->rdev | |
70fffd0b | 7448 | && tmp->rdev->recovery_offset == MaxSector |
b2d444d7 | 7449 | && !test_bit(Faulty, &tmp->rdev->flags) |
c04be0aa | 7450 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { |
6b965620 | 7451 | count++; |
43c73ca4 | 7452 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
7453 | } |
7454 | } | |
6b965620 | 7455 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7456 | mddev->degraded = raid5_calc_degraded(conf); |
6b965620 | 7457 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 7458 | print_raid5_conf(conf); |
6b965620 | 7459 | return count; |
1da177e4 LT |
7460 | } |
7461 | ||
b8321b68 | 7462 | static int raid5_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7463 | { |
d1688a6d | 7464 | struct r5conf *conf = mddev->private; |
1da177e4 | 7465 | int err = 0; |
b8321b68 | 7466 | int number = rdev->raid_disk; |
657e3e4d | 7467 | struct md_rdev **rdevp; |
1da177e4 LT |
7468 | struct disk_info *p = conf->disks + number; |
7469 | ||
7470 | print_raid5_conf(conf); | |
f6b6ec5c SL |
7471 | if (test_bit(Journal, &rdev->flags) && conf->log) { |
7472 | struct r5l_log *log; | |
c2bb6242 | 7473 | /* |
f6b6ec5c SL |
7474 | * we can't wait pending write here, as this is called in |
7475 | * raid5d, wait will deadlock. | |
c2bb6242 | 7476 | */ |
f6b6ec5c SL |
7477 | if (atomic_read(&mddev->writes_pending)) |
7478 | return -EBUSY; | |
7479 | log = conf->log; | |
7480 | conf->log = NULL; | |
7481 | synchronize_rcu(); | |
7482 | r5l_exit_log(log); | |
7483 | return 0; | |
c2bb6242 | 7484 | } |
657e3e4d N |
7485 | if (rdev == p->rdev) |
7486 | rdevp = &p->rdev; | |
7487 | else if (rdev == p->replacement) | |
7488 | rdevp = &p->replacement; | |
7489 | else | |
7490 | return 0; | |
7491 | ||
7492 | if (number >= conf->raid_disks && | |
7493 | conf->reshape_progress == MaxSector) | |
7494 | clear_bit(In_sync, &rdev->flags); | |
7495 | ||
7496 | if (test_bit(In_sync, &rdev->flags) || | |
7497 | atomic_read(&rdev->nr_pending)) { | |
7498 | err = -EBUSY; | |
7499 | goto abort; | |
7500 | } | |
7501 | /* Only remove non-faulty devices if recovery | |
7502 | * isn't possible. | |
7503 | */ | |
7504 | if (!test_bit(Faulty, &rdev->flags) && | |
7505 | mddev->recovery_disabled != conf->recovery_disabled && | |
7506 | !has_failed(conf) && | |
dd054fce | 7507 | (!p->replacement || p->replacement == rdev) && |
657e3e4d N |
7508 | number < conf->raid_disks) { |
7509 | err = -EBUSY; | |
7510 | goto abort; | |
7511 | } | |
7512 | *rdevp = NULL; | |
d787be40 N |
7513 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
7514 | synchronize_rcu(); | |
7515 | if (atomic_read(&rdev->nr_pending)) { | |
7516 | /* lost the race, try later */ | |
7517 | err = -EBUSY; | |
7518 | *rdevp = rdev; | |
7519 | } | |
7520 | } | |
7521 | if (p->replacement) { | |
dd054fce N |
7522 | /* We must have just cleared 'rdev' */ |
7523 | p->rdev = p->replacement; | |
7524 | clear_bit(Replacement, &p->replacement->flags); | |
7525 | smp_mb(); /* Make sure other CPUs may see both as identical | |
7526 | * but will never see neither - if they are careful | |
7527 | */ | |
7528 | p->replacement = NULL; | |
7529 | clear_bit(WantReplacement, &rdev->flags); | |
7530 | } else | |
7531 | /* We might have just removed the Replacement as faulty- | |
7532 | * clear the bit just in case | |
7533 | */ | |
7534 | clear_bit(WantReplacement, &rdev->flags); | |
1da177e4 LT |
7535 | abort: |
7536 | ||
7537 | print_raid5_conf(conf); | |
7538 | return err; | |
7539 | } | |
7540 | ||
fd01b88c | 7541 | static int raid5_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 7542 | { |
d1688a6d | 7543 | struct r5conf *conf = mddev->private; |
199050ea | 7544 | int err = -EEXIST; |
1da177e4 LT |
7545 | int disk; |
7546 | struct disk_info *p; | |
6c2fce2e NB |
7547 | int first = 0; |
7548 | int last = conf->raid_disks - 1; | |
1da177e4 | 7549 | |
f6b6ec5c SL |
7550 | if (test_bit(Journal, &rdev->flags)) { |
7551 | char b[BDEVNAME_SIZE]; | |
7552 | if (conf->log) | |
7553 | return -EBUSY; | |
7554 | ||
7555 | rdev->raid_disk = 0; | |
7556 | /* | |
7557 | * The array is in readonly mode if journal is missing, so no | |
7558 | * write requests running. We should be safe | |
7559 | */ | |
7560 | r5l_init_log(conf, rdev); | |
cc6167b4 N |
7561 | pr_debug("md/raid:%s: using device %s as journal\n", |
7562 | mdname(mddev), bdevname(rdev->bdev, b)); | |
f6b6ec5c SL |
7563 | return 0; |
7564 | } | |
7f0da59b N |
7565 | if (mddev->recovery_disabled == conf->recovery_disabled) |
7566 | return -EBUSY; | |
7567 | ||
dc10c643 | 7568 | if (rdev->saved_raid_disk < 0 && has_failed(conf)) |
1da177e4 | 7569 | /* no point adding a device */ |
199050ea | 7570 | return -EINVAL; |
1da177e4 | 7571 | |
6c2fce2e NB |
7572 | if (rdev->raid_disk >= 0) |
7573 | first = last = rdev->raid_disk; | |
1da177e4 LT |
7574 | |
7575 | /* | |
16a53ecc N |
7576 | * find the disk ... but prefer rdev->saved_raid_disk |
7577 | * if possible. | |
1da177e4 | 7578 | */ |
16a53ecc | 7579 | if (rdev->saved_raid_disk >= 0 && |
6c2fce2e | 7580 | rdev->saved_raid_disk >= first && |
16a53ecc | 7581 | conf->disks[rdev->saved_raid_disk].rdev == NULL) |
5cfb22a1 N |
7582 | first = rdev->saved_raid_disk; |
7583 | ||
7584 | for (disk = first; disk <= last; disk++) { | |
7bfec5f3 N |
7585 | p = conf->disks + disk; |
7586 | if (p->rdev == NULL) { | |
b2d444d7 | 7587 | clear_bit(In_sync, &rdev->flags); |
1da177e4 | 7588 | rdev->raid_disk = disk; |
199050ea | 7589 | err = 0; |
72626685 N |
7590 | if (rdev->saved_raid_disk != disk) |
7591 | conf->fullsync = 1; | |
d6065f7b | 7592 | rcu_assign_pointer(p->rdev, rdev); |
5cfb22a1 | 7593 | goto out; |
1da177e4 | 7594 | } |
5cfb22a1 N |
7595 | } |
7596 | for (disk = first; disk <= last; disk++) { | |
7597 | p = conf->disks + disk; | |
7bfec5f3 N |
7598 | if (test_bit(WantReplacement, &p->rdev->flags) && |
7599 | p->replacement == NULL) { | |
7600 | clear_bit(In_sync, &rdev->flags); | |
7601 | set_bit(Replacement, &rdev->flags); | |
7602 | rdev->raid_disk = disk; | |
7603 | err = 0; | |
7604 | conf->fullsync = 1; | |
7605 | rcu_assign_pointer(p->replacement, rdev); | |
7606 | break; | |
7607 | } | |
7608 | } | |
5cfb22a1 | 7609 | out: |
1da177e4 | 7610 | print_raid5_conf(conf); |
199050ea | 7611 | return err; |
1da177e4 LT |
7612 | } |
7613 | ||
fd01b88c | 7614 | static int raid5_resize(struct mddev *mddev, sector_t sectors) |
1da177e4 LT |
7615 | { |
7616 | /* no resync is happening, and there is enough space | |
7617 | * on all devices, so we can resize. | |
7618 | * We need to make sure resync covers any new space. | |
7619 | * If the array is shrinking we should possibly wait until | |
7620 | * any io in the removed space completes, but it hardly seems | |
7621 | * worth it. | |
7622 | */ | |
a4a6125a | 7623 | sector_t newsize; |
3cb5edf4 N |
7624 | struct r5conf *conf = mddev->private; |
7625 | ||
713cf5a6 SL |
7626 | if (conf->log) |
7627 | return -EINVAL; | |
3cb5edf4 | 7628 | sectors &= ~((sector_t)conf->chunk_sectors - 1); |
a4a6125a N |
7629 | newsize = raid5_size(mddev, sectors, mddev->raid_disks); |
7630 | if (mddev->external_size && | |
7631 | mddev->array_sectors > newsize) | |
b522adcd | 7632 | return -EINVAL; |
a4a6125a N |
7633 | if (mddev->bitmap) { |
7634 | int ret = bitmap_resize(mddev->bitmap, sectors, 0, 0); | |
7635 | if (ret) | |
7636 | return ret; | |
7637 | } | |
7638 | md_set_array_sectors(mddev, newsize); | |
b098636c N |
7639 | if (sectors > mddev->dev_sectors && |
7640 | mddev->recovery_cp > mddev->dev_sectors) { | |
58c0fed4 | 7641 | mddev->recovery_cp = mddev->dev_sectors; |
1da177e4 LT |
7642 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
7643 | } | |
58c0fed4 | 7644 | mddev->dev_sectors = sectors; |
4b5c7ae8 | 7645 | mddev->resync_max_sectors = sectors; |
1da177e4 LT |
7646 | return 0; |
7647 | } | |
7648 | ||
fd01b88c | 7649 | static int check_stripe_cache(struct mddev *mddev) |
01ee22b4 N |
7650 | { |
7651 | /* Can only proceed if there are plenty of stripe_heads. | |
7652 | * We need a minimum of one full stripe,, and for sensible progress | |
7653 | * it is best to have about 4 times that. | |
7654 | * If we require 4 times, then the default 256 4K stripe_heads will | |
7655 | * allow for chunk sizes up to 256K, which is probably OK. | |
7656 | * If the chunk size is greater, user-space should request more | |
7657 | * stripe_heads first. | |
7658 | */ | |
d1688a6d | 7659 | struct r5conf *conf = mddev->private; |
01ee22b4 | 7660 | if (((mddev->chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7661 | > conf->min_nr_stripes || |
01ee22b4 | 7662 | ((mddev->new_chunk_sectors << 9) / STRIPE_SIZE) * 4 |
edbe83ab | 7663 | > conf->min_nr_stripes) { |
cc6167b4 N |
7664 | pr_warn("md/raid:%s: reshape: not enough stripes. Needed %lu\n", |
7665 | mdname(mddev), | |
7666 | ((max(mddev->chunk_sectors, mddev->new_chunk_sectors) << 9) | |
7667 | / STRIPE_SIZE)*4); | |
01ee22b4 N |
7668 | return 0; |
7669 | } | |
7670 | return 1; | |
7671 | } | |
7672 | ||
fd01b88c | 7673 | static int check_reshape(struct mddev *mddev) |
29269553 | 7674 | { |
d1688a6d | 7675 | struct r5conf *conf = mddev->private; |
29269553 | 7676 | |
713cf5a6 SL |
7677 | if (conf->log) |
7678 | return -EINVAL; | |
88ce4930 N |
7679 | if (mddev->delta_disks == 0 && |
7680 | mddev->new_layout == mddev->layout && | |
664e7c41 | 7681 | mddev->new_chunk_sectors == mddev->chunk_sectors) |
50ac168a | 7682 | return 0; /* nothing to do */ |
674806d6 | 7683 | if (has_failed(conf)) |
ec32a2bd | 7684 | return -EINVAL; |
fdcfbbb6 | 7685 | if (mddev->delta_disks < 0 && mddev->reshape_position == MaxSector) { |
ec32a2bd N |
7686 | /* We might be able to shrink, but the devices must |
7687 | * be made bigger first. | |
7688 | * For raid6, 4 is the minimum size. | |
7689 | * Otherwise 2 is the minimum | |
7690 | */ | |
7691 | int min = 2; | |
7692 | if (mddev->level == 6) | |
7693 | min = 4; | |
7694 | if (mddev->raid_disks + mddev->delta_disks < min) | |
7695 | return -EINVAL; | |
7696 | } | |
29269553 | 7697 | |
01ee22b4 | 7698 | if (!check_stripe_cache(mddev)) |
29269553 | 7699 | return -ENOSPC; |
29269553 | 7700 | |
738a2738 N |
7701 | if (mddev->new_chunk_sectors > mddev->chunk_sectors || |
7702 | mddev->delta_disks > 0) | |
7703 | if (resize_chunks(conf, | |
7704 | conf->previous_raid_disks | |
7705 | + max(0, mddev->delta_disks), | |
7706 | max(mddev->new_chunk_sectors, | |
7707 | mddev->chunk_sectors) | |
7708 | ) < 0) | |
7709 | return -ENOMEM; | |
e56108d6 N |
7710 | return resize_stripes(conf, (conf->previous_raid_disks |
7711 | + mddev->delta_disks)); | |
63c70c4f N |
7712 | } |
7713 | ||
fd01b88c | 7714 | static int raid5_start_reshape(struct mddev *mddev) |
63c70c4f | 7715 | { |
d1688a6d | 7716 | struct r5conf *conf = mddev->private; |
3cb03002 | 7717 | struct md_rdev *rdev; |
63c70c4f | 7718 | int spares = 0; |
c04be0aa | 7719 | unsigned long flags; |
63c70c4f | 7720 | |
f416885e | 7721 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) |
63c70c4f N |
7722 | return -EBUSY; |
7723 | ||
01ee22b4 N |
7724 | if (!check_stripe_cache(mddev)) |
7725 | return -ENOSPC; | |
7726 | ||
30b67645 N |
7727 | if (has_failed(conf)) |
7728 | return -EINVAL; | |
7729 | ||
c6563a8c | 7730 | rdev_for_each(rdev, mddev) { |
469518a3 N |
7731 | if (!test_bit(In_sync, &rdev->flags) |
7732 | && !test_bit(Faulty, &rdev->flags)) | |
29269553 | 7733 | spares++; |
c6563a8c | 7734 | } |
63c70c4f | 7735 | |
f416885e | 7736 | if (spares - mddev->degraded < mddev->delta_disks - conf->max_degraded) |
29269553 N |
7737 | /* Not enough devices even to make a degraded array |
7738 | * of that size | |
7739 | */ | |
7740 | return -EINVAL; | |
7741 | ||
ec32a2bd N |
7742 | /* Refuse to reduce size of the array. Any reductions in |
7743 | * array size must be through explicit setting of array_size | |
7744 | * attribute. | |
7745 | */ | |
7746 | if (raid5_size(mddev, 0, conf->raid_disks + mddev->delta_disks) | |
7747 | < mddev->array_sectors) { | |
cc6167b4 N |
7748 | pr_warn("md/raid:%s: array size must be reduced before number of disks\n", |
7749 | mdname(mddev)); | |
ec32a2bd N |
7750 | return -EINVAL; |
7751 | } | |
7752 | ||
f6705578 | 7753 | atomic_set(&conf->reshape_stripes, 0); |
29269553 | 7754 | spin_lock_irq(&conf->device_lock); |
c46501b2 | 7755 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7756 | conf->previous_raid_disks = conf->raid_disks; |
63c70c4f | 7757 | conf->raid_disks += mddev->delta_disks; |
09c9e5fa AN |
7758 | conf->prev_chunk_sectors = conf->chunk_sectors; |
7759 | conf->chunk_sectors = mddev->new_chunk_sectors; | |
88ce4930 N |
7760 | conf->prev_algo = conf->algorithm; |
7761 | conf->algorithm = mddev->new_layout; | |
05616be5 N |
7762 | conf->generation++; |
7763 | /* Code that selects data_offset needs to see the generation update | |
7764 | * if reshape_progress has been set - so a memory barrier needed. | |
7765 | */ | |
7766 | smp_mb(); | |
2c810cdd | 7767 | if (mddev->reshape_backwards) |
fef9c61f N |
7768 | conf->reshape_progress = raid5_size(mddev, 0, 0); |
7769 | else | |
7770 | conf->reshape_progress = 0; | |
7771 | conf->reshape_safe = conf->reshape_progress; | |
c46501b2 | 7772 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7773 | spin_unlock_irq(&conf->device_lock); |
7774 | ||
4d77e3ba N |
7775 | /* Now make sure any requests that proceeded on the assumption |
7776 | * the reshape wasn't running - like Discard or Read - have | |
7777 | * completed. | |
7778 | */ | |
7779 | mddev_suspend(mddev); | |
7780 | mddev_resume(mddev); | |
7781 | ||
29269553 N |
7782 | /* Add some new drives, as many as will fit. |
7783 | * We know there are enough to make the newly sized array work. | |
3424bf6a N |
7784 | * Don't add devices if we are reducing the number of |
7785 | * devices in the array. This is because it is not possible | |
7786 | * to correctly record the "partially reconstructed" state of | |
7787 | * such devices during the reshape and confusion could result. | |
29269553 | 7788 | */ |
87a8dec9 | 7789 | if (mddev->delta_disks >= 0) { |
dafb20fa | 7790 | rdev_for_each(rdev, mddev) |
87a8dec9 N |
7791 | if (rdev->raid_disk < 0 && |
7792 | !test_bit(Faulty, &rdev->flags)) { | |
7793 | if (raid5_add_disk(mddev, rdev) == 0) { | |
87a8dec9 | 7794 | if (rdev->raid_disk |
9d4c7d87 | 7795 | >= conf->previous_raid_disks) |
87a8dec9 | 7796 | set_bit(In_sync, &rdev->flags); |
9d4c7d87 | 7797 | else |
87a8dec9 | 7798 | rdev->recovery_offset = 0; |
36fad858 NK |
7799 | |
7800 | if (sysfs_link_rdev(mddev, rdev)) | |
87a8dec9 | 7801 | /* Failure here is OK */; |
50da0840 | 7802 | } |
87a8dec9 N |
7803 | } else if (rdev->raid_disk >= conf->previous_raid_disks |
7804 | && !test_bit(Faulty, &rdev->flags)) { | |
7805 | /* This is a spare that was manually added */ | |
7806 | set_bit(In_sync, &rdev->flags); | |
87a8dec9 | 7807 | } |
29269553 | 7808 | |
87a8dec9 N |
7809 | /* When a reshape changes the number of devices, |
7810 | * ->degraded is measured against the larger of the | |
7811 | * pre and post number of devices. | |
7812 | */ | |
ec32a2bd | 7813 | spin_lock_irqsave(&conf->device_lock, flags); |
2e38a37f | 7814 | mddev->degraded = raid5_calc_degraded(conf); |
ec32a2bd N |
7815 | spin_unlock_irqrestore(&conf->device_lock, flags); |
7816 | } | |
63c70c4f | 7817 | mddev->raid_disks = conf->raid_disks; |
e516402c | 7818 | mddev->reshape_position = conf->reshape_progress; |
2953079c | 7819 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
f6705578 | 7820 | |
29269553 N |
7821 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
7822 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 7823 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
29269553 N |
7824 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
7825 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
7826 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
0da3c619 | 7827 | "reshape"); |
29269553 N |
7828 | if (!mddev->sync_thread) { |
7829 | mddev->recovery = 0; | |
7830 | spin_lock_irq(&conf->device_lock); | |
ba8805b9 | 7831 | write_seqcount_begin(&conf->gen_lock); |
29269553 | 7832 | mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks; |
ba8805b9 N |
7833 | mddev->new_chunk_sectors = |
7834 | conf->chunk_sectors = conf->prev_chunk_sectors; | |
7835 | mddev->new_layout = conf->algorithm = conf->prev_algo; | |
05616be5 N |
7836 | rdev_for_each(rdev, mddev) |
7837 | rdev->new_data_offset = rdev->data_offset; | |
7838 | smp_wmb(); | |
ba8805b9 | 7839 | conf->generation --; |
fef9c61f | 7840 | conf->reshape_progress = MaxSector; |
1e3fa9bd | 7841 | mddev->reshape_position = MaxSector; |
ba8805b9 | 7842 | write_seqcount_end(&conf->gen_lock); |
29269553 N |
7843 | spin_unlock_irq(&conf->device_lock); |
7844 | return -EAGAIN; | |
7845 | } | |
c8f517c4 | 7846 | conf->reshape_checkpoint = jiffies; |
29269553 N |
7847 | md_wakeup_thread(mddev->sync_thread); |
7848 | md_new_event(mddev); | |
7849 | return 0; | |
7850 | } | |
29269553 | 7851 | |
ec32a2bd N |
7852 | /* This is called from the reshape thread and should make any |
7853 | * changes needed in 'conf' | |
7854 | */ | |
d1688a6d | 7855 | static void end_reshape(struct r5conf *conf) |
29269553 | 7856 | { |
29269553 | 7857 | |
f6705578 | 7858 | if (!test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { |
05616be5 | 7859 | struct md_rdev *rdev; |
f6705578 | 7860 | |
f6705578 | 7861 | spin_lock_irq(&conf->device_lock); |
cea9c228 | 7862 | conf->previous_raid_disks = conf->raid_disks; |
05616be5 N |
7863 | rdev_for_each(rdev, conf->mddev) |
7864 | rdev->data_offset = rdev->new_data_offset; | |
7865 | smp_wmb(); | |
fef9c61f | 7866 | conf->reshape_progress = MaxSector; |
6cbd8148 | 7867 | conf->mddev->reshape_position = MaxSector; |
f6705578 | 7868 | spin_unlock_irq(&conf->device_lock); |
b0f9ec04 | 7869 | wake_up(&conf->wait_for_overlap); |
16a53ecc N |
7870 | |
7871 | /* read-ahead size must cover two whole stripes, which is | |
7872 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
7873 | */ | |
4a5add49 | 7874 | if (conf->mddev->queue) { |
cea9c228 | 7875 | int data_disks = conf->raid_disks - conf->max_degraded; |
09c9e5fa | 7876 | int stripe = data_disks * ((conf->chunk_sectors << 9) |
cea9c228 | 7877 | / PAGE_SIZE); |
dc3b17cc JK |
7878 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
7879 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
16a53ecc | 7880 | } |
29269553 | 7881 | } |
29269553 N |
7882 | } |
7883 | ||
ec32a2bd N |
7884 | /* This is called from the raid5d thread with mddev_lock held. |
7885 | * It makes config changes to the device. | |
7886 | */ | |
fd01b88c | 7887 | static void raid5_finish_reshape(struct mddev *mddev) |
cea9c228 | 7888 | { |
d1688a6d | 7889 | struct r5conf *conf = mddev->private; |
cea9c228 N |
7890 | |
7891 | if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
7892 | ||
ec32a2bd N |
7893 | if (mddev->delta_disks > 0) { |
7894 | md_set_array_sectors(mddev, raid5_size(mddev, 0, 0)); | |
fe67d19a HM |
7895 | if (mddev->queue) { |
7896 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
7897 | revalidate_disk(mddev->gendisk); | |
7898 | } | |
ec32a2bd N |
7899 | } else { |
7900 | int d; | |
908f4fbd | 7901 | spin_lock_irq(&conf->device_lock); |
2e38a37f | 7902 | mddev->degraded = raid5_calc_degraded(conf); |
908f4fbd | 7903 | spin_unlock_irq(&conf->device_lock); |
ec32a2bd N |
7904 | for (d = conf->raid_disks ; |
7905 | d < conf->raid_disks - mddev->delta_disks; | |
1a67dde0 | 7906 | d++) { |
3cb03002 | 7907 | struct md_rdev *rdev = conf->disks[d].rdev; |
da7613b8 N |
7908 | if (rdev) |
7909 | clear_bit(In_sync, &rdev->flags); | |
7910 | rdev = conf->disks[d].replacement; | |
7911 | if (rdev) | |
7912 | clear_bit(In_sync, &rdev->flags); | |
1a67dde0 | 7913 | } |
cea9c228 | 7914 | } |
88ce4930 | 7915 | mddev->layout = conf->algorithm; |
09c9e5fa | 7916 | mddev->chunk_sectors = conf->chunk_sectors; |
ec32a2bd N |
7917 | mddev->reshape_position = MaxSector; |
7918 | mddev->delta_disks = 0; | |
2c810cdd | 7919 | mddev->reshape_backwards = 0; |
cea9c228 N |
7920 | } |
7921 | } | |
7922 | ||
fd01b88c | 7923 | static void raid5_quiesce(struct mddev *mddev, int state) |
72626685 | 7924 | { |
d1688a6d | 7925 | struct r5conf *conf = mddev->private; |
72626685 N |
7926 | |
7927 | switch(state) { | |
e464eafd N |
7928 | case 2: /* resume for a suspend */ |
7929 | wake_up(&conf->wait_for_overlap); | |
7930 | break; | |
7931 | ||
72626685 | 7932 | case 1: /* stop all writes */ |
566c09c5 | 7933 | lock_all_device_hash_locks_irq(conf); |
64bd660b N |
7934 | /* '2' tells resync/reshape to pause so that all |
7935 | * active stripes can drain | |
7936 | */ | |
a39f7afd | 7937 | r5c_flush_cache(conf, INT_MAX); |
64bd660b | 7938 | conf->quiesce = 2; |
b1b46486 | 7939 | wait_event_cmd(conf->wait_for_quiescent, |
46031f9a RBJ |
7940 | atomic_read(&conf->active_stripes) == 0 && |
7941 | atomic_read(&conf->active_aligned_reads) == 0, | |
566c09c5 SL |
7942 | unlock_all_device_hash_locks_irq(conf), |
7943 | lock_all_device_hash_locks_irq(conf)); | |
64bd660b | 7944 | conf->quiesce = 1; |
566c09c5 | 7945 | unlock_all_device_hash_locks_irq(conf); |
64bd660b N |
7946 | /* allow reshape to continue */ |
7947 | wake_up(&conf->wait_for_overlap); | |
72626685 N |
7948 | break; |
7949 | ||
7950 | case 0: /* re-enable writes */ | |
566c09c5 | 7951 | lock_all_device_hash_locks_irq(conf); |
72626685 | 7952 | conf->quiesce = 0; |
b1b46486 | 7953 | wake_up(&conf->wait_for_quiescent); |
e464eafd | 7954 | wake_up(&conf->wait_for_overlap); |
566c09c5 | 7955 | unlock_all_device_hash_locks_irq(conf); |
72626685 N |
7956 | break; |
7957 | } | |
e6c033f7 | 7958 | r5l_quiesce(conf->log, state); |
72626685 | 7959 | } |
b15c2e57 | 7960 | |
fd01b88c | 7961 | static void *raid45_takeover_raid0(struct mddev *mddev, int level) |
54071b38 | 7962 | { |
e373ab10 | 7963 | struct r0conf *raid0_conf = mddev->private; |
d76c8420 | 7964 | sector_t sectors; |
54071b38 | 7965 | |
f1b29bca | 7966 | /* for raid0 takeover only one zone is supported */ |
e373ab10 | 7967 | if (raid0_conf->nr_strip_zones > 1) { |
cc6167b4 N |
7968 | pr_warn("md/raid:%s: cannot takeover raid0 with more than one zone.\n", |
7969 | mdname(mddev)); | |
f1b29bca DW |
7970 | return ERR_PTR(-EINVAL); |
7971 | } | |
7972 | ||
e373ab10 N |
7973 | sectors = raid0_conf->strip_zone[0].zone_end; |
7974 | sector_div(sectors, raid0_conf->strip_zone[0].nb_dev); | |
3b71bd93 | 7975 | mddev->dev_sectors = sectors; |
f1b29bca | 7976 | mddev->new_level = level; |
54071b38 TM |
7977 | mddev->new_layout = ALGORITHM_PARITY_N; |
7978 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
7979 | mddev->raid_disks += 1; | |
7980 | mddev->delta_disks = 1; | |
7981 | /* make sure it will be not marked as dirty */ | |
7982 | mddev->recovery_cp = MaxSector; | |
7983 | ||
7984 | return setup_conf(mddev); | |
7985 | } | |
7986 | ||
fd01b88c | 7987 | static void *raid5_takeover_raid1(struct mddev *mddev) |
d562b0c4 N |
7988 | { |
7989 | int chunksect; | |
6995f0b2 | 7990 | void *ret; |
d562b0c4 N |
7991 | |
7992 | if (mddev->raid_disks != 2 || | |
7993 | mddev->degraded > 1) | |
7994 | return ERR_PTR(-EINVAL); | |
7995 | ||
7996 | /* Should check if there are write-behind devices? */ | |
7997 | ||
7998 | chunksect = 64*2; /* 64K by default */ | |
7999 | ||
8000 | /* The array must be an exact multiple of chunksize */ | |
8001 | while (chunksect && (mddev->array_sectors & (chunksect-1))) | |
8002 | chunksect >>= 1; | |
8003 | ||
8004 | if ((chunksect<<9) < STRIPE_SIZE) | |
8005 | /* array size does not allow a suitable chunk size */ | |
8006 | return ERR_PTR(-EINVAL); | |
8007 | ||
8008 | mddev->new_level = 5; | |
8009 | mddev->new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
664e7c41 | 8010 | mddev->new_chunk_sectors = chunksect; |
d562b0c4 | 8011 | |
6995f0b2 | 8012 | ret = setup_conf(mddev); |
32cd7cbb | 8013 | if (!IS_ERR(ret)) |
394ed8e4 SL |
8014 | mddev_clear_unsupported_flags(mddev, |
8015 | UNSUPPORTED_MDDEV_FLAGS); | |
6995f0b2 | 8016 | return ret; |
d562b0c4 N |
8017 | } |
8018 | ||
fd01b88c | 8019 | static void *raid5_takeover_raid6(struct mddev *mddev) |
fc9739c6 N |
8020 | { |
8021 | int new_layout; | |
8022 | ||
8023 | switch (mddev->layout) { | |
8024 | case ALGORITHM_LEFT_ASYMMETRIC_6: | |
8025 | new_layout = ALGORITHM_LEFT_ASYMMETRIC; | |
8026 | break; | |
8027 | case ALGORITHM_RIGHT_ASYMMETRIC_6: | |
8028 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC; | |
8029 | break; | |
8030 | case ALGORITHM_LEFT_SYMMETRIC_6: | |
8031 | new_layout = ALGORITHM_LEFT_SYMMETRIC; | |
8032 | break; | |
8033 | case ALGORITHM_RIGHT_SYMMETRIC_6: | |
8034 | new_layout = ALGORITHM_RIGHT_SYMMETRIC; | |
8035 | break; | |
8036 | case ALGORITHM_PARITY_0_6: | |
8037 | new_layout = ALGORITHM_PARITY_0; | |
8038 | break; | |
8039 | case ALGORITHM_PARITY_N: | |
8040 | new_layout = ALGORITHM_PARITY_N; | |
8041 | break; | |
8042 | default: | |
8043 | return ERR_PTR(-EINVAL); | |
8044 | } | |
8045 | mddev->new_level = 5; | |
8046 | mddev->new_layout = new_layout; | |
8047 | mddev->delta_disks = -1; | |
8048 | mddev->raid_disks -= 1; | |
8049 | return setup_conf(mddev); | |
8050 | } | |
8051 | ||
fd01b88c | 8052 | static int raid5_check_reshape(struct mddev *mddev) |
b3546035 | 8053 | { |
88ce4930 N |
8054 | /* For a 2-drive array, the layout and chunk size can be changed |
8055 | * immediately as not restriping is needed. | |
8056 | * For larger arrays we record the new value - after validation | |
8057 | * to be used by a reshape pass. | |
b3546035 | 8058 | */ |
d1688a6d | 8059 | struct r5conf *conf = mddev->private; |
597a711b | 8060 | int new_chunk = mddev->new_chunk_sectors; |
b3546035 | 8061 | |
597a711b | 8062 | if (mddev->new_layout >= 0 && !algorithm_valid_raid5(mddev->new_layout)) |
b3546035 N |
8063 | return -EINVAL; |
8064 | if (new_chunk > 0) { | |
0ba459d2 | 8065 | if (!is_power_of_2(new_chunk)) |
b3546035 | 8066 | return -EINVAL; |
597a711b | 8067 | if (new_chunk < (PAGE_SIZE>>9)) |
b3546035 | 8068 | return -EINVAL; |
597a711b | 8069 | if (mddev->array_sectors & (new_chunk-1)) |
b3546035 N |
8070 | /* not factor of array size */ |
8071 | return -EINVAL; | |
8072 | } | |
8073 | ||
8074 | /* They look valid */ | |
8075 | ||
88ce4930 | 8076 | if (mddev->raid_disks == 2) { |
597a711b N |
8077 | /* can make the change immediately */ |
8078 | if (mddev->new_layout >= 0) { | |
8079 | conf->algorithm = mddev->new_layout; | |
8080 | mddev->layout = mddev->new_layout; | |
88ce4930 N |
8081 | } |
8082 | if (new_chunk > 0) { | |
597a711b N |
8083 | conf->chunk_sectors = new_chunk ; |
8084 | mddev->chunk_sectors = new_chunk; | |
88ce4930 | 8085 | } |
2953079c | 8086 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
88ce4930 | 8087 | md_wakeup_thread(mddev->thread); |
b3546035 | 8088 | } |
50ac168a | 8089 | return check_reshape(mddev); |
88ce4930 N |
8090 | } |
8091 | ||
fd01b88c | 8092 | static int raid6_check_reshape(struct mddev *mddev) |
88ce4930 | 8093 | { |
597a711b | 8094 | int new_chunk = mddev->new_chunk_sectors; |
50ac168a | 8095 | |
597a711b | 8096 | if (mddev->new_layout >= 0 && !algorithm_valid_raid6(mddev->new_layout)) |
88ce4930 | 8097 | return -EINVAL; |
b3546035 | 8098 | if (new_chunk > 0) { |
0ba459d2 | 8099 | if (!is_power_of_2(new_chunk)) |
88ce4930 | 8100 | return -EINVAL; |
597a711b | 8101 | if (new_chunk < (PAGE_SIZE >> 9)) |
88ce4930 | 8102 | return -EINVAL; |
597a711b | 8103 | if (mddev->array_sectors & (new_chunk-1)) |
88ce4930 N |
8104 | /* not factor of array size */ |
8105 | return -EINVAL; | |
b3546035 | 8106 | } |
88ce4930 N |
8107 | |
8108 | /* They look valid */ | |
50ac168a | 8109 | return check_reshape(mddev); |
b3546035 N |
8110 | } |
8111 | ||
fd01b88c | 8112 | static void *raid5_takeover(struct mddev *mddev) |
d562b0c4 N |
8113 | { |
8114 | /* raid5 can take over: | |
f1b29bca | 8115 | * raid0 - if there is only one strip zone - make it a raid4 layout |
d562b0c4 N |
8116 | * raid1 - if there are two drives. We need to know the chunk size |
8117 | * raid4 - trivial - just use a raid4 layout. | |
8118 | * raid6 - Providing it is a *_6 layout | |
d562b0c4 | 8119 | */ |
f1b29bca DW |
8120 | if (mddev->level == 0) |
8121 | return raid45_takeover_raid0(mddev, 5); | |
d562b0c4 N |
8122 | if (mddev->level == 1) |
8123 | return raid5_takeover_raid1(mddev); | |
e9d4758f N |
8124 | if (mddev->level == 4) { |
8125 | mddev->new_layout = ALGORITHM_PARITY_N; | |
8126 | mddev->new_level = 5; | |
8127 | return setup_conf(mddev); | |
8128 | } | |
fc9739c6 N |
8129 | if (mddev->level == 6) |
8130 | return raid5_takeover_raid6(mddev); | |
d562b0c4 N |
8131 | |
8132 | return ERR_PTR(-EINVAL); | |
8133 | } | |
8134 | ||
fd01b88c | 8135 | static void *raid4_takeover(struct mddev *mddev) |
a78d38a1 | 8136 | { |
f1b29bca DW |
8137 | /* raid4 can take over: |
8138 | * raid0 - if there is only one strip zone | |
8139 | * raid5 - if layout is right | |
a78d38a1 | 8140 | */ |
f1b29bca DW |
8141 | if (mddev->level == 0) |
8142 | return raid45_takeover_raid0(mddev, 4); | |
a78d38a1 N |
8143 | if (mddev->level == 5 && |
8144 | mddev->layout == ALGORITHM_PARITY_N) { | |
8145 | mddev->new_layout = 0; | |
8146 | mddev->new_level = 4; | |
8147 | return setup_conf(mddev); | |
8148 | } | |
8149 | return ERR_PTR(-EINVAL); | |
8150 | } | |
d562b0c4 | 8151 | |
84fc4b56 | 8152 | static struct md_personality raid5_personality; |
245f46c2 | 8153 | |
fd01b88c | 8154 | static void *raid6_takeover(struct mddev *mddev) |
245f46c2 N |
8155 | { |
8156 | /* Currently can only take over a raid5. We map the | |
8157 | * personality to an equivalent raid6 personality | |
8158 | * with the Q block at the end. | |
8159 | */ | |
8160 | int new_layout; | |
8161 | ||
8162 | if (mddev->pers != &raid5_personality) | |
8163 | return ERR_PTR(-EINVAL); | |
8164 | if (mddev->degraded > 1) | |
8165 | return ERR_PTR(-EINVAL); | |
8166 | if (mddev->raid_disks > 253) | |
8167 | return ERR_PTR(-EINVAL); | |
8168 | if (mddev->raid_disks < 3) | |
8169 | return ERR_PTR(-EINVAL); | |
8170 | ||
8171 | switch (mddev->layout) { | |
8172 | case ALGORITHM_LEFT_ASYMMETRIC: | |
8173 | new_layout = ALGORITHM_LEFT_ASYMMETRIC_6; | |
8174 | break; | |
8175 | case ALGORITHM_RIGHT_ASYMMETRIC: | |
8176 | new_layout = ALGORITHM_RIGHT_ASYMMETRIC_6; | |
8177 | break; | |
8178 | case ALGORITHM_LEFT_SYMMETRIC: | |
8179 | new_layout = ALGORITHM_LEFT_SYMMETRIC_6; | |
8180 | break; | |
8181 | case ALGORITHM_RIGHT_SYMMETRIC: | |
8182 | new_layout = ALGORITHM_RIGHT_SYMMETRIC_6; | |
8183 | break; | |
8184 | case ALGORITHM_PARITY_0: | |
8185 | new_layout = ALGORITHM_PARITY_0_6; | |
8186 | break; | |
8187 | case ALGORITHM_PARITY_N: | |
8188 | new_layout = ALGORITHM_PARITY_N; | |
8189 | break; | |
8190 | default: | |
8191 | return ERR_PTR(-EINVAL); | |
8192 | } | |
8193 | mddev->new_level = 6; | |
8194 | mddev->new_layout = new_layout; | |
8195 | mddev->delta_disks = 1; | |
8196 | mddev->raid_disks += 1; | |
8197 | return setup_conf(mddev); | |
8198 | } | |
8199 | ||
84fc4b56 | 8200 | static struct md_personality raid6_personality = |
16a53ecc N |
8201 | { |
8202 | .name = "raid6", | |
8203 | .level = 6, | |
8204 | .owner = THIS_MODULE, | |
849674e4 SL |
8205 | .make_request = raid5_make_request, |
8206 | .run = raid5_run, | |
afa0f557 | 8207 | .free = raid5_free, |
849674e4 SL |
8208 | .status = raid5_status, |
8209 | .error_handler = raid5_error, | |
16a53ecc N |
8210 | .hot_add_disk = raid5_add_disk, |
8211 | .hot_remove_disk= raid5_remove_disk, | |
8212 | .spare_active = raid5_spare_active, | |
849674e4 | 8213 | .sync_request = raid5_sync_request, |
16a53ecc | 8214 | .resize = raid5_resize, |
80c3a6ce | 8215 | .size = raid5_size, |
50ac168a | 8216 | .check_reshape = raid6_check_reshape, |
f416885e | 8217 | .start_reshape = raid5_start_reshape, |
cea9c228 | 8218 | .finish_reshape = raid5_finish_reshape, |
16a53ecc | 8219 | .quiesce = raid5_quiesce, |
245f46c2 | 8220 | .takeover = raid6_takeover, |
5c675f83 | 8221 | .congested = raid5_congested, |
16a53ecc | 8222 | }; |
84fc4b56 | 8223 | static struct md_personality raid5_personality = |
1da177e4 LT |
8224 | { |
8225 | .name = "raid5", | |
2604b703 | 8226 | .level = 5, |
1da177e4 | 8227 | .owner = THIS_MODULE, |
849674e4 SL |
8228 | .make_request = raid5_make_request, |
8229 | .run = raid5_run, | |
afa0f557 | 8230 | .free = raid5_free, |
849674e4 SL |
8231 | .status = raid5_status, |
8232 | .error_handler = raid5_error, | |
1da177e4 LT |
8233 | .hot_add_disk = raid5_add_disk, |
8234 | .hot_remove_disk= raid5_remove_disk, | |
8235 | .spare_active = raid5_spare_active, | |
849674e4 | 8236 | .sync_request = raid5_sync_request, |
1da177e4 | 8237 | .resize = raid5_resize, |
80c3a6ce | 8238 | .size = raid5_size, |
63c70c4f N |
8239 | .check_reshape = raid5_check_reshape, |
8240 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8241 | .finish_reshape = raid5_finish_reshape, |
72626685 | 8242 | .quiesce = raid5_quiesce, |
d562b0c4 | 8243 | .takeover = raid5_takeover, |
5c675f83 | 8244 | .congested = raid5_congested, |
1da177e4 LT |
8245 | }; |
8246 | ||
84fc4b56 | 8247 | static struct md_personality raid4_personality = |
1da177e4 | 8248 | { |
2604b703 N |
8249 | .name = "raid4", |
8250 | .level = 4, | |
8251 | .owner = THIS_MODULE, | |
849674e4 SL |
8252 | .make_request = raid5_make_request, |
8253 | .run = raid5_run, | |
afa0f557 | 8254 | .free = raid5_free, |
849674e4 SL |
8255 | .status = raid5_status, |
8256 | .error_handler = raid5_error, | |
2604b703 N |
8257 | .hot_add_disk = raid5_add_disk, |
8258 | .hot_remove_disk= raid5_remove_disk, | |
8259 | .spare_active = raid5_spare_active, | |
849674e4 | 8260 | .sync_request = raid5_sync_request, |
2604b703 | 8261 | .resize = raid5_resize, |
80c3a6ce | 8262 | .size = raid5_size, |
3d37890b N |
8263 | .check_reshape = raid5_check_reshape, |
8264 | .start_reshape = raid5_start_reshape, | |
cea9c228 | 8265 | .finish_reshape = raid5_finish_reshape, |
2604b703 | 8266 | .quiesce = raid5_quiesce, |
a78d38a1 | 8267 | .takeover = raid4_takeover, |
5c675f83 | 8268 | .congested = raid5_congested, |
2604b703 N |
8269 | }; |
8270 | ||
8271 | static int __init raid5_init(void) | |
8272 | { | |
29c6d1bb SAS |
8273 | int ret; |
8274 | ||
851c30c9 SL |
8275 | raid5_wq = alloc_workqueue("raid5wq", |
8276 | WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE|WQ_SYSFS, 0); | |
8277 | if (!raid5_wq) | |
8278 | return -ENOMEM; | |
29c6d1bb SAS |
8279 | |
8280 | ret = cpuhp_setup_state_multi(CPUHP_MD_RAID5_PREPARE, | |
8281 | "md/raid5:prepare", | |
8282 | raid456_cpu_up_prepare, | |
8283 | raid456_cpu_dead); | |
8284 | if (ret) { | |
8285 | destroy_workqueue(raid5_wq); | |
8286 | return ret; | |
8287 | } | |
16a53ecc | 8288 | register_md_personality(&raid6_personality); |
2604b703 N |
8289 | register_md_personality(&raid5_personality); |
8290 | register_md_personality(&raid4_personality); | |
8291 | return 0; | |
1da177e4 LT |
8292 | } |
8293 | ||
2604b703 | 8294 | static void raid5_exit(void) |
1da177e4 | 8295 | { |
16a53ecc | 8296 | unregister_md_personality(&raid6_personality); |
2604b703 N |
8297 | unregister_md_personality(&raid5_personality); |
8298 | unregister_md_personality(&raid4_personality); | |
29c6d1bb | 8299 | cpuhp_remove_multi_state(CPUHP_MD_RAID5_PREPARE); |
851c30c9 | 8300 | destroy_workqueue(raid5_wq); |
1da177e4 LT |
8301 | } |
8302 | ||
8303 | module_init(raid5_init); | |
8304 | module_exit(raid5_exit); | |
8305 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 8306 | MODULE_DESCRIPTION("RAID4/5/6 (striping with parity) personality for MD"); |
1da177e4 | 8307 | MODULE_ALIAS("md-personality-4"); /* RAID5 */ |
d9d166c2 N |
8308 | MODULE_ALIAS("md-raid5"); |
8309 | MODULE_ALIAS("md-raid4"); | |
2604b703 N |
8310 | MODULE_ALIAS("md-level-5"); |
8311 | MODULE_ALIAS("md-level-4"); | |
16a53ecc N |
8312 | MODULE_ALIAS("md-personality-8"); /* RAID6 */ |
8313 | MODULE_ALIAS("md-raid6"); | |
8314 | MODULE_ALIAS("md-level-6"); | |
8315 | ||
8316 | /* This used to be two separate modules, they were: */ | |
8317 | MODULE_ALIAS("raid5"); | |
8318 | MODULE_ALIAS("raid6"); |